[Andrew_R._A._Conway,_Christopher_Jarrold,_Michael(BookFi.org).pdf

8/10/2019 [Andrew_R._A._Conway,_Christopher_Jarrold,_Michael(BookFi.org).pdf

http://slidepdf.com/reader/full/andrewraconwaychristopherjarroldmichaelbookfiorgpdf 1/346



VARIATION IN WORKING MEMORY



This page intentionally left blank



Variation in Working Memory

Edited by

Andrew R. A. ConwayChristopher Jarrold

Michael J. Kane

Akira Miyake

John N. Towse

1 2007



3

Oxford University Press, Inc., publishes works that furtherOxford University’s objective of excellence

in research, scholarship, and education.

Oxford New York Auckland Cape Town Dar es Salaam Hong Kong KarachiKuala Lumpur Madrid Melbourne Mexico City Nairobi

New Delhi Shanghai Taipei Toronto

With offices in

Argentina Austria Brazil Chile Czech Republic France GreeceGuatemala Hungary Italy Japan Poland Portugal Singapore

South Korea Switzerland Thailand Turkey Ukraine Vietnam

Copyright# 2007 by Oxford University Press, Inc.

Published by Oxford University Press, Inc.198 Madison Avenue, New York, New York 10016

www.oup.com

Oxford is a registered trademark of Oxford University Press

All rights reserved. No part of this publication may be reproduced,stored in a retrieval system, or transmitted, in any form or by any means,

electronic, mechanical, photocopying, recording, or otherwise,

without the prior permission of Oxford University Press.

Library of Congress Cataloging-in-Publication Data

Variation in working memory / edited by Andrew Conway . . . [et al.].p. cm.

Includes bibliographical references.

ISBN-13: 978-0-19-516863-1

ISBN-10: 0-19-516863-11. Short-term memory. I. Conway, Andrew (Andrew R. A.)BF378.S54V37 2006

153.1'3—dc22 2006009858

9 8 7 6 5 4 3 2 1

Printed in the United States of America

on acid-free paper

http://www.oup.com/

http://www.oup.com/



Dedicated to our families



Preface

Working memory—the ability to keep importantinformation in mind while comprehending,thinking, and doing—changes dramatically overthe life span and varies considerably from per-son to person at a given age. Understandingsuch individual and developmental differencesis crucial because working memory is a keycontributor to general intellectual functioning.It has been demonstrated to be relevant to manyeveryday tasks, such as reading, making sense of spoken discourse, problem solving, and mentalarithmetic. As such, it is the focus of consider-able research efforts in cognitive psychologyand cognitive neuroscience. This research haswide-ranging implications for our general un-derstanding of cognitive processes in a variety of populations and settings.

Since Baddeley and Hitch (1974) intro-duced their seminal model of working mem-ory, there has been growing interest in thisarea. As a testament to the importance of thiswork, the work by Baddeley and Hitch wasidentified in a discipline-wide poll as one of the100 most influential works in cognitive science

(http://www.cogsci.umn.edu/OLD/calendar/past_events/millennium/top100.html). It has alsoserved as a catalyst for active research andtheoreticaldevelopment,asreflectedbythelargenumber of edited books and monographs cur-rently available on working memory (includingseveral edited volumes that originated as spe-cial issues of academic journals), some of which have documented recent theoreticaldevelopments in the area (e.g., Cowan, 2005;Miyake & Shah, 1999).

Despite the existence of various edited vol-umes and monographs on working memory, noprevious publication has properly explored theissue of the causes and consequences of varia-tion in working memory, the focus of this vol-ume and an issue of particular importance in

the field (e.g., see Jarrold & Towse, 2006).Since Daneman and Carpenter (1980) devel-oped the first useful measure of workingmemory capacity as an individual-differencesconstruct among adults (for a seminal devel-opmental perspective on working memory ca-pacity, see Case, Kurland, & Goldberg, 1982),

http://www.cogsci.umn.edu/OLD/calendar/past_events/millennium/top100.html






research on working memory variation hasmade critical contributions to theory on work-ing memory in general. Indeed, individual- andgroup-differences research has had more impact

on the development of working memory theorythan on any other cognitive construct. Workingmemory research thus represents a most suc-cessful integration of the experimental and cor-relational traditions or what Cronbach (1957)called ‘‘the two disciplines of scientific psychol-ogy.’’ It also serves as an exemplary instantiationof Underwood’s (1975) notion that individualdifferences may act as a ‘‘crucible’’ in which totest general theory (see also Kosslyn et al., 2002).

In addition, there are a number of otherreasons why a focus on individual differences isespecially important and timely. First, in thelast few years, new techniques developed inindividual-differences research, cognitive sci-ence and neuroscience have been applied tothe issue of variation in working memory.These new techniques include the statisticalapproach of latent variables analysis (e.g.,Conway, Cowan, Bunting, Therriault, &

Minkoff, 2002; Miyake et al., 2000) and func-tional neuroimaging techniques (e.g., Braver etal., 1997; Jonides, Smith, Marshuetz, Koeppe,& Reuter-Lorenz, 1998). Second, there hasbeen a growing realization that the dominanttheoretical conceptualization of workingmemory that depends solely on the ability toshare resources between information proces-sing and storage demands may be limited inexplanatory power. In particular, researchersare highlighting the need to also consider long-term memory constraints (e.g., Baddeley, 2000;Ericsson & Kintsch, 1995) and additional if notalternative theoretical mechanisms and pro-cesses for working memory performance (e.g.,Barrouillet, Bernadin, & Camos, 2004; Lustig,May, & Hasher, 2001; Saito & Miyake, 2004;Towse, Hitch, & Hutton, 1998). Third, de-spite the centrality of sentence comprehension

research for understanding the nature of in-dividual differences in working memory (Da-neman & Merikle, 1996; Just & Carpenter,1992), there has been controversy regarding thenature of working memory implicated duringlanguage comprehension (e.g., Caplan & Wa-

ters, 1999). Finally, this growth in interest invariation in working memory has been mat-ched by growth in the scope of the practicalapplication of this approach. Recent studies

have applied an individual-differences analysisto examine the importance of working memoryvariance in predicting learning disability (seediscussion in Jarrold, 2001) as well as intel-lectual functioning in typically developingchildren (e.g., Bayliss, Jarrold, Gunn, & Bad-deley, 2003; Fry & Hale, 1996) and adults(e.g., Engle, Kane, & Tuholski, 1999; Miyake,Friedman, Rettinger, Shah, & Hegarty, 2001).Moreover, advances in the understanding of the

neural basis of individual and age-related varia-tion in working memory (e.g., Kane & Engle,2003; Munakata, 2004; Oberauer & Kliegl,2001; Reuter-Lorenz et al., 2001; Roberts &Pennington, 1996) have also been important inshaping the present research field.

There is clearly a need to bring this recentwork together in a comprehensive yet cohesivevolume, particularly because existing modelsand theories of working memory variation are

quite diverse and straddle so many researchareas. Although some theories provide sophis-ticated and well-developed accounts of certainaspects of working memory variation, they alsotend to have different theoretical orientationsand at the same time leave unspecified someimportant aspects of working memory. In ad-dition, we view progress in this field as beinghindered by a prototypical research strategy of proposing one particular underlying source of working memory variation and confirmingpredictions based on that idea, without parallelconsideration of other mechanisms. Likewise,researchers often focus their empirical effortson working memory variation within one par-ticular target population, without consideringhow research with other populations mightcomplement or conflict with their findings. Inshort, there is a lack of overall cohesion among

different research findings, and this probleminhibits the ability to compare and contrastdifferent proposals about the nature of workingmemory variation.

The present volume attempts to offer anintegrative yet thorough approach by focusing

viii Preface



on explicit, detailed comparisons of currentmajor theoretical proposals on working mem-ory variation. Research groups have beendrawn from both the United States and Europe

to ensure that the rather different research per-spectives that operate on the two sides of the Atlantic are well represented. A particularstrength of the book is its coverage of working-memory research on a wide variety of popula-tions, such as healthy adults, children with andwithout learning difficulties, older adults, andneurological patients.

Another major feature of the volume is thateach research group has explicitly addressed the

same set of important theoretical questions—from the perspective of their own theoreticaland applied research, and from the perspectiveof competing views from within and beyondthis volume. We believe that this common-question approach was adopted successfully inMiyake and Shah’s (1999) edited volumeModels of Working Memory: Mechanisms of

Active Maintenance and Executive Control andensured that book had a coherent focus. We

hope this approach serves as a useful device toelucidate the commonalities and differencesamong different theoretical proposals on the na-ture of working memory variation. The ques-tions to be addressed by each research teamare as follows:

QUESTION 1: OVERARCHING

THEORY OF WORKING MEMORY

What is the theory or definition of workingmemory that guides your research on workingmemory variation?

QUESTION 2: CRITICAL SOURCESOF WORKING MEMORY VARIATION

What is your view on the critical source(s) of working memory variability within your targetpopulation(s) of study? Why do you focus onthe specific source(s) of variability in your re-search?

QUESTION 3: CONSIDERATION OFOTHER SOURCES OF WORKINGMEMORY VARIATION

Do you find other sources of working memoryvariability proposed in this volume to be appli-cable to your target population of study? Arethese sources of working memory variation com-patible or incompatible with your view of your target population?

QUESTION 4: CONTRIBUTIONSTO GENERAL WORKING

MEMORY THEORY

What does the variability within your targetpopulation of study tell us about the structure,

function, and/or organization of working mem-ory in general?

These questions were carefully chosen (throughextensive discussions among the five editorialteam members) not only to help each contrib-

uting team clearly outline their main theoreti-cal position in a larger theoretical context butalso to maximize the possibility of successfullyelucidating the commonalities and differencesamong the different theoretical proposals. Morespecifically, Question 1 was included to en-courage each contributing team to articulatetheir theoretical background directly (particu-larly some underlying assumptions that are notalways made explicit and, hence, often hindercomparisons among different theoretical pro-posals). Question 2 is perhaps the most centralone of the four common questions in that it askseach contributing team to specify their theo-retical proposals about the nature of variation inworking memory in the target population(s) theteam focused on. Question 3 was designedspecifically to alleviate the problem noted ear-lier of researchers often focusing on a single

theoretical construct or mechanism in theirwork and not considering other constructs ormechanisms proposed in the literature. Thisquestion essentially led each contributing teamto reflect upon other related theoretical pro-posals and clearly articulate the relationship to

Preface ix



those alternative proposals by either arguingagainst them or offering a theoretical synthesis.Finally, Question 4 raised a metatheoretical is-sue and asked each contributing team to discuss

the theoretical benefits (or even the necessityof ) studying variation in working memory. An-swers to this final question provided by differentresearchers in the volume will help us to betterunderstand the strengths (and likely weaknesses)of the individual-differences approach and, as awhole, offer an interesting case study (in thecontext of working memory research) of howstudies of interindividual and intraindividualvariation in general can inform or guide theory

development and practical applications.

References

Baddeley, A. (2000). The episodic buffer: a new

component of working memory. Trends in Cog-

nitive Sciences, 4, 417–423.

Baddeley, A. D., & Hitch, G. J. (1974). Working

memory. In G. H. Bower (Ed.), The psychology

of learning and motivation: Advances in research

and theory (Vol. 8, pp. 47–89). New York: Academic Press.

Barrouillet, P., Bernadin, S., & Camos, V. (2004).

Time constraints and resource sharing in

adults’ working memory span. Journal of Ex-

perimental Psychology: General, 133, 83–100.

Bayliss, D. M., Jarrold, C., Gunn, D. M., &

Baddeley, A. D. (2003). The complexities of

complex span: Explaining individual differ-

ences in working memory in children and

adults. Journal of Experimental Psychology:

General, 131, 71–92.

Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides,

J., Smith, E. E., & Noll, D. C. (1997). A

parametric study of prefrontal cortex involve-

ment in human working memory. NeuroImage,

5, 49–62.

Caplan, D., & Waters, G. S. (1999). Working

memory and sentence comprehension. Behav-

ioral and Brain Sciences, 22, 77–126.Case, R., Kurland, M., & Goldberg, J. (1982).

Operational efficiency and the growth of short

term memory span. Journal of Experimental

Child Psychology, 33, 386–404.

Conway, A. R. A., Cowan, N., Bunting, M. F.,

Therriault, D., & Minkoff, S. (2002). A latent

variable analysis of working memory capacity,

short-term memory capacity, processing speed,

and general fluid intelligence. Intelligence, 30,

163–183.

Cowan, N. (2005). Working memory capacity. Hove,UK: Psychology Press.

Cronbach, L. J. (1957). The two disciplines of

scientific psychology. American Psychologist,

12, 671–684.

Daneman, M., & Carpenter, P. A. (1980). Indivi-

dual differences in working memory and read-

ing. Journal of Verbal Learning and Verbal

Behavior, 19, 450–466.

Daneman, M., & Merikle, P. M. (1996). Working

memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review, 3,

422–433.

Engle, R. W., Kane, M. J., & Tuholski, S. W.

(1999). Individual differences in working mem-

ory capacity and what they tell us about

controlled attention, general fluid intelligence,

and functions of the prefrontal cortex. In A.

Miyake & P. Shah (Eds.), Models of working

memory (pp. 102–134). New York: Cambridge

University Press.Ericsson, K. A., & Kintsch, W. (1995). Long-term

working memory. Psychological Review, 102,

211–245.

Fry, A. F., & Hale, S. (1996). Processing speed,

working memory, and fluid intelligence. Psy-

chological Science, 7, 237–241.

Jarrold, C. (2001). Applying the working memory

model to the study of atypical development. In

J. Andrade (Ed.), Working memory in perspective

(pp. 126–150). Hove, UK: Psychology Press.

Jarrold, C., & Towse, J. N. (2006). Individual

differences in working memory. Neuroscience,

139, 39–50.

Jonides, J., Smith, E. E., Marshuetz, C., Koeppe,

R. A., & Reuter-Lorenz, P. A. (1998). Inhibition

in verbal working memory revealed by brain

activation. Proceedings of the National Academy

of Sciences USA, 95, 8410–8413.

Just, M. A., & Carpenter, P. A. (1992). A capacitytheory of comprehension: Individual differ-

ences in working memory. Psychological Re-

view, 99, 122–149.

Kane, M. J., & Engle, R. W. (2003). Working

memory capacity and the control of attention:

The contributions of goal neglect, response

x Preface



competition, and task set to Stroop interfer-

ence. Journal of Experimental Psychology: Gen-

eral, 132, 47–70.

Kosslyn, S. M., Cacciopo, J. T., Davidson, R. J.,

Hugdahl, K., Lovallo, W. R., Spiegel, W. R.,et al. (2002). Bridging psychology and biology:

The analysis of individuals in groups. American

Psychologist, 57, 341–351.

Lustig, C., May, C. P., & Hasher, L. (2001).

Working memory span and the role of proactive

interference. Journal of Experimental Psychol-

ogy: General, 130, 199–207.

Miyake, A., Friedman, N. P., Emerson, M. J.,

Witzki, A. H., Howerter, A., & Wager, T. D.

(2000). The unity and diversity of executivefunctions and their contributions to complex

‘‘frontal lobe’’ tasks: A latent variable analysis.

Cognitive Psychology, 41, 49–100.

Miyake, A., Friedman, N. P., Rettinger, D. A., Shah,

P., & Hegarty, M. (2001). How are visuo-spatial

working memory, executive functioning, and

spatial abilities related? A latent-variable anal-

ysis. Journal of Experimental Psychology: Gen-

eral, 130, 621–640.

Miyake, A., & Shah, P. (Eds.). (1999). Models of working memory: Mechanisms of active mainte-

nance and executive control. New York: Cam-

bridge University Press.

Munakata, Y. (2004). Computational cognitive neu-

roscience of early memory development. De-

velopmental Review, 24, 133–153.

Oberauer, K., & Kliegl, R. (2001). Beyond re-

sources: Formal models of complexity effectsand age differences in working memory. Euro-

pean Journal of Cognitive Psychology, 13, 187–

215.

Reuter-Lorenz, P. A., Marshuetz, C., Jonides, J.,

Smith, E. E., Hartley, A., & Koeppe, R. (2001).

Neurocognitive ageing of storage and executive

processes. European Journal of Cognitive Psy-

chology, 13, 257–278.

Roberts, R. J., & Pennington, B. F. (1996). An

interactive framework for examining prefrontalcognitive processes. Developmental Neuropsy-

chology, 12, 105–126.

Saito, S., & Miyake, A. (2004). On the nature of

forgetting and the processing-storage relation-

ship in reading span performance. Journal of

Memory and Language, 50, 425–443.

Towse, J. N., Hitch, G. J., & Hutton, U. (1998). A

reevaluation of working memory capacity in

children. Journal of Memory and Language, 39,

195–217.Underwood, B. J. (1975). Individual differences as a

crucible in theory construction. American Psy-

chologist, 30, 128–134.

Preface xi



Acknowledgments

We would like to thank those who helped stagethe Variation in Working Memory Conferenceheld at the University of Illinois at Chicago,July 24–26, 2003.

Sponsors of that meeting included the fol-lowing:

University of Illinois at Chicago, Depart-ment of Psychology

University of Illinois at Chicago, Office of the Vice Chancellor for Research

University of Illinois at Chicago, Center forthe Study of Learning, Instruction, andTeacher Development

University of Colorado at Boulder, Depart-ment of Psychology

University of North Carolina at Greensboro,

Department or Psychology

Jim Pellegrino (University of Illinois at Chi-cago) and Alan Baddeley (University of York)attended the conference and acted as discus-sants; they offered wise and constructive re-flections on the contributions, and they helped

to give a context and framework for the issuesthat emerged.

We thank Greg Colflesh for maintaining theWeb site where drafts of chapters were madeavailable for review and circulation.

The book has benefited from an exten-sive review process. Each of the contributors’chapters has been studied by relevant expertsand by graduate students, as well as by the edi-torial team. They were asked to assess a chap-ter in terms of coherence and cogency, andthey also had a specific brief to reflect on thereadability of the work (for a non-expert audi-ence), and the extent to which the text addressedthe book questions. We are grateful to the fol-lowing reviewers: Ivan Ash, Phil Beaman, GregColflesh, Rick Cooper, Margaret Cousins, Dale

Dagenbach, Eddy Davelaar, Rik Henson, Eli-zabeth Jeffries, Robert Kail, Susan Kemper, Alycia Kubat-Silman, Monica Luciana, DenisMareschal, Murray Maybery, Candice Morey,Tim Nokes, Harry Purser, Sarah Ransdell,

Alastair Smith, Craig Thorley, and Geoff Ward.



In addition, the book benefited from im-portant advice given by Nelson Cowan, whoacted as an overall reviewer.

Assembling a volume with five editors and

over 30 contributors has presented its ownchallenges, but it has also been fun. The abilityto communicate by e-mail has been crucial;one editor alone has a collection of over 1700

messages connected with this book, and thecamaraderie among the editors has helped toensure the success of the project. In this con-text, we are grateful to Catharine Carlin at

OUP for her steadfast belief in the idea from itsbeginning. We appreciate the work of all theOUP staff who have helped, especially NancyWolitzer, who has coordinated production.

xiv Acknowledgments



Contents

Contributors xvii

1. Variation in working memory: An introduction 3 Andrew R. A. ConwayChristopher JarroldMichael J. Kane

Akira MiyakeJohn N. Towse

PART I Working memory variationreflecting normal inter- andintra-individual differences

2. Variation in working memory

capacity as variation in executiveattention and control 21Michael J. Kane

Andrew R. A. ConwayDavid Z. HambrickRandall W. Engle

3. Individual differences in workingmemory capacity andreasoning ability 49Klaus Oberauer Heinz-Martin Su ßOliver WilhelmNicolas Sander

4. Explaining the many varieties of working memory variation: Dualmechanisms of cognitive control 76Todd S. Braver Jeremy R. GrayGregory C. Burgess

PART II Working memory variationdue to normal and atypicaldevelopment

5. Variation in working memory dueto normal development 109John N. TowseGraham J. Hitch



6. Variation in working memory dueto typical and atypical development 134Christopher JarroldDonna M. Bayliss

7. Developmental and computationalapproaches to variation inworking memory 162Yuko MunakataJ. Bruce MortonRandall C. O’Reilly

8. Variation in working memoryacross the life span 194Sandra HaleJoel MyersonLisa J. EmeryBonnie M. LawrenceCarolyn DuFault

PART III Working memory variationdue to normal and pathological aging

9. Inhibitory mechanisms and

the control of attention 227Lynn Hasher Cindy LustigRose Zacks

10. The executive is central to workingmemory: Insights from age,performance, and task variations 250Patricia A. Reuter-Lorenz

John Jonides

11. Specialized verbal workingmemory for languagecomprehension 272David CaplanGloria WatersGayle DeDe

Author Index 303

Subject Index 315

xvi Contents



Contributors

Donna M. Bayliss

School of Psychology

University of Western Australia

Crawley WA, Australia

E-mail: [email protected]

Todd S. Braver

Department of Psychology

Washington University

St. Louis, MO


Gregory C. Burgess


University of Colorado at Boulder

Boulder, CO


David Caplan

Neuropsychological Laboratory

Massachusetts General Hospital

Boston, MA


Andrew R. A. Conway


Princeton University

Princeton, NJ


Gayle DeDe

Department of Communication Disorders

Sargent College of Health and

Rehabilitation Sciences

Boston University

Boston, MA


Carolyn Dufault



St. Louis, MO


xvii



Lisa J. Emery


North Carolina State University

Raleigh, NC

E.mail: [email protected]

Randall W. Engle

School of Psychology

Georgia Institute of Technology

Atlanta, GA


Jeremy R. Gray


Yale University

New Haven, CT


Sandra Hale



St. Louis, MO


David Z. Hambrick

Department of Psychology,

Michigan State University

East Lansing, MI


Lynn Hasher


University of TorontoToronto, Ontario, Canada


Graham J. Hitch


University of York

York, UK


Christopher Jarrold

Department of Experimental

Psychology

University of Bristol

Bristol, UK


John Jonides


University of Michigan

Ann Arbor, MI

E-mail:[email protected]

Michael J. Kane


University of North Caroline at

Greensboro

Greensboro, NC


Bonnie M. Lawrence

Department of PsychologyCase Western Reserve University

Cleveland, OH

E.mail: [email protected]

Cindy Lustig

Department of Psychology,

University of Michigan

Ann Arbor, MI


Akira Miyake



Boulder, CO


J. Bruce Morton

Department of PsychologyUniversity of Western Ontario

London, Ontario, Canada.


Yuko Munakata



Boulder, CO


Joel Myerson



St. Louis, MO


xviii Contributors



Klaus Oberauer

Department of Experimental Psychology

University of Bristol

Bristol, UK


Randall C. O’Reilly



Boulder, CO


Patricia A. Reuter-Lorenz

Department of PsychologyUniversity of Michigan

Ann Arbor, MI


Nicolas Sander

University of Mannheim

Projekt Auswahlverfahren

Mannheim, Germany


Heinz-Martin Suß

Institute of Psychology

University of Magdeburg

Magdeburg, Germany


magdeburg.de

John N. Towse


Fylde College

Lancaster University

Lancaster, UKE-mail: [email protected]

Gloria Waters

Department of Communication Disorders

Sargent College of Health and

Rehabilitation Sciences

Boston University

Boston, MA


Oliver Wilhelm

Institute of Psychology

Humboldt University Berlin

Berlin, Germany


Rose Zacks


Michigan State University

East Lansing, MIE-mail: [email protected]

Contributors xix



VARIATION IN WORKING MEMORY



1

Variation in Working Memory:

An Introduction

ANDREW R. A. CONWAY, CHRISTOPHER JARROLD,

MICHAEL J. KANE, AKIRA MIYAKE,

and JOHN N. TOWSE

Individual differences have been an annoyancerather than a challenge to the experimenter. Hisgoal is to control behavior, and variation withintreatments is proof that he has not succeeded.Individual variation is cast into that outer dark-ness known as ‘‘error variance.’’ For reasons bothstatistical and philosophical, error variance is tobe reduced by any possible device. . . . The cor-

relational psychologist is in love with just thosevariables the experimenter left home to forget.He regards individual and group variation as im-portant effects of biological and social causes. Allorganisms adapt to their environments, but notequally well. His question is: what present char-acteristics of the organism determine its modeand degree of adaptation? (Cronbach, 1957,p. 674)

Neither group nor individual differences research

alone is sufficient; researchers need to combinethe two. Indeed, by combining the two, one maydiscover that the group results reflect the com-bination of several strategies, each of which drawson a different (or partially different) system. Thus,the group and individual differences findingsmutually inform each other, with the synergy be-

tween them illuminating the complex relationsbetween psychology and biology. (Kosslyn et al.,2002, p. 348)

The ability to mentally maintain informationin an active and readily accessible state, whileconcurrently and selectively processing newinformation, is one of the greatest accomplish-ments of the human mind; it makes possibleplanning, reasoning, problem solving, reading,and abstraction. Of course, some minds accom-plish these goals with more success than doothers. Working memory (WM) is the term thatcognitive psychologists use to describe the abil-ity to simultaneously maintain and process goal-relevant information. As the name implies, theWM concept reflects fundamentally a form of

memory, but it is more than memory, for it ismemory at work, in the service of complex cog-nition. As well, WM is a system with multiplecomponents, or a collection of interrelated pro-cesses, that carries out several important cog-nitive functions. Most WM theories argue thatthe system comprises mechanisms devoted to

3



the storage of information and mechanisms forcognitive control (see Miyake & Shah, 1999a).These mechanisms of active maintenance andexecutive control are thought to be involved in

most complex cognitive behaviors and so WMhas become a central construct in psychology. A fundamental characteristic of WM is that

it has a limited capacity, which constrains cog-nitive performance, such that individuals withgreater capacity typically perform better thanindividuals with lesser capacity on a range of cognitive tasks. For example, older childrenhave greater capacity than younger children,healthyadultshavegreatercapacitythanpatients

with frontal-lobe damage or disease, youngeradults have greater capacity than elderly adults,and in all such cases, those individuals withgreater WM capacity outperform individualswith lesser capacity in several important cog-nitive domains, including complex learning,reading and listening comprehension, and rea-soning. In short, we know that variation in WMcapacityexistsandthatthisvariationisimportantto everyday cognitive performance.

The central goal of this volume is to ad-vance understanding of WM variation and itsconsequences. In recent years, several theoriesand empirically based arguments have beenproposed to explain some form of inter- or intra-individual variation or another, yet no compre-hensiveaccountofvariationinWMhasemerged.Instead, the literature is cluttered with compet-ing claims and data sets that are inadequate atleveraging support for one theory or another.This state of affairs is the result of several unfor-tunate, yet not surprising, aspects of research onWM variation. First, no consensus has yetemerged regarding nomothetic WM theories ormodels (see Miyake & Shah, 1999a), and sovariation researchers start with different assump-tions about what WM is, how it plays a role incomplex cognition, and how best to measure itscapacity. Second, variation researchers study a

variety of subject populations—some are pri-marily interested in individual differences inyoung adults, others are interested in child de-velopment, while others are interested in cogni-tive aging, and so on. Not only does this createdivisions within the WM literature, but it alsocomplicates theories of variation because the

sources of variation in one research populationmay not be the same as the sources of variation inanother. Third, specific investigations often en-tertain only one or two potential sources of WM

variation, often pitting them against one another(e.g., processing speed vs. attentional inhibition),when many other mechanisms or processes arepotentially at play. Finally, much of the researchin this area has been plagued by poorly oper-ationalized constructs, and so the mapping fromconstructs to mechanisms to measures is oftenweak.

In this book we take a more comprehensivelook at variation in WM. Multiple models of

WM are considered, data from different sub-ject populations are presented, multiple sourcesof variation are discussed, and close attentionis paid to the relation between constructs, mech-anisms, and measurement. We hope that sucha comprehensive, diversified approach to onecommon issue will result in both a convergenceand a divergence of ideas. Ideally, some con-sensus can be achieved, while at the same timethis synthesis of perspectives should illuminate

current points of contention within the field,which should inspire investigation of the mostimportant empirical and theoretical questions.

To properly contextualize the chapters thatfollow, this Introduction will begin with a brief historicaloverviewofWMresearch,payingcloseattention to the role played by variation research,such as studies of individual differences, childdevelopment, aging, and neuropsychology. Inthat spirit, this chapter will also highlight theimportance of combining the experimental anddifferential disciplines in psychology and thebenefits of converging operations. Finally, wewillintroduceanddiscussthefourquestionsthatwere posed to each contributing research groupin this volume.

HISTORICAL OVERVIEW

The concept of a limited-capacity immediatememory system has a long history in psychol-ogy. Ebbinghaus (1885/1964) reported thathe could perfectly recall lists of 7 or fewer non-sense syllables upon a single presentation, butthat lists of 8, 9, and 10 syllables required

4 Variation in Working Memory: An Introduction



approximately 5, 9, and 12 repetitions, respec-tively (and the learning curve continued to growsteeply for still longer lists). However, in lightof modern views that distinguish immediate

memory from long-term memory, it is interest-ing that Ebbinghaus had so little to say aboutthis dramatic finding. He certainly proposed nospecial mental state or faculty associated withimmediate recall of short sequences.

Others soon did. In a review of Ebbinghaus’s(1885/1964) book, Jacobs (1885) predicted thatthe objective study of memory would allow forthe assessment of people’s mental abilities: ‘‘If this be visionary, we may at least hope for much

of interest and practical utility in the compari-son of the varying powers of different mindswhich can now be laid down to scale’’ (p. 456).Moreover, in his subsequent call for the forma-tion of a Society for Experimental Psychology,Jacobs (1886) cited Ebbinghaus’s immediate-memory findings to argue that a high priority forpsychology should be to understand variation inimmediate memory:

There is, I submit, a certain number of syllablesup to which each person can repeat a nonsenseword like borg-nap-fil-trip after only once hear-ing; and it is probable, though we cannot knowfor certain, that this number varies with differentpersons, giving a sort of test of their linguisticcapacity....But this law, if it is a law, has atpresent only been deduced from the observationof one man’s mind, and is therefore obviously nota law of mind in general, but at best a law of Dr. Ebbinghaus’s mind. (p. 53)

The following year, Jacobs (1887) reportedthe first empirical paper on the memory spantask and one of the first systematic individual-differences studies of memory. Students be-tween the ages of 8 and 20 years were presentedwith lists of auditory nonsense syllables, letters,or digits to repeat. The largest set that eachstudent perfectly reproduced was termed his or

her span of prehension. Jacobs found that spanincreased not only with chronological age butalso with higher school grades. In a supple-mental report to Jacobs, Galton (1887) notedthat the spans among institutionalized childrenandyoungadults(thenclassifiedas‘‘idiots’’)werequite limited, averaging only three to four

items. The capacity of immediate memory, asreflected by prehension span, thus appeared tobe a source of intellectual ability:

Under these circumstances we might expect that‘‘span of prehension’’ should be an important fac-tor in determining mental grasp, and its deter-mination one of the tests of mental capacity.(Jacobs, 1887, p. 79)

Following such reports, it may not be sur-prising that span tasks soon became a part of intelligence test batteries (e.g., Binet & Simon,1905; Burt, 1909; Cattell & Galton, 1890; Ebb-

inghaus, 1897). At about the same time, WilliamJames (1890) drew a theoretical distinction be-tween‘‘primary’’memoryand‘‘secondary’’mem-ory, or memory proper. According to James,primary memory is equated with the currentcontents of consciousness and the ‘‘rearwardportion of the present space of time,’’ and there-fore suffers from a severe capacity limit. Second-ary memory, in contrast, is thought to consist of memories of the distant past and to be unlimited

in capacity. James Mark Baldwin (1894), whowas jointly influenced by Wundt’s experimentalpsychology (see below) and Darwin’s theory of evolution, similarly argued that immediatememory is capacity limited and that the devel-opment of this capacity is central to the devel-opment of intelligence and cognitive abilities.

But what did this capacity represent? Al-though span tasks clearly required remember-ing and were soon referred to as ‘‘memory span’’tasks (e.g., Bennett, 1916; Humpstone, 1917),and although James used the phrase ‘‘primarymemory’’ to describe the underlying construct,it is quite clear that most early theorists con-sidered the span task to reflect the contents of consciousness or the capacity of attention. Ja-cobs (1887, p.79) described it as an index of what could be ‘‘taken on’’ by the mind. Cattelland Galton (1890, p. 377) considered the span

task to test ‘‘memory and attention.’’ Bolton(1892) argued that span tested ‘‘the power of concentrated and sustained attention’’ (p. 364),and Humpstone (1919) viewed it as ‘‘the abilityto grasp a number of discrete units in a sin-gle moment of attention.’’ In Wundt’s clas-sic (1912/1973) text, he viewed the span of

Variation in Working Memory: An Introduction 5



memory and the span of apprehension as re-flecting a common limit to focus of atten-tion, thus foreshadowing Miller’s (1956) famoushaunting by the magical number 7 as a general

limit to cognitive capacity.1

Moreover, muchlike the modern theory of Cowan (1988, 1995,2001), Wundt used the findings from span tasksto argue for a limited-capacity attentional focusof six or fewer impressions (termed appercep-tion), and a larger and more variably limited‘‘field of consciousness’’ (termed apprehension).

The historical perspectives we have discussedthus far can all be categorized as ‘‘capacity’’theories of immediate memory, for each of them

embraced, to one degree or another, the notionof an ‘‘amount’’ of information that can be ac-tively maintained at once (for a review, seeBlankenship, 1938). Other, subsequent theoristsoffered time-based or interference-based per-spectives on the limits on immediate memory,and so they also more firmly brought ‘‘memoryspan’’ from the conceptual realm of attentioninto that of memory (see Cowan, 1995). Forexample, Hebb (1949) introduced the concept

of transient ‘‘cell assemblies’’ by which incom-ing stimuli form a distributed pattern of activa-tion across neurons that quickly dissipates withtime, resulting in the quick loss of information.Similarly, Thorndike (1914) proposed the ‘‘lawof disuse,’’ arguing that memories are quicklylost over time if they are not used, or retrieved.McGeoch (1932), partly in rejection of Thorn-dike, argued that memory traces do not decayover time but become unavailable for retrievaldue to the forces of retroactive and proactiveinterference. In fact, McGeoch and subsequentinterference theorists argued that there is no dis-tinction between immediate memory and otherforms of memory and so the quest to understandthe capacity of immediate memory is misguided(Crowder, 1982; Melton, 1963; Nairne, 2002).

Neuropsychological case studies, however,do suggest a distinction between short-term and

long-term memory. Perhaps the most famous of these investigations involve the temporal lobec-tomy patient H.M., who demonstrated normalshort-term memory capacity but was unable toform new long-term memories (Milner, 1966).The opposite pattern of deficits has also been

demonstrated, i.e., normal long-term memoryperformance accompanied by verbal short-term memory deficits (Baddeley & Warrington,1970; Shallice & Warrington, 1970; Warring-

ton, Logue, & Pratt, 1971). Patients with long-term memory damage, such as H.M., typicallyhave acquired damage to temporal-lobe struc-tures, such as the hippocampus, whereas pa-tients with short-term memory damage typicallysuffer from more frontal and left parietal dam-age, suggesting different memory systems for theshort- and long-term retention of information.

Despite the obvious influences upon, andconnections to, contemporary research on WM,

most of the references we have discussed so faractually predate the introduction of the conceptof WM to the field of psychology. The first ref-erence to WM memory, as it is conceived today,was in the influential book Plans and theStructure of Behavior (Miller, Galanter, & Pri-bram, 1960).2 In sharp contrast to the behav-ioral psychologists that came before them,Miller et al. argued that a theory was needed tocapture the processes that occur between the

presentation of a stimulus and the execution of aresponse. In their terms, they were interested inhow knowledge is translated into action. Theyargued that human beings are capable of form-ing, hierarchically structuring, and executingplans. Importantly, plans were considered to beinternal knowledge representations that couldbe retrieved, or activated, into WM. Accordingto Miller et al.,

When we have decided to execute some partic-ular Plan, it is probably put into some specialstate or place where it can be remembered whileit is being executed. Particularly if it is a tran-sient, temporary kind of plan that will be usedtoday and never again, we need some specialplace to store it. The special place may be a sheetof paper. Or (who knows?) it may be somewherein the frontal lobes of the brain. Without com-mitting ourselves to any specific machinery, we

should like to speak of the memory we use for theexecution of our Plans as a kind of quick-access,‘‘working memory.’’ (p. 65)

The connection between this definition of WM and earlier concepts of capacity-limited




immediate memory is obvious,3 yet it alsorepresents an important departure. Note twokey phrases in the passage: a ‘‘special state orplace where it [a plan] can be remembered

while it is being executed,’’ and ‘‘the memorywe use for the execution of our Plans.’’ Thesephrases imply a unique system responsible notonly for the storage of plans but also for theirimplementation. Prior discussions of a limited-capacity immediate memory system generallyemphasized storage more than anything else.Here, the emphasis is on storage and proces-sing in the service of a complex cognitive goal.

While Plans, and the Structure of Behavior

can be credited with the first mention of WM,Baddeley and Hitch (1974) are duly creditedwith launching the empirical investigation of WM that continues today. Baddeley and Hitchfamously started their seminal chapter with thefollowing complaint:

Despite more than a decade of intensive researchon the topic of short-term memory (STM), we stillknow virtually nothing about its role in normal

human information processing. (p. 47)

Baddeley and Hitch acknowledged that aconsiderable amount of important research hadbeen conducted to address fundamental ques-tions about STM itself, such as how informationis coded in STM (Conrad, 1964; Conrad &Hull, 1964; Wickelgren, 1966), what the capac-ity of STM is (Miller, 1956; Waugh & Norman,1965), how information is retrieved from STM(Sternberg, 1966, 1969), and whether forget-ting from STM is due to decay or interference(Brown, 1958; Keppel & Underwood, 1962;Petersen & Petersen, 1959; Reitman, 1971,1974; Waugh & Norman, 1965). However, theyalsolamentedthefactthatverylittleresearchhadaddressed the role of STM in more complexcognitive behavior. This was indeed a strangestateofaffairs,especiallyconsideringtheprimary

role granted to STM in the most influentialinformation-processing models of the time (e.g., Atkinson & Shiffrin, 1968; Broadbent, 1958).

Baddeley and Hitch proposed a multicom-ponent WM model that consisted of domain-specific storage buffers (later referred to as the

‘‘slave systems’’; Baddeley, 1986) as well as acentral executive. They provided empirical ev-idence from dual-task studies showing that themental juggling required by complex cognitive

behaviors, such as reasoning, can be achievedby coordinated storage and processing betweenthe slave systems and the central executive. A certain amount of information can be held atbay in the slave systems while the executiveworks on new information.

As the influence of the Baddeley and Hitchstudy took hold in the late 1970s, a related threadof research started to emerge from developmen-tal psychology (see Case, 1978; Pascual-Leone,

1970). Particularly relevant here is the workof Robbie Case (1978, 1985) on the develop-ment of memory capacity in young children.Like Baddeley and Hitch (1974), he conceivedof a limited-capacity mental workspace that wasrequired in the service of complex cognition. Hereferred to this mental workspace as M-space andargued that the development of M-space is es-sential to the development of cognitive abilitiesin general. Importantly, Case developed an ob-

jective measure of M-space, called the countingspantask (see Case, Kurland, & Goldberg, 1982).In the task, children are presented with displays(index cards in the original version) consistingof an array of colored objects (e.g., green andyellow circles) and are instructed to count aparticular object-type (e.g., green circles) and re-member the count-total for later recall. A num-ber of displays are presented in succession and atthe end of the series the child is expected torecall all the count-totals in that series. This taskis thought to tap M-space, or WM capacity, be-cause it requires storage (remember the digits) inthe face of processing (count the objects in thedisplay).

Daneman and Carpenter (1980) developeda similar task to measure WM capacity in adults.In their reading span task, the subject is pre-sented with sentences and must read each sen-

tence aloud and remember the last word of thesentence for later recall. After a series of sen-tences the subject is expected to recall all thesentence-end words. Thus, like counting span,the reading span task requires storage (remem-ber the words) in the face of concurrent pro-




cessing (read the sentences aloud). Importantly,Daneman and Carpenter (1980) demonstratedthat scores on the reading span task predictedmeasures of reading comprehension better than

span tasks that did not have the secondary pro-cessing component (e.g., simple word span).Several WM span tasks have been developed

over the years (see Kane et al., 2004; Shah &Miyake, 1996; Turner & Engle, 1989) and theyconsistently show better predictive validity withrespect to complex cognitive behaviors, such asreasoning, reading comprehension, and prob-lem solving, than do simple span tasks thought tojust tap the capacity of a short-term store, a ` la

Miller (1956). In this respect, WM span tasks arethought to be a success. That is, they say some-thing about how well people perform real-worldcognitive tasks and they explain real-world varia-tion in performance. Working-memory span taskstherefore directly address the frustration evidentin the Baddeley and Hitch (1974) quote above.

In addition, the development of tasks de-signed specifically for the measurement of WMcapacity allowed for a surge in variation research.

WM span tasks, and similar measures of WMfunction, have now been extensively used toexamine individual differences in cognitive abil-ities in young adults, the development of WMand cognitive functioning in children, the de-clines in cognitive performance associated withaging, and the deficits experienced by patientswith brain damage or disease.

Despite this progress, several important ques-tions remain and the chapters in this book willaddress them. For instance, it still is not entirelyclear why WM span tasks, such as countingspan and reading span, predict complex cog-nitive behavior better than simple span tasksdo. Furthermore, the relation between spantasks and other measures of WM (such as the‘‘n-back’’ task, used often in neuroimaging stud-ies) has not been properly investigated. As a re-sult, the mechanisms underlying performance

of these tasks are not completely understood.Finally, and most imperative here, the sources of variation in WM performance—among chil-dren, among healthy young adults, among theelderly, and among patients—have not beenidentified, hence the need for a comprehensivereview of current research on WM variation.

VARIATION

The approach proposed here, the approach whichmakes individual-differences variables crucibles

in theory construction, will identify the processvariables as a fallout from nomothetic theory con-struction. (Underwood, 1975, p. 134)

All of the research presented in the subsequentchapters is influenced by classic experimentalpsychology; indeed, many sophisticated exper-imental manipulations are presented and dis-cussed. At the same time, each of these chapterstakes seriously and attempts to provide a theo-

retical accountof variation in performance. Thiscombination of experimental and differentialpsychology is an answer to Cronbach’s (1957)largely unheeded call for a unification of ‘‘thetwo disciplines of scientific psychology.’’ Cron-bach argued that a comprehensive account of human behavior could only be achievedthrough the synergy of experimental and differ-ential approaches to studying psychology. Heclaimed:

A true federation of the disciplines is required.Kept independent, they can only give wrong an-swers or no answers at all regarding certain im-portant problems. It is shortsighted to argue forone science to discover the general laws of mindor behavior and for a separate enterprise con-cerned with individual minds. (p. 673)

Sadly, despite Cronbach’s call for unifica-tion, most experimental psychologists still dis-regard individual differences as error variance;indeed,asimilarpleaforintegrationwasdeemednecessary, yet again, this decade (Kosslyn et al.,2002). A perusal of most mainstream cognitivepsychology and neuroscience journals will re-veal that individual differences analyses are stillfew and far between. Even more rare are studiesthat successfully combine experimental anddifferential approaches as Cronbach suggested.

Fortunately, the contributors to this volume area welcome exception.These contributors recognize that individual-

differences data can be used in combinationwith experimental data to test theories and sug-gest hypotheses (Kosslyn et al., 2002; Under-wood, 1975). More specifically, most models or




theories of WM make either explicit or implicitclaims about individual differences in perfor-mance and therefore a differential approach canprovide supportive or problematic data for a

theory, whether that theory concerns primarilybehavioral or neural mechanisms. As this vol-ume attests, the strategy of exploiting individ-ual differences to test theory can of course beapplied beyond individual differences in healthyadults, but also to normal and atypical devel-opment and aging, and even to interspecies vari-ation.

As well, an understanding of how experi-mental manipulations impact individual differ-

ences can lead to better measurement of psy-chologicalconstructs.BindraandScheier(1954)called upon psychometricians to make use of experimental design for this reason:

The knowledge of what experimental conditionsdo change a property may enable the psycho-metrician to control such conditions in futuremeasurements of that property, and it also maylead to the discovery of new properties not af-

fected by these conditions. Thus, the use of ex-perimental variation in psychometric researchmay be of considerable help both in makingpsychometric measurement more exact, that is,reliable, and in suggesting ways of experimentallychanging the organism’s ‘‘invariant’’ properties.(Bindra & Scheier, 1954, p. 70)

In sum, investigations of variation benefitboth the experimentalist and the psychometri-cian. Theories of the general laws of mindand behavior are enhanced by a considerationof individual-differences data, and individual-differences investigations are enhanced by knowl-edge of experimental outcomes.

CONVERGING OPERATIONS

The necessary condition which makes possible thedetermination of particular characteristics of anyconcept...istheuseofwhathavebeencalledcon-verging operations. Converging operations maybe thought of as any set of two or more experi-mental operations which allow the selection orelimination of alternative hypotheses or concepts

which could explain an experimental result. Theyare called converging operations because they arenot perfectly correlated and thus can converge ona single concept. (Garner, Hake, & Eriksen,1956, pp. 150–151)

One of the primary strengths of each of the re-search groups contributing to this volume istheir use of converging operations to test hy-potheses and compare theories. Rather than relyon one particular methodological approach,such as computational modeling, neuroimag-ing, or structural equation modeling, each of theresearch programs represented here employs a

combination of methods. Some combine imag-ing and modeling, others combine imaging andpatient data, still others combine multivariatestatistical techniques and modeling, and so on.Just as Underwood (1975) recognized the theo-retical leverage to be gained from a consider-ation of individual-differences data, the contrib-uting chapters illustrate the force of argumentgained by considering data garnered from dif-ferent methodological approaches. For exam-

ple, jointly considering neuroimaging andneuropsychological data yields at least twobenefits:

(1) Neuroimaging and neuropsychological stud-ies should provide converging information aboutthe locus of brain regions that contribute to a par-ticular task; failures to find convergence can di-rect attention to new brain regions and cognitiveprocesses that contribute to task performance,and they can provide important challenges to thetheoretical assumptions of each methodology;and (2) . . . neuroimaging and neuropsychologyshould provide converging information about as-sociations and dissociations between tasks; failuresto find convergence can direct attention to alter-native theoretical accounts of cognition, whileunexpected associations can direct attention tocognitive processes that cross traditional domainsof study. (Fiez, 2001, p.20)

As another example, consider the benefits of using computational modeling to complementbehavioral work:

Such models can be crucial for helping us tounderstand complex, nonlinear interactions of the




sort that characterize brain–behavior relations.Moreover, such models assist in theory compar-ison and evaluation by requiring theories to bespecific and plausible enough that they can leadto working models, and by generating testable

predictions. For these reasons and others, manyresearchers argue that such modeling work isessential for advancing theorizing about cogni-tive functioning. (Munakata, Morton, & O’Re-illy, Chapter 7, this volume)

Similar statements can be made about othercombinations of methodologies as well. Gener-ally speaking, the more converging operationsthe better. Sternberg and Grigorenko (2001,p. 1069) recently referred to this approach as‘‘unified psychology,’’ which is ‘‘the multipara-digmatic, multidisciplinary, and integrated studyof psychological phenomena through convergingoperations.’’ Sternberg and Grigorenko arguethat the vast majority of contemporary psychol-ogists are specialists and fail to consider researchquestions from multiple perspectives. They fur-ther argue that such specialization thwarts real

progress in psychological science and producesscientists who might be productive but not pro-vocative in their research.

The project presented here is an attempt toput WM variation in the crosshairs of unifiedpsychology.Acrossthesubsequentchapters,mul-tiple paradigms are presented, converging op-erations are used to test and contrast theories,and scientists from different subdisciplines arerequired to address a common set of theoretical

questions.Furthermore, thecommon setof ques-tions forces each contributing research group toconsider the responses of the other contributors,thus enhancing the degree of integration acrosschapters. It is to this set of shared questions thatwe now turn our attention.

THE FOUR DESIGNATED QUESTIONS

FOR THIS VOLUME

Each contributing research group was asked toaddress the same set of important theoreticalquestions in an attempt to make plain the basictenets that guide their current research on WM

variation. Through these common questions,each research group explicitly addressed thetheories and subject populations at the focusof ‘‘competing’’ research programs, presented in

this volume’s other chapters.By adopting this common-question approach,we are following the successful example set byMiyake and Shah (1999a). In their edited book,Models of Working Memory: Mechanisms of Ac-tive Maintenance and Executive Control, eachcontributing research group addressed eight the-oretical questions about their particular modelor theory of WM. We believe that this is an ex-tremely fruitful approach because it allows the

reader to compare and contrast alternative an-swers to the same questions, in turn allowing fora synthesis of information across chapters. This isespecially helpful for students, or researchersnew to the field, who are struggling to under-stand fundamental differences and similaritiesamong competing or complementary theories.

QUESTION 1: OVERARCHING

THEORY OF WORKING MEMORY

Whatisthetheoryordefinitionofworkingmemorythat guides your research on working memoryvariation? Miyake and Shah (1999) demon-strated that there are currently at least a dozennomothetic theories of WM that are competingin the intellectual marketplace. Our first ques-tion therefore encourages each contributingresearch group to articulate their theoreticalframework and to make explicit any funda-mental assumptions that underlie their ap-proach. As it turns out, some aspects of differentmodels that might appear at to be odds with oneanother at first glance are actually compatible(Kintsch, Healy, Hegarty, Pennington, & Salt-house, 1999; Miyake & Shah, 1999b). Otheraspects of different models, however, are indeedin opposition to one another. By making ex-

plicit the fundamental aspects of different def-initions of WM, these points of convergenceand divergence will more easily be revealed.Most current theories of WM differ in theirapproach to a limited number of key issues,which we address in turn below.




Cognitive Control

One of the most contentious issues with respectto WM is the nature of cognitive control and

the characterization of the ‘‘central executive.’’Several of the current chapters eschew the no-tion of a central executive altogether in favor of models that view cognitive control as an emer-gent property of a dynamic interactive network(see Chapters 3, 4, and 7). For example, Mu-nakata et al. (Chapter 7) claim that theirmodel ‘‘has the elements in place for a fully self-contained theory of both maintenance andcontrol of WM, without relying on unexplained

‘homunculi’ such as a central executive.’’ Otherchapters seem to accept at face value the exis-tence of a central executive in their model of WM, perhaps revealing the influence of theoriginal Baddeley and Hitch (1974; Baddeley,1986) model (see Chapters 8 and 11). Stillothers place executive processes and attentionat the center of their research program, thus noteliminating an ‘‘executive’’ system, but also notaccepting the original conception of a central

executive (see Chapters 2, 5, 6, 9, and 10).

Unitary vs. Multifaceted Approach

Each of the WM models endorsed in this bookembraces the idea that WM is a multicompo-nent system, or that it represents a multiplicityof mental and neurological processes, includinginteractive mechanisms of information storageand cognitive control. As well, each chapterrecognizes the contribution to WM function of both domain-general and domain-specific pro-cesses. However, the chapters differ in the extentto which they take a unitary or a multifacetedapproach to the characterization of variation inWM and with respect to their emphasis ondomain-general vs. domain-specific sources of variation in performance. For example, Braveret al. (Chapter 4), Hasher et al. (Chapter 9),

Kane et al. (Chapter 2), and Munakata et al.(Chapter 7) take primarily a unitary approach,their emphasis being on domain-general mech-anisms of cognitive control. In contrast, Haleet al. (Chapter 8), Jarrold and Bayliss (Chap-ter 6), and Caplan et al. (Chapter 11) take a

much more multifaceted approach, elucidatingall the components of the WM system and sug-gesting multiple sources of variation in perfor-mance.

The Nature of Capacity Limitationsin Working Memory

Each of the chapters recognizes that WM is alimited-capacity system. However, the chaptersdiffer in how they conceptualize this limit. Somemodels are consistent with the original notionof capacity, as an amount of information (e.g.,Jacobs, 1887; Miller, 1956; Woodworth, 1938;

Wundt,1912/1973).Forexample,Oberaueretal.(Chapter 3) argue that capacity refers to thenumber of simultaneous bindings of indepen-dent chunks of information that can be achievedat once. Other researchers attribute the capacitylimitation to an ‘‘attentional’’ system, such as ex-ecutive attention (Kane et al., Chapter 2) or at-tentional inhibition (Hasher et al., Chapter 9).Others argue that the capacity limit is simply anatural property of a highly interactive biological

system (Braver et al., Chapter 4; Munakata et al.,Chapter 7; Reuter-Lorenz & Jonides, Chapter10). Yet another set of researchers advocates thenotion of a limited pool of ‘‘resources’’ availablefor the storage and processing of information(Hale et al., Chapter 8; Jarrold & Bayliss, Chap-ter 6; Towse & Hitch, Chapter 5).

QUESTION 2: CRITICAL SOURCES OFWORKING MEMORY VARIATION

What is your view on the critical source(s) of work-ing memory variation within your target popula-tion(s)ofstudy?Whydoyoufocusontheparticular source(s) of variation in your research? This isperhaps the most central question of the four andis the one that motivated the current volume.Here we ask each contributing chapter to specify

a theoretical proposal about the nature of varia-tion in WM in some target population(s). Sev-eral different theoretical sources of variation areproposed throughout the book, including mentalspeed (Hale et al., Chapter 8); attentional inhi-bition (Hasher et al., Chapter 9); reduction in




the fidelity of memory when engaged in process-ing (Towse & Hitch, Chapter 5); executive atten-tion processes of goal maintenance and conflictresolution (Kane et al., Chapter 2); the isolation

and strength of representations (Munakata et al.,Chapter 7); the capacity of simultaneous bindingof independent chunks (Oberauer et al., Chapter3); the relative contributions of proactive andreactive cognitive control (Braver et al., Chapter4); and the nature of domain-specific content(Jarrold & Bayliss, Chapter 6; Reuter-Lorenz &Jonides, Chapter 10). Some of these proposalsare more in agreement than others. For example,there are rather subtle differences between Kane

et al.’s executive attention, Hasher et al.’s atten-tional inhibition, and Braver et al.’s proactivecontrol. Yet there are substantial differences hereas well. In particular, Hale et al.’s speed account,Oberauer et al.’s binding theory, and the multi-ple attentional approaches mentioned above arecertainly at odds with one another.

The chapters also differ with respect to whe-ther they view WM variation as coming fromprimarily one source or multiple sources. Some

chapters clearly argue in favor of the notion thatthere is a primary source, such as executive at-tention(Chapter2),attentionalinhibition(Chap-ter 9), or binding (Chapter 3), while others arguefor multiple sources of variation, all of equalimportance (Chapters 6 and 10).

QUESTION 3: CONSIDERATIONOF OTHER SOURCES OF WORKINGMEMORY VARIATION

Do you find other sources of working memoryvariation proposed in this book to be applicableto your target population of study? Are thesesources of working memory variation compatibleor incompatible with your view of your targetpopulation? The purpose of posing this ques-tion was to encourage each contributing re-

search group to carefully consider sources of WM variation proposed by ‘‘competing’’ con-tributors to this volume. This question addressesour concern that research in this area is too oftennarrowly focused on a single construct or mech-anism. This question also requires researchers toconsider directly the sources of variation in re-

search populations other than their own. Thatis, the main sources of variation in WM maydiffer across subject populations. For instance, itis possible that processing speed accounts for

more variation in the development of WM inchildren than it does in variation among healthyyoung adults (see Chapters 2 and 8). As well, it ispossible that abnormalities in domain-specificmemory representations may account for morevariationamongpatientpopulationsthanamongnormals (see Chapter 6). Similarly, strategicallocation of resources may account for varia-tion associated with aging more than it does fornormal variation among younger adults (see

Chapters 2 and 10). In essence, the subsequentchapters unmistakably illustrate that there is nosingle source (mechanism, process, or resource)that can account for all WM variation.

QUESTION 4: CONTRIBUTIONSTO GENERAL WORKINGMEMORY THEORY

What does the variability within your target pop-ulation of study tell us about the structure, func-tion, and/or organization of working memory in

general? How has the investigation of variationcontributed to WM theory in general? As thequestion implies, variation research has contrib-uted to knowledge about the structure of WMand its role in successful complex cognition. Interms of structure, experimental, psychometric,neuroimaging,andneuropsychological datahavesupported (1) the distinction between a memorysystem responsible for retention of information inthe short term and a system responsible for long-term retention, and (2) domain-specific as well asdomain-general components of the WM system.In terms of function, variation researchers havemade several important discoveries about whenand where WM does and does not play a role inother cognitive processes or functions (see Chap-

ters 2 and 11). Variation research has also exposed the dif-ficulty in operationalizing psychological con-structs and the importance of measurement toany scientific pursuit. Psychometric studies inparticular (Chapters 2, 3, and 11) have dem-onstrated the importance of having valid and




reliable measures of WM functioning. Simi-larly, neuropsychological, developmental, andaging studies have demonstrated that vali-dity and reliability can vary across research

populations—i.e., a measure that is valid andreliable for one group of subjects may not befor another group of subjects.

ORGANIZATION OF THE BOOK

The book is organized into three parts pertainingto the type of WM variation under investigation.The first part is devoted to variation reflected by

normal inter- and intra-individual differences.The second part covers variation due to normaland atypical development and the third sectionreviews variation due to normal and pathologicalaging. As a result of this organizational structure,chapters within a part tend to have more incommon with each other than do chapters acrosssections. However, there are several points of contact across parts as well. For example, Kaneet al.’s approach to individual differences in

young adults has much in common with Hasheret al.’s theory of cognitive aging. Also, Munakataet al.’s developmental work has a great deal of theoretical overlap with Braver et al.’s programof research on young adults. Readers will un-doubtedly discover more subtle points of con-vergence across chapters as well. Below we reviewthe key points made by subsequent chapters.

Working-Memory VariationReflecting Normal Inter- andIntra-individual Differences

The first part deals primarily with individualdifferences in young adults. Perhaps due to theinfluence of Daneman and Carpenter (1980),or perhaps because college students are morereadily available as research subjects than chil-dren,theelderly,orneuropsychologicalpatients,

investigations of individual differences in youngadults have dominated the variation landscapefor the last two and a half decades (see Miyake,2001). In Chapter 2, Kane et al. present theirexecutive-attention theory of WM capacity,which is largely based on investigations of in-dividual differences in WM capacity among

healthy young adults. Kane et al. argue for mul-tiple sources of variation in performance of WMtasks, including domain-specific skills and strat-egies as well as a domain–general attention ca-

pability. It is this latter attention factor, theyargue, that accounts for the predictive validityof WM span tasks and drives the strong rela-tionship between WM capacity and a range of important cognitive abilities, including generalfluid intelligence. In Chapter 3, Oberauer et al.review their psychometric research on WM andreasoning ability. From this work they arguethat individual differences in WM capacity areprimarily driven by a general factor and this

factor is strongly related to reasoning. They fur-ther argue that the mechanism underlying thisgeneral ability is a mechanism that simulta-neously binds independent chunks of informa-tion in the focus of attention. They also arguethat some measures of executive function, suchas task-set switching, are not related to WMcapacity, and therefore conclude that WM ca-pacity cannot be equated with executive atten-tion. Chapter 4 presents Braver et al.’s dual-

mechanism theory, according to which thereare two modes of cognitive control: proactiveand reactive. They argue that performance inmost cognitive tasks involves a mixture of thesetwo modes of control and that individual dif-ferences in WM capacity may be associatedwith the extent to which subjects are able toengage proactive control and/or the extent towhich they are able to efficiently move fromone mode of control to another.

Working Memory Variation Due toNormal and Atypical Development

The second part is primarily concerned with thedevelopment of WM capacity. As mentionedabove, the investigation into the developmentof immediate memory has a long history inpsychology, dating back to the work of Jacobs

(1887) and Baldwin (1894), and was continuedby prominent developmental psychologists, suchas Pascual-Leone (1970) and Case (1985). Thechapters in this section follow in that tradition.

Specifically, in Chapter 5, Towse and Hitchconsider in some detail contemporary theoreti-cal accounts of the development of children’s




WM capacity. They review data that have ledthem to propose their own view of WM in chil-dren; their task-switching model emphasizes thetemporal dynamics of span tasks, and they argue

that there are multiple sources of variation inWM. Moreover, they attempt to illustrate howexperimental data can be used in conjunctionwith the analysis of individual differences tomarshal theoretical arguments, i.e., how eachapproach can complement the other. In Chapter6, Jarrold and Bayliss look at the constraints onperformance in the complex-span paradigm, todetermine the causes of individual and devel-opmental differences in WM. They present evi-

dence to suggest that both storage capacity andprocessing efficiency are separable determinantsof WM performance, and that a third, potentiallyexecutive, source of variance arises from the needto combine these two requirements of the com-plex span task. They go on to argue that variationin rate of reactivation of to-be-remembered ma-terial and variation in speed of processing un-derlie the first two of these constraints, and thatthe third may reflect variation in the rate at which

individuals forget information while occupied byprocessing activities. In Chapter 7, Munakataet al. present a biologically plausible computa-tional model to account for the development of WM and cognitive control in young children.They focus their work on the complementaryprocesses of maintenance and updating and ar-gue that these simple processes can account for arange of phenomena related to WM (e.g., ‘‘exec-utive’’ processes, such as inhibition). In Chapter8, Hale et al. review a long line of research on thedevelopment of cognitive abilities in childrenand adolescents. According to their perspective,mental processing speed accounts for a large por-tion of developmental variance in WM, and thedevelopment of more efficient (i.e., faster) pro-cesses in turn results in greater capacity.

Working Memory Variation Due to

Normal and Pathological Aging

The third part covers variation due to aging. InChapter 9, Hasher et al. present an ‘‘inhibitory’’view of cognitive aging and WM. Accordingto their perspective, performance on tasks de-signed to measure WM is largely influenced by

an individual’s ability to cope with the effects of proactive interference. Hasher et al. argue thatindividuals differ in a general inhibitory abilityto cope with interference. This ability accounts

for a large portion of developmental and intra-individual variance, and is a primary source of WM variation in general. In Chapter 10, Reu-ter-Lorenz and Jonides emphasize the role of the executive processes in all measures of im-mediate memory and argue that the distinctionbetween measures of storage only and stor-ageþ processing is overstated. They take amultifaceted approach to the sources of varia-tion in WM associated with age and argue that

several factors are at play. Specifically, they il-lustrate how the elderly may allocate resourcesdifferently than younger adults as they performtests of WM. In Chapter 11, Caplan et al. reviewtheir work on the relation between WM capacityand language comprehension. Based on evi-dence of a disconnect between traditionalmeasures of WM and critical aspects of com-prehension, such as the assignment of syntacticstructure, Caplan et al. argue against a gen-

eral factor in WM (cf., Kane et al., Chapter 2;Oberauer et al., Chapter 3). Instead, they con-sider the role that long-term memory plays inthe performance of WM and suggest skill andexperience as potential sources of variation inWM (cf., Ericsson & Delaney, 1999).

In summary, the ability to process new informa-tion while simultaneously holding onto theprevious results of processing is essential in manydomains of higher-level cognition, such as lan-guage comprehension and production, mentalarithmetic, spatial thinking, complex reasoning,and problem solving. The cognitive mechanismthat supports this essential human capability isWM. Reflecting its important role in humancognition, WM has become a central topic incognitivepsychology,cognitivescience,andcog-nitive neuroscience. In this volume we explore

variation in WM, broadly construed to includeindividual differences in healthy adults, age-related changes due to normal development andaging, and the effects of cognitive pathologies.We hope that this diversified approach, whichunites the experimental and differential ap-proaches to psychology and uses the philosophy




of converging operations, will result in a morecomprehensive account of variation in WM.

Notes1. The span of apprehension refers to the number of

simultaneously presented stimuli, as opposed to

sequentially presented stimuli, that a person can

mentally grasp.

2. At least this seems to be the first use of the phrase

‘‘working memory.’’ A similar idea was expressed

earlier by Johnson (1955):

Whatever the items of the problem to be or-

ganized, manipulated, or otherwise dealtwith, there is a limit to the number of separateitems that can be thus grasped, retained, andmanipulated, and that limit is the span of im-mediate memory. If remote memory is thestorehouse of ideas, immediate memory is theworkshop wherein ideas are processed. (p. 82)

3. Indeed, 4 years earlier, Miller (1956) published his

famous Psychological Review report on memory

capacity, and all three of these authors were largely

influenced by James and Woodworth, amongothers.

References

Atkinson, R. C., & Shiffrin, R. M. (1971). The

control of short-term memory. Scientific Amer-

ican, 225, 82–90.

Baddeley, A. D. (1986). Working memory. New York:

Oxford University Press.

Baddeley, A. D., & Hitch, G. (1974). Working



and theory (Vol. 8, pp. 47–89). New York: Aca-

demic Press.

Baddeley, A. D., & Warrington, E. K. (1970). Amnesia

and the distinction between long- and short-term

memory. Journal of Verbal Learning and Verbal

Behavior, 9, 176–189.

Baldwin, J. M. (1894). Mental development in thechild and the race. New York: Macmillan.

Bennett, F. (1916). The correlations between differ-

ent memories. Journal of Experimental Psychol-

ogy, 1, 404–418.

Bindra, D., & Scheier, I. H. (1954). The relation

between psychometric and experimental re-

search in psychology. American Psychologist, 9,

69–71.

Binet, A., & Simon, T. (1905). Methodes nouvelles

pour le diagnostic du niveau intellectuel des

anormaux. L’Anne e Psychologique, 11, 191–244.Blankenship, A. B. (1938). Memory span: A review of

the literature. Psychological Bulletin, 35, 1–25.

Bolten, T. L. (1892). The growth of memory in

school children. American Journal of Psychol-

ogy, 4, 362–380.

Broadbent, D. E. (1958). Perception and communi-

cation. New York: Pergamon Press.

Brown, J. (1958). Some tests of the decay theory of

immediate memory. Quarterly Journal of Ex-

perimental Psychology, 10, 12–21.Burt, C. (1909). Experimental tests of general intel-

ligence. British Journal of Psychology, 3, 94–

177.

Case, R. (1978). Intellectual development from

birth to adulthood: A neo-Piagetian interpreta-

tion. In R. S. Siegler (Ed.), Children’s think-

ing: What develops? (pp. 37–71). Hillsdale, NJ:

Lawrence Erlbaum Associates.

Case, R. (1985). Intellectual development: Birth to

adulthood. New York: Academic Press.Case, R., Kurland, M. D., & Goldberg, J. (1982).

Operational efficiency and the growth of short-



Cattell, J. M., & Galton, F. (1890). Mental tests and

measurements. Mind, 15, 373–381.

Conrad, R. (1964). Acoustic confusion in immediate

memory. British Journal of Psychology, 55, 75–84.

Conrad, R., & Hull, A. J. (1964). Information,

acoustic information and memory span. British

Journal of Psychology, 55, 429–432.

Cowan, N. (1988). Evolving conceptions of memory

storage,selectiveattention,andtheirmutualcon-

straintswithin the human information processing

system. Psychological Bulletin, 104, 163–191.

Cowan, N. (1995). Attention and memory: An inte-

grated framework. Oxford, UK: Oxford Univer-

sity Press.

Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental

storage capacity. Behavioral and Brain Sciences,

24, 87–185.

Cronbach, L. J., (1957). The two disciplines of

scientific psychology. American Psychologist, 12,

671–684.




Crowder, R. G. (1982). The demise of short-term

memory. Acta Psychologica, 50, 291–323.

Daneman, M., & Carpenter, P. A. (1980). Individ-

ual differences in working memory and read-

ing. Journal of Verbal Learning and Verbal Be-havior, 19, 450–466.

Ebbinghaus, H. (1885/1964). Memory: A contribution

to experimental psychology. New York: Dover

Publications, Inc.

Ericsson, K. A., & Delaney, P. (1999). Long-term

working memory as an alternative to capacity

models of working memory in everyday skilled

performance. In A. Miyake & P. Shah (Eds.),

Models of working memory: Mechanisms of active

maintenance and executive control (pp. 257–297). New York: Cambridge University Press.

Fiez, J. A. (2001). Bridging the gap between neuro-

imaging and neuropsychology: Challenges and

potential benefits. Journal of Clinical and Exper-

imental Neuropsychology, 23, 19–31.

Galton, F. (1887). Supplementary notes on ‘‘pre-

hension’’ in idiots. Mind, 12, 79–82.

Garner, W. R., Hake, H. W., & Eriksen, C. W.

(1956). Operationism and the concept of per-

ception. Psychological Review, 63, 149–159.Hebb, D. O. (1949). Organization of behavior . New

York: Wiley.

Humpstone, H. J. (1917). Some aspects of the memory

span test: A study in associability. Philadelphia:

Psychological Clinic Press.

Humpstone, H. J. (1919). Memory span tasks. The

Psychological Clinic, 12, 196–200.

Jacobs, J. (1885). Ueber das Gedachtnis [Review of

Ebbinghaus (1885) Memory: A contribution to

experimental psychology]. Mind, 10, 454–459.

Jacobs, J. (1886). The need of a society for exper-

imental psychology. Mind, 11, 49–54.

Jacobs, J. (1887). Experiments on ‘‘prehension.’’

Mind, 12, 75–79.

James, W. (1890). The principles of psychology. New

York: Henry Holt.

Johnson, D. M. (1955). The psychology of thought

and judgment. New York: Harper & Brothers.

Kane, M. J., Hambrick, D. Z., Tuholski, S. W.,Wilhelm,O.,Payne,T.W.,&Engle,R.W.(2004).

The generality of working memory capacity:

A latent-variable approach to verbal and visuo-

spatial memory span and reasoning. Journal of

Experimental Psychology: General, 133, 189–217.

Keppel, G., & Underwood, B. J. (1962). Proactive

inhibition in short-term retention of single items.

Journal of Verbal Learning and Verbal Behav-

ior, 1, 153–161.

Kintsch, W., Healy, A. F., Hegarty, M., Pennington,B. F., & Salthouse, T. A. (1999). Models of work-

ing memory: Eight questions and some general

issues. In A. Miyake & P. Shah (Eds.), Models of

working memory: Mechanisms of active mainte-

nance and executive control (pp. 412–441).

New York: Cambridge University Press.

Kosslyn, S. M., Cacioppo, J. P., Davidson, R. J.,

Hugdahl, K., Lovallo, W. R., Spiegel, D., &

Rose, R. (2002). Bridging psychology and biol-

ogy: The analysis of individuals in groups. Amer-ican Psychologist, 57, 341–351.

McGeoch, J. A. (1932). Forgetting and the law of

disuse. Psychological Review, 39, 352–370.

Melton A. W. (1963). Implications of short-term mem-

ory for a general theory of memory. Journal of

Verbal Learning and Verbal Behavior, 2, 1–21.

Miller, G. A. (1956). The magical number seven,

plus or minus two: Some limits on our capacity

for processing information. Psychological Re-

view, 63, 81–97.Miller, G. A., Galanter, E., & Pribram, K. H. (1960).

Plans and the structure of behavior . New York:

Holt.

Milner, B. (1966). Amnesia following operation on

the temporal lobes. In C. W. M. Whitty & O. L.

Zangwill (Eds.), Amnesia. London: Butterworths.

Miyake, A. (2001). Individual differences in working

memory: Introductionto the special section.Jour-

nal of Experimental Psychology: General, 130,

163–168.

Miyake, A., & Shah, P. (1999a). Models of working

memory: Mechanisms of active maintenance and

executive control. New York: Cambridge Uni-

versity Press.

Miyake, A., & Shah, P. (1999b). Toward unified

theories of working memory: Emerging general

consensus, unresolved theoretical issues, and fu-

ture research directions. In A. Miyake & P. Shah

(Eds.), Models of working memory: Mechanismsof active maintenance and executive control (pp.

442–482).NewYork:CambridgeUniversityPress.

Nairne, J. S. (2002). Remembering over the short-

term: The case against the standard model. An-

nual Review of Psychology, 53, 53–81.




Pascal-Leone, J. (1970). A mathematical model for

the transition rule in Piaget’s developmental

stages. Acta Psychologica, 32, 301–345.

Peterson, L. R., & Peterson, M. J. (1959). Short-term

retention of individual verbal items. Journal of Experimental Psychology, 58, 193–198.

Reitman, J. (1971). Mechanisms of forgettingin short-

term memory. Cognitive Psychology, 2, 185–195.

Reitman, J. (1974). Without surreptitious rehearsal,

information in short-term memory decays. Jour-

nal of Verbal Learning and Verbal Behavior, 13,

365–377.

Shah, P., & Miyake, A. (1996). The separability of

working memory resources for spatial thinking

and language processing: An individual differ-ences approach. Journal of Experimental Psy-

chology: General, 125, 4–27.

Shallice, T., & Warrington, E. K. (1970). Indepen-

dent functioning of verbal memory stores: A

neuropsychological study. Quarterly Journal of

Experimental Psychology, 22, 261–273.

Sternberg, S. (1966). High-speed scanning in hu-

man memory. Science, 153, 652–654.

Sternberg, S. (1969). The discovery of processing

stages: Extensions of Donders’ method. Acta Psy-chologica, 30, 276–315.

Sternberg, R. J., & Grigorenko, E. L. (2001). Unified

psychology. American Psychologist, 56, 1069–

1079.

Thorndike, E. L. (1914). The psychology of learning.

New York: Teachers College.Turner, M. L., & Engle, R. W. (1989). Is working

memory capacity task dependent? Journal of


Underwood, B. J. (1975). Individual differences as a

crucible in theory construction. American Psy-

chologist, 30, 128–134.

Waugh, N. C., & Norman, D. A. (1965). Primary

memory. Psychological Review, 72, 89–104.

Warrington, E. K., Logue, V., & Pratt, R. T. C.

(1971). The anatomical localisation of selectiveimpairment of auditory verbal short-term mem-

ory. Neuropsychologia, 9, 377–387.

Wickelgren, W. A. (1966). Distinctive features and

errors in short-term memory for English conso-

nants. Journal of the Acoustical Society of Amer-

ica, 39, 388–398.

Woodworth, R. S. (1938). Experimental psychology.

New York: Holt.

Wundt, W. (1912/1973). An introduction to psychol-

ogy. Originally published by G. Allen, London.Reprinted by Arno Press, New York.




I

Working Memory Variation

Reflecting Normal Inter- and

Intra-individual Differences



2

Variation in Working Memory Capacity

as Variation in Executive Attention

and Control

MICHAEL J. KANE, ANDREW R. A. CONWAY,

DAVID Z. HAMBRICK, and RANDALL W. ENGLE

If, as many psychologists seem to believe, im-mediate memory represents a distinct system orset of processes from long-term memory (LTM),then what might it be for? This fundamental,functional question was surprisingly unanswer-able in the 1970s, given the volume of researchthat had explored short-term memory (STM),and given the ostensible role that STM wasthought to play in cognitive control (Atkinson &Shiffrin, 1971). Indeed, failed attempts to linkSTM to complex cognitive functions, such asreading comprehension, loomed large in Crow-der’s (1982) obituary for the concept.

Baddeley and Hitch (1974) tried to validateimmediate memory’s functions by testing sub-jects in reasoning, comprehension, and list-learning tasks at the same time their memory

was occupied by irrelevant material. Generally,small memory loads (i.e., three or fewer items)were retained with virtually no effect on theprimary tasks, whereas memory loads of sixitems consistently impaired reasoning, compre-hension, and learning. Baddeley and Hitchtherefore argued that ‘‘working memory’’ (WM)

is a flexible and limited-resource system withstorageand processingcapabilitiesthat aretradedoff as needed. In this system, small memoryloads are handled alone by a peripheral pho-nemic buffer, leaving central processing unaf-fected, whereas larger loads require additionalresources of a central executive. Thus, WMwas proposed to be a dynamic system that en-abled active maintenance of task-relevant in-formation in support of the simultaneous exe-cution of complex cognitive tasks.

As we will detail below, there are certainlyaspects of our theoretical perspective that can betraced to Baddeley and Hitch’s (1974) views. Butour approach to conducting WM research is alsostrongly influenced by another article that ap-peared at about the same time, entitled ‘‘Indivi-

dual differences as a crucible in theory construc-tion.’’ In this report, Underwood (1975) arguedthat psychological theories should be subjectedpromptly to an individual-differences test as ameans of falsification. Most nomothetic theoriesin psychology make predictions about individualdifferences, even if only implicitly, and so testing

21



these predictions is an efficient means to deter-mine whether a theory merits further pursuit.

Although Baddeley and Hitch (1974) formulatedandpursuedWMtheorybasedonexperiment,the

question of WM function obviously lent itself toindividual-differences predictions. Quite simply,if WM were a central mechanism to higher-ordercognition, then individuals with greater WMcapacity should perform better on complex cog-nitive tasks than those with lesser WM capacity.

These important predictions became test-able a half-decade later, when Daneman andCarpenter (1980) created the ‘‘complex span’’tasks that initiated an individual-differences

approach to WM research. These span measureswere dual tasks, requiring information storagein the context of simultaneous processing of other information. They therefore reflectedBaddeley and Hitch’s (1974) idea that the ex-ecutive component of WM must be measuredin a dual processing and storage context. Mostimportantly, scores on complex span tasks cor-related strongly with measures of languagecomprehension, and this provided important

validation for WM theory. Indeed, subsequentindividual-differences research has led the wayin fulfilling the theory’s greatest promise—toelucidate the function of immediate memory—by linking variation in WM to diverse aspectsof higher-order cognition, including languagelearning (e.g., Baddeley, Gathercole & Papagno,1998), comprehension (e.g., Daneman & Meri-kle, 1996), reasoning (e.g., Kyllonen & Christal,1990), and cognitive control (e.g., Miyake,Friedman, Rettinger, Shah, & Hegarty, 2001).These correlational findings have indicated thatWM plays an important role in a host of complexcognitivecapabilitiesandthatWMmeasureshavepractical value in assessing intellectual ability.

However, the magnitude and breadth of the correlations between WM span and othercognitive measures do not necessarily illuminatethe psychological source of those correlations.

We suggest that individual-differences researchwill have its greatest impact on basic WM the-ory only when it pursues questions of mecha-nism simultaneously with questions of function.Our research program has therefore addressedboth mechanism and function, in the spirit of Cronbach’s (1957) call to align the ‘‘two disci-

plines of scientific psychology’’ and his argumentthat scientific psychology should aim to under-stand individual minds as well as the general,nomothetic principles of mind. To do so, we

use both experimental and correlational meth-odologies and examine individual-by-treatmentinteractions. The central question that drives ourresearch, then, which is unapologetically tiedto individual differences, has clear ramificationsfor general WM theory: Why do WM capacity(WMC) measures so successfully predict perfor-mance across a range of cognitive abilities?

OVERVIEW OF AN ‘‘EXECUTIVEATTENTION’’ THEORY OFWORKING MEMORY CAPACITY

Our approach to understanding WMC and itsvariation emphasizes the synergy of ‘‘atten-tional’’ and ‘‘memorial’’ processes in maintain-ing and recovering access to information that isrelevant to ongoing tasks and in blocking accessto task-irrelevant information (e.g., Engle &

Kane, 2004; Engle, Kane, & Tuholski, 1999;Kane & Engle, 2002; Kane, Hambrick, & Con-way, 2005). Our theory, which follows in partfrom Cowan (1995), is depicted in Fig-ure 2.1. We view STM as a metaphorical‘‘store’’ represented by LTM traces activatedabove threshold. These traces may be main-tained in the limited focus of attention (con-scious awareness) or kept active and accessiblethrough domain-specific rehearsal and codingprocesses (e.g., inner speech, chunking, imag-ery). Domain-general executive attention pro-cesses may also be engaged to sustain activationof information beyond attentional focus, or toretrieve no-longer active information from out-side of conscious focus. These executive pro-cesses will be particularly useful when rehearsalor coding routines are relatively unpracticed ornot useful in a particular context (e.g., with

novel visuospatial materials, or in dual-task sit-uations). These same executive attention mech-anisms may also be deployed to block or inhibitgoal-irrelevant representations or responses eli-cited by the environment.

Wepropose that the extent towhich executiveattention is engaged by a task, for maintenance,

22 Variation from Inter- and Intra-individual Differences



retrieval, or for blocking, is critically deter-mined by the degree of interference or conflictpresented by the context. Proactive interfer-ence from prior events may, for example, slowthe search for one’s car in a familiar parkinglot. Or, the environment may induce compe-tition between habitual responses and morenovel ones when the context is ambiguous orunusual, such as when an American drives onthe wrong side of the road in Dublin. Our viewis that the presence of such interference orconflict makes the executive functions of WMmost helpful and readily measurable (Norman& Shallice, 1986). Thus, when we use the term

working memory capacity, we refer to the at-tentional processes that allow for goal-directedbehavior by maintaining relevant informationin an active, easily accessible state outside of conscious focus, or to retrieve that informationfrom inactive memory, under conditions of interference, distraction, or conflict.

Working Memory Capacity,Executive Attention, and WorkingMemory Span Tasks

Our perspective is closely tied to the complexspantasks wehaveusedtomeasure WMC,whichshow good reliability by internal-consistencyand test-retest measures (e.g.,Klein & Fiss, 1999;Turner & Engle, 1989; but see Chapter 11, thisvolume).1 These WM span tasks present subjectswith the traditional memory span demand to im-mediately recall short lists of unrelated stimuli.

Additionally, and critically, WM span tasks chal-lenge memory maintenance by presenting a sec-

ondary processing task in alternation with eachmemory item. Reading span (Daneman & Car-penter, 1980), for example, requires subjects toread series of sentences for comprehension andthen recall the sentence-ending words from theseries (or sometimes, to recall an isolated word orletter that followed each sentence); operation

Magnitude of this link is determined bythe extent to which the procedures forachieving and maintaining activation areroutinized or attention demanding. Thatis, it is assumed that in intelligent, well-educated adults, coding and rehearsal ina digit span task would be less attentiondemanding than in 4-year-old children

Grouping/chunking skills, codingstrategies, and rehearsal proceduresfor maintaining activation/accesswithin and outside of consciousness

a. Could be phonological, visual, spatial,motoric, auditory, etc.

b. More or less attention demanding dependingon the task, subject, and context

(working memory capacity, executive attention, controlledattention, supervisory attention system, anterior attention

system, common variance among diverse WM tasks, etc.)

a. Most important under conditions of interference or conflict

b. Achieve (and re-achieve) activation and conscious access viacontrolled retrieval

c. Maintain activation or access of stimulus representations, goalabstractions, or response productions either within or

outside of conscious focus

d. Block interfering or conflicting thoughts and actions viainhibition

a. Traces or representations active above threshold,with loss due to decay or interference

b. Some receive further activation by becomingfocus of attention/conscious awareness

c. Trace consists of a pointer to a region of LTM.Thus, the activated trace could be as simple as“look away from the flash” or as vast as the gistfor Crime and Punishment

Central Executive

Short-Term Memory (STM)

Any given WMC or STM task reflects all components to some extent

Long-Term Memory (LTM)

Figure 2.1. Measurement model of the working memory system, version 1.2. (Adapted from Engle,Kane et al., 1999 [version 1.0], and Engle & Kane, 2004 [version 1.1].)

Variation in Executive Attention and Control 23



span, in contrast, presents subjects with series of equations to verify, with each equation followedby an unrelated word to memorize (Turner &Engle, 1989). Less verbal tasks include count-

ing span, which presents series of arrays in whichto-be-counted target items are surrounded bydistractors and subjects must recall the countfrom each array in the series (Case, Kurland, &Goldberg, 1982), and spatial (rotation) span,which presents series of rotated letters that sub-jects judge to be normal or reversed while mem-orizing the letters’ original orientations (e.g.,Shah & Miyake, 1996).

Working memory span tasks are obviously

complex and multiply determined tasks, and sonone of them can be considered a process-puremeasure of ‘‘executive function.’’ Instead, WMspan tasks measure, in part, executive attentionprocesses that we believe are domain generaland contribute to WM span performance irre-spective of the skills or stimuli involved. In ad-dition, WM span tasks reflect the contributionsof rehearsal, coding, storage, processing skills,and strategies that are domain specific and vary

with the component tasks and stimuli presented(seealsoChapters5and6).OurviewisthatWMspan tasks reflect primarily general executiveprocesses and secondarily, domain-specific re-hearsal and storage processes. Moreover, thebroad predictive utility of WM span tasks derivesfrom the general, executive attention contribu-tions to performance. Short-term memory spantasks, in contrast, reflect domain-specific storageand rehearsal skills and strategies primarily andexecutive attention processes only secondarily.That said, we should emphasize that we do notclaim that STM tasks are pure measures of storage and rehearsal, without any influence of attention processes; nor do we claim that WMspan tasks measure or correlate with all possibleaspects of attentional processing. Instead, wethink that WM span tasks are reasonably goodmeasures of a domain-general attentional capa-

bility that is involved in the control of behaviorand thought and is important to many cognitiveabilities. Thus WM span tasks are generallybetter measures of the executive attention con-struct than STM span tasks (see Kane et al.,2005).

Working memory span tasks tap into exec-utive attention by requiring subjects to maintainor recover access to target information underproactive interference from prior trials (e.g.,

Lustig, May, & Hasher, 2001), while that ac-cess or retrieval is challenged by intermittentlyshifting attentional focus between the memoryand secondary processing tasks (e.g., Barrouillet,Bernadin, & Camos, 2004; Hitch, Towse, &Hutton, 2001). That is, interference encouragessubjects to rely on sustained, active access tothe memoranda, rather than on LTM retrieval,but subjects cannot easily maintain that accessbecause the processing task prevents them from

keeping target items in the focus of attention(the processing task also limits use of rehearsalor chunking strategies). Executive processes thushelp maintain or recover access to the targetitems in the absence of focal attention and ef-fective rehearsal procedures.

EXECUTIVE ATTENTION ASTHE CRITICAL SOURCE OF WORKING

MEMORY CAPACITY VARIATION

Our proposal, that WMC variation is drivenlargely by individual differences in executive at-tention processes, represents a web of inferenceacross correlational and experimental studies.Some of these studies, which we have describedas ‘‘macroanalytic’’ (Engle & Kane, 2004; seeSalthouse & Craik, 2001), have examined therelations between WMC and other hypotheticalconstructs, such as general fluid intelligence,using large subject samples and multiple tasks toidentify each construct. Two other kinds of studies, which we term ‘‘microanalytic,’’ take amore focused approach to analyzing span–abilityrelations. One line of microanalytic work com-bines correlational and experimental designs bymanipulating variables within WM span tasksto determine how those manipulations affect

the span–ability correlation. These are essen-tially task analyses of WM span that consider notonly the processes required by span tasks but alsothe processes shared between span and othermeasures. The second line of microanalytic re-search, using quasi-experimental designs, tests




for WM span–related differences by comparingindividuals with high WM span scores (highspans, from the upper quartile of a universitystudent distribution) to those with low scores

(low spans, from the lower quartile) in the per-formance of ‘‘elementary’’ cognitive tasks fromthe memory and attention literatures. We discussthese three sets of macro- and microanalyticfindings in detail below.

Macroanalytic Studies of WorkingMemory Capacity

The use of large-scale, structural equation mod-eling studies in WMC research has increasedrecently, influenced by the growing confluenceof the WMC and intelligence literatures (forreviews, see Conway, Kane, & Engle, 2003;Kane et al., 2005). An advantage of these tech-niques is that they permit the use of latentvariables, which reflect the shared varianceamong a number of tasks hypothesized to reflectthe same construct (e.g., WMC). As such, latent

variables are free from the measurement errorassociated with any one multiply determinedtask. Through use of latent variables and struc-tural equation modeling, research conclusionscan be shifted from the level of observed vari-ables, which always reflect some measurementerror, to the theoretical constructs of interest.

The Relation of Working Memory

Capacity to Short-Term Memory and Fluid Intelligence

Engle, Tuholski, Laughlin, and Conway (1999)tested 135 university subjects in WMC tasks(operation span, reading span, counting span),STM tasks (backward and forward word span),and tests of fluid intelligence, or psychometricGf (novel figural and spatial reasoning). Ourquestions for this study were whether WMC

and STM were dissociable constructs and, if so,whether WMC was the better predictor of Gf.In fact, WMC and STM were separable: thetwo latent variables correlated substantially(.68), but a model forcing a single factor ontothe span data did not fit well. We believe that

the correlation between STM and WMCwas driven primarily by the shared requirementamong span tasks to immediately recall shortlists of verbal items—that is, it reflected pri-

marily ‘‘storage’’ (although some shared vari-ance between STM and WMC will also reflectexecutive attention). The unique residual vari-ance in WMC reflected the dual-task demandin the WMC tasks only, that is, the increaseddemand they made on executive attention foractive maintenance outside of conscious focus.

Given that WMC and STM reflected bothshared and unique variance, which might con-tribute to general intellectual ability? We found

that the WMC factor, but not STM, predictedunique variance in Gf, suggesting that thegreater executive demands of WM span tasksare the source of WMC–Gf correlations, ratherthan the ‘‘simple’’ storage demands shared bySTM and WM tasks. However, as a further testof this idea, we constructed a hierarchical modelof our span data (illustrated in Fig. 2.2). Here,separate factors were derived for WMC andSTM, but in addition, the considerable variance

shared between WMC and STM was modeledas a second-order factor. This second-order‘‘common’’ factor ostensibly represented thestorage, coding, and rehearsal (and some exec-utive) processes involved in both WMC andSTM tasks, and it shared significant, uniquevariance with Gf. However, the residual, uniquevariance from WMC (i.e., WMC with STMfactored out) predicted Gf more strongly. What-ever WM span tasks demand beyond simplestorage seems to account primarily for the WM–Gf correlation. We interpret this residual WMCvariance to reflect the relatively strong executiveattention demands of WM span tasks, elicitedby their dual-task requirements.

Processing Speed and Working Memory Capacity–Gf Association

Conway, Cowan, Bunting, Therriault, andMinkoff (2002) replicated these findings whileadditionally testing the contribution of pro-cessing speed to predicting Gf and accountingfor the WMC–Gf correlation. Developmentalresearch clearly shows that speeded measures of




simple cognitive processes often account for alion’s share of age-related variance in higher-

order cognition, overwhelming the contribu-tion of WMC (Kail & Salthouse, 1994). Muchless clear, however, is whether processing-speedvariation within an age group can account forthe relation between WMC and intelligence.To find out, we tested 113 university subjects inthe WM span and Gf tasks used by Engle, Tu-holski et al. (1999), along with several STM andspeed tasks. The latter were paper-and-penciltasks requiring subjects to copy or compare listsof stimuli quickly and accurately. Because cor-relations between processing speed and Gf mea-sures typically increase with the complexity of speeded tasks (e.g., Jensen, 1998), thus cloud-ing the interpretation of what ‘‘processingspeed’’ reflects, we chose simple speed tasks as amost stringent test of their importance.

In order to examine the independent con-tributions of executive attention and storage,

coding, and rehearsal to the association betweenWMC and Gf, we used a nested structure inwhich all the span tasks loaded onto a common‘‘STM–storage’’ factor to represent their sharedstorage, coding, and rehearsal variance. WMCtasks also loaded onto a residual factor, reflectingthe additional executive attention processes en-

gaged by the dual-task nature of the WM spantasks. As shown in Figure 2.3, the shared ‘‘stor-

age’’ variance was a relatively weak predictor of Gf, and the residual WMC variance was stron-ger. These findings support the idea that ‘‘ex-ecutive’’ variance, tapped by WMC tasks to agreater degree than by STM tasks, drives theWMC–Gf relationship. It is also worth notinghere that not only did speed fail to predict Gf while controlling for WMC, but speed alsocorrelated more strongly with STM–storagethan WMC–executive processes. Among youngadults, then, with relatively ‘‘simple’’ tests of pro-cessing speed and with untimed measures of WMC, the two constructs share little variance,and only WMC is a significant source of vari-ation in general ability.

Domain Generality of Working Memory Capacity and Short-Term Memory

Our latent-variable research shows strong cor-relations between WMC and Gf, therefore sug-gesting WMC to be an important mecha-nism of general cognitive ability. Moreover, ourWMC latent variables were derived from ver-bal, symbolic WM span tasks and the Gf latent

operspan

readspan

counspan

ravens

cattellfword-a

WMC

STMfword-b

bword

Storage

.49*

.29*

.12

Gf

Figure 2.2. Structural-equation model, adapted from Engle, Tuholski et al. (1999). Circles representlatent variables and boxes represent individual tasks. Solid arrows with path coefficients with asterisksrepresent significant shared variance between constructs. Counspan¼Counting span task; Fword-a¼Forward Word span task, version a; Fword-b¼Forward Word span task, version b; bword¼Backwardword span task. Gf ¼ general fluid intelligence; Operspan¼Operation span task; Readspan¼Readingspan task; WMC¼working memory capacity; STM¼ short-term memory.




variables were created from nonverbal, figuralreasoning tasks. This cross-domain generalityindicates that the variance common to WMspan tasks cannot be substantially verbal. In-stead, we submit that the executive attentionvariance that is shared among WM span tasksreflects domain-general processes.

However, a few studies indicate domainspecificity in WM span, with low correlationsbetween individual verbal and visuospatialWM span tasks, low cross-domain correlationsbetween WM span and ability, or both (e.g.,Shah & Miyake, 1996). We suspect that re-stricted ranges of general ability and measure-ment error biased these studies toward findingexaggerated domain specificity in WMC. Allwere derived from samples of university stu-dents, some from prestigious schools, that might

represent a narrow range of general intellectualability relative to the population at large.Without suitable variation in general ability in asample, any variability in cognitive performancemustresultfromsomethingelse,suchasdomain-specific abilities, skills, or strategies. (Becauseuniversities are more likely to select students

from a narrow range of general ability thanfrom a narrow range of any one specific ability,samples drawn from these populations are morelikely to represent restricted ranges of generalability.) Moreover, with respect to measurementerror, these ‘‘domain-specific’’ studies used onlyone measure each of verbal and spatial WMC.With only one task per construct, the low cor-relations might result from dissociable con-structs (i.e., verbal and spatial WMC), or insteadfrom other, non-WMC abilities, skills, and pro-cesses tapped by complex span tasks.

To address these possibilities, Kane et al.(2004) tested 236 subjects from both competi-tive and comprehensive universities, as well asfrom two urban community samples, in mul-tiple tests of verbal and spatial WM and STMspan. We thus ensured some degree of varia-

tion in domain-general ability in our sampleand used latent-variable models to factor outsources of measurement error. We first con-trasted the fit of two kinds of models for theWM span data: unitary models derived fromall six WM span tasks and two-factor modelswith separate verbal and spatial WMC factors.

operspan

readspan

counspan

word1a

ravens

cattell

word1b

word2a

word2b

letter

pattern

digit

WMC

.18

.07

.40*

–.06

.60*

Storage Gf

PS

Figure 2.3. Structural-equation model, adapted from Conway et al. (2002). WMC¼workingmemory capacity; STM¼ short-term memory; PS¼ processing speed; Gf ¼ general fluid intelli-gence; Operspan¼Operation span task; Readspan¼Reading span task; Counspan¼Countingspan task.




Depending on the technical details of thesemodels, verbal and spatial WMC factors shared70%–85% of their variance (correlations from.84 to .93), demonstrating that WM span mea-

sures tap primarily general processes and abil-ities. In contrast, verbal and spatial STMmeasures shared only 40% of their variance,consistent with our view that complex WMspan tasks measure primarily the contributionof general executive processes and simple STMspan tasks measure primarily the contributionsof domain-specific ones. We also tested whetherthe domain generality of WMC and STMvaried with the range of Gf in the sample, by

dividing our subjects into different groups onthe basis of their matrix-reasoning perfor-mance: a high-Gf group, a low-Gf group, andtwo groups representing the full range of Gf.We found that both WMC and STM weremuch more domain specific in our high-Gf group than they were among our low-Gf sub-jects or our full Gf–range sample. Thus, as wesuspected, prior findings of domain-specific

WMC may have resulted from testing subjectsfrom a restricted range of high general ability.

Like Engle, Tuholski et al. (1999) and Con-way et al. (2002), we also tested whether the gen-

eral executive contributions to span, or domain-specific STM–storage contributions, were theprimary source of the WMC–Gf correlation. Inour nested-factor model, depicted on the leftside of Figure 2.4, all 12 of the verbal andspatial WMC and STM tasks loaded onto a com-mon factor reflecting their shared variance. Ad-ditionally, the six verbal and spatial span taskseach loaded onto a domain-specific residual fac-tor.Thelogic,again,wasthatbothWMandSTM

span tasks reflect joint contributions of generalexecutive attention and domain-specific STMand storage. At the same time, WM span tapsprimarily general executive attention processesand STM span taps primarily domain-specificstorage,coding,andrehearsalprocesses.However,in contrast to the Engle and Conway models,notethatweinterpretedthecommonfactortorep-resent general executive attention variance and

wordspan

lettspan

digtspan

operspan

readspan

counspan

navispan

symspan

rotaspan

ballspan

arrospan

matxspan

inference

analogy

readcomp

remoassoc

syllogism

ravens

wasi

beta3

spacerela

rotablock

surfdevel

formbord

paperfold

.40*

.16*

.29*

.52*

.25*

.54*

.49*

Storage

Verbal

Reasoning

Verbal

Storage

Spatial

Reasoning

Spatial

Executive

AttentionGf

Figure 2.4. Structural-equation model, adapted from Kane et al. (2004). Gf ¼ general fluid in-telligence. For descriptions of the individual span and reasoning tasks, see Kane et al. (2004).




the residual factors to represent domain-specificSTM–storage variance, rather than the reverse.

Our interpretation of these factors had ratio-nal and empirical grounding. First, given the

substantial dissociability between verbal andspatial STM span in our data and in others’ (seeJonides et al., 1996, for a review), it is unlikelythat shared variance among verbal and spatialspan measures reflected domain-specific storageand rehearsal abilities. Second, the WMC andSTM tasks loaded differently onto the ‘‘execu-tive’’ and ‘‘STM–storage’’ factors. The WMCtasks all had higher loadings on the common‘‘executive’’ factor than did the STM tasks, and

they had higher loadings on the executive factorthan on their respective ‘‘storage’’ factors. TheSTM tasks showed the opposite pattern. Thus,the executive attention factor captured morevariance from WMC than STM tasks, and thestorage factors showed the opposite pattern. Inthis data set, then, the common span variance re-flected domain-general executive attention andthe domain-specific residual variance reflectedSTM storage and rehearsal processes. Of most

importance, the executive attention factorstrongly predicted Gf (path coefficient¼ .52),with a similar magnitude to our previous stud-ies. Moreover, it was similar to the correlationwe found in a separate model where Gf waspredicted by a WMC factor derived from onlythe six WM span tasks (.64). This consistencyacross models and studies is compelling, giventhat we defined WMC and Gf much morebroadly here than in our prior work, with Gf reflecting shared variance among verbal andvisuospatial reasoning tests. Indeed, in a sepa-rate article we reanalyzed all the published la-tent variable data on the WMC–Gf correlationand found that WMC accounted for approxi-mately 50% of the variance in Gf (i.e., the me-dian correlation between WMC and Gf con-structs across studies was .72; Kane et al., 2005).This shared variance is not strong enough to

claim that WMC and Gf are synonymous. How-ever, we submit that WMC, which reflects pri-marily a general executive construct, is onecritical source of Gf variation. Short-term mem-ory, in contrast, reflects more domain-specificstorage and rehearsal processes that are lessimportant to general aspects of ability.2

Summary of Macroanalytic Research

As measured by a variety of span tasks, WMCand STM are strongly correlated constructs.However, despite this close relationship, the

attentional processes engaged primarily by WMspan tasks (and to a lesser extent by STM tasks)are responsible for WM span’s general and su-perior predictive utility. The executive atten-tion processes that contribute to WM span tasksare an important mechanism of fluid intelli-gence, and furthermore, these executive atten-tion processes are domain general. In contrast,variance associated with simple storage and re-hearsal activities, captured primarily by STMtasks, is relatively domain specific (for alterna-tive views on WMC’s domain generality, seeChapters 6 and 8).

Microanalytic Studies of WorkingMemory Capacity

Two lines of ‘‘microanalytic’’ research, usingsmaller scale quasi-experiments and regression-or ANOVA-based analytic approaches, have ad-dressed the mechanisms of span task perfor-mance and its relation to complex cognitiveabilities. One line has clarified which processesare not important to WMC variation and co-variation. We discuss this work first. We thenreview research that more closely and specifi-cally links WMC to the constructs of attentionand executive control.

Ruling out Some Mechanismsof Variation and Covariationin Working Memory Capacity

Task skill and processing efficiency The ideathat individual differences in task-specific skillsaffect the correlations between WM span andhigher-order cognitive measures is an old one.

Daneman, Carpenter, and colleagues (e.g.,Daneman & Carpenter 1980) proposed thatgood comprehenders could devote fewer WMresources to the reading and listening compo-nent of the span task than could poor compre-henders, thereby relieving more resources forthe simultaneous task of memory storage. Thus,




strong language skills lead to a larger functionalWMC for language, rather than a larger WMCleading to stronger language skills. In a similarvein, MacDonald and Christiansen (2002)

claimed that ‘‘the reading span task is simply ameasure of language processing skills’’ (p. 39).However, our research (Conway & Engle, 1996;Engle, Cantor, & Carullo, 1992) demonstratesthat partialing out subjects’ processing speedduring span tasks does not diminish the corre-lation between WM span and verbal ability, nordoes tailoring the difficulty of the span taskprocessing demand to each subjects’ individualskill level. If partialing out processing speed and

matching processing skills do not reduce span–ability correlations, then the relationship mustreflect more than domain-specific processingskills.

Strategy use Given the complexity of WMspan tasks, individual differences in strategiesmight contribute to WM span scores, as well asto the general patterns of WMC covariationwith other constructs. In fact, when McNamaraand Scott (2001) trained subjects to use a se-

mantic ‘‘chaining’’ strategy during simple wordspan tasks, they found that it substantially in-creasedsubsequentreadingspan(WMC)scores.Moreover, subjects who initially reported us-ing a semantic strategy to remember the STMtask words had higher reading span and verbalScholastic Aptitude Test (VSAT) scores thanthose of subjects who reported using only re-hearsal or no strategies.

However, despite the effects of strategyuse on WM span scores overall, we believe thatMcNamara and Scott’s (2001) data actuallyargue against the importance of strategy vari-ables to WMC variation and covariation. First,standard deviations in WM span were gener-ally larger after training than before training,indicating that training increased, rather thanreduced, span variability. Second, initially high-strategic subjects benefited more from strategy

training than did initially low-strategic subjects,and so WMC seems to determine how effec-tively people learn and use demanding strate-gies. Third, higher WM span scores were as-sociated with higher VSAT and math SAT(MSAT) scores, but high strategy use was asso-

ciated only with higher VSAT and not withMSAT. Strategy use therefore cannot accountfor the general predictive power of reading span.Finally, McNamara and Scott did not report

what should be critical evidence concerning astrategy hypothesis—if individual differences instrategy use account for WM–ability correla-tions, then a span–ability correlation should beweaker after strategy instruction than before it.

The few studies that have directly testedwhether indices of strategy use might accountfor the covariation between WMC and com-plex cognition have been negative. In two dif-ferent WM span tasks, Engle et al. (1992)

allowed subjects to control their study time foreach to-be-recalled word. Those subjects withhigh spans studied the target words for moretime than those with low spans, but whenEngle et al. partialed out study time from thespan–VSAT correlations, they actually increasednonsignificantly. Thus, strategic allocation of study time did not drive covariation betweenWMC and verbal ability; if anything, strategyuse suppressed measurement of the relation-

ship (see also Friedman & Miyake, 2004). A series of training studies by Turley-Ames

and Whitfield (2003) also found strategy use tosuppress correlations between WMC and verbalability. Following a pretest on operation span,subjects were instructed to engage in rehearsal,imagery, or semantic chaining to help them re-member the target words. Instructed subjectsgreatly improved their span scores relative to un-instructed control subjects, but scores follow-ing strategy instruction were more strongly cor-related with verbal ability than were controlscores. That is, matching subjects in their knowl-edge and encouragement of effective WM strat-egies did not make the span–ability correlationgo away, as it should have if strategic differencesaccounted for the correlation. Instead, normalvariation in strategy use (as reflected by controlsubjects) actually worked against finding corre-

lations between WM span and verbal ability. Variation in strategy use during WM span tasks,even if substantial, does not account for sharedvariance between WMC and ability measures.

Summary Microanalytic studies have notfound the contributions of processing efficiency,




processing skill, or strategy use to be compel-ling; they either fail to account for performanceof WM span tasks or, if they do affect WM spanscores, they do not contribute to the correlations

between WM span and higher-order abilities.Something else drives the predictive power of WMC measures. We can rule out STM storageand rehearsal as well, because macroanalyticstudies find that STM tasks are not as goodpredictors of general ability as WM tasks, andthe demands made by WM span tasks, beyondSTM storage and rehearsal, drive their associ-ation with complex cognition.

We think the studies described in this section

serve as good examples of the importance of combining experimental and correlational ap-proaches to WM research. Experimental taskanalyses of WM span tasks may suggest anynumber of variables that serve to either increaseor decrease span scores. Such findings may beinteresting in their own right, but it is a mistaketo infer from them that any of these variablesmust contribute to the correlations betweenWM span and other tasks. At least some vari-

ables that affect span scores clearly have no in-fluence on WM span correlations.

Variation in Working Memory Capacity and Individual Differences inExecutive Attention

With the elimination of processing skill or effi-ciency, strategy use, and STM storage as causesof the WMC–ability relationship, we infer thatthe attentional demands made by WM spantasks are most important in producing that rela-tionship. Moreover, our second line of micro-analytic work, discussed below, provides directevidence for an association between WMC andexecutive attention capabilities. First we reviewevidence linking individual differences in WMCto the attentional control of interference inmemory. We then discuss evidence from rela-

tively low-level attention tasks, not involvingmemory retrieval, that WMC predicts controlover goal-directed behavior. Specifically, we findthat WMC is associated with successful main-tenance of task goals and the attendant block-ing of strong but contextually inappropriate

responses, mechanisms that may be analogous tothe ‘‘proactive’’ and ‘‘reactive’’ control modes, re-spectively, proposed by Braver et al. (Chapter 4).

Working memory capacity and executive at-

tention in resolving memory interference Ourexecutive attention theory holds that the controlof memory retrieval in the face of interference iscentral to the attentional construct measured byWM tasks. The supporting evidence—a strongconnection between WMC and interferencevulnerability—comes from a variety of experi-mental preparations that pose different typesof interference and competition (e.g., retroac-tive, fan-type, output). In these studies, extreme

(quartile) groups of high– and low–WM spansubjects, identified via operation span, differ inrecall accuracy or latency under high-interferencebut not low-interference conditions (e.g., Conway& Engle, 1994).

Proactive interference, for example, affectslow- more than high-WMC individuals. Kaneand Engle (2000) tested subjects in a delayedfree-recall task presenting three consecutive listsof 10 words each, with recall preceded by a 15 s

rehearsal prevention task. To maximize inter-ference, the words from all three lists belongedto one semantic category (e.g., animals). OnList 1, in the absence of interference, the twospan groups showed equivalent free recall.However, by List 3, under proactive interfer-ence from prior lists, low-span subjects’ recalldropped more precipitously than that of high-span subjects (Ms50% vs. 30%, respec-tively). Rosen and Engle (1998) demonstratedspan differences in interference susceptibilityduring learning of paired associates. In List 1,all subjects learned 12 compound word pairs(e.g., bird-bath). In List 2, control subjectslearned 12 pairs of semantically related wordsthat were unrelated to the List 1 pairs (e.g., eye-tear), while interference subjects learned pairsusing the List 1 cues (e.g., bird-dawn). High andlow spans reached learning criterion equally

quickly for List 1, where interference was absent.However, in List 2, low spans required an aver-age of two more learning trials than did highspans overall, and low spans’ impairment wasespecially evident in the interference condition.Low spans under interference also committed




more overt List 1 intrusions during List 2learning than did high spans.

These studies additionally yielded evidencethat WMC-related variation in interference

arises from attention-control variation, consis-tent with theories linking interference resistanceto attentional inhibition (e.g., Anderson &Neely, 1996; Hasher & Zacks, 1988). Mostdirectly, Kane and Engle (2000) tested somesubjects under dual-task conditions, in whichthey continuously maintained a complex finger-tapping sequence during either study or recall of each list. Under both these dual-task conditions,the span groups showed equivalent interference.

The secondary task had no effect on the low-span group’s interference vulnerability but itincreased the high-span group’s interferencevulnerability to that of the low-span group.These counterintuitive findings indicate thatspan differences in proactive interference nor-mally result from high spans’ superior use of controlled processes to combat it (e.g., via in-hibition, blocking, source monitoring, etc.).Thwarting high spans’ control by imposing a

secondary task increased their interference sus-ceptibility. In contrast, low span subjects wereless effective in engaging controlled processingto limit interference in the first place, so divid-ing their attention was irrelevant—they couldnot lose what they were not already using.

Rosen and Engle (1998) inferred a role forattention in interference differences more indi-rectly. Their subjects attempted to relearn List1 (e.g., bird-bath) after learning List 2 (e.g.,bird-dawn). The critical dependent measurehere was cued-recall latency following the veryfirst learning trial for each list. Note that if highspans under interference conditions had blockedtheir List 1 associations in learning List 2 via aninhibitory process, then these pairs should sub-sequently suffer some residually impaired acces-sibility. Accordingly, high span subjects underinterference should be slower to recall List 1

than they had been in originally learning them,and they should be slower to recall List 1 thancontrol subjects. Low span subjects, by contrast,should show no evidence of inaccessibility forList 1. This is precisely what was found. Highspans’ relearning latencies in the interference

condition were significantly longer than those inthe control condition. In contrast, low-span in-dividuals’ response times (RTs) were actuallyshorter in the interference than in the control

condition. Moreover, high spans’ relearning la-tencies were significantly slower than their ownList 1 learning latencies; low spans’ latencieswerestatisticallyequivalenttooneanother.Thus,only high spans showed something analogous toa negative priming effect in relearning a list thathad previously been a source of interference, andso was inhibited.

High- and low-span subjects recall informa-tion from LTM equivalently quickly and ac-

curately in the absence of interference. Thus,WM span does not predict variability in all as-pects of remembering. Instead, WMC appearsimportant for learning and retrieval only whenthe environment presents a substantial sourceof interference and competition. Moreover, ourdual-task and RT results strongly suggest thathigh spans’ relative invulnerability to interfer-ence stems from a superior executive attentioncapability, whereby potentially competing in-

formation is blocked or inhibited. Only highspans show an effect of divided attention ontheir interference susceptibility, and only highspans show impaired access to initially learnedbut subsequently interfering associations.

Working memory capacity and executive at-tention in resolving response competition. If WMC reflects an attentional construct, as weclaim, then WMC differences should be observ-able in contexts that make no explicit memory-retrieval demands. That is, variation in WMCshould be associated with variation not only inmemory-interference tasks but also in simplerindices of attention control. In fact, high spans(as measured by operation span) are lessvulnerable to salient distractors in dichotic lis-tening than are low spans. Conway, Cowan,and Bunting (2001) had subjects shadow a listof unrelated words presented to their right ear

while distractor words were presented to theirleft ear. Either 4 or 5 min into the task, thesubject’s name was presented in the distractorchannel. Prior research indicated that approxi-mately 33% of subjects report hearing theirname in such contexts (e.g., Wood & Cowan,




1995). Here, only 20% of the high span sub-jects, but 65% of the low span subjects, reportedhearing their name.

WMC-related differences also arise in the

arguably simpler ‘‘antisaccade’’ task of attentioncontrol. In two experiments, Kane, Bleckley,Conway, and Engle (2001) tested high and lowspans in prosaccadic and antisaccadic versionsof a letter-identification task. Each trial brieflydisplayed a letter to the right or left of fixationfor identification. In prosaccade trial blocks thetarget location was always cued by a flashingstimulus near its upcoming location, so sub-jects could allow reflexive orienting responses

to guide, or ‘‘pull,’’ their attention and eyes tothe target. Here, both high- and low-span sub-jects identified targets equivalently quickly. Incontrast, antisaccade trial blocks always cuedthe target by presenting the flash to the oppositescreen location, and so subjects had to block,or quickly recover from, the orienting responseto the flash and endogenously ‘‘push’’ their at-tention and eyes toward the target. In bothexperiments, high spans identified antisaccade

targets more quickly than did low spans. More-over, Experiment 2 measured eye movementsacross hundreds of antisaccade trials, and highspan subjects showed fewer saccades towardthe cue, faster recovery from these saccade er-rors, and faster correctly guided saccades thandid low span subjects.

Both the dichotic-listening and antisaccadetasks make few demands on subjects beyondblocking a habitual orienting response in theservice of a novel goal. How do subjects actu-ally succeed? We hypothesize that a criticalaspect of preventing elicited but inappropriateresponse tendencies from controlling behavioris to actively maintain access to the novel goal.That is, to successfully block a prepotent re-sponse, such as looking toward a flash, onemust keep this goal especially accessible. Al-though it may be trivial to recall from LTM the

rules of a task, the rules of decorum, or the lawsof the land, it is often quite a bit more chal-lenging to behave, in the moment, accordingto these rules. Our view is that active goalmaintenance and the resolution of responsecompetition are interdependent processes of

executive control, therefore, ‘‘memory’’ is animportant determinant of ‘‘attentional’’ behav-ior (see also Chapter 4, this volume; De Jong,2001; for an alternative view, see Butler, Zacks,

& Henderson, 1999).To explicitly test the idea that WMC may betied to the executive acts of goal maintenanceand competition resolution, Kane and Engle(2003) tested subjects with high and low spansin several versions of the Stroop color-word task.The Stroop task is a paradigmatic example of an executive attention task—a habitual, over-learned reading response must be held in checkto allow the novel color-naming goal to control

behavior. In order to manipulate the require-ment to actively maintain access to task goals,we varied the proportion of congruent trials inthe task. In high-congruency contexts, mosttrials presented words that matched their colors(e.g., RED appearing in red), so the task envi-ronment did not reinforce the goal of ignoringthe word. Because the automatically elicitedresponse to most stimuli was correct, it shouldhave been easy to slip into word reading rather

than color naming. Here, then, accurate re-sponding on the rare incongruent trials, whichpresented conflicting color words (BLUE ap-pearing in red), required that subjects main-tained adequate access—in the moment—tothe task goal. Failures of executive controlshould therefore be evident in accuracy.

In contrast, in Stroop contexts that pre-sented few congruent trials and mostly incon-gruent trials, the stimuli reinforced subjects’goal. When every trial demands that the wordbe ignored, it may be unnecessary to do themental work required to actively maintain goalaccess; the task environment acts as an external‘‘executive.’’ Just as Americans are helped todrive on the correct side of the road in Londonby road signs, traffic patterns, and the ergonom-ics of the car’s controls, so too may subjects bekept on the desired path to color naming by

a preponderance of incongruent Stroop trials.Under these circumstances, Stroop interfer-ence is unlikely to reflect goal maintenance toany great degree, and it is also unlikely to bereflected primarily in overt errors. Instead, in-terference in low-congruency contexts should




reflect primarily the effectiveness of the com-petition resolution processes carried out by theexternally cued goal, and should therefore beevident primarily in response latencies—that

is, in slow but correct responses.In fact, when 75% or 80% of the trials werecongruent, Kane and Engle (2003) found that lowspans had substantially larger error-interferenceeffects than did high spans. These effects, acrossfour samples in three experiments, indicate alow-span deficit in goal maintenance. Althoughlow-span subjects understood the goal of thetask, and in some experiments even receivedaccuracy feedback after every trial, they none-

theless often ‘‘zoned out’’ and made word-readingerrors on incongruent trials (low-span subjectsalso responded faster to congruent trials thanhigh-span subjects, a finding suggesting thatthey periodically read the words aloud on thesetrials). In contrast, when only 0% or 20% of the trials were congruent, we found modestspan effects in RT interference, requiring largesamples to reach statistical significance. WMC-related differences were not found in errors in-

dicative of goal neglect but rather in latencies,suggesting a slowed resolution of conflict be-tween elicited and desired responses.

Summary Evidence from dichotic listen-ing, antisaccade, and Stroop tasks converges tosuggest that WMC predicts action control indeceptively ‘‘simple’’ attention tasks. Low spansare less able than high spans to act according tonovel goals when that action conflicts with well-learned, if not reflexive, response tendencies.Our view is that executive attention processeslargely determine performance of both WMspan and these attention-control tasks, and soWM per se does not cause attention differences.Instead, a third variable, representing a low-level executive attention capability, influencesfunctioning on all of these selective-attention,WM-span, and memory-retrieval tasks (and,presumably, on indices of Gf as well). More-

over, this executive attention capability has twoaspects, one engaged to keep goals of novel tasksaccessible in the face of conflict, and the otherto resolve the conflict presented by habitual andgoal-directed responses, or, in memory-retrievalcontexts, to resolve interference between mem-ories for similar events.

We see these goal-maintenance andcompetition-resolution functions of executivecontrol as being quite similar to the proactiveand reactive control modes, respectively, pro-

posed by Braver et al. (Chapter 4) in their dual-process theory of cognitive control. Like Braveret al., we are not yet sure how these dissociablesystems of control may interact with one an-other. On one hand, we propose that goal main-tenance is necessary for the proactive blockingof competition, as in high-congruency Strooptasks, and so here blocking or inhibition is de-pendent upon maintenance. On the other hand,the more reactive resolution of conflict seems

to be accomplished independently of goal main-tenance, and these mechanisms may also be theones required for the resolution of memory in-terference (e.g., Conway & Engle, 1994; Kane& Engle, 2000; Rosen & Engle, 1998).

CHALLENGES FOR AN EXECUTIVEATTENTION VIEW OF VARIATIONIN WORKING MEMORY CAPACITY

Evidence from a variety of macroanalytic andmicroanalytic studies indicates that normalvariation in WMC reflects primarily the func-tion of executive attention processes. We pro-pose that these executive processes keep rep-resentations of goal-relevant plans, responses,and stimuli in a highly accessible state in thepresence of interference from prior events anddistraction or conflict from the task environ-ment. However, in the sections that follow, wediscuss some current challenges for our theoryof WMC variation. We first discuss two recentfindings from our laboratories that may poseconstraints on our conceptualization of execu-tive attention. We then consider several com-plications that surround the measurement of the executive attention construct.

Boundary Conditions to the Relationbetween Memory Capacity andExecutive Attention?

When we began investigating the connectionbetween WMC and attention control, we hadthe naıve sense that most cognitive processes




widely agreed to be ‘‘controlled’’ or ‘‘executive’’would be sensitive to individual differences inWMC. However, two lines of research on visualsearch and task-set switching have shown that

simplistic view to be incorrect.

The Problem

Following the seminal work of Treisman andGelade (1980), visual search for targets amongperceptually similar distractors has been widelyconsidered a controlled process. That is, failingthe automatic ‘‘pop out’’ of a unique visualfeature from an array, attention is required to

serially integrate the independently processedfeatures into coherent object representations.We therefore reasoned that subjects with highspans should locate visual targets more quicklythan those with low spans when attention-demanding search is required to find a targetsharing features with its surround. We werewrong. After a pilot study indicated no spandifferences in visual search for either ‘‘auto-matic’’ or ‘‘controlled’’ search targets, Kane,

Poole, Tuholski, and Engle (2006) replicatedthis span equivalence in larger samples acrossseveral different tasks. In one experiment, high-and low- span subjects searched for a targetletter F among either Os (allowing more auto-matic, or efficient search) or Es (forcing morecontrolled, or inefficient search). Displays pre-sented 1, 4, or 16 stimuli, arranged either in aregular matrix or psuedorandomly on-screen.Regardless of the array characteristics, the spangroups showed identical search latencies andslopes across display sizes. In a second experi-ment, subjects searched for targets defined by aconjunction of features (Fs among Es and hor-izontally tilted Ts in one block, red vertical barsamidst red horizontal and green vertical bars inanother); here, again, high- and low-span sub-jects demonstrated equivalently large searchslopes. Whatever attentional processes are en-

gaged by typical instantiations of visual searchare not linked to those captured by WM span.Inefficient search may be commonly con-

sidered a ‘‘controlled’’ task, but it is not nearlythe gold-standard measure of executive controlthat task-set switching is thought to be (seeMonsell & Driver, 2001). In these tasks, sub-

jects regularly or unpredictably switch backand forth between two or more response sets forambiguous stimuli; in either case, task-switchsequences elicit an RT ‘‘switch cost’’ compared

to task-repeat sequences. We have so far failedtodemonstrateaconnectionbetweenWMCandswitch cost in two prototypical preparations(see also Chapter 3). In our first three experi-ments, Kane, Poole, Tuholski & Engle (2003)tested high and low spans in a numericalStroop task where subjects either identified orenumerated the digits in a horizontal string(e.g., 2222¼ ‘‘two’’ or ‘‘four’’). Within a cuedprime-probe procedure, half the trial pairs re-

peated the task between displays and half switched the task. High and low spans showedequivalent RT switch costs in all three experi-ments. We were surprised by these findings,but also concerned that the prime-probe pro-cedure might be measuring something differ-ent than the more typical ‘‘alternating-runs’’preparation (e.g., Rogers & Monsell, 1995). So,in a fourth experiment we tested subjects in oneof four different versions of the alternating-runs

task (differing from each other in task-cuing andresponse-mapping details). Each trial displayeda letter and number, and subjects classified ei-ther the letter as vowel or consonant or the num-ber as odd or even. In pure-trial blocks, the sametask repeated over trials; in mixed-trial bocks,the tasks alternated in an AABB sequence. In allfour versions of the task, high- and low-spansubjects showed equivalent RTs in all of thepure- and mixed-trial conditions.

Clearly, we find no evidence for a deficit invisual search or task-set switching in low spans.What should we conclude from these findings?Oberauer et al. (Chapter 3) suggest that our ex-ecutive attention view of WMC is falsified. Wedisagree, in part because we find robust WMCdifferences in a variety of other attention-controltasks. However, it may be that the attentionalprocesses engaged by WM span tasks are related

only to ‘‘inhibitory’’ attention tasks requiringprepotent responses to be withheld, as in Stroopand antisaccade tests (see Chapter 9). This ideadoes not appeal to us either, for a number of reasons. First, as we conceive it, goal mainte-nance and competition resolution should bequite generally important executive capabilities.




Second, the higher-order cognitive abilities thatWMC predicts, such as reading comprehensionand inductive reasoning, do not necessarily in-volve much response conflict or restraint of ha-

bit. Third, we do find WMC-related differencesin several visual-attention tasks that do not seemto fundamentally measure control of habit orprepotency. For example, Tuholski, Engle, andBayliss (2001) found that subjects with highspans could count the number of objects pre-sented in a disorganized visual array faster thansubjects with low spans when the tally exceededthe ‘‘subitizing’’ (pattern recognition) range of one to four items. Here there was no obvious

prepotent response to keep in check (which alsosuggests a problem for a purely inhibitory view of WMC variation). Likewise, in a very differentvisual task, Bleckley, Durso, Crutchfield, Engle,and Khanna (2003) asked high- and low-spansubjects to identify a letter presented briefly atcentral fixation. At the same time, another letterappeared in one of 24 locations along threeconcentric rings around fixation, and subjectstried to identify its location. The ring on which

the second letter would appear was cued (with80% validity) as ‘‘close,’’ ‘‘medium,’’ or ‘‘distant.’’

As expected from ‘‘spotlight’’ or ‘‘zoom lens’’theories of visual attention (e.g., Eriksen &Murphy, 1987), letters appearing outside thecued ring on invalid trials (i.e., outside thespotlight) were localized more poorly than lettersappearing along the cued ring. More interest-ingly, for high spans only, letters appearing in-terior to the cued ring were also localized morepoorly than letters along the cued ring. Thesefindings suggest that high-span subjects flexiblyconfigured attention discontiguously, focusingon the letter at fixation and on a ring beyondfixation, at the exclusion of intermediary rings of space. This pattern suggests that high spansadopted an object-based attentional focus. Low-span subjects, in contrast, showed a benefit forany location along or interior to a cued ring,

indicative of a spotlight configuration and aspace-based attentional focus.

A Solution?

Given these visual-attention findings that arenot obviously inhibitory in nature, how should

we reconcile our failures to link WMC tosearch and switching? We suggest that, unlikethe tasks that have yielded WMC correlations,prototypical search and switching methods do

not tap volitional, executive-control processes.Of course, if we want to avoid circularity in de-fining ‘‘executive attention’’ as simply anythingthat correlates with WMC, we must considermore closely what search and switching actuallyentail.

With respect to visual search, ‘‘guidedsearch’’ theory (Wolfe, 1994) proposes that at-tention is pulled across a master map of visuallocations, based on activation flowing pre-

attentively from multiple feature maps. That is,attention is probabilistically guided from thehighest activation peak to successively lowerpeaks, with activation summed from ‘‘bottom-up’’ and ‘‘top-down’’ sources. Bottom-up acti-vation accrues from physical differences amongstimuli: the more an object differs from itssurround, the greater the bottom-up activationto that location. In contrast, the top-down sig-nal represents the subject’s knowledge of the

features that specify the desired target, expressedas a verbal category (e.g., ‘‘red’’). If the subjectknows the target is red amidst blue and yellowobjects, then red features will prompt top-downactivation to their locations on the master map.

Despite the ‘‘top-down’’ label, we see littlerelation between this use of advance knowledgeand the attention-control processes we think arecentral to WMC, because search is proposedhere to be passively ‘‘pulled’’ rather than endog-enously ‘‘pushed.’’ However, there may be somecontexts in which top-down control is, in fact,more controlled. Wolfe (1994) proposes thattop-down effects may sometimes act to reducebottom-up contributions to the activation map.For example, if the target is a red horizontalbar amidst many red vertical bars and fewgreen horizontal bars, then color is less diag-nostic of the target than is orientation. Based

on this knowledge, the bottom-up contributionof orientation could be amplified or that of colorreduced. As evidence for this kind of volitionalmodulation, when experimenters manipulatethe proportions of particular non-target fea-tures, subjects use this information to speedtheir search (e.g., Bacon & Egeth, 1997). We




wonder whether this top-down ability to am-plify or dampen bottom-up influences mightvary with WMC, whereas typical conjunctionsearch prevents its expression by presenting

equal numbers of non-target types.With respect to task-set switching, we hadmany reasons to expect an association withWMC. For example, De Jong (2001) argues thatswitch costs result largely from periodic failuresto engage and maintain goal-related preparation(a parallel to our ‘‘goal maintenance’’ idea).This failure-to-engage hypothesis is supportedby findings that variables expected to affectsubjects’ ability to sustain goals in active mem-

ory also affect switch costs. Moreover, mixturemodels that assume switch trials to produce adistribution of fast RTs on one hand (due toadequate goal maintenance) plus a distributionof slow RTs on the other (due to engagementfailures) provide a good fit to the cumulativeRT functions from switch trials. As anotherreason to expect an association with WMC, All-port and Wylie (2000) propose that a proactiveinterference–like perseveration of task set con-

tributes to switch costs. Using Stroop-like stim-uli, they find asymmetrical switch costs thatdepend more on the difficulty of the task to beswitched from than the task to be switched to: forexample, the cost of switching from color nam-ing to word reading is larger than the reverse.Our findings of WMC span differences in proac-tive interference might therefore suggest WMC-related differences in switching.

At the same time, however, there are grow-ing concerns that task-set switching may not bethe ‘‘executive’’ measure it is widely assumed tobe (e.g., Altmann, 2002). To discuss just onespecific issue, most switching studies cue thetask set for each trial by presenting either itsname or an abstract symbol, and this cuingallows ostensibly non-executive encoding andretrieval processes to contaminate measurementof switch cost. Specifically, cuing paradigms

confound task and cue switches (Logan &Bundesen, 2003). That is, when subjects switchtasks, the new task is signaled by a cue that isdifferent from the immediately preceding one,whereas task-repeat trials always repeat the cue.Logan and Bundesen argue that these switchcosts actually reflect a benefit of repeating the

cue on task-repeat trials. Their idea is that thecue-plus-stimulus compound, by itself, pro-vides all the information needed to determinea response. No executive process is needed to

switch task set, so task-switching paradigmsactually require only one task: identify the cueand target and use them to retrieve prior stim-ulus and cue episodes. Critically, when cuesrepeat, cue identification and retrieval are facil-itated, and so task-repeat trials yield faster re-sponses than those of task-switch trials.

Evidence for these ideas comes from ex-periments in which cues and tasks repeated orswitched independently (e.g., Logan & Bun-

desen, 2003). For example, subjects were cuedto make digit-magnitude judgments followingthe cues Magnitude or HighLow and parityjudgments following the cues Parity or Even-Odd. When both the cue and task repeatedacross trials (e.g., Parity!Parity), RTs weresubstantially faster than when the cue switchedbut the task repeated (e.g., EvenOdd!Parity),indicating a cue-switch cost in the absence of atask switch. Moreover, RTs for cue-switch trials

were virtually identical to actual task-switchRTs (e.g., Magnitude!Parity). Data like thesesuggest that cue switches are responsible formost of the switch cost observed in cued pro-cedures. Thus, in order to tap truly volitionaland executive processes (and to be sensitive toWMC variation), task-switching proceduresmust eliminate the roles of cue encoding andcue-based retrieval processes.

Summary

It may be surprising, initially, that WMC isunrelated to search slopes and switch costs.

And, taken at face value, these null findingsmay seem to call our ‘‘executive attention’’ the-oryofWMCintoquestion(seeChapter3).How-ever, we suggest that our null effects are lesssurprising when we appreciate the complexity

of visual search and task switching. Most task-switching and visual-search paradigms aresimply not good indices of executive control,and just because researchers label search andswitching tasks as ‘‘controlled’’ or ‘‘executive’’does not make them so. More broadly, given thenumber of WMC–attention associations we




have observed, we submit that the mere failureto find a correlation between WMC and anyparticular executive task does not necessarilyfalsify our view. However, such failures do point

to obviously important questions for future re-search on the nature of both WMC and exec-utivecontrol,andthe tasksthatweusetomeasurethem.

Measuring Working Memory Capacityand Executive Attention

We view WM span tasks as reasonably goodmeasures of executive attention (along with a

host of other processes) because their dual-taskrequirements challenge subjects to maintainaccesstoinformationoutsideofconsciousaware-ness, recover access to information that wasoutside of awareness, or both, all in the face of proactive interference. However, a WMCmeasure need not be a dual task in order totax attention control. Indeed, Oberauer and col-leagues have shown that a latent factor derivedfrom various ‘‘coordination’’ tasks can be in-

distinguishable from a ‘‘storage-plus-proces-sing’’ WM span factor (see Chapter 3). Thesecoordination tasks generally require subjects tokeep track of a large number of stimuli at thesame time, or to rapidly switch attention amongthese active stimuli, and so they also requiremaintained access to information that is mo-mentarily outside the focus of attention.

Indeed, prototypical STM span tasks shouldalso make non-negligible demands on executivecontrol, particularly when routinized rehearsaltechniques are made ineffective or when thelists are too long to be maintained entirely inconscious focus. In fact, spatial STM tasks,which do not afford phonological rehearsal,seem to be quite good measures of the WMC–executive construct and correlate strongly withgeneral ability (e.g., Kane et al., 2004; see alsoChapters 3 and 8, this volume). Moreover,

Unsworth and Engle (2006) found that longlists from verbal STM span tasks correlatestrongly with all list lengths from verbal WMCspan tasks, and long STM lists predict similarvariance in Gf, as do long and short WMClists. Because the focus of attention comprises

only about 4 ± 1 items (Cowan, 2001), andbecause the phonological loop can hold onlywhat can be spoken in about 2 s (Baddeley,1986), verbal STM lengths larger than four

items or so will require some degree of exec-utive attention to be maintained or recoveredfrom outside conscious focus.

By this view, span tasks (or any other mem-ory tasks) cannot be dichotomized as reflectingeither STM or WMC, or either storage or ex-ecutive control, because all immediate memorytasks are complex and determined by a host of factors, including both storage and executiveattention. The challenge for researchers of WM

variation is to assess the contributions of thesevarious processes to the associations betweenmemory tasks and measures of higher-ordercognition (or between these memory tasks andage, or personality, or psychopathology, etc.).We have focused our research on WM spanbecause these tasks have undergone more para-metric and task-analytic work than the alterna-tives, and we suggest that researchers conductsimilar explorations of other candidate WMC

tasks to move the field forward.Before leaving a consideration of WMC

measurement, we must accept a shortcomingof our view, pointed out in Chapter 3. Ourresearch has demonstrated a strong associationbetween WMC and Gf on one hand, and sta-tistically significant differences between high-and low-span subjects in attention performanceon the other hand. From these findings, wehave inferred that executive attention processestapped by WMC tasks are responsible for itscovariation with Gf and cognitive ability. Note,however, that there are two important infer-ences here that require more explicit support,preferably from a latent-variable approach. First,we have not yet established the strength of thecorrelation between WM span and attention-control measures; our extreme-group designstesting high- vs. low-span subjects may overesti-

mate the WMC effect size. Second, if executiveattention is responsible for the covariation be-tween WMC and cognitive ability, then a latentfactor comprised of the shared variance amongWMC and attention-control tasks should predictsubstantial variance in Gf. At the same time,




any remaining, unique WMC and attention-control variance should not correlate with Gf asstrongly. If WMC and attention-control con-structs correlate weakly, or if shared WMC–

attention variance is not a strong predictor of cognitive ability, then our theory is in trouble.

OUR EXECUTIVE ATTENTION VIEWIN RELATION TO OTHER THEORIES

General Theories ofWorking Memory

Our view is that WM span tasks are complexand determined by many general and domain-specific processes, skills, and strategies. However,variation in WMC, as measured by individualdifferences in WM span, reflects primarily ex-ecutive attention capabilities. These executiveactivities are general and important to a rangeof intellectual functions, from controlling inap-propriate actions, to learning and recalling in-

formationamidstcompetingmemories,tosolvingcomplex verbal and nonverbal problems. Beforewe reflect upon particular ‘‘competing’’ perspec-tives on WMC variation, we first consider ourviews in light of important nomothetic WM the-ories, such as Baddeley’s ‘‘multiple-component’’WM theory (Baddeley, 1986, 2000), and Nairne’svery different, more process-oriented approach(Nairne, 2002).

The Multiple-Component Working Memory Model

Our theory of WM variation is inspired byBaddeley and Hitch’s (1974) demonstrationthat the cognitive problem of balancing mem-ory storage and ongoing mental activity is cen-tral to a range of intellectual capabilities. More-over, our distinction between general attention

processes and domain-specific storage processesis consistent with Baddeley’s (1986, 2000) sep-aration of the central-executive (attentional)component from the phonological-loop andvisuospatial-sketchpad (storage/rehearsal) com-ponents of the WM system.

However, our view differs from themultiple-component model in emphasizingfunction and process over structure. That is, weview the domain specificity of ‘‘STM storage’’

to reflect different perceptual bases of, and re-hearsal activities afforded by, different stimuli. As we stated earlier, our process-oriented viewis more akin to Cowan’s (1995) conception of immediate memory. ‘‘STM’’ is represented bygraded activation of LTM traces (with ‘‘focalattention’’ representing the limited, consciousportion of activated LTM), along with routin-ized and executive processes that maintainactivation. Cowan’s model, in turn, closely

resembles Wundt’s (1912/1973) conceptionof consciousness. Wundt distinguished appre-hension, the graded entrance of objects intoconsciousness, from apperception, the entranceof apprehended objects into awareness. Intoday’s terms, Wundt argued that informationcould remain accessible (or activated) outsideattentional focus. Moreover, Wundt claimed,like Cowan, that the focus of attention is strictlylimited, whereas above-threshold activation is

more broad and variable in scope. These ideasresonate with our claims regarding WMC,particularly that executive processes maintainor recover access to ‘‘apprehended’’ represen-tations of goals, response productions and stim-uli in the absence of focal attention or skilledrehearsal routines, and in the presence of in-terference or conflict.

The Baddeley (1986, 2000) model is mostobviously characterized by its structural focus,that is, by its separation of WM into distinctcomponents with different attributes and func-tions. In general, we are unenthusiastic aboutsuch neo-structuralist approaches to memorytheorizing. Although Baddeley has been morerestrained in proposing structures to accountfor new dissociations than have LTM research-ers, WM structures have begun to proliferate.Theorists now pose separate buffers for seman-

tic information (e.g., Haarmann, Davelaar, &Usher, 2003), visual imagery vs. visual rehearsal(Pearson, 2001), and assigning syntactic struc-tures (Caplan, Waters, and DeDe, Chapter 11).

As in LTM research, consensual and specificcriteria for proposing new immediate-memory




systems are lacking, and we therefore envisionan undisciplined explosion of WM buffers as‘‘explanations’’ for behavior.

Baddeley’s (2000, 2001) most recent incar-

nation of the WM model makes a structuralclaim that is most relevant to our work. A newsubsystem, the ‘‘episodic buffer,’’ is proposed tohandle some problems for the model, such ashow verbal material is maintained under ar-ticulatory suppression. The episodic bufferis essentially an immediate-memory version of episodic memory, a mnemonic store for main-tenance of integrated, multidimensional rep-resentations of objects and events. Of primary

concern here, Baddeley (2001) speculates thatthe episodic buffer underlies performance of WM span tasks: by the multiple-componentview, the phonological loop cannot supportverbal WM span when the processing-taskstimuli are read aloud (and thus provide artic-ulatory suppression). From our perspective,however, the episodic buffer currently offerslittle to research on WM variation. Baddeleyhas not yet clarified the buffer’s importance to

the predictive power of WM span: is it inci-dental, with the executive driving the correla-tions, or does the buffer’s multimodal naturemake it critical to cognition broadly? Further-more,the ‘‘constrained-sentencespan’’ task,desi-gned to measure the capacity of the episodicbuffer (Baddeley, 2001), seems a minor varia-tion on the reading span task. We are thereforeskeptical that individual-differences researchwill soon clarify the nature or utility of this newWM component, or vice versa.

Functionalist and Process-Oriented Approaches to Immediate Memory

Our view of immediate memory is less structuralthan Baddeley’s (1986, 2000) model. None-theless, we do identify the executive attentionprocesses tapped by WM span tasks with partic-

ular brain systems, particularly the circuitry of the dorsolateral prefrontal cortex (dPFC; Kane &Engle, 2002). Moreover, like many cognitivepsychologists, we retain a conceptual distinctionbetween immediate memory and LTM, regard-ing the former as an activated portion of thelatter.

Our comfort with the dichotomy of activeand inactive memory is not for lack of agood alternative. Research deriving from theverbal-learning and functionalist traditions

generally assumed a unitary memory, and neo-functionalist and proceduralist views drawupon thisheritage (e.g., Crowder, 1982; Melton,1963; Toth & Hunt, 1999). According to theseaccounts, one set of processing rules governsremembering: memory is an activity, not athing, and remembering over the ‘‘short term’’and ‘‘long term’’ is identical, despite the phe-nomenological and folk-psychological distinc-tion. Most recently, Nairne (2002) questioned

the activation metaphor of immediate memory,arguing that evidence is wanting for such aspecial memory state, its loss through decay,and its protection via rehearsal. By Nairne’sview, which is widely agreed upon in the studyof LTM, memories have no special status out-side of a given constellation of cues. Remem-bering is not determined by the strength of atrace, but rather by the discriminability of thetarget event amidst competitors, given a spe-

cific task environment. In Nairne’s ‘‘featuremodel,’’ specifically, all retrieval is governed bythe match between environmental cues and thefragile (non-conscious) ‘‘processing records’’ of recent events that are vulnerable to interfer-ence. Forgetting thus occurs when the cue and/ or processing records fail to uniquely promptrecollection of the target event.

The activation and decay of representationsare appealing metaphors for WM research, in-cluding that on WM variation. Indeed, our re-search relies heavily on the activation metaphorto describe the heightened accessibility of in-formation, resulting from rehearsal or executiveattention processes, which contributes to variouscomplex cognitive tasks. Nairne (2002) providescompelling arguments and evidence against de-cay and the protective powers of rehearsal, butwe still find the activation concept useful. Heu-

ristically, it provides language with which todescribe the ways in which goals may controlbehavior in the face of conflict, as in Stroop tasks,as well as to conceive of the relation betweenmeasures of WM span and attention controlmore broadly. The feature model, emphasizingcue-driven discrimination among stored process-




ing records, is not easily extended into the do-main of WM variation. If WMC reflected suchmemory-discrimination processes, even if closelytied to interference, we do not yet see how they

should relate to reading comprehension, spatialvisualization, or moving one’s eyes away from aflash. In contrast, the idea that WMC reflects anability to maintain information in an activated oraccessiblestateduringongoingprocessingismoreeasily applied across simple and complex cogni-tive tasks.3

Moreover, recent research suggests a stronglink between WMC and dPFC functioning, andin the realm of neuroscience, the activation

metaphor is less of a metaphor. For example,individual dPFC cells that are ‘‘tuned’’ to par-ticular locations, objects, rules, or their combi-nations maintain a pattern of sustained firing overmemory delays for their preferred stimuli. What’smore, these dPFC cells, unlike those in posteriorbrain areas, maintain their activity when dis-tracting stimuli are presented during the delay(for a review, see Kane & Engle, 2002). Suchtarget-specific, delay-related activity is difficult to

interpret from a neo-functional perspective thatdenies a special state of activity tied to immediatememory. It fits quite well, however, with our viewthat executive attention involves the active main-tenance of goal-relevant information in the faceof interference and distraction.

As an ‘‘attentional’’ example, sustained dPFCactivity that is related to goal maintenance pre-dicts Stroop interference. Under functional mag-netic resonance imaging (fMRI), MacDonald,Cohen,Stenger,andCarter(2000)presentedsub-jects with word-reading and color-naming Strooptasks, unpredictably cued 11 s before each stim-ulus. Fifty percent of the trials were congruent,and so in combination with the frequent word-readingdemand,theoveralltaskenvironmentdidnot reinforce the color-naming goal. On color-naming trials, which demanded an anti-habitualresponse,dPFCactivityincreasedsteadilyoverthe

cue-to-target delay and this increase correlatednegatively with interference magnitude (r ¼.63). Subjects who were better able to activateand sustain the ‘‘ignore-the-word’’ goal were alsobetter able to resist interference from the words.

Across memory and attention studies in theneuroscience literatures, then, we see a parallel

between neural activity and WM maintenance,and we view these more literal demonstrationsof activation as license to use the activationmetaphor in describing normal behavioral var-

iation in WMC.

Theories of Variation in WorkingMemory Capacity

The most popular alternative views of WMCvariation that we analyze below, processing-speed and attentional-inhibition theories, havebeenmostwidelyandsuccessfullyappliedtostud-ies of life span cognitive development. Indeed,

we will claim that, whereas processing speedseems to be important to age-related variation inWMC, it has yet to prove its mettle in account-ing for within-age variation. Our discussion of attentional inhibition will be more nuanced andless decisive because, in our view, the inhibitionapproach is only subtly different from our own.First, however, we briefly consider the over-lapping, recently developed views of Oberauer etal. (Chapter 3) and Cowan (2004).

Capacity of Attentional Focusand Region of Direct Access

Although Oberauer et al. (Chapter 3) acceptCowan’s (1995) conceptualization of the WMsystem, as do we, our view of WMC variation isdistinguishable from both of these views. Cowan(2005a, 2005b) recently suggested that WMCvariation and covariation may reflect the size, orcapacity, of attentional focus. Thus, WMC is atrue ‘‘capacity’’ by Cowan’s view, correspondingto a structural limit in the amount of materialthat can be held in a particular state, in parallelto the 7 ± 2 capacity limit for STM proposedby Miller (1956). Our view, in contrast, is thathigh-WMC individuals are not necessarily ableto hold more discreet representations in con-sciousness than are low-WMC individuals, but

high spans are better able to actively maintaintask-relevant information outside of conscious-ness and to do the mental work necessary toquickly recover information from inactive mem-ory despite interference.

We see evidence against Cowan’s view intwo findings of WMC equivalence in ostensible




signatures of focused-attention capacity: sub-itizing and primary memory. Tuholski et al.(2001) found that high- and low-span subjectscould enumerate in parallel (or ‘‘subitize’’)

an equal number of visual objects (3.35and 3.25, respectively), but low spans showed amuch steeper counting slope beyond the sub-itizing range than that of high spans. Engle,Tuholski et al. (1999) tested subjects in an im-mediate free-recall task, and their derived esti-mates of primary memory capacity did notcorrelate with WMC; only estimates of sec-ondary memory capacity did. Both results sug-gest that high and low spans can keep a similar

number of representations available in theconscious focus of attention. Where they differ,instead, is in processing and recovering repre-sentations outside attentional focus. BecauseOberauer et al. (Chapter 3) equate their ‘‘direct-access’’ component of WM with Cowan’s focusof attention, their view that individual differ-ences in WMC and reasoning ability centrallyreflect variation in direct access also seems to becontradicted by our findings.

General Processing Speed

General processing-speed (PS) theories broadlypropose to account for variation in higher-ordercognition via the measurement of latenciesfrom simple cognitive tasks (e.g., Jensen, 1998).The idea is that people with low PS completefewer mental operations per unit time, and thisleads to a failure in completing some criticaloperations, a greater likelihood of losing theproducts of processing through decay, or a re-duced ability to keep multiple processingstreams active via rehearsal or switching. Withrespect to life span cognitive development, akey finding is that age-related variance in com-plex cognitive activity, and in WMC, is reduceddramatically after statistically controlling forvariance in mean PS (Kail & Salthouse, 1994;

see also Chapters 6 and 8, this volume).However, we have already reported findingsfrom our research indicating that PS is not apromising mechanism for WMC variationamong young adults. First, PS measures nei-ther correlate strongly with WMC nor account

for the shared variance between WMC and Gf (Conway et al., 2002). Second, studies of re-trieval interference find that high and lowspans’ recognition latencies are equivalent in the

absence, but not the presence, of responsecompetition (Conway & Engle, 1994). Indeed,high-span subjects’ recall latencies are actuallylonger than those of low-span subjects when thetarget information had previously been sup-pressed when related information was learned(Rosen & Engle, 1998). Third, high and lowspans’ letter-identification latencies are equiv-alent in the prosaccade task when it was pre-sented before antisaccade (Kane et al., 2001),

and span differences in baseline Stroop RTscome and go across experiments independentlyof interference differences (Kane & Engle,2003). Fourth, and finally, in visual search andtask switching, high and low spans show equiv-alent RTs in relatively complex conditions withlong mean latencies.

Processing-speed theory cannot accommo-date these results. However, even if it could bemodified to do so without incorporating our

theoretical premises, we would remain unen-thusiastic because PS theory has yet to provide areasonably specific psychological account for itsvariation or covariation. Conway, Kane, andEngle (1999) suggested that researchers considerthe role of variation in attention control in pro-ducing variation in PS within and across agegroups (see also Chapter 9). For example, in-creases in PS–ability correlations with PS taskcomplexity and decreases in PS–ability correla-tions with task practice seem fit for an attentionalexplanation. Furthermore, in studies of youngadults, individuals’ variability in RT is oftenmore strongly correlated with cognitive abilitythan is median RT (Jensen, 1998). Given thatRT variability may reflect failures of sustainedattention, RT variability may reflect executive-control difficulties. At least among young adults,then, WMC–attention variation might drive PS

variation, rather than the reverse.

Attentional Inhibition

Finally, an influential research program byHasher, Zacks, and colleagues suggests that WM




variation derives from the operation of inhibitoryattentional mechanisms (e.g., Hasher & Zacks,1988; see Chapter 9). Their claim is that whatappears to be a structurally reduced WMC in

some individuals (e.g., the elderly, or youngadults at their off-peak hours) actually reflects a functional reduction due to the intrusion andpersistence of irrelevant information. Inhibitoryattention mechanisms, which control cognitionby restricting access to and deleting informationfrom WM, often fail with aging and circadianvariation. Thus, individuals with ineffective in-hibition due to aging, normal individual differ-ences, variation in circadian arousal, or what

have you, suffer disproportionately from memoryinterference, language production and compre-hension difficulties, and contextually inappropri-ate responding. Interference and distractionare thus central problems of control and WMvariation.

These ideas have significantly influencedour view of WMC variation, and indeed, thedifference between the inhibitory view andours is quite subtle. Hasher and colleagues

argue that WMC variation is driven largely byvariation in inhibitory control. As evidence,their microanalytic work has investigated therole of proactive interference in determiningWM span scores. In typical ‘‘ascending’’ ad-ministrations of reading span, large sets areencountered only after interference from priortrials builds up. Thus, interference-vulnerableindividuals, such as the elderly, are disadvan-taged on the large sets that are most critical tothe span score. Lustig et al. (2001) thereforetested some subjects with an ascending task,proceeding from set-size 2 to 4, and others in adescending task, proceeding from size 4 to 2.

Additional young subjects were tested on adescending task with filled breaks between ev-ery set to reduce interference further. Lustiget al. found significant age differences in WMspan in the ascending but not the descending

condition—reducing proactive interference re-duced age differences. Furthermore, within agegroups, the span condition with the least inter-ference(descendingforolderadults,descending-with-breaks for younger adults) showed no cor-relation with reading comprehension. Thus,

interference resistance, assumed to reflect in-hibition, mediated span correlations with ageand ability (see also Bunting, 2006).

We have characterized the executive-attention

and attentional-inhibition views as providing a‘‘chicken–egg’’ dilemma, where one assumesWMC to determine inhibitory control and theother, the reverse (Kane et al., 2001). However,upon further reflection we do not think thisis quite right. Instead, whereas the inhibitoryview assumes inhibition to determine WMC,we submit that a third variable causes both. Ourview is that executive attention processes blocksources of interference and competition, as

well as keep information active in interference-and conflict-rich contexts and in the serviceof ongoing cognitive processes. Thus, WMCand inhibition are strongly linked, but indi-rectly through a more basic attentional con-struct.

Although interference clearly contributes toparticular tests of reading comprehension, WMspan predicts performance in many tasks whereinterference or inhibition are not obviously rel-

evant but active maintenance should be, such asmental rotation, verbal analogies, or counting vi-sual objects (Kane et al., 2004; Tuholski et al.,2001). In addition, our Stroop findings, as well asthose reviewed from the neuroscience literature,suggest that PFC maintenance of stimuli andgoals allows for effective inhibition under someconditions. We are unsure how inhibition mightaccount for sustained memory-related activity of dPFC cells, increasing dPFC activation promp-ted by Stroop-task cues, or congruency effects onspan differences in Stroop interference. More-over, if resistance to proactive interference actu-ally reflects inhibition (but for alternative views,see Hasher & Johnson, 1975; Underwood &Ekstrand, 1967), we find that high spans’ in-hibitory control is impaired by secondary tasks,which suggests that some more fundamentalcontrol process governs inhibition (Kane &

Engle, 2000). Thus,we believe our more general,‘‘executive attention’’ view to be more compre-hensive than the inhibitory view in accountingfor the breadth of the cognitive and neurosciencefindings regarding the covariation of WMC withother cognitive activities and abilities.




Notes

1. Demonstrations of low reliability reported by Ca-

plan et al. (Chapter 11) are troubling. However,

Turley-Ames and Whitfield (2003) showed opera-

tion span to be reliable over minutes (r s¼ .7 –.8),

Klein and Fiss (1999) showed it to be reliable over

weeks and months (r s¼ .7 – .8), and an automated

version of operation span had a test-retest reliability

of .83 with an average lag of 13 days (Unsworth,

BOX 2.1. SUMMARY ANSWERS TO BOOK QUESTIONS

1. THE OVERARCHING THEORY

OF WORKING MEMORY

Following Cowan (1995), we view WM as an

integrated memory and attention system, com-

prised of long-term memory representations

(for stimuli, goals, or action plans) activated

above threshold, procedural skills for rehearsal

and stimulus coding, and executive attention

processes. Activated representations represent

the contents of ‘‘short-term memory,’’ and a

very limited subset of these are experienced as

the focus of conscious awareness, or ‘‘focused

attention.’’ Procedural skills and executive at-

tention are engaged to maintain activation or

access to goal-relevant representations, par-

ticularly those outside of focused attention,

which would otherwise return to baseline as a

result of decay or interference.

2. CRITICAL SOURCES OF

WORKING MEMORY VARIATION

Variation may occur in any WM component.

However, in healthy adults, WM capacity’s

covariation with general cognitive ability stems

from variation in executive attention. Executive

attention maintains activation to goal-relevant

representations outside of conscious focus,

recovers access to non-active representations

against interference, and resolves competition

between co-active representations or betweenhabitual and goal-appropriate actions. Macro-

analytic , latent-variable studies suggest that

WMC and Gf share substantial variance while

controlling for STM. As well, microanalytic,

quasi-experimental studies indicate that WMC

variation predicts individual differences in a

variety of memory-interference and attention-

control tasks.

3. OTHER SOURCES OF WORKING

MEMORY VARIATION

Our executive attention view, emphasizing

goal maintenance and competition resolution,

parallels the dual mechanisms of cognitive

control proposed by Braver et al. (Chapter 4),

and is only subtly different from the inhibitory

view of Hasher et al. (Chapter 9), who suggest

that inhibition drives WMC. We argue that a

common attention-control capability under-

lies WMC and inhibition. However, our view

is incompatible with those of others. Although

measures of processing speed account for age-

related variance in WMC and cognitive abil-

ity (see Chapter 8), they do not account for

WMC–ability covariation in young adults. In

addition, WMC differences do not correspond

to measures of focused-attention capacity, but

rather to processing beyond conscious focus,

contradicting proposals by Cowan (2005a,b)

and Oberauer et al. (Chapter 3).

4. CONTRIBUTIONS TO GENERAL

WORKING MEMORY THEORY

The study of WMC variation has led the way in

fulfilling WM theory’s original and greatest

promise—to illuminate the functions of imme-

diate memory. This work shows that the domain-

general, executive components of the WM sys-

tem support a broad range of cognitive abilities,and may even provide the scaffolding for a

cognition-based understanding of intelligence.

Correlational research also provides support for

the dissociability of domain-specific storage and

rehearsal processes and for the idea that domain-

specific memory processes and domain-general

attention processes are intimately linked (if not

synthesized) within the WM system.




Jensen on Intelligence-g-Factor. Psycoloquy, 10

(074), Article 16.

Conway, A.R.A., Kane, M.J., & Engle, R.W. (2003).

Working memory capacity and its relation

to general intelligence. Trends in CognitiveSciences, 7, 547–552.

Cowan, N. (1995). Attention and memory: An

integrated framework. Oxford, UK: Oxford Uni-

versity Press.

Cowan, N. (2001). The magical number 4 in short-

term memory: A reconsideration of mental


24, 87–185.

Cowan, N. (2005a). Understanding intelligence: A

summary and an adjustable-attention hypothe-sis. In O. Wilhelm & R. W. Engle (Eds.), Hand-

bookofunderstandingandmeasuringintelligence

(pp. 469–488). Thousand Oaks, CA: Sage.

Cowan, N. (2005b). Working memory capacity. New

York, New York: Psychology Press.

Cronbach, L. J., (1957). The two disciplines of

scientific psychology. American Psychologist, 12,

671–684.

Crowder, R. G. (1982). The demise of short-term

memory. Acta Psychologica, 50, 291–323.Daneman, M., & Carpenter, P. A. (1980). Indivi-



Behavior, 19, 450–466.


memory and language comprehension: A meta-

analysis. Psychonomic Bulletin & Review, 3,

422–433.

De Jong, R. (2001). Adult age differences in goal

activation and goal maintenance. European

Journal of Cognitive Psychology, 13, 71–89.

Engle, R. W., Cantor, J., & Carullo, J. J. (1992).

Individual differences in working memory and

comprehension: A test of four hypotheses.

Journal of Experimental Psychology: Learning,

Memory, and Cognition, 18, 972–992.

Engle, R.W., & Kane, M.J. (2004). Executive atten-

tion, working memory capacity, and a two-factor

theory of cognitive control. In B. Ross (Ed.) Thepsychology of learning and motivation (pp. 145–

199). New York: Academic Press.

Engle, R. W., Kane, M. J., & Tuholski, S. W.


ory capacity and what they tell us about con-

trolled attention, general fluid intelligence and

functions of the prefrontal cortex. In A. Miyake

& P. Shah (Eds.), Models of working memory:

Mechanisms of active maintenance and executive

control (pp. 102–134). New York: Cambridge

University Press.Engle, R. W., Tuholski, S. W., Laughlin, J. E., &

Conway, A. R. A. (1999). Working memory,

short-term memory and general fluid intelli-

gence: A latent variable approach. Journal of Ex-


Eriksen, C. W., & Murphy, T. D. (1987). Move-

ment of attentional focus across the visual field:

A critical look at the evidence. Perception and

Psychophysics, 42, 299–305.

Friedman, N. P., & Miyake, A. (2004). The readingspan test and its predictive power for reading

comprehension ability. Journal of Memory and

Language, 51, 136–158.

Haarmann, H. J., Davelaar, E. J., & Usher, M. (2003).

Individual differences in semantic short-term

memory capacity and reading comprehension.

Journal of Memory and Language, 48, 320–345.

Hasher, L., & Johnson, M. K. (1975). Interpretive

factors in forgetting. Journal of Experimental

Psychology: Human Learning and Memory, 1,567–575.

Hasher, L., & Zacks, R. T. (1988). Working

memory, comprehension, and aging: A review

and a new view. In G. H. Bower (Ed.), The

psychology of learning and motivation: Advances

in research and theory (Vol. 22, pp. 193–225).

San Diego: Academic Press.

Hitch, G. J., Towse, J. N., & Hutton, U. M. Z.

(2001). What limits children’s working memory

span? Theoretical accounts and applications for

scholastic development. Journal of Experimen-

tal Psychology: General, 130, 184–198.

Jensen, A. R. (1998). The g factor: The science of

mental ability. Westport, CT: Praeger.

Jonides, J., Reuter-Lorenz, P. A., Smith, E. E., Awh,

E., Barnes, L. L., Drain, M., Glass, J., Lauber,

E. J., Patalano, A. L., & Schumacher, E. H.

(1996). Verbal and spatial working memory in

humans. Psychology of Learning and Motiva-tion, 35, 43–88.

Kail, R., & Salthouse, T. A. (1994). Processing speed

as a mental capacity. Acta Psychologica, 86,

199–225.

Kane, M. J., Bleckley, M. K., Conway, A. R. A., &

Engle, R. W. (2001). A controlled-attention view




of working-memory capacity. Journal of Experi-

mental Psychology: General, 130, 169–183.

Kane, M. J., & Engle, R. W. (2000). Working memory

capacity, proactive interference, and divided

attention: Limits on long-term memory retrieval.Journal of Experimental Psychology: Learning,


Kane, M. J., & Engle, R. W. (2002). The role of

prefrontal cortex in working-memory capacity,

executive attention, and general fluid intel-

ligence: An individual-differences perspective.

Psychonomic Bulletin & Review, 9, 637–671.

Kane, M. J., & Engle, R. W. (2003). Working-


The contributions of goal neglect, responsecompetition, and task set to Stroop interfer-


eral, 132, 47–70.

Kane, M. J., Hambrick, D. Z., & Conway, A. R. A.

(2005). Working memory capacity and fluid

intelligencearestronglyrelatedconstructs:Com-

ment on Ackerman, Beier, and Boyle (2005).

Psychological Bulletin, 131, 66–71.

Kane, M. J., Hambrick, D. Z., Tuholski, S. W.,

Wilhelm, O., Payne, T. W., & Engle, R. W.(2004). The generality of working memory

capacity: A latent-variable approach to verbal

and visuo-spatial memory span and reasoning.

Journal of Experimental Psychology: General,

133, 189–217.

Kane, M. J., Poole, B., Tuholski, S. W., & Engle,

R. W. (2003). Working memory capacity and

executive control in search and switching. Pre-

sented at the annual meeting of the Psycho-

nomic Society, Vancouver, Canada.

Kane, M. J., Poole, B. J., Tuholski, S. W., & Engle,

R. W. (2006). Working memory capacity and

the top-down control of visual search: Explor-

ing the boundaries of ‘‘Executive attention.’’



Klein, K., & Fiss, W. H. (1999). The reliability and

stability of the Turner and Engle working mem-

ory task. Behavior Research Methods, Instru-ments & Computers, 31, 429–432.

Kyllonen, P. C., & Christal, R. E. (1990). Reasoning

ability is (little more than) working-memory

capacity?! Intelligence, 14, 389–433.

Logan, G. D. & Bundesen, C. (2003). Clever

homunculus: Is there an endogenous act of con-

trol in the explicit task-cuing procedure? Journal

of Experimental Psychology: Human Perception

and Performance, 29, 575–599.


Working memory span and the role of proactiveinterference. Journal of Experimental Psycho-

logy: General, 130, 199–207.

MacDonald, A. W., III, Cohen, J. D., Stenger, V. A., &

Carter, C. S. (2000). Dissociating the role of

the dorsolateral prefrontal and anterior cingulate

cortex in cognitive control. Science, 288, 1835–

1838.

MacDonald, M. C., & Christiansen, M. H. (2002).

Reassessing working memory: Comment on Just

and Carpenter (1992) and Waters and Caplan(1996). Psychological Review, 109, 35–54.

McNamara, D. S., & Scott, J. L. (2001). Working

memory capacity and strategy use. Memory &

Cognition, 29, 10–17.

Melton, A. W. (1963). Implications of short-term

memory for a general theory of memory. Journal

of Verbal Learning and Verbal Behavior, 2, 1–21.

Miller, G. A. (1956). The magical number seven,

plus or minus two: Some limits on our capacity

for processing information. Psychological Re-view, 63, 81–97.


P., & Hegarty, M. (2001). How are visuospatial




eral, 130, 621–640.

Monsell, S., & Driver, J. (Eds.). (2001). Attention

and performance XVIII: Control of cognitive

processes. Cambridge, MA: MIT Press.

Nairne, J. S. (2002). Remembering over the short

term: The case against the standard model. An-

nual Review of Psychology, 53, 53–81.

Norman, D. A., & Shallice, T. (1986). Attention to

action:Willed and automatic control of behavior.

In R. J. Davidson, G. E. Schwartz, & D. Shapiro

(Eds.), Consciousness and self-regulation: Ad-

vances in research and theory (vol. 4, pp. 1–18).

New York: Plenum Press.Pearson, D. G. (2001). Imagery and the visuo-spatial

sketchpad. In J. Andrade (Ed.), Working mem-

ory in perspective (pp. 33–59). Hove, UK:

Psychology Press.

Rogers, R. D., & Monsell, S. (1995). The cost of a

predictable switch between simple cognitive




tasks. Journal of Experimental Psychology: Gen-

eral, 124, 207–231.

Rosen, V. M., & Engle, R. W. (1998). Working

memory capacity and suppression. Journal of

Memory and Language, 39, 418–436.Salthouse, T.A., & Craik, F.I. M.(2001). Closing com-

ments. In F. I. M. Craik & T. A. Salthouse (Eds.),

The handbook of aging and cognition (2nd ed.).

Mahwah, NJ: Lawrence Erlbaum Associates.



and language processing: An individual differ-

ences approach. Journal of Experimental Psy-


Toth, J. P., & Hunt, R. R. (1999). Not one versusmany, but zero versus any: Structure and func-

tion in the context of the multiple memory

systems debate. In J. K. Foster & M. Jelicic (Eds.),

Memory: Systems, process, or function? (pp. 233–

272). Oxford, UK: Oxford University Press.

Treisman, A. M., & Gelade, G. (1980). A feature-

integration theory of attention. Cognitive Psy-

chology, 12, 97–136.

Tuholski, S. W., Engle, R. W., & Baylis, G. C.

(2001). Individual differences in working mem-ory capacity and enumeration. Memory & Cog-

nition, 29, 484–492.

Turley-Ames, K. J., & Whitfield, M. M. (2003).

Strategy training and working memory task

performance. Journal of Memory and Language,

49, 446–468.

Turner, M. L., & Engle, R. W. (1989). Is work-

ing memory capacity task dependent? Jour-

nal of Memory and Language, 28, 127–154.

Underwood, B. J. (1975). Individual differences as a

crucible in theory construction. American Psy-chologist, 30, 128–134.

Underwood, B. J., & Ekstrand, B. R. (1967). Studies

of distributed practice XXIV: Differentiation

and proactive inhibition. Journal of Experimen-

tal Psychology, 74, 574–580.

Unsworth, N., & Engle, R. W. (2006). Verbal working

and short-term memory spans and their relation

to fluid abilities: Evidence from list-length

effects. Journal of Memory and Language, 54,

68–80.Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle,

R. W. (2005). An automated version of the

operation span task. Behavioral Research Meth-

ods, 37, 498–505.

Wolfe, J. M. (1994). Guided search 2.0: A revised

model of visual search. Psychonomic Bulletin

and Review, 1, 202–238.

Wood, N., & Cowan, N. (1995). The cocktail party

phenomenon revisited: How frequent are at-

tention shifts to one’s name in an irrelevantauditory channel? Journal of Experimental Psy-

chology: Learning, Memory, and Cognition, 21,

255–260.

Wundt, W. (1912/1973). An introduction to psychol-

ogy. Originally published by G. Allen, London.

Reprinted by Arno Press, New York.




3

Individual Differences in Working Memory

Capacity and Reasoning Ability

KLAUS OBERAUER, HEINZ-MARTIN SUß,

OLIVER WILHELM, and NICOLAS SANDER

A substantial number of studies have shown thatworking memory capacity (WMC) is the bestsingle predictor identified so far of reasoningability as measured by intelligence tests (Ack-erman, Beier, & Boyle, 2002; Colom, Flores-Mendoza, & Rebollo, 2003; Conway, Cowan,Bunting, Therriault, & Minkoff, 2002; Engle,Tuholski, Laughlin, & Conway, 1999; Kyllonen& Christal, 1990; Oberauer, 1993; Suß, Ober-auer, Wittmann, Wilhelm, & Schulze, 2002; forreviews see Ackerman, Beier, & Boyle, 2005;Oberauer, Schulze, Wilhelm, & Suß, 2005;Kane, Hambrick, & Conway, 2005). This robustfinding is an important step toward under-standing psychometric intelligence in terms of theories from cognitive psychology. How muchwe gain from such a link depends on how well

we can specify WMC itself. Much of ourgroup’s research, therefore, has been devoted tocapture more precisely—both on a conceptualand a measurement level—the variance of WMC that is responsible for its tight relation-ship to reasoning ability.

On the measurement level we feel that it isimportant to operationalize WMC by a varietyof heterogeneous indicators (i.e., tasks). Thishelps to delineate the generality and bound-aries of the construct and to investigate itsfactorial structure. When we conducted afactor-analytic study on a large sample of tasksused in the literature to assess WMC and re-lated constructs (Oberauer, Suß, Schulze,Wilhelm,&Wittmann,2000),weweresurprisedby the large commonality among many of thesetasks. This is matched by the high generality of the reasoning factor in structural models of intelligence. Our analyses show that the rela-tionship between measures of WMC and of reasoning is highest on a high level of aggre-gation. This implies that the common variance

among a broad set of different working mem-ory tasks, not the specific variance of one taskor task family, is related to reasoning ability,where reasoning ability also stands for thecommon variance of a diverse set of reasoningtasks.

49



On a conceptual level, we are looking fora theoretically meaningful characterization of WMC as a limiting factor for reasoning ability. Itshould capture the requirements shared by rea-

soning tasks and those tasks in the workingmemory literature that are good predictors of reasoning, while at the same time excludingtasks only weakly related to reasoning ability.Our current working hypothesis is this: work-ing memory capacity reflects the ability to keepseveral chunks of information simultaneouslyavailable for direct access. The critical under-lying mechanism for this ability is the flexible,temporary binding of chunks to positions in a

common cognitive coordinate system. The bind-ing to a position provides a cue, or ‘‘address,’’ bywhich the chunk can be accessed as input for acognitive operation. The common coordinatesystem provides the basis for the construction of new relations between the chunks. We believethat the construction of new structures out of representational elements is the common re-quirement of all reasoning tasks that sets themapart from other tasks not loading highly on the

reasoning factor. Working memory capacity is ageneral limiting factor for the generation of newstructural representations; therefore, it is a criti-cal source of variance for all reasoning tasks.

THEORETICAL FRAMEWORK

Our theoretical model of working memory ismainly inspired by the work of Cowan (1988;1995) and of Halford and colleagues (Halford,Wilson, & Phillips, 1998). We think of workingmemory as a system responsible for makingrepresentations in long-term memory (LTM)available for intentional (i.e., goal-directed) pro-cessing. As such, working memory is not a mem-ory system in itself, but a system for attention tomemory, an idea shared by several other authors(Baddeley, 1993; Engle, Kane, & Tuholski,

1999; Moscovitch, 1992). A schematic illustra-tion of the model is given in Figure 3.1.Memory representations are selected for pro-

cessing on three levels (Oberauer, 2002). Onthe first level, representations in LTM are ac-tivated above baseline. Activation has its source

in perceptual input and in representationscurrently held in the central parts of workingmemory, and it spreads along associations inLTM. The degree of activation can be regardedas a quick on-line ‘‘estimate’’ of the expectedrelevance of each representation for the currentgoal (Anderson & Lebiere, 1998). There is noinherent limit to the amount of activation al-located to memory representations, but activat-ing too many representations (includingirrelevant ones) can impair performance asmuch as activating too few.

On a second level, a subset of the activated

representations is held in the region of directaccess. This means that they are temporarilybound to positions in a cognitive coordinatesystem, which could be a literal representationof space, as when a chess player representsthe position of several pieces on the board in

Figure 3.1. The concentric model of working mem-ory. Nodes connected by continuous lines representthe network of associated representations in long-term memory; activated representations are high-lighted in black. The large, broken circle delineatesthe region of direct access, in which a few elementsare bound to positions in a cognitive coordinate sys-tem. In addition to the pre-existing associations, new

relations (dotted lines) can be established betweenthem. The small, continuous circle represents thefocus of attention that selects one element as theobject of a cognitive operation.




working memory. It can also be a temporal di-mension, as when a participant in a workingmemory experiment memorizes the temporalorder of words read to her. Much evidence on

the short-term recall of temporal serial ordersupports the assumption that elements of a se-ries are bound to positions on a temporal con-text representation (Burgess & Hitch, 1999;Henson, Norris, Page, & Baddeley, 1996). Veryoften a quasi-spatial medium is used to repres-ent nonspatial relations between elements, forexample, when comparative relations (e.g., ‘‘Jimis taller than John’’) are represented as spatialarrays of tokens, as was first suggested by De

Soto, London, and Handel (1965). The ideathat people use representations of space in ametaphorical way to capture nonspatial rela-tions has been elaborated on and supported bynumerous examples in cognitive linguistics(Lakoff & Johnson, 1980; Langacker, 1986). Notall relations between objects, events, and per-sons, however, can be mapped onto spatial re-lations, and so we assume that working memoryalso uses schemata to organize its contents.

Schemata are abstract templates for structuralrepresentations that contain placeholders forelements that can be bound to them to fill spe-cific roles. For example, elements can be boundto the roles of agent and object in causal-forceschemata (Talmy, 1988). Language processingalso involves temporary binding of sentenceconstituents into syntactic roles in a schema.This form of binding, however, seems to drawon a specialized mechanism that is not limitedby general WMC (see Chapter 11).

Currently, we have no fixed position on howthis binding in working memory is accomplished(for various proposals see Halford et al., 1998;O’Reilly, Busby, & Soto, 2003; Raffone & Wol-ters, 2001; Usher, Haarmann, Cohen, & Horn,2001). All current mechanisms for temporarybinding suggest that there is a limit to thenumber of independent information elements

(i.e., chunks) that can be kept separate andbound to different positions or argument slots atthe same time. We see this as the source of thecapacity limit of working memory. The limitingfactor for reasoning tasks thus arises from thelimited capacity to keep up several temporary

bindings simultaneously in the region of directaccess. The idea that WMC primarily limits thecomplexity of new structures has been forcefullyadvanced by Halford et al. (1998) on the basis of

a connectionist model of binding in workingmemory. Robin and Holyoak (1995) and Waltzet al. (1999) have argued for a link between re-lational integration and the prefrontal cortex (seealso Christoff et al., 2001). We borrow their termhere to refer to the ability to coordinate elementsinto new structures in the direct-access region of working memory.

A third component of the model is the focusof attention. Its role is to select one chunk at a

time from the contents of the direct-access re-gion as the object of a cognitive operation. Forinstance, when the task is to add two three-digitnumbers, the role of the direct-access region is tolink the six digits to their respective roles (e.g.,the ones of the first number, the tens of thesecond number). The role of the focus is to pickout one digit, increment it by the amount indi-cated by the second digit, and thereby transformits value to obtain the sum (Oberauer, 2003).

Our model is similar to that of Cowan (1995,1999) in that it distinguishes between the acti-vated part of LTM and a more central compo-nent with limited capacity. Unlike Cowan, wedifferentiate the central part of working mem-ory into the direct-access region, in which sev-eral elements can be held and related to eachother, and the focus of attention, which is lim-ited to a single element.1 We also make moreexplicit assumptions about the functions of the various components of the model and theirqualitative differences.Activationin LTMmarksa representation as likely to be relevant, thusmaking it easier to retrieve it (i.e., bring it intothe region of direct access). Moreover, activatedrepresentations can prime or bias cognitiveoperations and overt responses, for instance, bygenerating a feeling of familiarity in a recogni-tion task (Oberauer, 2001). Only the contents of

the direct-access region, however, provide re-lational representations as input to cognitiveoperations. For example, when a chess player’sdecision depends on whether the queen is tothe left or in front of the king, the two figuresmust be encoded in the direct-access region.

Differences in Working Memory Capacity and Reasoning Ability 51



The focus of attention serves to pick out the oneobject that is actually manipulated by a cogni-tive operation or an action—e.g., the playerneeds to decide whether to move the queen or

the king, and this selection is done by takingonly one figure into the focus of attention. Onefurther point on which we differ from Cowan(2001) is that we are not committed to the‘‘magical number 4’’ as a constant to describethe capacity of the direct-access region.

ANSWERS TO THEFOUR QUESTIONS

We now provide concise answers to the fourquestions posed by the editors, which will beelaborated in the remainder of this chapter.

1. The theoretical framework we found use-ful to integrate our results is the three-layer model of working memory outlinedabove (Oberauer, 2002), which is an ex-tension of the model proposed by Cowan

(1995). A second important source of ourwork is facet theory (Canter, 1985), inparticular the two-facet structure of theBerlin Intelligence Structure (BIS) model(Jager, Suß, & Beauducel, 1997). Facettheory can be regarded as the equivalentof an experimental design in correla-tional research: it is assumed that thevariance of any manifest (i.e., measured)variable is affected by the variance of several latent (i.e., hypothetical) variablesthat are levels of orthogonal factors(called facets). Our factor model of work-ing memory assumes two facets, one re-lated to content and the other to cognitivefunctions. On the content facet, we dis-tinguish between two sources of variance,one related to verbal-numerical contentand the other to visuospatial content. On

the functional facet, we differentiate stor-age and processing, relational integration,and executive functions. The latent var-iables can be represented as factors instructural equation models. Each mani-fest variable is assumed to reflect variance

from (at least) one content factor and onefunctional factor. A facet model providesa matrix for classifying existing manifestvariables and rules for guiding the con-

struction of new variables (i.e., test tasks)for specific cells of the matrix (for in-stance, theconstruction of a verbal storage-and-processingtask).Testtaskscanbecon-structed by varying their features accord-ing to the facets, just as experimentalconditions are varied according to a de-sign factor (e.g., changing the content tobe memorized from digits to spatial ma-trices corresponds to a variation in the

content facet). Whether manipulations of tests according to facet have an effect canbe investigated by testing whether themanipulation changes a task’s correlationwith other tasks. For instance, a storage-and-processing task with verbal contentshouldloadonaverbalfactor,whereasonewith spatial matrices as content shouldload on a spatial factor. For a review of facet models of intelligence see Suß and

Beauducel (2005).2. We believe that the critical source of

individual differences in WMC is the abil-ity to provide direct access to several inde-pendent information elements (chunks)at the same time. This capacity rests ona mechanism that quickly establishes anddissolves temporary bindings betweenthese elements and positions in a cogni-tive coordinate system, or placeholders ina schema. Direct access to a multitude of separate elements is necessary to constructnew relations between them and integratethem into new structural representations.The limited capacity for relational inte-gration is the most important limitingfactor for reasoning ability. We focus onthis source of variance because it seems tobe a parameter of the cognitive system that

affects a large number of different tasks,thereby explaining the common varianceof many experimental working memorytasks, reasoning tasks from intelligencetests, and potentially complex cognitiveachievements in everyday life.




3. One other source of individual differencesin cognition that has attracted much at-tention recently is the efficiency of exec-utive functions. Our results (Oberauer,

Suß, Wilhelm, & Wittmann, 2003) sug-gest that at least one prototypical execu-tive function, switching between task sets,is not strongly related to WMC. We ven-ture that this is not an exception. We thinkthat the capacity of working memorycannot be reduced to the efficiency of ex-ecutive functions, and currently there isno compelling evidence that executivefunctions contribute substantial variance

to reasoning ability. Another construct, related to executive

functions but not identical to it, is the abil-ity to inhibit irrelevant representations(Hasher, Zacks, & May, 1999, see alsoChapter 9). We believe there is someevidence that this source of variancecontributes to individual differences inworking memory and other complexcognitive functions (e.g., reading ability).

The role of the inhibition functionwithin our model is to reduce the activa-tion of irrelevant representations in LTM,thereby reducing the amount of intrusioninto the more central parts of workingmemory (Oberauer, 2001). Inhibition ef-ficiency, however, cannot account for thecommon variance of all working memorytasks.

4. Factor-analytic studies of individual dif-ferences can help us to figure out whichcognitive functions belong closely to-gether and which are relatively indepen-dent of each other; this is an importantcontribution to the shaping and sharpen-ing of concepts such as working memory.Our research thus far has yielded twoimportant insights. First, working mem-ory has a limited capacity that constrains

cognitive performance in a wide varietyof tasks. Their common feature is prob-ably that they all require simultaneousaccess to several independent elementsof information. Second, WMC cannot bereduced to the efficiency of executive

functions. Therefore, we argue that work-ing memory and executive functions betreated as separate constructs.

THE FACTOR STRUCTUREOF WORKING MEMORY

The factor-analytic approach to individual dif-ferences provides a tool to simultaneously iden-tify associations and dissociations between indi-cators of cognitive functions. These indicatorscan be raw performance scores in tasks designedto tap a particular function, or derived measures

(e.g., task-switching costs or interference ef-fects) used to assess specific functions. Our ap-proach is to model the correlational structure of large sets of indicators by theoretically specifiedstructural equation models. We regard this asan important complement to experimental andneuropsychological research, which usually fo-cuses on dissociations of cognitive functions.Through factor analysis one can test the as-sumptions of generalizability, which is implicit

in all experimental work that uses a specificparadigm. By investigating which indicators of cognitive functions are highly correlated andwhich are not, we can assess whether a phe-nomenon observed in an experimental para-digm is in fact representative of the construct itis meant to operationalize.

In two large factor-analytic studies (Ober-auer et al., 2000; Oberauer et al., 2003) wetested a facet model of the structure of workingmemory. The basic framework was borrowedfrom the Berlin Intelligence Structure (BIS)model (Jager et al., 1997). It specifies two di-mensions by which cognitive performances, orthe indicators used to measure them, can beclassified. One distinguishes content (verbal,numerical, and spatial-figural materials), theother, cognitive functions. In the BIS, cognitivefunctions are described as processing capacity

(i.e., reasoning), creativity (i.e., fluency of ideas),memory (i.e., remembering supra-span sets of items for brief periods), and processing speed. Inthe working memory model, we distinguishedbetween three functions: simultaneous stor-age and processing, coordination (i.e., relational




integration),andsupervision (i.e.,executivefunc-tions).

In the first study (Oberauer et al., 2000) weused a large pool of tasks sampled from the

literature (up to 1995) in an attempt to cover thewhole range of measures proposed by scholars tocapture working memory and executive func-tions. Among the tasks were classics such asreading span (Daneman & Carpenter, 1980) andrandom generation (Baddeley, 1986), tasks usedin research on cognitive aging such as memoryupdating (Salthouse, Babcock, & Shaw, 1991),and tasks developed to measure relational inte-gration (i.e., ‘‘coordination’’; Oberauer, 1993),

together with indicators of executive functionssuch as task-set switching (Allport, Styles, &Hsieh, 1994).

Since most of these tasks were complexmixtures of various cognitive functions, we at-tempted a more analytical approach in the sec-ondstudy(Oberaueretal.,2003).Weconstructednew tasks designed to reflect specifically thethree functions proposed in our model: simul-taneous storage and processing, relational inte-

gration, and supervision of cognitive processes.We do not claim that we succeeded in con-structing pure indicators of the constructs weintended to measure, but our new tasks argu-ably reflected the cognitive functions we tar-geted more than other functions.

First, we chose to reduce the measures of ‘‘simultaneous storage and processing’’ to the es-sence of what this construct description requires,that is, processing some information while con-currently memorizing a briefly presented list of items. We constructed dual-task combinationsof memory for serial order and choice reaction-time (RT) tasks. First, the memory lists werepresented, then participants worked throughseveral trials of a two-choice RT task for 5 s, andfinally they reproduced the memory list. Thistask schema was realized with verbal, numeri-cal, and spatial materials.

We attempted to construct tasks that requirerelational integration but no storage, in that noinformation had to be kept available when itwas no longer present in the environment. Inthese so-called structure-monitoring tasks, par-ticipants observed a set of objects on the com-puter screen, which changed independently

from time to time. The task was to detect theemergence of specific relations between theobjects. In the finding squares task, for exam-ple, participants saw 10 dots randomly placed

in fields of a 1010 matrix. Every 1.5 s two dotsjumped to a new position. Participants had toindicate by a key press when four of the dotsformed a square somewhere in the matrix (seeFig. 3.2, left panel). This task required re-presenting the ever-changing relations betweenthe dots and integrating them into larger struc-tures to detect patterns that form a square (or apartialsquare,suchthatattentioncanbedirectedto the critical positions in which the appear-

ance of a dot would complete a square). Weconstructed structure-monitoring tasks withverbal, numerical, and spatial content. For eachtask there was one version without need to mem-orize any information, and a complementaryversion in which some information was erasedfrom the screen briefly after presentation sothat it had to be memorized.

As a measure of executive function we usedfour versions of the task-set switching paradigm

introduced by Rogers and Monsell (1995). Thisparadigm is used to compare pure blocks, inwhich a single task is repeated, with mixedblocks, in which two tasks alternate every secondtrial. We derived two indicators for the efficiencyof executive function. Specific switch costs weredefined as the RT difference between trials fol-lowing a switch and trials following no switchwithin the mixed blocks. General switch costswere defined as the RT difference between no-switch trials in the mixed blocks and trials in thepure blocks.

The major results of the two studies werethe following:

1. The factor representing storage and pro-cessing and the relational-integration fac-tor were highly correlated and in one case(Oberauer et al., 2000) indistinguishable.

Most of their variance was shared, and weregard this shared variance as the core of working memory capacity.

2. The factors representing indicators of su-pervision or executive functions were onlyweakly correlated with the other two work-ing memory factors.




3. Verbal and numerical working memoryloaded on the same factor. The factors forverbal-numerical and for spatial workingmemory were highly correlated, but asmall proportion of their variance seemsto be specific for each domain.

4. As in other studies (Ackerman et al.,2002; Kyllonen & Christal, 1990; Kyllo-nen, 1994, but see Conway et al., 2002),WMC was highly correlated with factorsreflecting processing speed. Nonetheless,speed variables always formed a distinctseparate factor.

One representative model is displayed inFigure 3.3 (for factor loadings see Table 3.1).This model is based on the data from Oberaueret al. (2003), integrating two models reported inthat article into one. The model had a satisfac-tory fit, with w

2 (141)¼ 198.7, CFI¼ .966, and

RMSEA ¼ .056, despite the lack of content-specific factors, which did little to improvethe fit. The right part of the model representsthe choice RT variables from the pure and themixed blocks. A general speed factor capturesthe variance common to all choice RT mea-sures. A more specific factor captures the re-sidual variance shared by all variables takenfrom the mixed blocks; it thereby represents thecommon variance of general switch costs. A third, even more specific factor represents thevariance shared by only the switch RTs, thusreflecting the common variance associated with

specific switch costs. The two factors reflect-ing general and specific switch costs have sub-stantial loadings from most of their variables,supporting the hypothesis that these switchcosts have something in common beyondparticular tasks and even across content do-mains.

Reproduce

7

4

8

?

?

... (more probes)

... (more displays)hit

... (continued)

Figure 3.2. Schematic exampletrials from working memory tasks.Left column: Finding squares task(Oberauer et al., 2003). Middlecolumn: Spatial short-term memorytask (Oberauer, 1993). Note that acorrect response requires onlyreproduction of the relationsbetween dots, not their absolute

locations in the matrix. Rightcolumn: Memory updating(numerical) task, STM version(modified from Salthouse et al.,1991).




The tasks we used to measure relationalintegration mostly required establishing a struc-ture within a spatial coordinate system. For ex-ample, two tasks developed by Oberauer (1993),called spatial short-term memory (STM) andspatial coordination, display a sequence of dotsin different cells of a 1010 matrix (see Fig.3.2, middle panel). The dots are displayed oneby one, and the participants are asked toimagine how they would look if they appeared

simultaneously and reproduce the resulting pat-tern in an empty matrix. This can be doneeither by an absolute coding of each dot’s po-sition, based on bindings between each dot andits position in a representation of the matrix, orby a relative coding, based on binding dots toargument slots in schemata such as TWO-

CELLS-LEFT-OF(x,y). Both approaches work,because reproduction of an absolute position isnot asked for in the instructions. Both ap-proaches require quickly establishing tempo-

rary bindings.In the verbal monitoring task designed byOberauer et al. (2003), participants see a 3 3matrix with one word displayed in each cell.Every 2 s, one of the words is replaced by a newone. Participants have to signal with a key presswhenever three words in a horizontal, vertical, ordiagonal row rhyme with each other. To detectthis structure, one has to bind the words to theirspatial positions in the matrix. It is not sufficient

to activate the representations of all currentlypresented words and to detect that some of themrhyme. It is also necessary to be aware of therhyming words’ (relative or absolute) positions.Once again, accurate performance depends onbinding elements to positions in a cognitive co-ordinate system, and efficient performance re-quires that subjects build up and resolve bindingsquickly.

Storage and Processing Tasks

Performance on tasks measuring storage andprocessing is usually assessed by the accuracyof recall of a list in serial order. This requiresmemory not only for the particular items in thelist but also for their ordinal position in thelist. One way to accomplish this is to bind listelements temporarily to positions in a mentalcoordinate system in which one dimension rep-resents temporal order. Several computationalmodels of serial recall use a mechanism that linkslist elements to a context representation chang-ing over time (Brown, Preece, & Hulme, 2000;Burgess & Hitch, 1999; Henson, 1998). The ef-ficiency of a mechanism that establishes tempo-rary bindings can therefore be expected to becritical for the accuracy of serial recall.

This account raises one question: there is

ample evidence that serial recall tasks combinedwith a concurrent processing task (so-calledcomplex spans) are better predictors of reason-ing and other complex tasks than serial recallalone (so-called simple spans) (Ackerman et al.,2005;Daneman&Merikle,1996;Engle,Tuholskiet al., 1999; Kane et al., 2005; Kane et al., 2004;

TABLE 3.1. Factor Loadings of Variables inthe Model in Figure 3.3

Variable Speedgen.SC

spec.SC S&P Rel.Int

Switch-v .72 .18 .29Switch-n .65 .34 .53Switch-s1 .66 .18 .42Switch-s2 .69 .17 .31Noswitch-v .76 .34Noswitch-n .69 .57Noswitch-s1 .68 .27Noswitch-s2 .73 .10CRT-v .78CRT-n .76CRT-s1 .77

CRT-s2 .71S&P-v .47S&P-n .50S&P-s1 .76S&P-s2 .64Rel.Int-v .60Rel.Int-n .69Rel.Int-s1 .61Rel.Int-s2 .60

All loadings were significant with p< .05, except the loading of switch-s1 on the general switch cost factor (p¼ .052). N¼ 131,

w

2¼

198.7, df ¼

141. Errors were uncorrelated except S&P-v withS&P-n (.38) and Rel.Int-v with Rel.Int-n (.41). Switch¼ choice

reaction times from switch trials in mixed blocks; v¼ verbal; n¼numerical; s1, s2¼ two versions of spatial material; Noswitch¼choice reaction times from no-switch trials in mixed blocks;

CRT¼ choice reaction times from pure blocks; S&P¼memoryperformance from dual-task combinations of storage and concur-rent processing; Rel.Int¼ structure-monitoring tasks used to mea-

sure relational integration.




Oberauer, 1993; Oberauer et al., 2005), at leastin the verbal domain. (with spatial material theavailable data are more ambiguous—Miyake,Friedman, Rettinger, Shah, & Hegarty, 2001,

could not differentiate simple and complex spa-tial span tasks; Kane et al., 2004, could separatethem, but spatial simple span tasks contributedunique variance to explaining fluid intelligenceover and above a domain-general complex-span factor). Why should this be the case? Oneplausible explanation is that there is a specializedmechanism for recall in forward serial order,which is disrupted by concurrent processing, sothat the system can rely on it for simple span tasks,

but has to fall back onto the more generalmechanism of temporary binding for complexspan tasks.

One such specialized mechanism could be aprimacy gradient, as proposed by two furthercomputational models of serial recall (Farrell &Lewandowsky, 2002; Page & Norris, 1998). Inthese models, activation or encoding strength of each successive list item gradually decreases asthe list progresses, and the resulting gradient is

used to reconstruct the order of encoding. Sucha mechanism requires no binding. Its limita-tion, however, is that every event interveningbetween two list items or occurring after listpresentation could easily leave behind a repre-sentation with its own activation (or strength)level.Theactivationof representations decreasesonly slowly once they become irrelevant (Ober-auer, 2001), therefore, their activation is likelyto distort the primacy gradient at recall. Hence,at recall, keeping the relevant list items from thenow irrelevant material separate from the in-tervening processing task requires the bindingmechanism after all, at least to bind list items toone context and representations from the pro-cessing task to another.

A variant of this explanation is to assume aspecialized system for keeping the serial orderonly of phonemes—a ‘‘phonological loop’’ in the

model of Baddeley (1986). The processing task,introducing additional phonological material(e.g., reading sentences aloud in reading span),would interfere with this system, leaving recall of the memory list at the mercy of the generalbinding mechanism (Oberauer, Lange, & Engle,

2004). This variant would be specific to verbalmaterial, consistent with Miyake et al.’s (2001)findings, who did not obtain a distinction be-tween simple and complex spans with spatial

material.Our theory predicts that the addition of a con-current processing component is just one way toturn a simple span task into a working memorytask (i.e., a powerful predictor of complexcognitive performance), and it might not evenbe the best way. Another way to disrupt thespecialized mechanism for forward recallwould be to require recall of a list in randomorder. This is what we did in an adaptation of

the (numerical) memory updating task de-signed by Salthouse et al. (1991). In what wecalled the ‘‘STM version’’ of this task, partici-pants saw a number of digits, each in a separateframe on the screen. Later a subset of theframes was probed for recall in random order(see Fig. 3.2, right panel). No processing had tobe done in the meantime. What makes this taskdifferent from ordinary serial recall is the al-location of the list elements to spatial positions

and the probing of recall in an order that dif-fered from presentation order. This task had veryhigh loading on a working memory factor and afirst-order correlation with the BIS reasoningscaleof r ¼ .40 (Oberaueret al., 2000), not muchless than tasks combining storage and proces-sing (e.g., reading span: r ¼ .56).

Although this result is certainly preliminaryand in need of replication, it gains additionalsupport by the fact that several spatial tasks re-quiring only storage but no processing also didvery well as predictors of reasoning. For exam-ple, the spatial STM task (Fig. 3.2, middlepanel) was one of the best single predictors of the BIS reasoning factor in three studies(Oberauer, 1993, Suß et al., 2000, 2002), withfirst-order correlations ranging from .54 to .59.Likewise, Miyake et al. (2001) found that onespatial task measuring short-term retention

without additional processing, the dot memorytask, was as highly correlated with two spatialreasoning tasks (termed ‘‘spatial visualization’’)as were the complex span tasks combiningstorage and processing (r .40). In both thespatial STM and dot memory tasks, participants




have to memorize the locations of several dots ina matrix, presented either sequentially or si-multaneously. Arguably, this taps directly intothe ability to integrate spatial relations between

the dots within a common coordinate system.Future research will have to disentangle whe-ther it is the spatial nature of these tasks or thefact that they require relational integration thatmakes them good predictors of reasoning ability.

WORKING MEMORY CAPACITYAND EXECUTIVE FUNCTIONS

Several authors believe there is a very stronglink between WMC and the efficiency of ex-ecutive functions. Within the framework of Baddeley’s (1986) theory, the capacity of thecentral executive is the most natural candidateto account for the relationship between WMCand reasoning ability (Bayliss, Jarrold, Gunn, &Baddeley, 2003). A strong association betweenWMC, reasoning ability (or fluid intelligence,which is almost the same in practice), and the

executive functions of the prefrontal cortex hasbeen postulated by several authors (see Chap-ters 2 and 10).

The term executive function is often used ina loose and very encompassing way, sometimesincluding all cognitive operations performedon some stimuli or memory contents. Here weuse the term for the set of cognitive processesthat serve to keep cognition and action in linewith the current primary goal. Thus, we dis-tinguish between primary cognitive operations,which work on the representation currentlyheld in the focus of attention according to a setof parameters (called a task set), and executiveprocesses that ensure that the primary opera-tions function properly in the service of theperson’s overarching goal (cf. Chapter 7). Forthe present purpose, we distinguish betweentwo kinds of executive processes: (1) cognitive

processes that control and manipulate the tasksets, and (2) processes that inhibit irrelevant,potentially distracting representations. Brieflystated, the former ensure that the right task isexecuted, and the latter make sure that it isexecuted with the right information. The two

categories can, of course, overlap, for instance,when inhibition is applied to task sets them-selves (Mayr & Keele, 2000). A third categoryof processes that is often subsumed under ex-

ecutive functions is the updating of workingmemory (Miyake et al., 2001). The tasks usedto measure this process combine storage andprocessing in working memory (in fact, theyare virtually identical to the ‘‘memory updat-ing’’ tasks used in our studies), so we regardperformance on these tasks as reflecting WMCand not executive functions.

Control of Task Sets

Task-set switching (Allport et al., 1994; Rogers &Monsell, 1995) is clearly a prototypical case of the first category of executive functions, themanipulation of task representations. Currentlythere are two major theoretical interpretationsof the time costs of switching between two tasks(i.e., specific switch costs). One, advanced byRogers and Monsell (1995), is that these costsreflect the time needed for a process of task-set

reconfiguration. Thus, specific switch costs di-rectly reflect the speed of one executive process.The other view, first proposed by Allport et al.(1994), sees these costs as a reflection of proac-tive interference from the old task set when ithas to be replaced by a new one. Specific switchcosts can then be interpreted as the efficiencywith which the cognitive system overcomes thisproactive interference. Overcoming proactiveinterference from no-longer relevant task setscan be accomplished by inhibiting the old taskset, boosting activation of the new task set, or amixture of both (for a computational model of task-set switching using these mechanisms seeGilbert & Shallice, 2002). This reads like aperfect job description for the ‘‘supervisory at-tentional system’’ (Norman & Shallice, 1980;Shallice & Burgess, 1991), which Baddeley(1986) used as the blueprint for his ‘‘central

executive.’’Therefore, we regard specific task-set switch-ing costs as one of the least disputable means of measuring executive functions available today.Moreover, general switch costs can best be in-terpreted as reflecting the need to coordinate




two task sets held available at the same time,whichwouldappeartobeatypicalassignmentforthe central executive. The finding that neitherspecific nor general switching costs correlate

substantially with WMC factors is therefore aserious challenge for theories assuming a stronglink between executive functions and workingmemory.

One might dismiss this finding as beingexceptional. First, it could be argued that thespecific way in which we implemented task-setswitching was responsible for our low correla-tions. This seems not to be the case. Other au-thors who investigated individual or age differ-

ences in switching costs, using different variantsof the paradigm, also reported weak to non-existent correlations of switching costs withworking memory tasks such as operation span(Miyake et al., 2000), backward digit span (Ce-peda, Kramer, & Gonzalez de Sather, 2001), or acomposite of three complex span tasks (Kray &Lindenberger, 2000). Recently, Friedman et al.(2006) have shown that task-set switching(‘‘shifting’’ in their terminology) is not signifi-

cantly related to fluid intelligence, furtherstrengthening the case that WMC and task-setswitching measure different constructs.

Second, it could be argued that switching isan atypical representative of the concept of ex-ecutive functions, and other indicators of thisconstruct correlate better with WMC mea-sures. There is indeed an impressive number of studies showing that complex span tasks arerelated to indicators of executive functions suchas speed and errors in the antisaccade task (Kane,Bleckley, Conway, & Engle, 2001; Unsworth,Schrock, & Engle, 2004), Stroop interference inblocks with few conflict trials (Kane & Engle,2003), and verbal fluency tasks (Rosen & Engle,1997). Establishing a strong link between WMCand executive functions, however, requiresmuch more. First, one would have to show thatdifferent measures of executive function corre-

late high enough to warrant conceptualizingthem through a single overarching construct(Salthouse, Atkinson, & Berish, 2003). The stud-ies by Miyake et al. (2000) and Salthouse et al.(2003) have provided some preliminary evi-dence in that direction, but others have failed tofind substantial associations between indicators

ofexecutivefunction(Duncan,Johnson,Swales,& Freer, 1997; Kramer, Humphrey, Larish,Logan, & Strayer, 1994; Shilling, Chetwynd, &Rabbitt, 2002; Ward, Roberts, & Phillips, 2001).

Second, one would have to demonstrate that afactor representing this construct is correlatedwith a WMC factor to a degree that goes beyondthe moderate positive correlation to be expectedfrom the ‘‘positive manifold,’’ that is, the ten-dency toward positive correlations among allcognitive tasks. The case for such a strong con-nection between WMC and a general executivefactor has not been made so far, and we doubtthat it can be made.

Inhibition of Irrelevant Information

One potential source of individual differences incomplex cognition is the efficiency with whichirrelevant or no-longer relevant information isinhibited. The inhibition hypothesis was firstadvanced by Hasher and Zacks (1988) in thecontext of cognitive aging, but was soon appliedto individual differences within one age group as

well (e.g., Conway & Engle, 1994). There is nowconsiderable evidence consistent with the as-sumption that the efficiency of inhibiting irrele-vant information is related to measures of WMC(Conway, Cowan, & Bunting, 2001; Gray,Chabris, & Braver, 2003; Lustig, May, & Hasher,2001; May, Hasher, & Kane, 1999). Moreover,De Beni and colleagues (De Beni, Palladino,Pazzaglia, & Cornoldi, 1998; De Beni & Palla-dino, 2000) linked inhibition of irrelevant wordsin a working memory task to performance in textcomprehension (see also Meiran, 1996). Therehas been some progress toward establishing aninhibition factor that captures the common var-iance of several measures of cognitive inhibi-tion (Friedman & Miyake, 2004), but it has notyet been established that this factor is substan-tially related to a factor reflecting WMC. In thestudy of Friedman et al. (2006), a latent factor

measuring inhibition was not correlated withfluid intelligence, a finding that should dampenour expectations of finding a strong relationshipwith WMC.

One scenario that deserves serious consider-ation is that correlational links between mea-sures of WMC and indicators of inhibition exist




on a relatively low level of generality. For in-stance, it is theoretically plausible that the abil-ity to inhibit irrelevant memory representationsis helpful in a task such as reading span, where

one must recall the sentence-final words butnot words from the remainder of the sentence(DeBenietal.,1998).Butthesameabilitymightbe much less important in a task such as ourspatial STM task (Oberauer, 1993), in whichall information presented on each trial hasto be recalled. On the other hand, the abilityto inhibit an already initiated action (Logan &Cowan, 1984) might contribute to variance inthe structure-monitoring tasks of Oberauer et al.

(2003) by helping to reduce false alarms (i.e.,pressing the key when the required structure isabsent), but be unrelated to complex span tasksthat require no suppression of action. This high-lights the importance of distinguishing levelsof generality when it comes to relating con-structs (Wittmann, 1988), which we will returnto again in the section on working memory andreasoning.

INTERLUDE: WORKING MEMORYCAPACITY AND SPEED

In many studies, WMC was found to be stronglyrelated to measures of processing speed (Acker-man et al., 2002; Kyllonen & Christal, 1990;Kyllonen, 1994; Salthouse, 1991, 1994, see alsoChapter 8), and this is also apparent in Fig-ure 3.3. One obvious potential explanation isthat many working memory tasks are complexspan tasks that involve a processing componentthemselves, and the speed of performing thiscomponent is one source of variance in com-plex span tasks (see Chapters 5 and 6). How-ever, this cannot be the whole story, becausenot all WMC tasks are complex span tasks, andspeed measures are also correlated to workingmemory tasks lacking a processing component.

A clear example of this is the study by Conwayet al. (2002), who found a substantial correla-tion of speed with a factor that extracted thecommon variance of short-term and workingmemory tasks, but none with a second factorcapturing the specific variance of workingmemory. This is the opposite of what one would

expect if speed is related to complex spanthrough the processing component of the latter.One might even speculate on the basis of Conway et al.’s findings that the WMC–speed

relationship is due to the STM component incomplex span tasks. A high correlation withspeed indicators, however, was also obtainedwith relational integration tasks (see Fig. 3.3),part of which have no STM component. Takentogether, these findings suggest that the linkbetween speed and WMC cannot be reduced toshared components between tasks used tomeasure the two constructs. We should look foran explanation for this relationship on a higher

level of generality.We believe that the binding account of

WMC advanced here can offer such an expla-nation, although at present it is largely specu-lative. Speed tasks require the fast execution of arule mapping categories of stimuli onto cate-gories of responses. Often, these rules must beimplemented ad hoc according to the instruc-tions (e.g., when a left key press is to be made inresponse to the letter A and a right key press in

response to a B). Several attention theorists haveargued that these tasks are executed as a ‘‘pre-pared reflex’’ (Logan, 1978) once a task set oraction plan incorporating the rule has been setup. The task set specifies all parameters for theaction to be executed in advance, except thosethat depend on the stimulus, and then the stim-ulus can immediately trigger the action by ‘‘di-rect parameter specification’’ (Neumann &Klotz, 1994).

The task set or action plan implements thestimulus–response (S-R) mapping in a way thatmakes it immediately executable in reactionto an appropriate stimulus. Thus, the task setdiffers from a representation of the rule oneacquires, for instance, from reading the in-structions. Knowledge of a rule does not in itself makeanorganismpreparedtoreacttoastimulusaccording to the rule. Once a task set is estab-

lished, however, an appropriate stimulus acti-vates the response mapped to it automatically(Logan & Schulkind, 2000), although this ac-tivation might not be sufficient for execution(Lien & Proctor, 2002). For a fast and accurateresponse to a stimulus it is therefore critical thatthe task set be firmly established, such that an




incoming stimulus can ‘‘pass through’’ withoutrequiring extra cognitive work.

A task set is a temporary binding between rep-resentations of the relevant stimulus categories

and representations of the corresponding re-sponses (c.f. Hommel, 1998), which is probablyheld in working memory. The bindings must betemporary because they can overcome estab-lished long-term associations (e.g., one can re-spond to the letter A by saying ‘‘B’’ and the otherway round), without having to unlearn these as-sociations (i.e., after executing the A-B mapping athousand times, the letter A is still associatedmore with saying ‘‘A’’ than with saying ‘‘B’’). The

ability to set up temporary bindings between ar-bitrary representations, which we argue underliesWMC, should therefore also be highly relevantfor efficient execution of speeded responses tostimuli according to S-R-mapping rules. A highcorrelation between WMC measures and speedin such tasks is therefore to be expected.

This argument suggests one important pre-diction: thecorrelation between WMC tasks andspeed tasks should depend on the degree of S-R

compatibility in the speed task. According toKornblum, Hasbroucq, and Osman (1990), S-R compatibility is high when (a) there is a highdegree of dimensional overlap between stimuliand responses, and (b) stimuli are mapped tothose responses that correspond to them on thedimensions they share. For instance, when thetask is to press a button in response to one of fourlights arranged in a horizontal line, an arrange-ment of the four response keys in a horizontalline would be one with high dimensional over-lap. Given this overlap, a mapping in which theleft–right order of the lights corresponds tothe left–right order of the keys to be pressed isone with a maximum of S-R compatibil-ity. A random mapping would be an example of low S-R compatibility, although the dimen-sional overlap is still high. Another way to havelow S-R compatibility is to eliminate dimen-

sional overlap, for example, by having partici-pants call out one of four arbitrarily chosenfemale names in response to each light.

Kornblum et al. (1990) assume two process-ing paths from a stimulus representation to aresponse. One is mediated by pre-existing as-sociations between stimulus and response rep-

resentations (e.g., between the printed letter Aand saying ‘‘A’’) and enables a stimulus to auto-matically activate its corresponding response.This path is operative only in cases with di-

mensional overlap between stimulus and re-sponse. The second path is mediated by a repre-sentation of the task rule, and can implementany arbitrary S-R mapping. The second path isalways operative and determines which re-sponse is correct. When the S-R mapping iscompatible, this is just a quick verification of the already activated response; when it is incom-patible, the activated response must be sup-pressed and replaced by the correct one. In

case of no dimensional overlap, the secondpath does all the S-R translation on its own. Inour interpretation, the second path is mediatedthrough a task set implemented by temporarybindings. Tasks with low S-R compatibility willhave to rely heavily on this task set, whereastasks with high S-R compatibility will be lessdependent on a well-established task set, be-cause it is needed just to confirm the alreadyactivated response. This line of reasoning leads

to the hypothesis that tasks with high S-R compatibility will correlate less (although notnecessarily zero) with a measure of WMC thantasks with low S-R compatibility.

Two of us have recently tested this hypoth-esis (Wilhelm & Oberauer, 2006). In a studyinvolving four-choice RT tasks with compatibleand with arbitrary S-R mapping we identifiedtwo latent factors, one general RT factor cap-turing the common variance of all RT tasks,and an arbitrary-mapping factor capturing theresidual variance shared by all tasks with arbi-trary S-R mapping. The general RT factorwas correlated substantially with a latent factorreflecting WMC, but the specific arbitrary-mapping factor had an even higher correlationwith WMC.

The only other study we are aware of thatcontrasts high and low S-R compatibility within

the same task is the study by Kane et al. (2001),who compared groups of people with highand low WMC on the speed and accuracy of executing prosaccades (i.e., eye movements to-ward a suddenly appearing cue in the periphery)and antisaccades (i.e., eye movements awayfromthecue).Obviously,theprosaccadetaskhas




high S-R compatibility, whereas the antisac-cade has lower compatibility. Kane et al. (2001)found that the two capacity groups differed inperformance on the antisaccade task but not on

the prosaccade task.Unfortunately, there is an alternative ex-planation, the one advanced by Kane et al.(2001): the better performance of high-capacitypeople on the antisaccade task could reflecttheir greater ability to suppress the saccade to-ward the cue, which is a strongly prepotentresponse. One way to decide between the in-hibition interpretation and an interpretation interms of S-R bindings would be to compare

high- and low-capacity groups on the speed andaccuracy of saccades in response to arbitrarycues presented centrally (e.g., when the fixa-tion point turns into a square, move the eyes tothe left; if it turns into a circle, move the eyes tothe right). An account based on inhibition of prepotent responses would not predict a cor-relation of this task with WMC, because thereis no prepotent action to be inhibited. Ourbinding account, in contrast, predicts that this

arbitrary saccade task correlates as strongly withWMC as the antisaccade task.

WORKING MEMORY CAPACITYAND REASONING

What it is about working memory capacity thatmakes it such an important prerequisite forsuccessful reasoning? Reasoning tasks found inintelligence tests tend to load on a single factor,even though they share hardly any cognitiveoperations apart from those common to vir-tuallyalltasks.Forexample,theanalysesofcom-ponent processes of inductive- and deductive-reasoning tasks performed by Sternberg (1985)reveal hardly any common processes beyond‘‘encoding’’ and ‘‘response selection.’’ None-theless, Wilhelm (2005) was unable to obtain

separate factors for inductive and deductivereasoning. The most important limiting factorfor these tasks must be one they have in com-mon, independent of the diverse cognitive op-erations used to perform each of them.

We think that what is common to all reason-ing tasks is the fact that their solutions require

the construction of new structural representa-tions. This means that given elements must becombined by new relations. The complexity of the new structures is limited by the capacity of

working memory. The capacity of the direct-access region sets a limit on the number of ele-ments that can be placed simultaneously withina common cognitive coordinate system, andthereby be integrated into a new structure. Wewill demonstrate this for a few prototypical ex-amples in the following section.

Reasoning and RelationalIntegration

Three categories of reasoning tasks are com-monly found in intelligence tests as well as inexperimental studies of human reasoning. Thefirst, deductive reasoning,canbedefinedasdraw-ing inferences from premises with logical ne-cessity. The premises are given as verbal state-ments (as in syllogisms) or in mathematicalform (as in equations to be solved), but spatial-figural formats can also be designed (c.f., Wil-

helm, 2000). Inductive reasoning is defined asdrawing a plausible, but not necessary, infer-ence from given information. In psychologicaltests and experiments this involves inferring arule or category from particular instances (e.g.,categorization or series completion) or trans-ferring information from one instance to another(e.g., analogical reasoning). Transformationscan be defined as tasks in which an initial stateof some object or situation must be mentallytransformed into a final state by the applicationof given operators. Problem solving can be seenas a subset of this category because it is usuallydefined as transforming an initial state into agoal state by means of a set of operators (Newell& Simon, 1972). Problem-solving tasks loadinghigh on reasoning factors are typically from thewell-defined variety for which the states andoperators are given, and one has to look for an

optimal combination of operators to attain thedesired transformation.Currently the most successful theory of

deductive reasoning is the theory of mentalmodels( Johnson-Laird & Byrne,1991; Johnson-Laird, 2001). According to this theory, deduc-tive inferences are drawn from mental models




representing the truth conditions of the prem-ises. This means that representations of theobjects or events referred to in the premises arearranged according to the relations expressed in

the premises. For example, in reasoning abouttemporal relations (Schaeken et al., 1996), to-kens representing individual events must bearranged into an array corresponding to theirtemporalorder.Insyllogisticreasoning(Johnson-Laird & Bara, 1984), tokens representing in-dividual members of the categories must bearranged such that the category relations statedin the premises are true. Mental models makeuse of geometrical constraints even when they

represent nonspatial relations. Consider, forexample, the mental model of the syllogism‘‘All A are B. No B is C’’ (each line representsone individual with its features A, B, C, or anycombination of them):

A¼B

A¼B

B

C

Constructing the model according tothe premises results in a spatial arrangement of the tokens, which represent the individuals. Thenew relationship between the ‘‘end terms’’ A and C emerges through the geometric con-straints of the medium: no C token can lie onthe same line as an A token, and therefore themodel supports the conclusion ‘‘No A is C.’’Thus, the deductive inference relies on bind-ing the tokens to places in a quasi-spatial men-tal coordinate system and making use of theemerging relations not explicitly mentioned inthe premises.

Typical inductive reasoning tasks are seriescompletion, matrices, and analogies. All of these tasks depend on detecting relations be-tween relations, that is, mapping relations onto

each other. This has been spelled out mostelaborately for analogical reasoning (e.g.,Gentner, 1989). Gentner defined an analogy asa mapping of the structure of a base domainonto a target domain. Analogical reasoning thusrequires identifying relations in the base domain

and at least in part also in the target domain,then identifying mappings (i.e., relations of identity or similarity) between these relations.

A good analogy is characterized by systemati-

city, that is, preferentially mapping coherentsystems of relations. Finding the most system-atic mapping between two domains requiresrepresenting complex structures in both thebase and the target domains simultaneously. Itshould be obvious that this requires extensivebinding of elements into the argument place-holders of the relations in question. A compu-tational model of analogies demonstrating thecentral role of bindings was developed by

Hummel and Holyoak (1997).Matrices and series completions can be an-

alyzed as special cases of analogical mapping.In series-completion tasks, the rule generatingthe series is repeated after n items. In order todetect the rule, together with the value of n,one must try to map the first n elements of theseries onto the next n elements (for varyingvalues of n). In matrices (such as Raven’s) therelations between elements in a row must be

mapped across columns, and the relationshipsbetween elements in a column must be map-ped across rows. According to this analysis, thelimiting factor for Raven’s matrices is not mem-orizing several hypotheses or rules, as Carpenter,Just, and Shell (1990) assumed, but simplyrepresenting three structures, each one linkingthree elements, at the same time to figure outtheir mapping. Partly supporting this conten-tion, Unsworth and Engle (2005) found thatthe correlation between a WMC task (operationspan) and Raven items did not increase withthe number of rules to be applied for solvingthe Raven item.

A prototypical problem-solving task inves-tigated in numerous studies is the Tower of Hanoi (e.g., Carpenter et al., 1990; Kotovsky,Hayes, & Simon, 1985). Severaldiscs are stackedon one of three rods, ordered by size. The tower

has to be moved to another rod, with the con-straints that only one disc can be moved at atime, and a larger disc must never be placed ontop of a smaller one. Problem-solving tasksusually require the construction of a sequenceof operations (e.g., movements of individual




discs) that transform the given state into thegoal state. If the task is not to be solved by blindtrial and error, at least part of the sequence mustbe planned in advance. This means building a

mental structure to bring individual steps into atemporal order. Often the plan is also hierar-chical because of the goal–subgoal structure,extensively analyzed for the Tower of Hanoiproblem (e.g., Karat, 1982).

Other transformation tasks used in intelli-gence tests such as the BIS test involve themental rotation, translation, or folding of ob-jects or their parts in space (e.g., paper-foldingtasks, puzzle-assembly tasks) or the recombina-

tion of symbols (e.g., anagram tasks). These taskshighlight a second aspect of relational integra-tion in mental transformations: the states are of-ten structures of objects, object parts, or sym-bols, and the operations are rearrangements of the structure’s elements (e.g., the discs of theTower of Hanoi, the sides of a cube to be foldedfrom a two-dimensional cut-out, the letters in ananagram tasks). To keep track of the currentstate in planning, one has to continuously up-

date the structure of the state one transforms.This again requires the flexible, temporary bind-ing of representations of the independent ob-jects or parts (e.g., the discs) to their places in acognitive coordinate system (e.g., the rod posi-tions).

This analysis shows that many reasoningtasks share the requirement of constructing newrelational representations, often with consider-able complexity, and updating them quicklyand efficiently. Constructing new structural rep-resentations requires a mechanism that flexiblysets up and dissolves temporary bindings of el-ements into argument slots of a schema or po-sitions in a mental coordinate system. Accordingto our model, a schema or coordinate system ismade available by the direct-access region of working memory, such that new elements canbe bound to it. Updating structural represen-

tations often involves picking out selectivelyone element in the structure to manipulate itby a cognitive operation (e.g., moving one discwhile leaving the others on their rod). This isthe role of the focus of attention in workingmemory. An important feature of structural

representations is that individual elements canbe changed without affecting the other ele-ments. This is accomplished by the division of labor between the direct-access region and the

focus of attention, thus enabling the system toattend to several elements concurrently. A structural representation is thus formed, and atthe same time one element is selected exclu-sively as the object of a cognitive operation(Oberauer, 2002).

The above analysis of reasoning tasks shouldalso explain why some tasks used to measureexecutive functions have been observed to cor-relate highly with measures of WMC and

spatial reasoning (Miyake et al., 2001). Thesetasks are characterized by a mixture of demandson cognitive functions, among which may in-deed be executive functions. Importantly, theyall also place heavy demands on relational in-tegration. Miyake et al. (2001) operationalizedexecutive function through use of the Tower of Hanoi and a random-generation task (i.e., thegeneration of a sequence of digits that fulfilscriteria of randomness such as equal frequency

of all possible digits and digit transitions). Asshown above, the Tower of Hanoi is a problem-solving task that relies heavily on the construc-tion of new structures. The random-generationtask requires constant monitoring of the se-quence produced thus far, with careful attentionto systematic patterns, that is, relations andstructures that repeat themselves inadvertently.This comes down to a series completion taskperformed on the memory of one’s own output—again, a task that relies considerably on relationalintegration. It is therefore not surprising thatthese tasks are strongly correlated with WMCand reasoning factors, and it proves in no way thecrucial role of executive functions.

Our reinterpretation of some putative exec-utive tasks leads to the prediction that remov-ing the executive aspect from these tasks shouldnot diminish their correlation with indicators

of WMC and reasoning. The Tower of Hanoiproblem, for example, is assumed to have ex-ecutive demands because problem solvers mustmanage goals and subgoals and plan an appro-priate sequence of moves (Miyake et al., 2001).

A version without this requirement could be




constructed in which participants receive aninitial state and a sequence of operations, andtheir task is to determine what the final statewill look like. Random generation is assumed to

require the central executive because partici-pants must suppress prepotent schemata andswitch between generation strategies repeatedly(Baddeley, Emslie, Kolodny, & Duncan, 1998).

A version with much reduced executive demandcould be constructed in which participants areasked to monitor a sequence of digits and detectcertain deviations from randomness (e.g., therepetition of a sequence of n digits or theemergence of a constant generation rule). We

predict that these versions with diminished ex-ecutive demand will correlate as highly as theoriginals with factors reflecting WMC andreasoning ability.

Relating Reasoning to WorkingMemory—Levels of Generalityand Symmetry

In both our factor-analytic studies mentioned

above, we also asked participants to take a testfor the BIS ( Jager et al., 1997). Consistent withprevious studies, we found a strong correla-tion between a factor representing WMC andthe reasoning factor extracted from the BIStest (Suß, Oberauer, Wilhelm, & Wittmann,2000; Suß et al., 2002). Because of the facet-theoretical framework of both the BIS and ourstructural model of working memory, we wereable to investigate the relation between WMCand intelligence on different levels of general-ity. This investigation provided an opportunityto apply the multivariate reliability theory de-veloped by Wittmann (1988) to the predictionof intelligence test scores from measures of working memory (and related constructs).

The central idea of multivariate reliabilitytheory is that each measure is a composite of wanted and unwanted variance. Wanted vari-

ance is the variance due to the construct oneintends to measure. Unwanted variance consistsof variance due to other psychological variablesone is not interested in (e.g., content-relatedvariance when the intention is to measure acontent-independent construct, or speed vari-ance in a speeded reasoning test), task-specific

variance (which one is usually not interested ineither), and random error. Only random errorcan be suppressed by increasing the number of items in a test; the first two components of un-

wanted variance are systematic and thereforedon’t drop out through longer testing with itemsof the same type. So in order to obtain a rela-tively pure estimate of the intended construct(i.e., the wanted variance), it is necessary toaggregate over several indicators of this con-struct. These indicators should not be clonesof each other, distinguished only by trivial var-iations; heterogeneity is important to suppresssystematic unwanted variance in the com-

posite.This requirement raises the question of the

right amount of heterogeneity, or, in otherwords, the right level of generality. For an abilityconstruct such as WMC, a minimum require-ment is that its indicators are positively corre-lated, such that their shared variance can beinterpreted as manifestation of the ability. Thus,one obvious means to assess what level of gen-erality is adequate for an ability construct—that

is, what set of indicators has an adequate level of heterogeneity—is to investigate the correlationsin a large set of potential indicators, as we didabove for WMC and related constructs. A sec-ond, complementary approach is to investigatethe relationship to external criteria on differentlevels of generality. The rationale is given by theprinciples of symmetry in the Brunswik lensmodel (Tucker, 1964), illustrated in Figure 3.4(after Wittmann, 1988). The figure shows apredictor and a criterion construct, each mea-sured on three levels of generality (illustrated asthe manifest variables and two levels of latentvariables in a structural equation model). As-suming that the constructs share most (or evenall) of their true variance, the highest correla-tions will be measured when symmetricalmeasures are related to each other. Symmetri-cal measures are measures on the same level of

aggregation that reflect corresponding con-structs with the same scope, represented by thethick, continuous lines in the figure. Asym-metrical relations (broken lines) lead to smallercorrelations, because either the predictor or thecriterion measure contains variance that has nocounterpart on the other side. An example for




an asymmetric relationship would be to predictgeneral intelligence (a highly general criterion)from a single measure of WMC such as readingspan (path 1 in Fig. 3.4). An example of the

converse asymmetry would be to use a battery of working memory tests to predict success insolving the Tower of Hanoi problem (path 2 inFig. 3.4).

On the basis of this logic, we can use thecorrelations between WMC (and related con-structs) and intelligence test measures on dif-ferent levels of aggregation to search empiricallyfor the most symmetric level for their relation-ship. Here we illustrate this approach with areanalysis of data from our two studies (Ober-auer et al., 2003; Suß et al., 2002).

The first level of generality is to corre-late individual working memory tasks with the

reasoning scale from the BIS. Table 3.2 showsthree examples using classical WMC tasks. Thecorrelations on this level are already substantialbut leave enough room for doubting that WMCand reasoning ability are actually identicalpsychometric constructs (e.g., Ackerman et al.,2005). Many studies in the literature report re-lationships between single measures of WMCand some criterion variable, and the correla-tions are rarely larger than r ¼ .50.

For a second level of generality we averagedthe z scores of all tasks measuring predominantlystorage and processing and all tasks measuringpredominantly relational integration, separately

Pred Crit

1

3

2

Figure 3.4. Principles of symmetry in predictor–criterion relations (after Wittmann, 1988). Thinlines represent factor loadings; thick lines represent

predictor–criterion paths. Solid paths are symmet-rical, broken paths are asymmetrical. Three kinds of asymmetry are illustrated: (1) the predictor variable istoo specific relative to the criterion variable; (2) thecriterion variable is too specific relative to the pre-dictor variable; (3) predictor and criterion vari-ables measure different subsets of the varianceshared between predictor construct and criterion con-struct.

TABLE 3.2. Correlations between Working Memory Measures and Reasoning Abilityand General Intelligence, on Four Levels of Generality

Reasoning*

General

Intelligence* Reasoning

{

General

Intelligence

{

Reading span .57 .64 .47 .51Computation span .54 .52 .24 .29Spatial STM .59 .55 .58 .44

S&P-vn (7,5) .69 .66 .53 (5) .52S&P-spat (3,2) .68 .59 .58 (2) .50Rel.Int-vn (3,4) .61 .61 .67 (4) .64Rel.Int-spat (5,7) .65 .64 .67 (7) .57

WMC (18, 18) .77 .75 .76 (18) .69Speed/Exec (6, 16) .59 .71 .55 .54

WMCþ

Speed/Exec (24, 34) .76 .82 .77 .72*Data from Suß et al. (2002), N ¼ 129.{Data from Suß et al. (2000), for task descriptions see Oberauer et al. (2003); N ¼ 133. Numbers in parentheses behind

composites represent the number of individual tasks entering into the respective composite (first entry¼ study 1, secondentry¼ study 2). The Speed/Exec composite of Suß et al. (2002) consists of the tasks loading on the speed/executivefunctions factor in Oberauer et al. (2000); the Speed/Exec composite of Suß et al. (2000) consists of the choice response

time tasks in single blocks and task-set switching blocks (i.e., the 16 manifest variables on the right side of Figure 3.3).STM¼ short-term memory; WMC¼working memory capacity. For other abbreviations see Table 3.1.




for verbal-numerical and for spatial contents.The composite WMC scores correlate some-what higher with reasoning ability. This is thelevelofgeneralityattainedbyanumberofstudies

using a latent-variable approach to measuringWMC and intelligence constructs. Most of thesestudies operationalized WMC by composites of complex span tasks with verbal or numericalcontent (e.g., reading span, operation span,counting span). This corresponds to the variablestorage and processing with verbal or numericalmaterial (S&P-vn) in Table 3.2.

On the third level of generality, we averagedthe four composite variables to form a single

indicator of WMC. The correlations with rea-soning ability were higher still. When we moveto an even higher level of generality on thepredictor side, including variables reflectingprocessing speed and executive functions (e.g.,reaction times from task-set switching para-digms), the correlations with reasoning level off.This suggests that the gain in reliability obtainedthrough the larger number of test items is nowoffset by a reduction in symmetry.

Table 3.2 also contains correlations of all theWMC variables with a general intelligencescore, obtained from the BIS test by aggregatingthe four functional scales (reasoning, creativity,memory, and speed). This illustrates a move toa higher level of generality on the criterion side.Relative to the correlations obtained with thereasoning scale, this move doesn’t increase thecorrelations with the predictor variables, withone exception: the combination of WMC withspeed and executive function was related moreto general intelligence than to reasoning in thestudy of Suß et al. (2002).

Taken together, these data suggest that themost symmetric relationships are between themost comprehensive composite of WMC andreasoning ability, and, on a higher level of gen-erality,betweentheevenmoreencompassingcom-posite of WMC, speed, and executive functions

and general intelligence. The more elaborateanalyses reported in Suß et al. (2002) show thatthe relationship on the highest level of generalitycan be summarized as (1) a strong associationbetween WMC and reasoning, as shown above,and (2) a strong association between speed and

executive functions with the processing speedand memory factors of the BIS test.

A third kind of asymmetry (path 3 in Fig.3.4) is obtained when two measures on the

same level of generality are correlated that re-flect different portions of the variance sharedbetween the two constructs. An example fromcognitive abilities is relating measures fromdifferent content domains. Although we andoth-ers found that content-specific WMC factorstend to be highly correlated, a separation of verbal-numerical from figural-spatial workingmemory factors is often warranted (Kyllonen,1994; Oberauer et al., 2000; see Chapters 6 and

8), and at least in one study their correlationturned out to be quite low (Shah & Miyake,1996). Therefore, correlating WMC measuresconsisting exclusively (Conway et al., 2002) ormostly (Ackerman et al., 2002) of verbal-nu-merical tasks with measures of fluid intelligenceor reasoning based only on spatial-figural mate-rial (such as the Raven matrices or Cattell’sCulture Fair Test) will result in an underesti-mation of the true relationship between WMC

and fluid intelligence and reasoning.2 A structural equation model reported in Suß

et al. (2002) illustrates this. In that model (Fig-ure 10 in Suß et al., 2002), we used separatefactors for spatial-figural WMC and for verbal-numerical WMC to predict the three contentscales from the BIS test (verbal, numerical, andspatial-figural). The spatial-figural WMC factorpredicted the spatial-figural factor of the BISwith a path coefficient of .70, whereas the pathcoefficients to the numerical factor (.35) and theverbal factor (.26) were low. Conversely, theverbal-numerical WMC factor predicted verbalabilities in the BIS (.87) and numerical abilities(.52), but not spatial-figural abilities (.12).

To conclude, many correlations obtained inpublished studies on the relationship betweenWMCandreasoningorfluidintelligence,thoughimpressively high, might actually underestimate

the true amount of shared variance between thetwo constructs. Still, we do not believe thatthe two can be identified. Evidence that fluidintelligence cannot be completely reduced toWMC comes again from the content dimension.Whereas verbal and numerical content could




never be factorially separated in sets of WMCtasks (Kyllonen & Christal, 1990; Oberauer et al.,2000, 2003), they are clearly different on the sideof intelligence in general (Carroll, 1993) and

reasoning ability in particular (Jager et al., 1997).Reasoning is, after all, a little bit more thanworking memory.

CONCLUSIONS

We see converging evidence that WMC is ahighly general parameter of the cognitive sys-tem, one that acts as an important limiting

factor for a wide variety of complex cognitivetasks. In this chapter we argued that the com-mon denominator of all cognitive functionslimited by WMC is the temporary binding of representations. The more a performance onany task relies on the establishment of multipletemporary bindings maintained simultaneouslyand consistently, the more we expect it to loadon a general WMC factor. First direct supportfor this prediction comes from a study showing

that short-term recognition tasks (such as theSternberg task and the n-back task) correlatewith WMC only if they involve a high demandon bindings (Oberauer, 2005a).

Working memory capacity, in turn, is the bestsingle predictor of reasoning ability, explaining atleast half of the systematic variance in tests of reasoning or fluid intelligence. This relationshipseems to exist on a high level of generality. It isnot due to associations between particular WMCtasks and particular reasoning tasks, but rather is amanifestation of a highly general limiting factorthat affects both working memory and reasoning.We hypothesize that this factor is the limitedability of our nervous system to establish multipletemporary bindings at the same time, therebyenabling the construction of new relational rep-resentations. The construction of new relationalrepresentations is a requirement shared by most

tasks commonly used in established reasoningtests and in experiments on reasoning.Temporary bindings are also needed to es-

tablish task sets implementing arbitrary linksbetween stimulus and response categories. Al-though speculative at the moment, this para-

digm can offer an explanation for the strongassociation between WMC measures and manyspeed tasks, a relationship largely ignored in theworking memory literature so far. At the same

time, WMC cannot be reduced to mental speed;factors representing WMC and factors repre-senting speed have always been separable incomprehensive studies.

Finally, our view differs in one respect fromthe majority opinion in the field, in that wedon’t assume a special link between WMCand executive function. The evidence availablethus far suggests that different executive func-tion measures have little variance in common.

Some of them are related to WMC, but wethink this is not because they reflect executivefunctions but because of some other feature.For example, choice RTs obtained in task-setswitching paradigms correlate with WMC fac-tors, but no more than do RTs in the same taskswithout switching. Planning tasks such as Towerof Hanoi correlate with WMC because of theirdemand on relational integration, not becauseof their demand on executive functions to se-

lect individual moves.The challenge to theories that link working

memory and executive functions is perhapsthe most significant contribution of our studiesto theories of working memory. It is just oneexample of the general potential inherent inindividual-difference studies. They can serve tovalidate, and invalidate, experimental tasks andvariables as indicators of theoretical constructs,and they can test theories about associations, aswell as dissociations, between constructs (cf.Oberauer, 2005b). Factor analyses of largesamples of tasks have, for example, helped toestablish working memory as an importanttheoretical concept in cognitive psychology.Factor-analytic research, however, raises seri-ous doubts about the usefulness of a unitaryconstruct subsuming all executive functions.Experimental research helps us to understand

the cognitive processes going on when a personis confronted with a specific situation or task.Individual-differences research tells us whetherthese processes are in fact representative of thetheoretical concepts the experimentalist as-sumes them to represent.




Notes

1. For this reason, we renamed ‘‘focus of attention’’

in the model of Cowan (1995) as ‘‘region of direct

access’’ in our model.

2. In light of this, it is impressive how high the re-

lationships obtained in these two studies were, as

Conway et al. (2002) pointed out.

Acknowledgments

The research reported in this chapter was supported

by Deutsche Forschungsgemeinschaft (DFG, grants

WI 1390/1, KL 955/4, and OB 121/3).

References

Ackerman, P. L., Beier, M. E., & Boyle, M. O.


ory within a nomological network of cognitive

and perceptual speed abilities. Journal of Exper-

imental Psychology: General, 131, 567–589.

Ackerman, P. L., Beier, M. E., & Boyle, M. O.

(2005). Working memory and intelligence: The

same or different constructs? Psychological Bul-

letin, 131, 30–60.

Allport, A., Styles, E. A., & Hsieh, S. (1994). Shifting

intentional set: Exploring the dynamic control

of tasks. In C. Umilta & M. Moscovitch (Eds.),

Attention and Performance (Vol. XV, pp. 421–

452). Cambridge, MA: MIT Press.

Anderson, J. R., & Lebiere, C. (1998). The atomic

components of thought. Mahwah, NJ: Erlbaum.Baddeley, A. D. (1986). Working memory. Oxford:

Clarendon Press.

Baddeley, A. D. (1993). Working memory or working

attention? In A. Baddeley & L. Weiskrantz

(Eds.), Attention: Selection, awareness, and con-

trol (pp. 152–170). Oxford: Clarendon Press.



OF WORKING MEMORY

We think of working memory as a system that

provides access to memory representations for

goal-directed processing on three successive

levels: (1) activation in LTM, (2) temporary

binding of content representations to positions

in a cognitive coordinate system or place-

holders in a schema in the region of direct

access, and (3) selection as the object of ma-

nipulation by the focus of attention. Further,

we use facet theory to organize the sources of

variance among individuals that affect perfor-

mance on particular tasks.



The capacity limit of working memory is a limit

on the simultaneous binding of several inde-

pendent chunksin a common coordinate sys-tem or schema in the region of direct access.

This limit constrains how many independent

chunks can be accessed simultaneously, and

thereby constrains the ability to construct new

relational representations between them, and

the ability to take existing relations into ac-

count in cognitive operations. People with lowworking memory capacity are assumed to

have a specific problem with establishing and

maintaining robust bindings.


MEMORY VARIATION

We consider two other sources of variability:

the efficiency of executive function and the

efficiency of inhibition. We believe that both

these functions can be dissociated from work-

ing memory capacity.



Our research shows that the limited capacity

of working memory affects a wide variety of

tasks, its scope reaching beyond the classicalcharacterization as subserving simultaneous

storageandprocessing. Working memory, how-

ever, cannot be equated with executive func-

tion.




Baddeley, A. D., Emslie, H., Kolodny, J., & Duncan,

J. (1998). Random generation and the executive

control of working memory. Quarterly Journal of

Experimental Psychology, 51A, 819–852.

Bayliss, D. M., Jarrold, C., Gunn, D. M., &Baddeley,A.D.(2003).Thecomplexitiesofcom-

plex span: Explaining individual differences in

working memory in children and adults. Jour-

nal of Experimental Psychology: General, 132,

71–92.

Brown, G. D. A., Preece, T., & Hulme, C. (2000).

Oscillator-based memory for serial order. Psy-

chological Review, 107, 127–181.

Burgess, N., & Hitch, G. J. (1999). Memory for

serial order: A network model of the phonolog-ical loop and its timing. Psychological Review,

106, 551–581.

Canter, D. (1985). Facet theory. Berlin: Springer.

Carpenter, P., Just, M. A., & Shell, P. (1990). What

one intelligence test measures: A theoretical

account of the processing in the Raven Progres-

sive Matrices Test. Psychological Review, 97,

404–431.

Carroll, J. B. (1993). Human cognitive abilities: a

survey of factor-analytic studies. New York: Cam-bridge University Press.

Cepeda, N. J., Kramer, A., & Gonzalez de Sather,

J. C. M. (2001). Changes in executive con-

trol across the life span: Examination of task-

switching performance. Developmental Psychol-

ogy, 37, 715–730.

Christoff, K., Prabhakaran, V., Dorfman, J., Zhao,

Z., Kroger, J. K., Holyoak, K. J., et al. (2001).

Rostrolateral prefrontal cortex involvement in

relational integration during reasoning. Neuro-

Image, 14, 1136–1149.

Colom, R., Flores-Mendoza, C., & Rebollo, I. (2003).

Working memory and intelligence. Personality

and individual Differences, 34, 33–39.

Conway, A. R. A., Cowan, N., & Bunting, M. F.

(2001). The cocktail party phenomenon revis-

ited: The importance of working memory ca-

pacity. Psychonomic Bulletin & Review, 8, 331–

335.Conway, A. R. A., Cowan, N., Bunting, M. F.,

Therriault, D. J., & Minkoff, S. R. B. (2002).

A latent variable analysis of working memory

capacity, short-term memory capacity, process-

ing speed, and general fluid intelligence. Intel-

ligence, 30, 163–183.

Conway, A. R. A., & Engle, R. W. (1994). Working

memory and retrieval: A resource-dependent

inhibition model. Journal of Experimental Psy-


Cowan, N. (1988). Evolving conceptions of memorystorage, selective attention, and their mutual

constraints within the human information-

processing system. Psychological Bulletin, 104,

163–191.

Cowan, N. (1995). Attention and memory: An

integrated framework. New York: Oxford Uni-

versity Press.

Cowan, N. (1999). An embedded-process model of

working memory. In A. Miyake & P. Shah

(Eds.), Models of working memory. Mechanismsof active maintenance and executive control

(pp. 62–101). New York: Cambridge University

Press.

Cowan, N. (2001). The magical number 4 in short-



24, 87–185.



ing. Journal of Verbal Learning and VerbalBehavior, 19, 450–466.

Daneman, M., & Merikle, P. M. (1996). Work-

ing memory and language comprehension: a

meta-analysis. Psychonomic Bulletin & Review,

3, 422–433.

De Beni, R., & Palladino, P. (2000). Intrusion errors

in working memory tasks. Are they related to

reading comprehension ability? Learning and

Individual Differences, 12, 131–143.

De Beni, R., Palladino, P., Pazzaglia, P., & Cornoldi,

C. (1998). Increases in intrusion errors and

working memory deficit of poor comprehenders.

Quarterly Journal of Experimental Psychology,

51A, 305–320.

DeSoto, C., London, M., & Handel, S. (1965). Social

reasoning and spatial paralogic. Journal of Per-

sonality and Social Psychology, 2, 513–521.

Duncan, J., Johnson, R., Swales, M., & Freer, C.

(1997). Frontal lobe deficits after head injury:Unity and diversity of function. Cognitive Neuro-

psychology, 14, 713–741.

Engle, R. W., Kane, M. J., & Tuholski, S. W. (1999).

Individual differences in working memory

capacity andwhat they tell us about controlled at-

tention, general fluid intelligence, and functions




of the prefrontal cortex. In A. Miyake & P. Shah

(Eds.), Models of working memory. Mechanisms

of active maintenance and executive control (pp.

102–134). New York: Cambridge University

Press.Engle, R. W., Tuholski, S. W., Laughlin, J. E., &

Conway, A. R. A. (1999). Working memory, short

term memory and general fluid intelligence: A

latent variable approach. Journal of Experimen-


Farrell, S., & Lewandowsky, S. (2002). An endoge-

nous distributed model of ordering in serial

recall. Psychonomic Bulletin & Review, 9, 59–79.

Friedman, N. P., & Miyake, A. (2004). The relations

among inhibition and interference control func-tions: a latent variable analysis. Journal of Exper-


Friedman, N. P., Miyake, A., Corley, R. P., Young,

S. E., DeFries, J. C., & Hewitt, J. K. (2006). Not

all executive functions are related to intelligence.

Psychological Science, 17, 172–179.

Gentner, D. (1989). The mechanisms of analogi-

cal learning. In S. Vosniadou & A. Ortony

(Eds.), Similarity and analogical reasoning (pp.

199–241). Cambridge: Cambridge UniversityPress.

Gilbert, S. J., & Shallice, T. (2002). Task switching: A

PDP model. Cognitive Psychology, 44, 297–337.

Gray, J. R., Chabris, C. F., & Braver, T. S. (2003).

Neural mechanisms of general fluid intelli-

gence. Nature Neuroscience, 6, 316–322.

Halford, G. S., Wilson, W. H., & Phillips, S. (1998).

Processing capacity defined by relational com-

plexity: Implications for comparative, develop-

mental, and cognitive psychology. Behavioral

and Brain Sciences, 21, 803–864.




psychology of learning and motivation (Vol. 22,

pp. 193–225). New York: Academic Press.

Hasher, L., Zacks, R. T., & May, C. P. (1999). Inhib-

itory control, circadian arousal, and age. In D.

Gopher & A. Koriat (Eds.), Attention and perfor-mance (pp. 653–675). Cambridge, MA: MIT

Press.

Henson, R. N. A. (1998). Short-term memory for

serial order: The Start-End Model. Cognitive

Psychology, 36, 73–137.

Henson, R. N. A., Norris, D. G., Page, M. P. A., &

Baddeley, A. D. (1996). Unchained memory:

Error patterns rule out chaining models of im-

mediate serial recall. Quarterly Journal of Exper-

imental Psychology, 49A, 80–115.Hommel, B. (1998). Event files: Evidence for

automatic integration of stimulus-response ep-

isodes. Visual Cognition, 5, 183–216.

Hummel, J. E., & Holyoak, K. J. (1997). Distributed

representations of structure: A theory of ana-

logical access and mapping. Psychological Re-

view, 104, 427–466.

Jager, A. O., Suß, H.-M., & Beauducel, A. (1997).

Test fu r das Berliner Intelligenzstrukturmodell

(BIS). Gottingen: Hogrefe.Johnson-Laird, P. N. (2001). Mental models and de-

duction. Trends in Cognitive Sciences, 5, 434–

442.

Johnson-Laird, P. N., & Bara, B. G. (1984). Syllo-

gistic inference. Cognition, 16, 1–61.

Johnson-Laird, P. N., & Byrne, R. M. J. (1991).

Deduction. Hillsdale, NJ: Erlbaum.



of working-memory capacity. Journal of Exper-imental Psychology: General, 130, 169–183.

Kane, M. J., & Engle, R. W. (2003). Working-


The contributions of goal neglect, response com-

petition, and task set to Stroop interference.


132(47–70).

Kane, M. J., Hambrick, D. Z., & Conway, A. R. A.

(2005). Working memory capacity and fluid

intelligence are strongly related constructs: com-

ment on Ackerman, Beier, and Boyle (2004).

Psychological Bulletin, 131, 66–71.


Wilhelm,O.,Payne,T.W.,&Engle,R.W.(2004).

The generality of working-memory capacity: A

latent-variable approach to verbal and visuo-

spatial memory span and reasoning. Journal of Ex-


Karat, J. (1982). A model of problem solving withincomplete constraintknowledge. Cognitive Psy-

chology, 14, 538–559.

Kornblum, S., Hasbroucq, T., & Osman, A. M.

(1990). Dimensional overlap: Cognitive basis

for stimulus-response-compatibility—a model




and taxonomy. Psychological Review, 97, 253–

270.

Kotovsky, K., Hayes, J. R.,& Simon, H. A. (1985). Why

are some problems hard? Evidence from Tower

of Hanoi. Cognitive Psychology, 17, 248–294.Kramer, A. F., Humphrey, D. G., Larish, J. F.,

Logan, G. D., & Strayer, D. L. (1994). Ag-

ing and inhibition: Beyond a unitary view of

inhibitory processing in attention. Psychology &

Aging, 9, 491–512.

Kray, J., & Lindenberger, U. (2000). Adult age

differences in task switching. Psychology and

Aging, 15, 126–147.

Kyllonen, P. C. (1994). Aptitude testing inspired by

information processing: A test of the four-sources model. Journal of General Psychology,

120, 375–405.


ability is (little more than) working-memory


Lakoff, G., & Johnson, M. (1980). Metaphors we live

by. Chicago: University of Chicago Press.

Langacker, R. W. (1986). An introduction to cogni-

tive grammar. Cognitive Science, 10, 1–40.

Lien, M.-C., & Proctor, R. W. (2002). Stimulus-response compatibility and psychological refrac-

tory period effects: Implications for response

selection. Psychonomic Bulletin & Review, 9,

212–238.

Logan, G. D. (1978). Attention in character-

classification tasks: Evidence for the automa-

ticity of component stages. Journal of Experi-


Logan, G. D., & Cowan, W. B. (1984). On the ability

to inhibit thought and action: A theory of an

act of control. Psychological Review, 91, 295–

327.

Logan, G. D., & Schulkind, M. D. (2000). Parallel

memory retrieval in dual-task situations: I. Se-

mantic memory. Journal of Experimental Psy-

chology: Human Perception and Performance, 26,

1072–1090.


Working memory span and the role of proactiveinterference. Journal of Experimental Psycho-

logy: General, 130, 199–207.

May, C. P., Hasher, L., & Kane, M. J. (1999). The

role of interference in memory span. Memory &

Cognition, 27, 759–767.

Mayr, U., & Keele, S. W. (2000). Changing internal

constraints on action: The role of backward

inhibition. Journal of Experimental Psychology:

General, 129, 4–26.

Meiran, N. (1996). Is reading ability related toactivation dumping speed? Evidence from im-

mediate repetition priming. Memory & Cogni-

tion, 24, 41–59.

Miyake, A., Friedman, N. P., Emerson, M. J., Witzki,

A. H., Howerter, A., & Wager, T. D. (2000). The

unity and diversity of executive functions and

their contributions to complex ‘‘frontal lobe’’

tasks: A latent variable analysis. Cognitive Psy-

chology, 41, 49–100.

Miyake, A., Friedman, N. P., Rettinger, D. A., Shah,P., & Hegarty, M. (2001). How are visuospa-

tial working memory, executive functioning,

and spatial abilities related: A latent-variable

analysis. Journal of Experimental Psychology:

General, 130, 621–640.

Moscovitch, M. (1992). Memory and working-with-

memory: A component process model based on

modules and central systems. Journal of Cogni-

tive Neuroscience, 4, 257–267.

Neumann, O., & Klotz, W. (1994). Motor responses tononreportable, maskedstimuli:Where is the limit

of direct parameter specification? In C. Umilta &

M. Moscovitch (Eds.), Attention and Perfor-

mance (Vol. XV, pp. 124–150). Cambridge: MIT

Press.

Newell, A., & Simon, H. A. (1972). Human problem

solving. Englewood Cliffs, N.J.: Prentice Hall.


action. Willed and automatic control of behavior

(CHIP Report No. 99): University of California

San Diego.

O’Reilly, R., Busby, R. S., & Soto, R. (2003). Three

forms of binding and their neural substrates:

Alternatives to temporal synchrony. In A. Clee-

remans (Ed.), The unity of consciousness: bind-

ing, integration, and dissociation (pp. 168–192).

Oxford, UK: Oxford University Press.

Oberauer, K. (1993). Die Koordination kognitiver

Operationen. Eine Studie zum Zusammenhangvon ‘‘working memory’’ und Intelligenz. Zeits-

chrift fu r Psychologie, 201, 57–81.

Oberauer, K. (2001). Removing irrelevant informa-

tion from working memory. A cognitive aging

study with the modified Sternberg task. Journal




of Experimental Psychology: Learning, Memory,

and Cognition, 27, 948–957.

Oberauer, K. (2002). Access to information in work-

ing memory: Exploring the focus of attention.

Journal of Experimental Psychology: Learning,Memory, and Cognition, 28, 411–421.

Oberauer, K. (2003). Selective attention to elements

in working memory. Experimental Psychology,

50, 257–269.

Oberauer, K. (2005a). Binding and inhibition in

working memory individual and age differences

in short-term recognition. Journal of Experi-


Oberauer, K. (2005b). The measurement of working

memory capacity. In O. Wilhelm & R. W.Engle (Eds.), Handbook of understanding and

measuring intelligence (pp. 393–408). Thou-

sand Oaks: Sage.

Oberauer, K., Lange, E., & Engle, R. W. (2004). Work-

ing memory capacity and resistance to interfer-

ence. Journal of Memory and Language, 51, 80–

96.

Oberauer, K., Schulze, R., Wilhelm, O., & Suß, H.-M.

(2005). Working memory and intelligence—

their correlation and their relation: A commendon Ackerman, Beier, and Boyle (2005). Psycho-

logical Bulletin, 131, 61–65.

Oberauer, K., Suß, H.-M., Schulze, R., Wilhelm,

O., & Wittmann, W. W. (2000). Working mem-

ory capacity—facets of a cognitive ability con-

struct. Personality and Individual Differences,

29, 1017–1045.

Oberauer, K., Suß, H.-M., Wilhelm, O., & Witt-

mann, W. W. (2003). The multiple faces of

working memory—storage, processing, supervi-

sion, and coordination. Intelligence, 31, 167–193.

Page, M. P. A., & Norris, D. (1998). The primacy

model: A new model of immediate serial recall.

Psychological Review, 105, 761–781.

Raffone, A., & Wolters, G. (2001). A cortical mech-

anism for binding in visual working memory.

Journal of Cognitive Neuroscience, 13, 766–785.

Robin, N., & Holyoak, K. J. (1995). Relational com-

plexity and the functions of prefrontal cortex.In M. S. Gazzaniga (Ed.), The cognitive neurosci-

ences (pp.987–997).Cambridge, MA: MIT Press.

Rogers, R. D., & Monsell, S. (1995). Costs of a

predictable switch between simple cognitive

tasks. Journal of Experimental Psychology: Gen-

eral, 124, 207–231.

Rosen, V. M., & Engle, R. W. (1997). The role of

working memory capacity in retrieval. Journal

of Experimental Psychology: General, 126, 211–

227.

Salthouse, T. A. (1991). Mediation of adult agedifferences in cognition by reductions in work-

ing memory and speed of processing. Psycho-

logical Science, 2, 179–183.

Salthouse, T. A. (1994). The aging of working

memory. Neuropsychology, 8, 535–543.

Salthouse, T. A., Atkinson, T. M., & Berish, D. E.

(2003). Executive functioning as a potential

mediator of age-related cognitive decline in

normal adults. Journal of Experimental Psychol-

ogy: General, 132, 566–594.Salthouse, T. A., Babcock, R. L., & Shaw, R. J.

(1991). Effects of adult age on structural and

operational capacities in working memory.

Psychology and Aging, 6, 118–127.

Schaeken, W., Johnson-Laird, P. N., & d’Ydewalle,

G. (1996). Mental models and temporal rea-

soning. Cognition, 60, 205–234.



and language processing: An individual differ-ences approach. Journal of Experimental Psy-


Shallice, T., & Burgess, P. (1991). Higher-order

cognitive impairments and frontal lobe lesions

in man. In H. S. Levin, H. M. Eisenberg &

A. L. Benton (Eds.), Frontal lobe function and

dysfunction (pp. 125–138). New York: Oxford

University Press.

Shilling, V. M., Chetwynd, A., & Rabbitt, P. M. A.

(2002). Individual inconsistency across mea-

sures of inhibition: an investigation of the con-

structvalidityofinhibitioninolderadults.Neuro-

psychologia, 40, 605–619.

Sternberg, R. J. (1985). Beyond IQ. New York:

Cambridge University Press.

Suß, H.-M., & Beauducel, A. (2005). Faceted

models of intelligence. In O. Wilhelm &

R. W. Engle (Eds.), Handbook of understanding

andmeasuringintelligence (pp. 313–332). Thou-sand Oaks, CA: Sage.

Suß, H.-M., Oberauer, K., Wilhelm, O., & Witt-

mann, W. W. (2000). Can working memory ca-

pacity explain reasoning ability? Paper presented

at the XXVII International Congress of Psychol-

ogy, Stockholm, Sweden.




Suß, H.-M., Oberauer, K., Wittmann, W. W.,

Wilhelm, O., & Schulze, R. (2002). Working

memory capacity explains reasoning ability and

a little bit more. Intelligence, 30, 261–288.

Talmy, L. (1988). Force dynamics in language andcognition. Cognitive Science, 12, 49–100.

Tucker, L. R. (1964). A suggested alternative formu-

lation in the developments by Hursch, Ham-

mond and Hursch, and by Hammond, Hursch

and Todd. Psychological Review, 77, 528–530.

Unsworth, N., & Engle, R. W. (2005). Working

memory capacity and fluid abilities: Examining

the correlation between operation span and

Raven. Intelligence, 33, 67–81.

Unsworth, N., Schrock, J. C., & Engle, R. W.(2004). Working memory capacity and the

antisaccade task: Individual differences in vol-

untary saccade control. Journal of Experimental

Psychology: Learning, Memory, and Cognition,

30, 1302–1321.

Usher, M., Haarmann, H. J., Cohen, J. D., & Horn,

D. (2001). Neural mechanism for the magical

number 4: Competitive interactions and non-

linear oscillation. Behavioral & Brain Sciences,

24, 151–152.

Waltz, J. A., Knowlton, B. J., Holyoak, K. J., Boone,

K. B., Mishkin, F. S., de Menezes Santos, M.,

et al. (1999). A system for relational reasoning

in human prefrontal cortex. Psychological Sci-

ence, 10, 119–125.Ward, G., Roberts, M. J., & Phillips, L. H. (2001).

Task-switching costs, Stroop-costs, and executive

control: A correlational study. Quarterly Journal

of Experimental Psychology, 54A, 491–511.

Wilhelm, O. (2005). Measuring reasoning ability. In

O. Wilhelm & R. W. Engle (Eds.), Handbook

of understanding and measuring intelligence (pp.

373–392). Thousand Oaks: Sage.

Wilhelm, O., & Oberauer, K. (2006). Why are

reasoning ability and working memory capac-ity related to mental speed? An investigation

of stimulus-response compatibility in choice-

reaction-time tasks. European Journal of Cog-

nitive Psychology, 18, 18–50.

Wittmann, W. W. (1988). Multivariate reliability

theory. Principles of symmetry and successful

validation strategies. In J. R. Nesselroade &

R. B. Cattell (Eds.), Handbook of multivariate

experimental psychology (pp. 505–560). New

York: Plenum.




4

Explaining the Many Varieties of

Working Memory Variation: Dual

Mechanisms of Cognitive Control

TODD S. BRAVER, JEREMY R. GRAY,

and GREGORY C. BURGESS

Virtually all working memory (WM) theoristsagree that control processes are a critical com-ponent of WM function. Some set of internalmechanisms must be responsible for (1) se-lecting information for active maintenance inWM; (2) ensuring that it can be stored for anappropriate length of time; (3) protecting itagainst sources of interference; (4) updating itat appropriate junctures; and (5) using it to in-fluence other cognitive systems (i.e., percep-tion, attention,memory andaction). Yet, equallyclear to most theorists is the observation that theability to exert control over WM varies sub-stantially, both within individuals (across timeand task situations) and across individuals. Insome sense, this observation poses perhaps thecore paradox regarding cognitive control: why is

cognitive control so important, yet simulta-neously so fragile and vulnerable to disruption?Why does it appear that our ability to exertcontrol is so strong in some cases but so weak inothers? If exerting cognitive control seems to bethe optimal response in many situations, whydoes it seem as if behavior is suboptimally

controlled much of the time in many individ-uals, and at least some of the time in all indi-viduals?

In this chapter, we put forth a theory of cognitive control in WM that attempts to ex-plain this variability. Our central hypothesis isthat cognitive control operates via two distinctoperating modes: proactive control and reactivecontrol. We will present arguments suggest-ing that these two modes are dissociable on anumber of dimensions, such as computationalproperties, neural substrates, temporaldynamics,and consequences for information processing.We will suggest that although most formulationsof cognitive control in WM only considerproactive control, reactive control mechanismsmay be more dominant. We will further suggest

that by distinguishing between these two modeswe will be able to (1) resolve some of the ap-parent inconsistencies in the existing WM liter-ature; (2) understand how and why the impact of cognitive control processes in WM can vary sostrongly within individuals across time and tasksituations; (3) gain insight into the nature of

76



cognitive control impairments found in healthyaging (and possibly in other populations suffer-ing from neuropsychiatric disorders); (4) under-standsomeofthecriticalunderlyingmechanisms

related to individual differences in WM func-tion; and (5) account for potentially surprisingdata indicating that putatively ‘‘noncognitive’’variables such as mood states and personalitytraits (e.g., extraversion, neuroticism) may alsoinfluence WM function.

The general theoretical framework that weadvance here for understanding the sources of variation that affect WM and cognitive controlis termed the dual mechanisms of control, or

DMC account. It is worth noting that, althoughwe have been developing this framework forseveral years now, this chapter marks the firstcomprehensive treatment of the theory and itsempirical support. As such, we combine discus-sion of both published and not-yet-publishedexperimentaldatainthesectionsbelow,tobettermakethecaseforhow the DMC theoryprovidesafully integrated account of a variety of cognitive-control phenomena. Moreover, before turning

to experimental findings, we first provide im-portant theoretical background that motivatedthe development of this new theory.

A GENERAL THEORETICAL MODELOF WORKING MEMORY

In this section, we describe the overarching the-oretical framework that guides our work (i.e.,Question #1). Our theoretical model attemptsto specify critical WM components in terms of underlying neurobiologically based computa-tional mechanisms (Botvinick, Braver, Barch,

Carter, & Cohen, 2001; Braver & Cohen, 2000;O’Reilly, Braver, & Cohen, 1999) (see Fig. 4.1for schematic of model). The central hypothesisis that the core of WM is controlled processing:the ability to flexibly adapt behavior to partic-ular task demands, favoring the processingof task-relevant information over competingsources of information and emphasizing goal-compatible behavior over habitual or otherwisedominant responses. This definition is fairly

Learning Sensory

Input

Domain-SpecificKnowledge

(Posterior Cortex)

Response(Motor Cortex)

Goal/Context (PFC)

Active Maintenance

Bias

RewardPrediction / Gating

(DA)

Selection / Updating

Reward

Control Recruitmen

ConflictDetection

(ACC)

Episode(MTL)

Binding

Figure 4.1. Schematic diagram of theoretical framework of working memory. Lines with singlearrowheads reflect excitatory interconnection; bold lines with double arrowheads reflect encodingand retrieval of traces via episodic memory. The excitatory connection from prefrontal cortex(PFC) back to itself represents recurrent connectivity in PFC that mediates active maintenance.The line ending in a square reflects the ability of dopaminergic (DA) projections to modulate orgate inputs into PFC. The line with a circle reflects gradual learning and plasticity involved inreward prediction. ACC¼ anterior cingulate cortex; MTL¼medial temporal lobe.

Dual Mechanisms of Cognitive Control 77



similar to others that have been put forth inthe literature (e.g., Duncan, Emslie, Williams,Johnson, & Freer, 1996; Norman & Shallice,1986) and in this volume (e.g., Chapters 2 and

7). We suggest that controlled processing is anemergent phenomenon, arising from dynamicinteractions between specialized processingsubsystems in the brain. Considering eachsubsystem in isolation cannot fully account forany of the mechanistic properties of WMfunctionality. Nevertheless, we argue that anumber of core brain systems play critical rolesin WM function because of their specializedcomputational properties. A central feature of

our theoretical framework is that these special-izations arise in neural tissue as a means of optimizing a fundamental computational trade-off. The interaction between these specializedsystems enables a kind of global constraintsatisfaction process to occur, whereby maximalflexibility is ensured across the entire range of information-processing situations.

In particular, we suggest that the prefrontalcortex (PFC) is an especially influential struc-

ture in WM because of its extensive connec-tivity with other brain regions and specializedprocessing capabilities. As discussed in muchgreater detail below, the PFC is hypothesizedto play a central role in the active maintenanceof internally represented context information,allowing it to bias processing in other neuralsystems in accordance with this maintainedinformation (e.g., goals, instructions, intermedi-ate products of mental computation). The PFCis aided in this function by its interactive con-nectivity with the hippocampus–medial tem-poral lobe complex (MTL), the midbraindopamine (DA) system, and the anterior cin-gulate cortex (ACC). The MTL complex aug-ments PFC functions through rapid, associativebinding of active representations throughoutthe brain, which can then serve as an auxiliaryform of storage in WM tasks (O’Reilly et al.,

1999; a similar type of mechanism is describedin Chapter 3). The DA system is postulated toregulate the contents of PFC via a dynamicupdating mechanism sensitive to reinforcementcontingencies (Braver & Cohen, 2000). The

ACC is postulated to modulate the general

responsiveness of PFC through a performance-monitoring mechanism that continuously in-dexes the need for top-down control via acomputation of ongoing processing conflict

(Botvinick et al., 2001). The PFC also interactsextensively with posterior brain systems, whichstore domain-specific content knowledge. Thecontent-specific computational specializationswithin posterior cortex may contribute todomain-specificcomponentsofWM,suchasthestorage of sequential-order information associ-ated with verbal WM, and the storage of con-figural and/or movement-based representationsassociated with visuospatial WM.

THE PREFRONTAL CORTEX ANDPROACTIVE COGNITIVE CONTROL

Our research has primarily focused on under-standing the role of the PFC in cognitive con-trol and WM. There is general agreement thatPFC plays a critical role in control functions,but much less agreement concerning the par-

ticular computational and neural mechanismsthat enable it to play such a role. A great dealof data on PFC function has come from single-cell recording studies in nonhuman primatesduring simple WM tasks, such as the delayed-response paradigm. In these studies, a highlyreliable finding is that PFC neurons (particu-larly in dorsolateral regions of PFC) show ele-vated and sustained activity during the retentioninterval (Fuster, 1997). This pattern of activity,along with related results, has been taken asevidence that PFC subserves storage functionsin WM. Human neuroimaging data have pro-vided some support for this notion, showingsustained lateral PFC responses during WMmaintenance periods (e.g., Cohen et al., 1997).However, a growing trend within the neuro-imaging literature has been to emphasize PFCinvolvement in cognitive control rather than

WM per se (Smith & Jonides, 1999). Indeed, anumber of imaging studies have suggested thatWM tasks involving simple storage may notengage PFC at all, or at least not the dorsolateralregions targeted in primate neurophysiologystudies(e.g.,Postle,Berger,&D’Esposito,1999).




Yet, at the same time, there is no clear consensusas to what the specific control functions are thatengage PFC.

Our own work has been guided by the hy-

pothesis that PFC is central to cognitive controland WM because it is specialized to enablethe representation and active maintenance of context information. Context is defined as task-relevant information that is internally repre-sented in such a form that it can bias processingin the pathways responsible for task perfor-mance. Representations of context are similarto the goal representations of other theoreticalformulations (e.g., Norman & Shallice, 1986),

but are flexible enough to influence not onlyaction systems but also perception, attention,memory, and emotion. Likewise, some contextrepresentations should be considered micro-goals, in that they operate over short timescalesor act to bias a narrow range of target repre-sentations.

Representations of context are particularlyimportant in situations where there is strongcompetition for response selection. These situa-

tions may arise when the appropriate response isrelatively infrequent (such as the color name inthe Stroop task). Because context representa-tions are maintained on-line, in an active state,they are continually available to influence pro-cessing. Consequently, context can be thought of as one component of WM. Specifically, contextcan be viewed as the subset of representationswithin WM that govern how other representa-tions are used. In this manner, context repre-sentations simultaneously subserve both mne-monic and control functions. This aspect of themodel differentiates it from classical models of WM (e.g., Baddeley, 1986), which postulate astrict separation of representations for storage vs.control (but for an updated view, see Baddeley,2003).

An important component of our accountof PFC function concerns the interaction

between PFC and the DA neurotransmitter sys-tem, which projects strongly to this region of thebrain. We suggest that active maintenance of context within PFC occurs via local recurrentconnectivity, resulting in a stable, self-sustainingpattern of neural activity (i.e., an attractor, incomputational terms). The DA system is postu-

lated to regulate active maintenance withinPFC, by gating the entrance of information intoPFC, such that only task-relevant context will beactively maintained. We claim that this regula-

tory action occurs in response to phasic bursts of DA release within PFC, which produce a neu-romodulatory effect on PFC neurons, enablingthem to update and actively maintain afferentinputs arriving from other brain regions (Braver& Cohen, 2000). Without such a synchronousburst of DA activity at the time of external inputs,they will only be transiently represented withinPFC, decaying shortly after the external inputstops. For this reason, actively maintained rep-

resentations of task-relevant context in PFC willbe relatively robust to interference from task-irrelevant inputs.

Importantly, the DA system is postulated toalso play a critical role in learning based on pre-dictions of expected reward (i.e., reinforcement-based learning; Schultz, Dayan, & Montague,1997). Because of this learning role, the DA system can self-organize to develop the appro-priate timing of gating signals to enable the

appropriate updating and maintenance of rele-vant context. As such, the system is not a ‘‘ho-munculus,’’ in that it uses simple principles of learning to dynamically configure and adap-tively regulate its own behavior. Moreover, ourhypotheses regarding the functional roles of PFC and DA and their interaction have beenstudied within implemented computationalmodels (e.g., Braver & Cohen, 2000).

For example, we have developed a model of PFC function in a simple delayed-response par-adigm known as the AX-CPT, in which contex-tual cues must be actively maintained over aretention interval to bias processing to a subse-quent probe item. A key aspect of the task is thatin some trial conditions (termed BX ), the con-textual information must be used to inhibit adominant response tendency, whereas in othertrials (termed AY ) context serves an attentional

biasing function. In our computational model of the AX-CPT, the representation and mainte-nance of the context provided by the cue is pos-tulated to occur within PFC. The DA systemregulates the access of this context informationto PFC, such that context can be appropriatelyupdated on a trial-by-trial basis and sustained




control depends upon the use of context infor-mation, the activation of such information byreactive mechanisms occurs transiently ratherthan in a sustained fashion, and thus decaysaway quickly. As a consequence, in situationswhen the same context must be repeatedly ac-cessed, this must occur through full reactivationof the information each time it is needed.

The distinction between proactive and reac-tive control can be thought of as a distinction

between early selection and late correction(Jacoby, Kelley, & McElree, 1999). Concreteexamples can help illustrate the proactive–reactive distinction. A real-world example mightbe the typical prospective memory situation inwhich an intention is formed about a behavioralgoal to be completed at some later point, such asstopping at the dry cleaners after leaving work,before they close. A proactive control strategywould require the goal information to be ac-tively sustained from the time the intention isformed until the goal is satisfied (e.g., the end of the day). The usefulness of such a proactivestrategy is that plans and behaviors can be con-tinually adjusted to facilitate optimal comple-tion of the goal (e.g., not scheduling a latemeeting). In contrast, with a reactive controlstrategy the goal would only be transiently ac-tivated at the time of intention, and then need to

be reactivated again by an appropriate triggerevent (e.g., opening the car door). Because of this need for repeated reactivation, there isgreater dependence on the trigger events them-selves, since if these are insufficiently salient ordiscriminative they will not drive reactivation(e.g., the dry cleaning errand might only be re-

membered because of the cleaning ticket left onthe car seat).

The AX-CPT task, described above, pro-vides another example. A proactive control strat-egy would result in a context representationbeing activated by the cue stimulus, and main-tained at full strength over the intervening de-lay prior to the probe. During this delay,cognitive control would be achieved by prim-ing the perceptual and response systems in

accordance with cue-driven attentional expec-tancies. In contrast, a reactive control strategywould result in context information being onlytransiently represented following the cue. Dur-ing the delay between cue and probe, contextactivation and, as a result, response-related prim-ing would be minimal. Upon presentation of the probe, context would need to be reactivatedvia retrieval. Once contextual representationsreached full activation strength they could beused to overcome interference that had occurredin the interim, due to probe-related biases. Thus,in the AX-CPT, proactive control means controlengaged by the cue, whereas reactive controlmeans control driven by the probe.

The hypothesized distinction between pro-active and reactive control extends to neuralmechanisms. In particular, we have suggestedabove that proactive control requires that con-

text representations be sustained over extendedperiods, whereas in reactive control the repre-sentation of context occurs only transiently, asneeded. Our theory assumes that the repre-sentation and active maintenance of contextoccur in PFC, and most specifically in lateral(rather than medial) PFC regions. Thus, when

TABLE 4.1. Distinctions between Proactive and Reactive Control

Proactive Control Reactive Control

Computational properties Future-oriented, early selection,

preparatory attention

Past-oriented, late correction,

interference resolutionInformation processing Strong goal-relevant focus,global control effects

Increased goal-irrelevant processing,item-specific control

Temporal dynamics Sustained, activation prior toimperative stimulus

Transient, activation afterimperative stimulus

Neural substrates Lateral PFC, midbrain DA (phasic activity)

ACC, lateral PFC (transientresponse), MTL, others

ACC¼ anterior cingulate cortex; DA ¼ dopamine; MTL¼medial temporal lobe; PFC¼ prefrontal cortex.




proactive control is engaged, sustained activityshould be found in these PFC regions duringthe interval between the initial presentation of context and the point at which it is used. Under

proactivecontrolconditions,PFCactivityshouldbe present reliably across events, and not juston those in which it is most needed. In con-trast, under conditions of reactive control, PFCactivity will be (1) transient rather than sus-tained; (2) present only for those events thatdirectly require the reactivation of context tomediate appropriate performance; and (3) ac-tivated after rather than before the onset of an imperative stimulus. Additionally, the two

control mechanisms should differ in terms of the involvement of the DA system. We havesuggested that the ability to sustain inputs inPFC requires a phasic DA-mediated gatingsignal occurring at the time of context presen-tation. Without such a gating signal, PFC canonly be transiently activated. Thus, our hy-pothesis is that under conditions of proactivecontrol, presentation of contextual input is ac-companied by a phasic change in DA, whereas

under reactive conditions there is no DA-mediated gating signal. In the absence of DA gating, PFC can only be transiently activated,and only in situations where there is a strongenough association between the context repre-sentation and the triggering stimulus to pro-duce spreading activation.

Under reactive conditions, we would also ex-pect that other brain systems in addition to PFCwould be more strongly involved in mediatingperformance. For example, if reactive controlcan be achieved through the activation of long-term memory traces or through retrieval of ep-isodic information, then we would expect to seeengagement of either posterior cortical regionsor the hippocampal–MTL complex. Anotherbrain system that might be critical for reactivecontrol is the ACC. A currently influential ac-count of ACC function postulates that this brain

region indexes the demand for cognitive controlby detecting the presence of response conflict oruncertainty due to either interference, weak re-sponse strength, the activation of an erroneousresponse, or the estimated high-likelihood of making an erroneous response (Botvinick et al.,

2001; Brown & Braver, 2005). Critically, con-flict-related ACC signals are postulated tomodulate activation in lateral PFC regions thatcan implement an increase in top-down control

to resolve such conflict. Thus, the conflict signalin ACC might be used to increase the tendencyto use proactive control on subsequent trials, ashas been postulated by current theory (Botvinicket al., 2001). However, it might also be the casethat the ACC serves as a core component of reactive control processing, by rapidly signalingthe need for increased control on the currenttrial, to resolve interference, increase responsestrength, or correct an impending error. In pre-

liminary computational modeling work, we havebeen exploring the hypothesis that the ACC mayserve a dual role in proactive and reactive controlthrough outputs to different PFC systems (De-Pisapia & Braver, 2006).

COSTS AND BENEFITS OF PROACTIVEAND REACTIVE CONTROL

We have just described some of the func-tional and neural characteristics that distinguishreactive from proactive control. Yet an obvi-ous question is the following: why postulatesuch a dual-process account at all, given thevirtues of parsimony and the wisdom of Oc-cam’s razor? This question can best be an-swered by considering that there may be bothcosts and benefits associated with proactive andreactive control, such that a computationaltrade-off exists. By using both mechanisms tovarying degrees through a dual-process controlarchitecture, the cognitive system is best able toovercome these trade-offs, and in so doing op-timize behavioral performance across a widerange of environments and task demands. In-deed, this type of dual-process architecture, in-volving a mixture of proactive and reactivecontrol mechanisms, is one that tends to be

present in many existing computer systems.Specifically, most computer operating systemstend to operate with a standard top-down flow of control driven by a stored program (i.e.,proactive control), but with a separate built-inmechanism to deal with interrupts (i.e., reactive




control). Likewise, general-purpose symboliccomputational cognitive architectures, like Soar(Newell, 1990), frequently operate in two dis-tinct control modes: one in which problem

spaces are traversed according to the dictates of apre-existing goal stack (similar to proactivecontrol), and a second mode that initiates toresolve unexpected impasses or conflicts (simi-lar to reactive control). Thus, on purely compu-tational grounds it is sensible to argue that adual-process or mixture model mechanism of control is one that serves to optimize informa-tion processing. Nevertheless, it is important toconsider these computational trade-offs be-

tween proactive and reactive control more ex-plicitly, since they provide insights into thefactors that should affect which control mode isdominant in specific situations and for specificindividuals.

Below, we list some of the limitations anddisadvantages of proactive control, followed bythe negative consequences of reactive control:

Proactive control requires the presence of pre-

dictive contextual cues. Many times predic-tive contextual information is not presentin the environment, and as such, controlcannot be prepared in advance. In thesecircumstances, the only possible controlstrategy is a reactive one.

Proactive control requires predictive contex-tual cues to be highly reliable. In situationswhere predictive cues turn out to be in-valid, there can be a strong cost if the cue-based contextual information is used asa basis for proactive control. Such cueinvalidity costs can be seen in a rangeof cognitive situations (e.g., the Posnerspatial cued reaction time task; Posner,Snyder, & Davidson, 1980). Thus, adop-tion of a proactive control strategy is onlylikely in situations where contextual cuesserve as highly reliable predictors of up-

coming events or required actions.Proactive control is metabolically costly. Ac-cording to our theoretical model, theactive maintenance of goal-relevant infor-mation requires a high and sustained levelof neuronal activity in lateral PFC during

theentireretentioninterval.Suchextendedperiods of high firing are likely to requireadditional metabolic resources (e.g., forglucose consumption, waste removal, neu-

rotransmitter recycling, etc.) that may notalways be available, or at the minimum,reduce the amount available for other pur-poses.Evenwithoutconsideringmetabolicrequirements directly, it seems clear thatproactive control is capacity demanding,since only a small number of goals can beactively maintained in the focus of atten-tion (Cowan, 2001). Thus, proactive con-trol draws away resources from other active

maintenance demands. Consequently, it islikely to only be used if sufficient capacityis available (i.e., other WM demands arelow, general cognitive resources are high,and cortical arousal is optimal).

Proactive control is prohibitive with very longretention intervals. Because the sustained,activecontextmaintenanceassociatedwithproactive control is so resource demand-ing, the longer the interval between the

maintenance initiation and context utili-zation, the less feasible this strategy be-comes. Thus, proactive control is unlikelywith retention intervals longer than a fewminutes. Certain prospective memorytasks are the best example of situationsin which the interval between goal forma-tion and goal realization can be hours ordays. In such situations, actions are ac-complished through reactive control—that is, by transiently reactivating the goal(via episodic retrieval) at the appearanceof an appropriate trigger stimulus suchthat it can bias the goal-relevant behavior.This idea is consistent with the bulk of theprospective-memory literature, which hassuggested that retrospective processes arethe primary mechanisms guiding delayed-intention behavior (Einstein & McDaniel,

1996). Nevertheless, recent studies havebegun to suggest that in certain experi-mental prospective-memory paradigms,preparatory control may be occurring,even across longer timescales of retention(Smith, 2003). It will be important to




determine more conclusively whether,and under what constraints, active main-tenance processes are being used in suchparadigms.

Proactive control is less sensitive to changes inreward–punishment contingencies. Becauseenvironments are typically non-stationary,contingencies can often change withoutwarning. The active representation of goalsduring proactive control modes causes thesystem to be biased to attend primarily togoal-relevant features of the environment,and to be predisposed to interact withthese features in a goal-driven manner.

This leads to a reduction in inciden-tal encoding of goal-irrelevant or goal-incongruent features, which may, in fact,serve as cues that the environment ischanging. Theorists have suggested thatcontinuous monitoring of environmental(or internal) background information is acritical function of motivationally orientedneural systems. For example, such mech-anisms can lead to optimal detection of

low-probability but potential threats(Goschke, 2003). Thus, high demands ora pre-existing bias for background moni-toring (such as when vigilance towardpotential threats is required) will make theuse of proactive control less likely.

Proactive control impedes the natural pro- gression toward automatization. There is afundamental tension between the exer-tion of cognitive control and the devel-opment of automaticity, which has beentermed the ‘‘control dilemma’’ (Goschke,2003). Because automatic processes arerobust, fast, and efficient, it is likely thatthere is an inherent computational pres-sure or bias on the cognitive system toautomatize processing wherever possible,via strengthening of internal associationsand stimulus–response bindings. Proac-

tive control processes oppose such mech-anisms by providing a sustained top-downflow of information that enables contex-tual goals to override default processing.Indeed, it is reactive rather proactive con-trol that allows for the best optimizationof the control dilemma, by introducing a

highly transient and minimalist (i.e.,only-as-needed) form of intervention, thatallows habits, skills, and procedures to belearned while still enabling the system to

override these forces if necessary.Reactive control is more susceptible to proac-tive interference. Control mechanisms arenecessary because many times the effectsof past experience conflict with currentgoals. However, such sources of proac-tiveinterference (PI)cannot be completelycounteracted by a reactive control strat-egy. This is because reactive control is ini-tiated by post-stimulus processing, such

that potentially interfering stimulus-basedassociations will already be activated bythe time control mechanisms are engaged.In contrast, proactive control may lead tocomplete suppression of PI, via optimalattentional configuration. Thus, in con-ditions where PI effects are very strongand the costs of interference are high, thedisadvantages of reactive control will bemost apparent.

Reactive control is suboptimal when stimulus-driven processing is insufficient. Becausereactive control is stimulus driven ratherthan preparatory, it is always a suboptimalcontrol strategy. However, the limitationsof reactive control are most prominent inconditions where perceptual informationis weak, response selection parameters areunderdetermined, and/or when there is apremium on optimal performance (i.e.,high speed and accuracy constraints).

Reactive control does not maximize rewards.Maximizing reward often depends uponthe ability to predict its occurrence andmagnitude. Proactive control aids in max-imizing rewards through the use of predic-tive contextual cues that can bias actionselection. The strong link between proac-tive control and reward prediction can be

seen in the phasic DA signals postulatedto engage proactive control processes inPFC, according to the DMC account,and also appear to signal reward-relatedsalience of predictive cues, according toinfluential reinforcement learning models(Schultz et al., 1997). In contrast, reactive




control is not geared toward maximizationof rewards but rather toward resolvinginterference and facilitating the transitionto automaticity. Thus, in conditions where

processing is oriented toward reward max-imization, and where reward attainmentdepends on precise focusing of attentionor optimal response preparation, a reactivecontrol strategy will be highly disadvan-tageous.

As the above discussion indicates, there area number of advantages and limitations asso-ciated with both reactive and proactive control,

thus successful cognition may depend on somemixture of both proactive and reactive controlstrategies. Indeed, it may be the case that thetwo systems are fully independent, and thusmay be both engaged simultaneously. Never-theless, there is likely to be some bias favoringone type of control strategy over the other. It isour hypothesis that a default mode for thecognitive system is one favoring reactive con-trol, given its greater applicability (i.e., use in

more situations), lower demands on metabolicresources, and maximal compatibility with thedevelopment of automatization. However, otherfactors may be present that exert pressures onthe system to engage in proactive control. Thesefactors can be characteristics of the task situa-tion, but may also be characteristics of the in-dividual. Indeed, we believe that the DMCaccount provides a unifying framework forunderstanding both intra-individual and inter-individual variability in normal WM functionin terms of shifting biases toward proactive vs.reactive control. Likewise, because of the pu-tative dependence of proactive control on spe-cific neural mechanisms (e.g., DA interactionsin lateral PFC), the DMC framework also pro-vides a coherent explanation as to why specificpopulations suffering from breakdown or dys-function in these neural systems might also ex-

perience specific changes in WM and cognitive-control function. Finally, the DMC accountsuggests that there should be an importantrole for ‘‘noncognitive’’ factors in producingWM variation. This is due to the role playedby constructs such as reward prediction andbackground-threat monitoring in altering the

balance between proactive and reactive con-trol. These constructs are likely to be impactedby related state (e.g., mood) and trait (e.g., per-sonality) variables.

In the sections that follow, we elaborateeach of these points in turn, discussing howthe DMC account makes specific predictionsregarding different sources of variability onWM and cognitive-control function: (a) within-individual variation; (b) cognitive individualdifferences; (c) neural dysfunction; and (d) non-cognitive factors. Moreover, we describe resultsof recent studies designed to provide empiricalsupport for each of these components of the

DMC account.

Within-Individual Variation

A central assumption of the DMC framework isthat a change in situational factors will result inalteration of the weighting between proactiveand reactive control strategies. These situa-tional factors include (1) the availability andreliability of predictive information; (2) length

of time provided to engage in preparation;(3) demands on speed and accuracy; (4) lengthof retention interval; (5) strength of habit orresponse biases; (6) expectation of proactiveinterference; (7) arousal level; (8) motivationalfocus (reward vs. punishment); (9) expectedWM load; and (10) available capacity. Changesin any or all of these factors are thus predictedto produce a change in cognitive-control strat-egy. Thus, the DMC account naturally leadsto the idea that there will be considerable var-iability in the control strategies employed byhealthy individuals across different task situa-tions. Indeed, it is possible that potentiallysubtle differences between otherwise similartasks might lead to large changes in an individu-al’s preferred cognitive-control strategy. Thesecontrol-mode differences would be expected toresult in shifts in both behavioral performance

characteristics and brain activation profiles. Inrecent work, we have examined this aspect of the DMC account, by directly manipulatingfactors expected to influence cognitive-controlstrategy during the performance of WM (andother cognitive) tasks. These experimentswere all conducted using a within-participants




effects will impact the engagement of controleven on the six-item trials with matched load(the role of expectancy was enhanced by pre-senting memory items sequentially, such that

the memory load for the current trial was onlyknown at the end of the encoding period).The results of the study were just as pre-

dicted. Probe decisions on six-item trials wereboth faster and more accurate in the low-loadexpectancy condition. This finding is consis-tent with the hypothesis that in the low-loadcondition probe decisions were based on a ra-pid target-detection process associated withproactive control. However, in the high-load

condition, it appeared as if probes were moredeeply encoded (as assessed by a surprise de-layed-recognition test for non-target probes),consistent with the hypothesis that the probeswere used as retrospective retrieval or matchingcues to achieve reactive control. More impor-tantly, we observed a strong double dissociationin the pattern of activity dynamics within PFC(Fig. 4.2, see color insert). First, in a number of medial and mid-lateral PFC regions, we ob-

served that activity in the low-load conditiontended to progressively increase throughout thememory-set encoding and delay period of thesix-item trials. However, in the high-load con-

dition the same six-item trials, evoked an ac-tivity pattern of rapid increase at the beginningof the memory-set encoding period, but nofurther increase (and, in fact, a tendency to-

ward decreased activity) during the delay. Incontrast, within a more anterior PFC region(an area that has previously been associatedwith episodic retrieval processes; see, e.g. Le-page, Ghaffar, Nyberg, & Tulving, 2000), ac-tivation was found to increase only during theperiod of probe judgment, and was signifi-cantly greater for six-item trials in the high-loadcondition. These observed activity patterns areconsistent with the hypothesis that the mid-

lateral PFC regions reflect the engagement of proactive control processes that are increasedin the low-load condition, while the anteriorPFC is associated with a reactive control pro-cess that is preferentially engaged in the high-load condition.

The central message of the Sternberg studyis that a subtle task factor, such as the expectedmemory load, can dramatically influence bothperformance and brain activation during the

performance of WM tasks. Indeed, the sub-tlety of the manipulation demonstrates that thecognitive-control processes engaged to achievesuccessful performance might be highly vari-

DorsolateralPFC

Low-Load

Expectancy

0.8

0.6

0.4

0.2

–0.2

–0.4

0

0 5

A v e r a g e r e s p o n s e c h a n g e

A v e r a g e r e s p o n s e c h a n g e

10

High-LoadExpectancy

Low-LoadExpectancy

15 20

Time (seconds)

25

Early Enc Late Enc/Del Probe

30 35

AnteriorPFC

High-Load

Expectancy

0.8

0.6

0.4

0.2

–0.2

–0.4

0

0 5 10 15 20

Time (seconds)

25


30 35

Figure 4.2. Memory strategy effects on prefrontal cortex (PFC) activity (Speer et al., 2003). Leftpanel: Left dorsolateral PFC region showing anticipatory, delay-related activation in low-loadexpectancy condition. Right panel: Left anterior PFC region showing increased probe-relatedactivity in high-load expectancy condition. X-axis refers to the time course of activity. Y-axis refersto average percentage fMRI signal change (from baseline).




able even across very similar WM conditions.In particular, some WM conditions might beassociated with a preferential engagement of proactive control, which would be reflected in

sustained (or increasing) lateral PFC activityduring retention periods, while other condi-tions might be associated with a stronger biastoward reactive control, which would be re-flected in reduced retention-period activity inPFC, but potentially a greater probe-driven re-sponse in different brain regions (includingother areas of PFC).

Significantly, these results have importantimplications for resolving some of the apparent

inconsistencies and controversies in the re-cent neuroimaging literature investigating therole of PFC in WM task performance. Thecurrent controversy is whether PFC, and spe-cifically dorsolateral PFC, actually subservesactive maintenance functions in WM, or whe-ther it instead performs ‘‘executive’’ operations(i.e., manipulation, selection) on maintaineditems, but does not subserve maintenance itself (e.g., D’Esposito et al., 1998). Although there is

quite a bit of evidence for the active-maintenanceview from both primate neurophysiologic stud-ies (Fuster, 1997) and human neuroimaging(Cohen et al., 1997), findings from some morerecent studies have directly called into questionthe maintenance account. The strongest evi-dence comes from fMRI studies using event-related designs that isolate PFC activity to dif-ferent periods of a trial, in which the activationof PFC has been observed to be transient ratherthan sustained and most prominent during theencoding or response period rather than themaintenance interval (Rowe, Toni, Josephs,Frackowiak, & Passingham, 2000).

Our results suggest such cross-study differ-ences in the dynamics of PFC activity might ac-tuallybeduetocross-studydifferencesinwhetherproactive vs. reactive control processes werepreferred for task performance. Thus, the lack of

sustained delay–period activity in lateral PFCmight reflect the presence of tasks or subtle taskfactors that encourage reactive rather thanproactive control strategies on the part of par-ticipants. Indeed, undetected cross-study differ-ences in the use of proactive vs. reactive controlmay constitute an uncontrolled source of

variability that can confound clear interpreta-tions of the results of WM studies.

Critically, the DMC account suggests thatthere are a number of task factors, in addition

to expected WM load, that couldlead to variabil-ity in the type of cognitive control strategypreferredacrossdifferenttasksituations.Further-more, we postulate that such condition-relatedvariability in cognitive control may not onlyimpact WM tasks but also other cognitive do-mains that place a high demand on controlprocesses. Indeed, in other studies that we arecurrently conducting, we have observed sub-stantial within-individual variation in proactive

vs. reactive control associated with interferenceexpectancy in the Stroop task (Braver & Hoyer,2006), trial-by-trial fluctuations during taskswitching (Braver, Reynolds, & Donaldson,2003; Reynolds et al., 2006), and motivationalincentives in the AX-CPT (Locke & Braver,2006). Such work will help to provide a betterfoundation on which to design and interpretWM studies by taking into account potentialsources of cognitive-control variation.

Cognitive Individual Differences

An important factor that likely has an influenceon the selection of control strategy is individualdifferences in cognitive abilities. We make thisclaim because many of the psychological fac-tors that should influence cognitive controlstrategies are likely to vary in a stable man-ner across individuals. These include availablecognitive resources, arousal level, and motiva-tional orientation. In terms of the effects of cognitive resources, we have suggested thatengaging in proactive control is more resourcedemanding than engaging in reactive control.Thus it is likely that individuals possessinggreater cognitive resources will be those mostwilling and able to adopt a proactive mode.Indeed, there has been a great deal of research

suggesting that the construct of cognitive re-sources may index the same underlying mech-anism indexed by the constructs of WMcapacity and fluid intelligence (Kane & Engle,2002). Individual differences in these constructshave been shown to have high validity in pre-dicting performance on tasks that place a strong




demand on proactive cognitive control (Dun-can et al., 1996).

It is not yet clear how to directly translatethese constructs into their underlying compu-

tational and neural mechanisms. Nevertheless,there is growing evidence that these constructsare closely linked to PFC function, which pro-vides support for the idea that they have a rela-tionship to proactive cognitive control (Duncanet al., 1996). In particular, we are highly influ-enced by the work of Kane, Engle and col-leagues, who have suggested that the constructsof WM capacity and general fluid intelligence(gF) jointly index the efficacy of PFC function,

and in particular the ability to actively main-tain goal-relevant information in the face of interference (Kane & Engle, 2002; see alsoChapter 2, this volume). In our framework,individuals with high WM-span and high gFshould thus show an increased tendency to useproactive control strategies, but only in the taskdemands that most require and benefit fromsuch strategies.

In a first test of this hypothesis, we examined

the role of gF in predicting performance andbrain activity in the well-known n-back WMparadigm (Gray, Chabris, & Braver, 2003). Wefound that gF was positively correlated withincreased activation in lateral PFC and parietalcortex regions. Moreover, we found that thisrelationship was selective to trial conditionshaving the highest levels of interference (so-called lure nontargets, in which the currentitem is a repeat of a recent trial, but not thecritical n-back trial). Most strikingly, we foundthat the increased PFC and parietal activationon high-interference trials statistically explainedthe facilitated performance that high-gF indi-viduals exhibited on these trials. Together thesefindings provide important new evidence thatindividual differences in gF are associated withindividual differences in the ability to activatecontrol processes in lateral PFC and parietal

cortex that enable the successful managementof interference. However, the Gray et al. (2003)n-back study provides only an indirect test of the DMC account. The DMC account pre-dicts that gF-related individual differencesshould reside primarily in the ability to useproactive control processes. In particular, evi-

dence of increased proactive control in high-gFindividuals should be reflected in increasedand sustained lateral PFC activity prior to onsetof a target event. It is difficult to test such a

hypothesis within the context of the n-back task,since the task design involves a continuous WMload (i.e., active maintenance is continuouslyrequired across each trial and intertrial inter-val). Thus, there is no clear way to distinguishbetween pre-target maintenance and prepara-tion vs. post-target interference resolution.

To more directly test hypotheses based on theDMC account for individual differences in gF,we recently conducted a second study using the

Sternberg paradigm instead of the n-back task(Burgess & Braver, 2004). The Sternberg para-digm enjoys a conceptual advantage over the n-back design, in that it enables the temporaldecomposition of cognitive effects occurring atthe time of the retrieval probe from those oc-curring during encoding or delay periods. Wewere interested in examining whether high-gFindividuals would show an increased tendencyto use proactive control during Sternberg per-

formance. Our specific hypothesis was thathigh-gF individuals would preferentially engageproactive rather than reactive control mecha-nisms during WM performance under high-interference conditions. To examine this hy-pothesis within the context of the Sternbergtask, we used a version that has been popular-ized by Jonides and colleagues (Jonides, Badre,Curtis, Thompson-Schill, & Smith, 2002). Inthis ‘‘recent-negative’’ Sternberg task, probes onsome trials were ‘‘negative,’’ in that they werenot present in the current trial memory-set, butthey were also ‘‘recent,’’ in that they were pres-ent in the memory set from the previous trial.These recent-negative probes produce increasedinterference, since their high familiarity inducesa bias to make an erroneous target response. Assuch, recent-negative trials in the Sternberg taskcan be thought of as being formally similar to

the high-interference lure trials in the n-backtask. This is consistent with the results of a se-ries of neuroimaging studies demonstrating thatrecent-negative trials are associated with in-creased activity in left lateral PFC (Jonideset al., 2002). Consequently, a straightforwardprediction, based on the Gray et al. (2003)




and showed direct sensitivity to recent negativetrials. Our motivation for examining this aspectof the data stemmed from the prior neuroimag-ing literature. In previous studies, Jonides andcolleagues, as well as other groups, have foundthat left ventrolateral PFC (VLPFC) regions areselectively activated by recent-negative trials inthe Sternberg task, suggesting that this region isrecruited to successfully resolve interference(Jonides & Nee, 2006). Moreover, using event-related fMRI, D’Esposito and colleagues de-termined that the increased recent-negativeactivity in left VLPFC occurred at the time of probe onset (D’Esposito, Postle, Jonides, &

Smith, 1999). Such an effect suggests that theleft VLPFC activity indicates the presence of areactive control process that is mobilized fol-lowing the detection of interference to helpresolve potential processing conflict. This issupported by the finding that older adults showincreased behavioral interference on recent-

negative trials, accompanied by decreased left VLPFC activity (Jonides & Nee, 2006)

Based on the results of this literature, onemight expect that it would be effective reactivecontrol rather than proactive control that wouldbe most strongly associated with successful in-terference management on recent-negative tri-als. However, our results did not provide strongsupport for that hypothesis. When collapsingacross conditions and gF, we did find increasedactivation in left VLPFC on recent-negative tri-als (relative to novel negatives) at the time of theprobe (F(1,17)¼ 8.2, p< .05). However, we didnot observe the recent-negative activation to be

reliably greater for high-gF individuals. In fact, ina left VLPFC region very near that observed toshow control effects, recent-negative activity ac-tually tended to be lower in high-gF individualsduring the high-expectancy condition (albeitnonsignificantly; gF Recency: F(1,17)¼ 1.65,p> .1; see Fig. 4.3, see color insert). At first

Figure 4.3. Interactions between general fluid intelligence (gF) and interference expectancy inlateral prefrontal cortex (PFC) activity (Burgess & Braver, 2004). Top panel: High-gF group shows

that high interference expectancy leads to increased delay-related activity in left ventrolateral PFC,whereas the low gF group shows no expectancy effect. Bottom panel: In high-expectancy condition,low-gF group shows increased probe-related activity in a nearby left ventrolateral PFC region forrecent negatives, but high-gF group does not.




blush, this finding is somewhat surprising, sincethe high-interference expectancy condition isassociated with reduced interference effects inthe high-gF individuals. However, the results are

consistent with the DMC account, which sug-gests that high-gF individuals shifted from aprimarily reactive to a primarily proactive con-trol strategy in the high–interference expectancycondition. Since proactive control is thought tobe a more effective strategy for preventing in-terference, a switch toward such mechanismswould improve performance selectively in thehigh-gF group, while simultaneously reducingthe need to engage reactive control mechanisms.

Thus, the results of this study, like our otherwork, provide a clear and coherent interpretationof what might otherwise be counterintuitive re-sults on the role of PFC and individual differ-ences in mediating performance on tasks placinghigh demands on cognitive control processes. Infuture studies, it would be useful to determinewhether the DMC account could also explainthe effects of related individual-difference con-structs on WM and cognitive control (e.g., WM

span). We have recently begun to provide suchevidence, by demonstrating, in a replication of the Gray et al. (2003) study, that lure inter-ference effects in lateral PFC were associatedwith WM span as well as gF (Burgess et al.,2006).

Neural Dysfunction

The previous section discussed situational fac-tors that are likely to influence control biases inhealthy young adults. However, another im-portant factor that will affect control strategy isthe structural integrity of the neural systemssupportingeachmechanism.Ifonesystemisdys-functional, there will be a strong bias towardadopting the other, intact control mode. Ourtheoretical framework suggests that the proac-tive control system will be most vulnerable to

disruption, given its dependence on precisedynamics (i.e., sustained activation of PFC rep-resentations, strong phasic DA response tocontextual cues, moderate tonic DA activity). Infact, the available evidence suggests that manyneuropsychiatric disorders involving cognitive

control impairment, such as schizophrenia, Par-kinson disease, and attention-deficit hyperac-tivity disorder (ADHD), are also associated withdysfunction in PFC and/or DA systems (Arn-

sten & Robbins, 2002). Consequently, our the-ory predicts that these populations will showprimary impairments in the use of proactivestrategies. Importantly, in some cases, it may bethat the control impairment is completely se-lective, such that the reactive control system isintact. A population that likely fits this scenariois healthy older adults. The evidence is accu-mulating that healthy aging is associated withdeclines in both PFC and DA function and

with impairments in cognitive control (Braveret al., 2001). However, given that healthy agingis by definition nonpathological, it is likely thecase that these biological changes are relativelymild (at least in relation to clinical populationssuffering dysfunction in the same systems).

From a theory-testing and validation per-spective it would be ideal if there were evi-dence that other clinical populations showedevidence of the reverse form of impairment—

intact proactive control, but impaired reactivecontrol. However, this pure double dissociationmay be unlikely given the argument that proac-tive control processes are the ones most vul-nerable to disruption by brain dysfunction.Nevertheless, it may be the case that there arecertain populations that show an overrelianceon proactive control, even under conditions thatshould normally favor reactive strategies. Al-though purely speculative at this point, one pop-ulation that may fit this description is patientssuffering from obsessive-compulsive disorder(OCD). In particular, some theorists have sug-gested that OCD can be characterized as ‘‘hy-per-activation’’ of the executive control system(Tallis, 1995). Further work will be needed toinvestigate this idea more directly.

In our own prior work, we have examinedchanges in cognitive control strategy in healthy

older adults. This work has demonstrated thatolder adults show a reduced tendency to en-gage in proactive control, but still show theability to effectively engage reactive controlmechanisms. Specifically, in two studies withthe AX-CPT task, older adults displayed a rel-




ative impairment on inhibitory (BX) trials, interms of disproportionately slowed responding,yet nevertheless showed only a slight increasein error rate in this condition (Braver et al.,

2001,Braver et al., 2005). The fact that olderadults did not make many inhibitory errors sug-gests that they are able to appropriately repres-ent context. However, the greatly increased RTinterference on these trials suggests that contextrepresentation occurred in a reactive ratherthan a proactive fashion. That is, we hypothe-size that under conditions where control isengaged reactively, context information is notrepresented prior to probe onset, and instead

must be reactivated following the appearanceof the probe. The context-activation processmust happen quickly when it occurs reactively,so that it can suppress the priming effect of theprobe before an error is committed. However,even in this case it is still likely that slowing of performance will occur, since during the timeof context reactivation the probe has an op-portunity to prime inappropriate responsepathways. The low error rates but high inter-

ference in older adults indicates that they wereable to achieve control, but that such controlmay have necessitated a more intense engage-ment of reactive control mechanisms on inhib-itory trials (relative to the intensity needed forproactive control). In other words, it is likelythat, because of their increased dependence onreactive control mechanisms, older adults hadto exert compensatory effort to achieve suc-cessful inhibition.

This suggestion is consistent with recentobservations from neuroimaging studies. Inthese studies, older adults have been found toshow increases as well as decreases in brainactivation during performance of difficult cog-nitive control tasks (Cabeza, 2001). Our hypoth-esissuggeststhatashiftfromproactivetoreactivecontrol would result in both the activation of brain regions not typically activated in young

adults (i.e., those subserving reactive control)and in a different pattern of activity dynamicsin regions activated by young adults (i.e., greateractivation in conditions most dependent oncontrol, reduced activity in conditions withlower control demands).

In a recently completed neuroimaging studywith the AX-CPT, we found evidence sup-porting the hypothesis that older adults showedincreased neural activity in conditions low in

control demands, but decreased activity in theconditions associated with proactive control(Paxton et al., 2006). In this study, 20 older(range: 66–83; mean age¼ 73 years) and 21younger adults (range: 18–31; mean age¼23 years) performed the AX-CPT task underboth short (1s) and long (7.5s) delay condi-tions (with total trial duration held constantacross conditions at 10 s). The delay manipu-lation (which was blocked) enabled the isola-

tion of brain regions involved in activelymaintaining cue information, since the onlyvariable manipulated was the proportion of thetrial in which the delay occurred. In addition,task blocks alternated with control (fixation)blocks, which allowed the identification of regions generically (i.e., nonspecifically) acti-vated by task performance. We predicted thatolder adults would show decreased delay-relatedactivity in dorsolateral PFC, indicating a re-

duction in proactive control, while at the sametime showing generalized (i.e., brain-wide) in-creases in task-related activation, indicatinggreater activation of reactive control processes.

The results confirmed the predictions (Fig.4.4, see color insert). Younger adults showed asignificant delay-related increase in dorsolat-eral PFC activity, in a region very similar inlocation to previous studies (Braver et al., 2002).In contrast, older adults actually showed a de-lay-related decrease in the activity of this region,producing a significant AgeDelay interaction(F(1,39)¼ 5.6, p< .05). Interestingly, the inter-action was of the crossover form, such that olderadults showed greater activity than young adultsin the short-delay condition, but less activity inthe long-delay condition. Moreover, in terms of general task-related activation, older adultsshowed a strong trend toward greater activation

in a number of brain regions, including otherregions of PFC.Taken together, these results provide initial

support for the hypothesis that healthy agingproduces a shift from proactive to reactivecontrol that is observable in terms of a




changing pattern of activity in PFC and otherbrain regions. Under conditions of high demandfor proactive control (long delay), older adultsshowed reduced activation, while more gener-ally showing increased activity, consistent with agreater reliance on less effective reactive controlprocesses. Yet more direct investigation of thishypothesis is required. Specifically, these resultswere observed with a block-design study, whichprovided no information on the temporal dy-namics of activity. The use of event-related fMRIwould enable a test of the DMC hypothesis that

older adults would show reduced activity duringthe cue and delay period but increased activationduring the probe (specifically on inhibitory BX trials). Indeed, recent work in our lab, involvingjust this type of event-related design, has begunto provide more conclusive support for the DMCmodel (Paxton et al., 2006).

Noncognitive Factors

A unique aspect of the DMC account is that itprovides a potential means for understandinghow noncognitive factors might influence cog-nitive control. As described earlier, we believethere is a close linkage between proactive con-trol and reward prediction. Conversely, a rela-tionship might exist between reactive controland background threat monitoring and detec-tion. These constructs of reward prediction andthreat detection might be primarily affective in

nature. Indeed, personality theorists have sug-gested that constructs related to reward sensi-tivity and threat sensitivity might represent thetwo fundamental affective dimensions of per-sonality. For example, the theory of J. A. Gray(1994) has described these personality dimen-sions in terms of neural systems that trigger

Figure 4.4. Age-related changes in prefrontal cortex (PFC) activity in the AX-CPT task (Paxtonet al., 2006). Top panel: Right dorsolateral PFC region showing reduction in delay-related acti-vation in older adults in long delay. Bottom panel: Brain regions (including PFC) showing gen-eralized age-related increase in activation.




motivational and goal-directed behaviors. Thebehavioral approach system (BAS), which isroughly linked with extraversion, is motivatedby reward-associated cues, and works to achieve

appetitive outcomes. In contrast, the behavioralinhibition system (BIS), associated with neu-roticism, is driven by threat cues to withdrawfrom potentially aversive outcomes. These traitvariables are also linked to affective states, withhigh-BIS individuals being more susceptible tonegative mood inductions and high-BAS indi-viduals being more susceptible to positive moodinductions (Larsen & Ketelaar, 1991).

Our hypothesis is that a reward-focused mo-

tivational orientation (high BAS sensitivity) isintrinsically proactive, in that achieving complexreward goals requires anticipatory planning andattentional focusing. Conversely, a punishment-focused orientation (high BIS sensitivity) maybias a more reactive state, in which attention isdiffusely vigilant and aroused, monitoring forpotential threats, so that the individual can reactappropriately when any threat appears. Simi-larly, positive moods may increase the tendency

toward a reward-focused, proactive orientation,while negative moods may increase the ten-dency toward a punishment-focused reactiveorientation. The magnitude of these effects islikely to interact with trait sensitivity (e.g., high-BIS individuals will likely have a different re-sponse to situations promoting a negative moodfrom that of low-BIS individuals).

The DMC theoretical framework also sug-gests a possible mechanistic basis on which tointegrate and explain these relationships be-tween affect or personality and cognitive controlin terms of underlying neurobiology. A recentinfluential theoretical analysis of the neurobi-ology of personality has suggested that BAS/ extraversion is directly related to variability inDA function (Depue & Collins, 1999) Likewise,in an extensive series of studies, Davidson andcolleagues have persuasively argued that both

BAS and BIS traits and positive and negativemoods are associated with hemispheric shifts inlateral PFC activity (Davidson, 1995). Theseaccounts are strikingly consistent with the re-ward-prediction aspects of the DMC model. Aswe have described above, the phasic DA re-sponse signals the reward-related salience of envi-

ronmental cues. Encoding these cues as contextin PFC helps to maximize the achievement of reward. Thus, according to the DMC model,high-BAS individuals should be more able to

achieve the precise neural activity dynamicsrequired for proactive control.The DMC account does not provide as rich

a mechanistic framework for understanding theneurobiology of the BIS trait. However, if BIS/ neuroticism is associated with a heightenedsensitivity to threats, this should be associatedwith greater reactivity in a conflict-monitoringsystem used to detect the presence of suchthreats. Thus, it is noteworthy that a number of

studies have reported that BIS/neuroticism isassociated with increased resting-state activitywithin the ACC (e.g., Zald, Mattson, & Pardo,2001), the brain region most strongly associ-ated with conflict detection. However, theseprior studies were not conducted during cog-nitive task performance, which makes it hard todetermine whether the activity reflects in-creasedconflictmonitoringperse. Nevertheless,we speculate that high BIS may be associated

with an increased bias toward reactive controlstrategies.

In our first preliminary studies investigatingthese hypotheses, we examined the role of af-fective states and affect-related personality traitsin modulating behavioral performance andbrain activity during performance of the n-backWM task (Gray & Braver, 2002b; Gray, Braver,& Raichle, 2002). Through initial behavioralstudies, we found that inductions of positive andnegative mood (through viewing of emotionallyevocative video clips) had a striking influenceon performance that selectively interacted withthe n-back task condition (Gray & Braver,2002a). Thus, when participants were inducedinto a positive mood, performance was facili-tated when they were doing a version of the taskinvolving verbal materials, but impaired whenperforming a nonverbal task variant. Conversely,

when a negative mood was induced, the oppo-site pattern of performance modulation was ob-served (improved performance on the nonverbalversion, impaired performance on the verbalversion). This initial finding was replicated andextended in an fMRI study, where we observedthat the crossover interaction effect of mood and




those associated with bodily states. This moni-toring function is postulated to get shut off during cognitive task performance so as to freeup resources for optimal performance. Thus,

the failure of high-BIS individuals to deactivaterostral ACC is consistent with the idea that theycontinue to monitor potential threats during n-back task performance to more effectively reactto such threats if they appear. However, the in-creased background-monitoring activity divertsresources away from proactive control pro-cesses, making them less effective.

The studies just discussed suggest an impor-tant way in which ‘‘noncognitive’’ factors such

as personality and affect might impact cogni-tive control strategy. Although work in this areais just in its infancy, we strongly believe thatconsideration of affective influences on cogni-tive control is essential for a full mechanisticunderstanding of WM and its sources of vari-ation. The DMC framework provides one ac-count for synthesizing and understanding waysin which affect and personality might contrib-ute to WM variation, by linking the more

proximal constructs of reward prediction andthreat detection (and their associated neuralmechanisms) to proactive and reactive controlmodes, respectively. Nevertheless, the relation-ship between affect–personality and cognitivecontrol is likely to be somewhat more complexthan our original formulation, as these initialstudies indicate (i.e., we had initially predictedthat BIS would be associated with increasedtrial-specific activity in caudal ACC, but actu-ally found an association with tonic activity inrostral ACC). Thus, further progress in thisarea may require more focused attention ondesigning WM experiments in a manner thatwill provide enhanced sensitivity to influencesof affect and personality on cognitive controlstrategy, via manipulation of factors such asmotivational incentives, ego threats, or emo-tionally evocative stimuli.

Other Sources of Variationin Working Memory

In the preceding sections we have providedevidence that many aspects of variation in WM

function—within-individual effects, individualdifferences, effects of age and other certaintypes of neural dysfunction, and noncognitiveinfluences—can all be explained by appeal to a

dual-process architecture of cognitive controlthat we term the DMC account. In this respect,we share a common goal with the other con-tributors to this volume of trying to understandthe common or distinct mechanisms under-lying the many kinds of WM variation. In thissection, we discuss how the DMC account re-lates to the other accounts of WM variation putforth in this volume, and more generally con-sider whether additional sources of variation are

needed to account for the full range of empir-ical phenomena related to WM function(i.e., Question #4). In particular, we consider anumber of outstanding issues related to varia-tion in WM: neural and computational mech-anisms, other dual-process accounts, inhibition,domain specificity vs. domain generality, de-velopment, and genetics.

Neural and Computational Mechanisms

Our view of WM variation is highly similar tothe account put forward by Munakata et al.(Chapter 7), in terms of relating active main-tenance and updating to the interactive func-tion of the PFC and DA system. Likewise, weshare with them a deep interest in translatingWM constructs into explicit neural and com-putational mechanisms. The similarity of ourframeworks and approach is no coincidence, aswe are frequent collaborators and have devel-oped our central theoretical ideas, for the mostpart, in tandem. Nevertheless, Munakata et al.place a central emphasis on the role of the basalganglia (BG) in mediating WM updating viaphasic gating signals (O’Reilly & Frank, 2006).This idea is very similar to our own ideasregarding DA and gating, but is also subtly

different, focusing on how the architecture of the BG can enable hierarchical updating tooccur within complex WM tasks. This is a po-tentially important direction for understandingWM and its potential breakdown in diseasessuch as Parkinson disease. It is not yet clear




whether such distinctions will have any impli-cations for the development of the DMCaccount.

Reuter-Lorenz and Jonides (Chapter 10)

also have a neural systems-oriented view of WM variation. In their account, lateral PFCareas serve attentional-control functions inWM tasks, but they also postulate that all WMtasks require some degree of control (whichexplains the near ubiquity of lateral PFC ac-tivity during WM). Additionally, Reuter-Lorenzand Jonides call attention to the fact that evenvery subtle task manipulations can lead to sub-stantial changes in behavioral performance and

PFC activity, by altering control demands.These views are highly similar to our own, ex-cept that we further suggest that it is importantto distinguish between the demands placed onproactive vs. reactive control, and to specifywhich mechanism is likely to be dominant in agiven task context. Such distinctions in controlwill have important ramifications on the way inwhich different PFC regions are engaged inWM tasks—with either sustained or transient

dynamics, and in either a cue-specific or moreglobal manner.

Other Dual-Process Frameworks

The executive attention account described byKane et al. (Chapter 2) is also highly compati-ble with our own, and has highly influencedour thinking on the role of cognitive individualdifferences in WM. Like our group, Kane et al.posit that variation in WM performance is pri-marily linked to the active maintenance of goal-relevant information rather than storage capac-ity per se. Kane et al. also argue that active goalmaintenance, which in our terminology wouldrefer to proactive control, is most critical whenPI effects are prominent and must be overcome.Interestingly, Kane and colleagues have alsobegun to develop a dual-process account of

executive attention, which contrasts active goalmaintenance with conflict resolution. This ac-count seems very similar to our own distinctionbetween proactive and reactive control, sincewe argue that reactive control occurs preciselyunder conditions where high conflict is de-

tected, but not anticipated or prevented viaproactive control. Kane et al. make their dis-tinction on the basis of data from the Strooptask that indicate low-WM span individuals

show heightened interference (and facilitation)effects selectively under low-interference expec-tancy conditions. This finding seemingly paral-lels our own results from the recent-negativeSternberg task (Burgess & Braver, 2004), inwhich low-gF individuals exhibited increasedinterference effects related to expectancy. How-ever, in one way our results are exactly oppo-site those of Kane et al., since in our studygF differences were present selectively in the

high-interference expectancy rather than low-expectancy condition. The discrepancy be-tween these two sets of results is puzzling, butmay be due to subtle cross-study differencesin tasks (Stroop vs. Sternberg) and individual-difference constructs (WM span vs. gF). Nev-ertheless, further investigation of the issue seemswarranted.

Inhibition An important component of the DMC model isthat both proactive and reactive control are of-ten invoked in the service of interference man-agement. This view is similar to that espousedby Hasher et al. (Chapter 9), who highlightchanges in the efficacy of inhibitory controlprocesses as a major source of both age-relatedand within-individual (e.g., time-of-day effects)WM variation. Like Kane et al., Hasher andcolleagues have convincingly demonstratedthrough a series of studies that differences inWM function are unlikely to be due to storagecapacity. However, Hasher et al. go further bypostulating three distinct forms of inhibitorycontrol relevant to WM: access, deletion, andrestraint. Although we are sympathetic to theappeal of these types of constructs, an importantdistinction between the account of Hasher et al.

and our own (and those of Kane et al. andMunakata et al. as well) is the causal primacyattributed to inhibitory control. In the DMCaccount, proactive control mechanisms serve toprevent interference, while reactive controlprocesses can detect and suppress interference




when it occurs. However, the control mecha-nisms themselves are not inhibitory per se, butrather achieve inhibition as an emergent con-sequence of active goal maintenance exerting a

top-down bias on local competition withinposterior brain systems. These types of emer-gent effects are most easily understood throughthe use of computational models that translatepsychological constructs into specific under-lying mechanisms, in often a non–one-to-onemanner. Thus, we might argue that the ‘‘access’’and ‘‘restraint’’ inhibitory functions might bothrelate to consequences of proactive control, butunder different task contexts. Similarly, ‘‘dele-

tion’’ might relate to effective DA-mediatedupdating in PFC. However, explicit simula-tions of tasks and behavioral phenomena wouldbe required to determine whether such map-pings are truly applicable.

Domain Specificity vs.Domain Generality

The DMC account can be considered a

domain-general model of WM variation, sincewe postulate that a single source of variability,between proactive and reactive control, canaccount for empirical phenomena under awide range of circumstances. There are otheraccounts of WM variation that are also domain-general, but which posit very distinct sources of variability. For example, a domain-general ac-count of WM variation discussed by Hale et al.in Chapter 8 is the widely known processing-speed model. Much attention has been given tothe processing-speed construct in the WM lit-erature, especially with regard to aging, and itappears to be highly successful as a source of explanationforWMvariability(Salthouse,1994).It is our hope that the construct can be fleshedout further into mechanistic terms so that itsinteractions with other dimensions of informa-tion processing can be better appreciated. For

example, in other recent work, Hale, Myerson,and colleagues present a mathematical modelthat specifies how computational speed mightinteract with individual differences, group dif-ferences, and task complexity to produce thediversity of speed-related behavioral phenom-

ena that have been observed in cognitive per-formance (Myerson, Hale, Zheng, Jenkins, &Widaman, 2003).

It is clear that any comprehensive account

of WM variation will also have to deal with theobvious variation that occurs across individuals(and developmentally, within individuals) inregard to domain-specific knowledge. The brainis an organ defined by experience-dependentplasticity. Thus, individuals with different ex-periences are surely to have different knowledgestoredinthevariousbrainsystemsspecializedforcoding these experiences. Likewise, just as someindividuals have greater perceptual acuity than

others, likely as a result of domain-specific ge-netic variation, so might there be domain-specific differences in cognitive abilities, suchas verbal and spatial reasoning, that will likelyimpact WM function. Indeed, these domain-specific individual differences may be asimportant a source of WM variation as domain-general factors. A number of chapters have pro-vided convincing evidence supporting such aposition (e.g., Chapters 6 and 8). A critical

future direction for studies of WM will be todiscover and explore the nature of interactionbetween domain-specific and domain-generalsources of variability.

Development

In addition to the accounts and views of WMvariation with which we hope our work inter-sects, we also recognize that there are likely othersources of important variation that are relevantfor understanding WM function but are notwithin the purview of our current model. Forexample, developmental approaches to WM areextensively considered by a number of investi-gators within this volume (see Chapters 5, 6, 7,and 8). The developmental maturation of thecognitive system and the concomitant neuralchanges that occur with it undoubtedly serve as

both important sources of variation and impor-tant tools for discovering the causal mechanismsthat underlie cognitive developmental variabil-ity. We have not explored yet whether the DMCaccount would be useful for understanding thenature of cognitive control in young children,




the traditional purview of cognitive scien-tists, that can influence WM task perfor-mance, by shifting which control strategy isdominant. An example is the influence of

the personality traits BIS and BAS on ACCand lateral PFC activity during perfor-mance of the n-back task (Gray et al., 2005).

Probably the most important take-homemessage of our theoretical account to general

WM theory is that of an appreciation of WMvariability itself (see Box 4.1 for a summarystatement of our answers to the four centralchapter questions). We suggest that an under-

standing of the variability between proactiveand reactive control is fundamental to under-standing the core mechanisms of WM. Thecritical point of the DMC account is thatvariability between dual control mechanismscan be a naturally occurring part of cognition,



OF WORKING MEMORY

Our general approach is to link psychological

constructs of WM to underlying neural and

computational mechanisms. We suggest that

WM is an emergent phenomenon arising from

the interaction of multiple mechanisms (ac-

tive context representation, dynamic updating,

conflict detection, and binding). However, wegive particular focus to the influential role of

lateral PFC in mediating top-down biases over

processing via actively maintained context or

goal representations.

2. CRITICAL SOURCES OF WORKING

MEMORY VARIATION

We suggest that the distinction between

proactive and reactive cognitive control maybe the core source of WM variation. Proactive

control enables the optimal task preparation

and prevention of interference via sustained

goal maintenance in lateral PFC. Reactive

control provides an as-needed, just-in-time

form of interference resolution or context re-

trieval via transient activation of the PFC or

related brain systems (e.g., MTL, ACC). Be-

cause these dual mechanisms of control each

have computational advantages and disad-vantages, shifts in the dominant cognitive

control mode can arise from situational factors

(intra-individual variation), individual differ-

ences, neural dysfunction, and noncognitive

factors such as mood and personality.


MEMORY VARIATION

Many other sources of variation should be in-

cluded in a comprehensive WM theory. We

consider some of the differing sources dis-

cussed by other contributors to this volume:

neural mechanisms (e.g., the basal ganglia,

Chapter 8), other dual-process accounts (Chap-

ter 2), inhibition (Chapter 9), processing speed(Chapter 8), domain-specific mechanisms

(Chapter 6), and development (Chapter 5).

Additionally, we think that genetic variation

will be an important focus of future WM re-

search.



We believe that the DMC model put forward

here can provide a unifying framework for un-

derstanding the many varieties of WM variation.

The model (a) indicates the task and individual-

difference factors that should influence WM task

performance; (b) clarifies the nature and dy-

namics of PFC activity in WM tasks; and (c) links

specific forms of neural dysfunction to stable

shifts in cognitive control strategy. The concep-

tualization of cognitive control mechanisms interms of computational specialization and trade-

offs provides a coherent causal explanation for

the occurrence of variability in complex cogni-

tive activities, such as WM.




but can also become more prominent undervarious changes in internal states and externalsituations. Moreover, the DMC account doesnot posit a distinction between the types of

WM variation that occur on an intra-individualbasis from those that occur on an interindividualbasis. In other words, regardless of the sourceof WM variation—task factors, state factors,cognitive individual differences, personalitydifferences, or population differences—theproximal mechanisms of variation remain thesame and have the same impact on brain ac-tivity and behavior. Thus, the DMC frameworkprovides a unifying account that has the po-

tential to synthesize and integrate a large bodyof literatures on WM function. By recognizingthat there are multiple alternative routes to cog-nitive control, investigators may be in a betterposition to explore and investigate the com-plexity of empirical findings and thus moreeffectively manage the previously impossiblydifficult task of defining the unifying latentWM constructs that will replicate across tasks,individuals, and cognitive domains.

References

Arnsten, A. F. T., & Robbins, T. W. (2002). Neuro-

chemical modulation of prefrontal cortical

function in humans and animals. In D. T. Stuss

& R. T. Knight (Eds.), Principles of frontal lobe

function (pp. 51–84). New York: Oxford Uni-

versity Press.



Baddeley, A.D. (2003). Working memory: Looking

back and looking forward. Nature Reviews Neu-

roscience, 4, 829–839.

Botvinick, M. M., Braver, T. S., Barch, D. M.,

Carter, C. S., & Cohen, J. C. (2001). Conflict

monitoring and cognitive control. Psychological

Review, 108, 624–652.

Braver, T. S., Barch, D. M., Keys, B. A., Carter,

C. S., Cohen, J. D., Kaye, J. A., et al. (2001).Context processing in older adults: Evidence

for a theory relating cognitive control to neurobi-

ology in healthy aging. Journal of Experimental

Psychology: General, 130, 746–763.

Braver, T. S., & Cohen, J. D. (2000). On the control

of control: The role of dopamine in regulating

prefrontal function and working memory. In

S. Monsell & J. Driver (Eds.), Attention and Per-

formance XVIII (pp. 713–738). Cambridge,

MA: MIT Press.

Braver, T. S., Cohen, J. D., & Barch, D. M. (2002).The role of the prefrontal cortex in normal and

disordered cognitive control: A cognitive neuro-

science perspective. In D. T. Stuss & R. T. Knight

(Eds.), Principles of frontal lobe function (pp.

428–448). New York: Oxford University Press.

Braver, T. S., Reynolds, J. R., & Donaldson, D. I.

(2003). Neural mechanisms of transient and

sustained cognitive control during task switch-

ing. Neuron, 39, 713–726.

Braver, T.S., Satpute, A.B., Rush, B.K., Racine,C.A., Barch, D.M. (2005). Context processing

and context maintenance in healthy aging and

early stage dementia of the Alzheimer’s type.


Braver, T. S., & Hoyer, C. M. (2006). Neural

mechanisms of proactive and reactive cognitive

control. Submitted manuscript.

Brown, J.W., and Braver, T.S. (2005). Learned

predictions of error likelihood in the anterior

cingulate cortex. Science, 307, 1118–1121.Burgess, G. C., & Braver, T. S. (2004). Dual mech-

anisms of cognitive control over interference.

Presented at the Cognitive Neuroscience Society,

Eleventh Annual Meeting, San Francisco, CA.

Burgess, G.C., Braver, T.S., and Gray, J.R. (2006).

Exactly how are fluid intelligence, working

memory, and executive function related? Cog-

nitive neuroscience approaches to investigating

the mechanisms of fluid cognition. Behavioral

and Brain Sciences, 29, 128–129.

Cabeza, R. (2001). Functional neuroimaging of

cognitive aging. In R. Cabeza & A. Kingstone

(Eds.), Handbook of functional neuroimaging of

cognition. Cambridge, MA: MIT Press.

Cohen, J. D., Perstein, W. M., Braver, T. S.,

Nystrom, L. E., Noll, D. C., Jonides, J., et al.

(1997). Temporal dynamics of brain activation

during a working memory task. Nature, 386,

604–608.Cowan, N. (2001). The magical number 4 in short-



24, 87–185.

Davidson, R. J. (1995). Cerebral asymmetry, emo-

tion, and affective style. In R. J. Davidson &




K. Hugdahl (Eds.), Brain asymmetry (pp. 361–

387). Cambridge, MA: MIT Press.

DePisapia, N., & Braver, T. S. (2006). A model of dual

control mechanisms through anterior cingulate

and prefrontal cortex interactions. Neurocom-puting, 69, 1322–1326.

Depue, R. A., & Collins, P. F. (1999). Neurobiology

of the structure of personality: Dopamine,

facilitation of incentive motivation, and extra-

version. Behavioral and Brain Sciences, 22,

491–569.

D’Esposito, M., Aguirre, G. K., Zarahn, E., Ballard,

D., Shin, R. K., & Lease, J. (1998). Functional

MRI studies of spatial and nonspatial working

memory. Cognitive Brain Research, 7, 1–13.D’Esposito, M., Postle, B. R., Jonides, J., & Smith,

E. E. (1999). The neural substrate and tempo-

ral dynamics of interference effects in working

memory as revealed by event-related functional

MRI. Proceedings of the National Academy of

Sciences USA, 96, 7514–7519.

Diamond, A. (2002). Normal development of prefron-

tal cortex from birth to young adulthood:

Cognitive functions, anatomy, and biochemis-

try. In D. T. Stuss & R. T. Knight (Eds.),Principles of frontal lobe function (pp. 466–

503). New York: Oxford University Press.

Drevets, W. C., & Raichle, M. E. (1998). Reciprocal

suppression of regional cerebral blood flow

during emotional versus higher cognitive pro-

cesses: Implications for interactions between

emotion and cognition. Cognition and Emo-

tion, 12, 353–385.

Duncan, J., Emslie, H., Williams, P., Johnson, R., &

Freer, C. (1996). Intelligence and the frontal

lobe: The organization of goal-directed behav-

ior. Cognitive Psychology, 30, 257–303.

Einstein, G. O., & McDaniel, M. A. (1996). Re-

trieval processes in prospective memory: The-

oretical approaches and some new empirical

findings. In M. Brandimonte, G. O. Einstein,

& M. A. McDaniel (Eds.), Prospective memory:

Theory and applications (pp. 115–142). Mah-

wah, NJ: Lawrence Erlbaum Associates.Fuster, J. (1997). The prefrontal cortex (3rd ed.).

New York: Lippincott-Raven.

Goschke, T. (2003). Voluntary action and cognitive

control from a cognitive neuroscience perspec-

tive. In S. Maasen, W. Prinz, & G. Roth (Eds.),

Voluntary action: An issue at the interface of

nature and culture (pp. 49–85). Oxford, UK:


Grasby, P. M., Frith, C. D., Friston, K. J., Simpson,

J., Fletcher, P. C., Frackowiak, R. S. J., et al.(1994). A graded task approach to the func-

tional mapping of brain areas implicated in

auditory-verbal memory. Brain, 117, 1271–

1282.

Gray, J. A. (1994). Personality dimensions and

emotion systems. In P. Ekman & R. J. Davidson

(Eds.), The nature of emotion (pp. 329–331).

New York: Oxford University Press.

Gray, J. R., & Braver, T. S. (2002a). Integration of

emotion and cognitive control: A neurocom-putational hypothesis of dynamic goal regula-

tion. In S. C. Moore & M. R. Oaksford (Eds.),

Emotional cognition (pp. 289–316). Amster-

dam: John Benjamins.

Gray, J. R., & Braver, T. S. (2002b). Personality

predicts working memory related activation

in the caudal anterior cingulate cortex. Cogni-

tive, Affective, & Behavioral Neuroscience, 2, 64–

75.

Gray, J. R., Braver, T. S., & Raichle, M. E. (2002).Integration of emotion and cognition in the

lateral prefrontal cortex. Proceedings of the

National Academy of Sciences USA, 99, 4115–

4120.

Gray, J. R., Burgess, G. C., Schaefer, A., Yarkoni, T.,

Larsen, R. L., & Braver, T. S. (2005). Affective

personality differences in neural processing effi-

ciency confirmed using fMRI. Cognitive, Affec-

tive, & Behavioral Neuroscience, 5, 182–190.

Gray, J. R., Chabris, C. F., & Braver, T. S. (2003).


gence. Nature Neuroscience, 6, 316–322.

Jacoby, L. L., Kelley, C. M., & McElree, B. D.

(1999). The role of cognitive control: Early

selection versus late correction. In S. Chaiken &

E. Trope (Eds.), Dual process theories in social

psychology (pp. 383–400). New York: Guilford

Press.

Jonides, J., Badre, D., Curtis, C., Thompson-Schill,S. L., & Smith, E. E. (2002). Mechanisms of

conflict resolution in prefrontal cortex. In D. T.

Stuss & R. T. Knight (Eds.), Principles of frontal

lobe function (pp. 233–245). New York: Oxford

University Press.




Jonides, J. & Nee, D.E. (2006). Brain mechanisms

of proactive interference in working memory.

Neuroscience, 139, 181–193.


prefrontal cortex in working-memory capacity,executive attention and general fluid intelli-

gence: An individual-differences perspective.


Larsen, R. J., & Ketelaar, E. (1991). Personality and

susceptibility to positive and negative emo-

tional states. Journal of Personality and Social


Lepage, M., Ghaffar, O., Nyberg, L., & Tulving, E.

(2000). Prefrontal cortex and episodic memory

retrieval mode. Proceedings of the National Academy of Sciences USA, 97, 506–511.

Lieberman, M. D., & Rosenthal, R. (2001). Why

introverts can’t always tell who likes them:

Multitasking and nonverbal decoding. Jour-

nal of Personality and Social Psychology, 80,

294–310.

Locke, H. S., & Braver, T. S. (2006). Motivational

influences on cognitive control: Behavior, brain

activity, and individual differences. Submitted

manuscript.Mattay, V. S., Goldberg, T. E., Fera, F., Hariri, A. R.,

Tessitore, A., Egan, M. F., et al. (2003).

COMT genotype and individual variation in

the brain response to amphetamine. Proceed-

ings of the National Academy of Sciences USA,

100, 6186–6191.

Myerson, J., Hale, S., Zheng, Y., Jenkins, L., &

Widaman, K. F. (2003). The difference engine:

A model of diversity in speeded cognition.


Newell, A. (1990). Unified theories of cogntion. New

York: Cambridge University Press.


action: Willed and automatic control of behav-

ior.InR.J.Davidson,G.E.Schwartz,&D.Shapiro

(Eds.), Consciousness and self-regulation (Vol. 4,

pp. 1–18). New York: Plenum Press.

O’Reilly, R. C., Braver, T. S., & Cohen, J. D.

(1999). A biologically based computationalmodel of working memory. In A. Miyake &

P. Shah (Eds.), Models of working memory:

Mechanisms of active maintenance and execu-

tive control (pp. 375–411). New York: Cam-

bridge University Press.

O’Reilly, R.C. & Frank, M.J. (2006). Making

working memory work: A computational

model of learning in the frontal cortex and

basal ganglia. Neural Computation, 18, 283–

328.Paxton, J.L., Barch, D. M., Racine, C. A., & Braver,

T. S. (2006). Prefrontal function in healthy

aging: Context processing and proactive versus

reactive cognitive control. Submitted manu-

script.

Posner, M. I., Snyder, C. R. R., & Davidson, B. J.

(1980). Attention and the detection of signals.


109, 160–174.

Postle, B. R., Berger, J. S., & D’Esposito, M. (1999).Functional neuroanatomical double disso-

ciation of mnemonic and executive control

processes contributing to working memory

performance. Proceedings of the National Acad-

emy of Sciences USA, 96, 12959–12964.

Reynolds, J.R., Braver, T.S., Brown, J.W., Van der

Stigchel, S. (2006). Computational and neural

mechanisms of task-switching. Neurocomput-

ing, 69, 1332–1336.

Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R. S. J.,& Passingham, R. E. (2000). The prefrontal cor-

tex: Response selection or maintenance within

working memory? Science, 288, 1656–1660.

Salthouse, T. A. (1994). The aging of working

memory. Neuropsychology, 8, 535–543.

Schultz, W., Dayan, P., & Montague, P. R. (1997).

A neural substrate of prediction and reward.

Science, 275, 1593–1599.

Smith, E. E., & Jonides, J. (1999). Storage and

executive processes in the frontal lobes. Science,

283, 1657–1661.

Smith, R. E. (2003). The cost of remembering to

remember in event-based prospective memory:

Investigating the capacity demands of delayed

intention performance. Journal of Experimental


29, 347–361.

Speer, N. K., Jacoby, L. L., & Braver, T. S. (2003).

Strategy-dependent changes in memory: Effectson brain activity and behavior. Cognitive,

Affective, & Behavioral Neuroscience, 3, 155–

167.






Tallis, F. (1995). Obsessive compulsive disorder: A

cognitive and neuropsychological perspective.

Chichester, UK: John Wiley & Sons.

Visscher, K. M., Miezin, F. M., Kelly, J. E., Buckner,

R. L., Donaldson, D. I., McAvoy, M. P., et al.(2003). Mixed block/event-related designs sep-

arate transient and sustained activity in fMRI.

NeuroImage, 19, 1694–1708.

Zald, D., Mattson, D., & Pardo, J. (2001). Brain

activity in ventromedial prefrontal cortex pre-

dicts negative temperament. NeuroImage,13(6), S490.




II

Working Memory Variation

Due to Normal and

Atypical Development



QUESTION 1: OVERARCHINGTHEORY OF WORKING MEMORY

What is the theory or definition of working mem-

ory that guides your research on working memoryvariation? Ourresearchinitiallystemmedfromthe general framework for working memoryoutlined by Baddeley and Hitch (1974) and themore specific model proposed by Baddeley(1986). Furthermore, we acknowledge the rel-evance of working memory for understandingphenomena among children (for example, seeHalliday & Hitch, 1988; Hitch & Halliday,1983). Accordingly, we take the position that

working memory is a multicomponent, limited-capacity system responsible for retaining as wellas transforming fragile representations. Thecomponents of working memory are assumedto comprise a central executive, a phonologicalloop, and a visuospatial sketchpad. The cen-tral executive is the hub of these components,controlling and orchestrating the operation of dedicated, modality-specific memory systems.More recently Baddeley (2000) has proposed a

fourth component, a multimodal episodic buf-fer that is closely connected to the central ex-ecutive. Although the episodic buffer has notbeen a direct stimulus for our research, wesuggest that it is quite compatible with severalconclusions that come from it.

In addition to making assumptions about thearchitecture of working memory, and thereforethe ways in which temporary information is re-presented, our theoretical approach stresses theimportance of processing operations. In partic-ular, we take the view that the interaction be-tween memory and processing lies at the heartof understanding the limitations of workingmemory.

Our program of research has focused onworking memory span. This task was first devel-oped within a slightly different theoreticalframework in which working memory is con-

ceived as an undifferentiated general resource(Daneman & Carpenter, 1980). According tothis approach, working memory span measuresthe ability to commit mental resources to mem-ory as well as concurrent processing events. Interms of the multicomponent working memory

model we have just articulated, working memoryspan is regarded as involving the central execu-tive. However, like some others, we suspect thatcomplex working memory tasks only partially

overlap with central executive functioning. Oneissue is that these complex tasks likely incorpo-rate some characteristics of slave system func-tioning. An additional concern is that centralexecutive functioning is not just about servic-ing the requirements of these span tasks, but hasindependent responsibilities too.

With this in mind, we argue that variation inworking memory performance can lie on severaldistinct dimensions. Performance on tasks in-

volving working memory will vary as a result of which components of its architecture are ac-cessed, for example, the degree to which verbalrehearsal contributes to the memory representa-tion. Performance can also vary as a result of differences in the use of these components—forexample, articulation speed affects the efficiencyof rehearsal. Critically for us, the interplay be-tween the processing and memory demands of complex working memory tests contributes to

performance. These demands go beyond theconstraints of the individual slave systems, and inthis sense our theoretical account is an attempt todevelop the account of working memory limita-tions provided by the multicomponent model.

THE THEORETICAL VISIONOF WORKING MEMORY

Baddeley and Hitch’s (1974) initial study onworking memory represents something of aback-to-the-drawing-board approach in termsof the conceptualization of short-term memoryprocesses. Rather than take on and attempt todevelop a specific and well-worn paradigm, oradapt a theoretical model prevalent at the time,Baddeley and Hitch described a varied series of experiments into the functional characteris-

tics of short-term memory. These were used toprovide a perspective on short-term memoryas a limited-capacity working memory, the‘‘workbench of cognition’’ (Klatzky, 1980) or‘‘the interface between memory and cognition’’(Baddeley, 1994). Thus, it was recognized that

110 Variation Due to Normal and Atypical Development



short-term memory is not just a system for theretention of information. Instead, immediatememory processes are an integral part of cog-nitive activities, shaping and constraining how

thought processes take place. This idea forms acornerstone of what Baddeley (1986) subse-quently referred to as working memory—general(WMG), the general framework within whichworking memory could be understood.

Baddeley and Hitch (1974) also proposed thatworking memory involved the combination of specialized, modality-based memory traces andmore generic, flexible memory processes (part of what Baddeley, 1986, referred to as WMS, or

working memory—specific). In particular, theyfocused on the idea that working memory in-corporated a verbally based system that allowedindividuals to remember a few items for a shortperiod of time, supported by rehearsal of thoseitems. They also hinted at the idea that a spe-cialized visual memory system might also con-tribute to working memory. These proposalshave been influential in leading to the study of the characteristics of these specialized memory

systems(referredtoasslavesystems),oftenthroughthe use of dual-task interference paradigms.These paradigms include articulatory suppres-sion (repeating an irrelevant phrase during amemory trial; see Baddeley, Lewis, & Vallar,1984), concurrent spatial tracking (Baddeley &Lieberman, 1980), and dynamic visual noise(Quinn & McConnell, 1996), where a changingvisual display is watched while trying to remem-ber a sequence of items. Different tasks causedifferent patterns of interference and these effectscanbe interpreted in terms of selective disruptionof different components of working memory.There has been a steady accumulation of knowl-edge about these slave memory systems, and theyform part of the specific model or implementa-tion of working memory outlined by Baddeley(1986).

In the present chapter we consider the ap-

plicability of the multicomponent model to thedevelopment of working memory in children.We focus on working memory span, as this iswidely used as a measure of individual differ-ences in children and adults, and yet our un-derstanding of this task is incomplete.

WORKING MEMORY ANDCOGNITIVE DEVELOPMENT

Although the multicomponent model of work-

ing memory was proposed as a theoretical ac-count of adult memory performance, it hasbeen fruitfully applied to a rangeof developmen-tal issues. In several cases, research has shownthat changes in memory among primary-schoolchildren can be attributed to variations in thestrategies that children use. Verbal recoding of visually presented material (whether images orwords) is not ubiquitous (see Halliday & Hitch,1988). At around the age of 8 years, children

become increasingly consistent in their sensi-tivity to phenomena such as word length effectsand phonological similarity effects even whenmaterial is presented in a nonverbal form. Con-vergent with these results, children below about7 years of age are sensitive to visual similarityeffects when remembering pictorial stimuli(Hitch, Woodin, & Baker, 1989). They showconfusions between items with visually over-lappingfeatures,whichhasbeentakentosuggest

that their memory may be based on relativelyuntransformed visual representations of the ini-tial stimuli. Exploring this last idea in moredetail, Walker, Hitch, Dewhurst, Whiteley, andBrandimonte (1997) have shown that visualmemory does not maintain a veridical copy of visual stimuli but rather is object based, storingboth surface descriptions of objects and moreabstract structural descriptions.

The multicomponent model of workingmemory has not just been employed to accountfor qualitative developmental shifts. Quantita-tive changes have also been explored. For ex-ample, studies of the word length effect inimmediate recall in adults have demonstratedthe importance of pronunciation time as a deter-minant of performance (Baddeley, Thomson, &Buchanan, 1975). Similarly, developmentalchangesinarticulationspeedmayformonecom-

ponent of improved memory (Hitch, Halliday,& Littler, 1989; Hulme, Thomson, Muir, &Lawrence, 1984). Pronunciation speed is astrong predictor of short-term memory span,especially at the group level (that is, the extent towhich memory ability for different age groups is

Variation in Working Memory Due to Normal Development 111



predicted by their pronunciation speed). Thus,there are proportionate changes in memoryperformance that accompany changes to thepronunciation speed of words, and this can be

understood within the notion that the phono-logical loop is constrained by rehearsal time.

CHILDREN’S WORKING MEMORYAS DUAL-TASK PERFORMANCE

From the perspective that working memoryinvolves an interaction between memory andcognition, it is not surprising that there has

been interest in whether working memory ser-vices the ability to carry out two tasks at thesame time, and whether the development of working memory in childhood at least partlyreflects the development of skills in combiningseparable cognitive activities; see comments inChapter 7 that working memory span reflects adual-task situation. We consider three aspects of this perspective.

The first aspect concerns working memory

span, sometimes known as complex span. Work-ing memory span was devised as a single taskthat embodied dual-task characteristics (Dane-man & Carpenter, 1980). Participants are askedto carry out a processing task and retain someinformation associated with each processingepisode. Their span is a measure of the upperlimit on how much information they can retainwhile continuing to carry out the processingtask. In these terms, working memory span in-volves processing and memory. In the mostwidely held theoretical account these two func-tions compete against each other. Case (1985)articulated this perspective lucidly. He arguedthat all children are endowed with an executiveprocessing space, the capacity of which is lar-gely invariant across age. While capacity per seremained constant, the use of this capacity un-derwent substantial developmental change. In

particular, cognitive development was thoughtto involve large increases in the efficiency withwhich processing operations were accom-plished. As the resource demands of processingdropped, a progressively increasing proportionof the EPS capacity could be allocated to other

activities, such as immediate memory. As Casenotes, this model implies ‘‘one capacity thatcan be flexibly allocated to either of twofunctions’’ (Case, 1985, p. 120). Working mem-

ory in these terms reflects the balance betweentwo competing activities, each of which re-quires access to a common pool of resources.We refer to this account as the resource-sharinghypothesis.

Although the present discussion focuses onchildren, it is important to note that, contem-poraneously, Daneman and Carpenter (1980)proposed a similar theoretical view, based onan influential study of individual differences in

college students. They found that working mem-ory span was a good predictor of Scholastic

Aptitude Test (SAT) scores and also that digitspan, the standard measure of short-term mem-ory, was a much poorer predictor. These find-ings were taken as support for the idea thatworking memory serves the function of combin-ing processing and storage in cognition, whereasshort-term memory is concerned merely withstorage in isolation. They led to a surge of in-

terest in identifying more precisely what limitsworking memory span.

The second aspect of dual-task capabilitiesstems from work by Towse and Houston-Price(2001), who suggested that one contributoryfactor to the predictive success of working mem-ory tasks over short-term memory tasks couldbe the requirement to combine different taskelements. To explore this idea, they developeda memory task that involved two separate com-ponents (a digit span task and Corsi span task,the latter requiring a sequence of spatial loca-tions to be remembered). By examining indi-vidual differences in the elemental skills aswell as the task in which these are broughttogether (a so-called combination span task, inwhich numbers were remembered at specificspatial locations), they tried to assay whetherthe combined task was more than the sum of its

parts (see also Emerson, Miyake, & Rettinger,1999; Yee, Hunt, & Pellegrino, 1991). Indeed,Towse and Houston-Price (2001) found thatthe combination span task was related to chil-dren’s cognitive ability (as measured by theirword-reading and number skills) once the




information in the counting span test, and didso in a highly lawful manner. With counting ratetaken as an index of the difficulty of counting,so that younger children are regarded as slower

to count because it is harder for them, theresults are really quite striking. They suggestthat as processing efficiency improves duringdevelopment, resources are freed up for otheroperations, and this is reflected in improvedcounting-span performance across development.Thus, the data support the theoretical accountoutlined earlier, whereby memory and process-ing functions trade off against each other and thebalance of this trade-off changes through devel-

opment.It is worth pointing out that in separate ex-

periments, Case et al. (1982) investigated short-term memory span and obtained a similarlinear developmental relationship against thetime taken to perceptually identify the memoryitems, adding to the weight of argument that ageneral system is being tapped. Case et al. alsoprovided an experimental test of the trade-off hypotheses. A group of adults completed

a modified counting span test, counting witha sequence of non-words instead of numbers, toincrease processing difficulty. Unsurprisingly,adults counted slowly with this novel sequence.More importantly, as Figure 5.1 shows, countingspan was observed at a level that was in propor-tion to their reduced processing ability. That is,adult performance fell along the developmentalfunction that related span and counting speed.Thus, both measures of their span and efficiencymatched those of 7-year-olds and fell within theconfidence interval for the regression line.1 Itshould be apparent that this result is just whatwould be predicted by the resource-sharing hy-pothesis articulated above: by changing the num-ber words and increasing adults’ processingdemands, their memory ability had been com-promised. The critical conclusion, then, is thatthe change in task demands produced not only a

drop in memory but also a drop in performancethat provided a quantitative fit with the devel-opmental function. In passing it is worth notingthat Case et al. (1982) appear to have success-fully harnessed both a correlational and an ex-perimental approach. With respect to the former,individual differences indicated a close link be-

tween counting efficiency and counting span,and Case et al. suggested that age differences inspan could be explained by corresponding dif-ferences in task-processing efficiency. With re-

spect to the latter, Case et al. manipulated taskdifficulty experimentally, to buttress the argu-ment that there is a direct relationship betweenmemory and task difficulty.

For some time, much of the subsequent re-search into working memory tasks among chil-dren focused on individual differences, andexamined the predictive strength of workingmemory span tasks for other cognitive skills. Forexample, Daneman and Blennerhassett (1984)

administered a listening-span test to 4-year-oldchildren in two related experiments. In thelistening-span test, children listened to and at-tempted to comprehend sentences. They werealso asked to remember the sentences (with thisage group, individual words may not have beensalient and therefore remembering individualwords might not be straightforward). Childrenwere also administered a more usual word spantest and standardized assessment of listening

comprehension (which involved matching pic-tures to auditorily presented sentences). A keyfinding was that working memory span was astrong predictor of comprehension ability, sig-nificantly more so than word span. Danemanand Blennerhassett (1984) argued on the basisof the differential sensitivity of word and lis-tening span tests that working memory is inti-mately involved in those cognitive processesrequiring integration and resource sharing.

Leather and Henry (1994) studied the rela-tive contributions to early reading skills of short-term memory (word span, or ‘‘simple span’’),working memory (listening span and countingspan, or ‘‘complex span’’) and phonological-awareness tasks (e.g., tasks that required chil-dren to strip the initial consonant from words,and make sound blends). A group of 7-year-oldchildren were recruited and reading ability was

assessed in terms of reading accuracy and com-prehension. An important finding was thatworking memory measures were able to accountfor significant variance in reading over andabove the contribution from the word span task.Phonological-awareness tasks shared consid-erable variance with the complex span tasks,




though they were also distinguishable. Leatherand Henry (1994) speculated that the overlapmight arise because both variables reflectedsimultaneous processing and retention demands

(moreover, phonological-awareness tasks mighthave memory demands built into them, inhaving to remember and compare differentstimuli). In any case, the results support theidea that working memory (that is, the reten-tion and transformation of information) isimportant for developmental skills such as read-ing, and that complex span tasks, which theyassumed tapthe central executive,are better mea-sures of working memory than simple span tasks.

As was the case for Daneman and Blennerhas-sett, the question of interest is in the uniqueproperties of working memory span (over short-term memory span) as an index of a develop-mentally sensitive cognitive ability.

The data from Leather and Henry (1994) fo-cus on individual differences among a normalsample. Siegel and Ryan (1989) had also arguedthat reading and mathematical skills involved themaintenance and transformation of information,

characteristics attributed to working memoryspan tasks. Siegel and Ryan considered whetherchildren with learning difficulties exhibitedproblems on working memory span tasks. Theypresented working memory tests in the form of both listening span and counting span, permit-ting an evaluation of the generality of any defi-cits. Their sample comprised children with read-ing difficulties (RD), arithmetic difficulties (AD),and attentional difficulties (ADD).

The results for the listening span test showedthat RD children were impaired relative tocontrol children. Children with AD were notsignificantly different from controls, althoughtheir scores were depressed. On the countingspan test, RD children again showed deficitsrelative to normally achieving controls, and onthis task AD children showed significant spandeficits as well. Children with ADD did not

show significant impairments on either the lis-tening- or the counting-span test. From theseresults, Siegel and Ryan (1989) argued that notall children with learning difficulties are alike;they exhibited different patterns of deficits onworking memory tasks. In the present context,these results add to the evidence that working

memory span is not a domain-free measure of ability.

Theoretical Accounts of Children’s

Working Memory Span

One point of emphasis in the work reportedabove is that working memory span tests are dif-ferent from short-term memory tasks in terms of their psychometric characteristics (Daneman &Blennerhassett, 1984; Daneman & Carpenter,1980; Leather & Henry, 1984). Working mem-ory span tests are able to account for variance inabilities such as, but not limited to, reading and

arithmetic, which short-term memory tasks donot do (see Alloway, Gathercole, Willis & Ad-ams, 2004). A second point of emphasis is thatvariation in working memory span arises fromthe way that a system such as the central exec-utive allocates resources to the processing andmemory components of the task. Changes inone can influence the other. Thus childrenwith learning disabilities may find the process-ing requirements particularly taxing, and this

can compromise their memory performance.However, whether this trade-off arises from adomain-specific system, or a domain-generalsystem at the core of working memory, has beena contested question. In this section, we considerthe arguments relevant to theories of children’sworking memory, and in particular discuss therole of experimental and correlational researchin this process.

While we have described working memoryspan so far in terms of resource sharing, thedirect evidence for a relationship between pro-cessing demands and memory ability is sparse.The data from Case et al. (1982) seem the mostcompelling, in that there was a linear rela-tionship in several age groups between the ef-ficiency of array counting and span level, as wellas a child-like level of memory performanceamong adults when processing demands were

high. Nonetheless, both these findings rely onthe assumption that resource demand can bemeasured accurately by asking participants tocount arrays as quickly as possible. Yet, the ef-ficiency of counting also delivers an indicationof how long counting span trials last. If thecounting trials last a long time, the totals of each




count—the information held in memory—mustbe retained longer. This raises the question of whether counting span varies as a function of processing resource demand (resource sharing)

or just the processing duration. One reason thatthe duration of processing may be important isthat representations in working memory aresubject to rapid forgetting. For example, Hitch(1978) found that errors in mental arithmeticcould be predicted from a knowledge of thesequence of processing operations and its im-plications for the times over which temporaryinformation (such as partial results) had to bemaintained. The idea that working memory

span is limited because representations in work-ing memory undergo rapid forgetting during thetime spent processing has been termed the task-switching account (Towse, Hitch, & Hutton,1998).

According to the task-switching view, thedevelopmental function for working memoryspan would occur because younger children,being slower to count, have more opportunityto forget information up to the point of recall.

Asking adults to count with non-words wouldalso put them at a disadvantage because theslower counting would increase the time pe-riod over which non-words could be forgotten.Likewise, this task-switching approach could ex-plain the pattern of data from learning-disabledchildren (Hitch & McAuley, 1991; Siegel &Ryan, 1989), since their processing deficitswould place them at a disadvantage on workingmemory span trials.

Since both resource-sharing and task-switching hypotheses are able to account formuch the same set of data, Towse and Hitch(1995) attempted to develop an experimentalsituation where the two approaches made dif-ferent predictions for performance. Countingspan is useful in this regard because processingis determined by the difficulty of executing thecounting operation (the visual identification of

targets and the articulation of an appropriateverbal label) and is also affected by the numberof iterations required, that is, the number of objects to count. These factors make it possibleto develop counting span materials that varyindependently along the dimensions of taskcomplexity and task duration.

Two counting span conditions labeled hereeasy (short) and difficult (long) differed in pro-cessing difficulty (the ease with which itemscould be isolated and identified for enumera-

tion), and the time taken to complete counting. A third easy (long) condition was constructed,by increasing the array sizes and therefore thecompletiontime requirements of theeasy (short)condition so as to match that of the difficult(long) condition. Both resource-sharing andtask-switching hypotheses suggest that countingspan will be larger in the easy (short) conditionthan in the difficult (long) condition. How-ever, the resource-sharing account also predicts

higher span in the easy (long) condition thanthe difficult (long) condition, because of thedifference in the difficulty of processing op-erations. In contrast, the task-switching hy-pothesis predicts equivalent performance inthese two conditions because of the experimen-tally equated duration of the processing oper-ations.

Towse and Hitch (1995) found that theprediction derived from the task-switching ac-

count was borne out in each of four differentage groups of children (ranging between justunder 5 and a half years of age to over 11 yearsof age). Counting span with easy (short) ma-terials was significantly higher than with diffi-cult (long)materials,buttherewasnodiscerniblespan difference between the difficult (long) andthe easy (long) condition. There was a reliabledifference in the rate of counting errors betweenthe latter two conditions, confirming that pro-cessing resource demands were indeed greaterin the difficult (long) condition.

Individual differences were also examinedin this study. Counting span was found to cor-relate with counting speed, a result anticipatedby both accounts—by resource sharing becausespeed is an index of processing demand, andtherefore residual memory ability, and by taskswitching because fast counters can reach the

recall cue more swiftly and therefore limit theirretention interval. Partialing out the effect of age left the correlation nonsignificant. However,there are reasons to believe that there was var-iability in the speed measure and subsequentstudies show that speed can predict span evenafter age is partialed out (see below).




Following up this result, Towse et al. (1998)presented children with working memory spantasks in which the trials involved one of two

sequences of processing requirements, as shownin the schematic time line of Figure 5.2. Eachcard in the figure refers to a problem requiring asolution, and all solutions had to be recalled atthe end of the trial. Both of the sequences, (a)and (b), contain the same problems. What dif-ferentiates them is the completion order for theproblems. In set (a) there is a lengthy problem atthe beginning of the trial but a short problem atthe end, while the reverse occurs in set (b),

where there is a short problem at the beginningof the trial and a lengthy problem at the end. Asa result of this ordering, the two arrangementsinvolve the same processing work on each trial(because the same total set of operations is in-volved) but differing retention times (because inset (a) the retention requirements, resultingfrom the completion of each problem, com-mence at a later point in the trial).

Towse et al. (1998) also examined the effi-ciency of processing operations throughouteach trial. They noted that on the first card, thesubject performs the task with no memoryload—it is only the completion of the card thatgenerates the first memory item. On the secondcard, the processing occurs with a concurrentload of one item. On the third card, if there isone, processing is accompanied by a memoryload of two items, and so on. Thus as the trial

progresses, the memory load increases. From aresource-sharing perspective, then, processingefficiency should decline throughout the trial.

A task-switching account gives no reason tosuppose that this fade in efficiency will occur.

To summarize, the paradigm offers two linesof evidence to discriminate between theoretical

accounts of working memory. The manipula-tion of temporal order of cards should affectspan according to task switching but should not

affect span according to resource sharing. Card-processing efficiency within a trial should notvary according to task switching but should de-cline according to resource sharing.

Towse et al. (1998) reported three experi-ments, overall using children between 6 and 11years of age, in which the above paradigm wasused with counting span, operation span (sumswere presented and the answers formed thememoranda), and reading span (children read

incomplete sentences and generated a suitableend word, which formed the memoranda).They found that in each experiment, span scoreswere reliably affected by the order in which theensemble of cards appeared. Where the overallretention time was smaller because memoryrequirements started late, spans were larger. Asexplained above, this fits with the predictionmade by the task-switching model (for addi-tional convergent findings with a longitudinalcomponent, see Ransdell & Hecht, 2003).There was also no consistent evidence that par-ticipants were slower to count cards at the end of the trial rather than at the beginning; some an-alyses found no effects, others found a reliablespeeding up, and still others, a reliable slowingdown but only with some test administrationorders. Overall, within-trial analyses give theimpression that processing times arise from a

number of flexible strategies.Individual differences in memory were stron-ger in these experiments, with respect to age—older children reached higher span levels—andprocessing ability—span scores correlated withthe speed at which the processing was accom-plished. The correlation between span and

Figure 5.2. Schematicrepresentation of

memory requirementsfollowing a manipula-tion of the order of com-pleting an ensemble of cards. (Adapted fromTowse et al., 2005.)




processing speed persisted after partialing outthe effect of age. Among children, then, mem-ory performance is affected by the rate at whichthe processing operations are completed. Work-

ing memory span scores reflect the influence of processing speed in addition to any influ-ence from immediate memory skills. In turn,this provides a basis for explaining why workingmemory span might involve a combination of domain-specific and domain-general abilities.Thus, if there are multiple skills that contributeto performance, some of these might be idio-syncratic to a paradigm, one that is domainspecific, whereas others reflect abiding traits

relevant to many related tasks.From the final experiment in which children

were tested on both operation span and read-ing span, there was also evidence that these twotasks were correlated, albeit modestly. Indepen-dent measures of processing speed had also beenadministered (from the Kit of Factor-referencedTests; Ekstrom, French, Harman, & Dermen,1976); consequently, it was possible to considerwhether the relationship between working mem-

ory speed and span related to the task-specificprocesses or more global processing parameters.Reading span was predicted by reading speedbut was not uniquely predicted by general speed,while operation span was predicted by generalspeed. The data further underline the view thatworking memory variation in children com-prises both general and specific skills.

Hitch, Towse, and Hutton (2001) were ableto reassess children from the final study of Towseet al. (1998) on operation span and reading span1 year after their initial assessment, and collectadditional measures of scholastic ability. Thispermits replication of the previous work on thesame sample of children and introduces severalnew dimensions to the research. In particular itallows assessment of relationships betweenworking memory span, reading attainment, andnumerical competence as well as longitudinal

analysis of performance across a 12-month in-terval.Hitch et al. (2001) replicated their previous

finding that span scores were reliably affected bythe order in which the ensemble of cards ap-peared. Once again, spans were larger when theresults of processing operations had to be main-

tained over shorter intervals, consistent with theprediction from task switching. However, analy-sis of changes in the time taken to perform pro-cessing operations within a trial gave a different

result from before. When data were combinedacross the two studies for greater power, a sig-nificant increase in processing time from the startto the end of a trial was evident in both span tasks.Such an increase is consistent with the predic-tion from resource sharing (although Hitch et al.,2001, suggest other explanations might be of-fered too; see also Saito & Miyake, 2004) and isnot predicted by task switching. Thus, while it isclear that task switching has robust effects on

working memory span, it is equally clear that taskswitching does not provide a full account. This issomething we return to later.

The data collected by Hitch et al. (2001)also allowed a test of the generality of the devel-opmentalrelationshipbetweenworkingmemoryspan and processing speed plotted by Case et al.(1982). To recapitulate, Case and colleaguesobserved a linear relationship between meancounting span and the mean time to perform

counting operations across age groups, suchthat older children’s superior spans were pre-dictable from their faster processing speed (seeFig. 5.1). In our study three age groups weretested on two separate occasions, generatingsix points on the graph for each span task. Figure5.3 shows reading span plotted against readingtime and operation span plotted against the timeto perform arithmetic operations. Two observa-tions are immediately clear. One is that, to a firstapproximation, developmental changes in eachtype of span are linearly related to changes in thespeed of the relevant processing operations, as forcounting span. Thus, a range of very differenttasks reflects in general a common developmen-tal relationship between working memory spanand processing speed.

A second observation is that the slopes andintercepts of the best-fitting linear functions

for each span task are different. This is furtherevidence for differences between working mem-ory span tasks. Once again, observations that arecommon to different span tasks suggest domain-general aspects of working memory whereasdifferences between tasks suggest the role of domain-specific factors. As regards theoretical




interpretation of the relationship between spanand speed, we have already noted that it can beexplained in terms of either resource sharing, asCase et al (1982) originally proposed, or task

switching (Towse et al., 1998). However, Hitchet al. (2001) pointed out that task switchinggives a more parsimonious account of the dif-ferent functions for different tasks shown inFigure 5.3, because it does not require addi-tional assumptions in the crossover of linearfunctions.

Hitch et al. (2001) also analyzed individualdifferences in reading span and operation spanin relation to children’s performance on stan-

dardized tests of number skills and word read-ing (as assessed by a single-word reading task).There was substantial overlap in the varianceattributable to reading span and operation spanwith respect to each of these two measuresof scholastic attainment. However, there wasalso evidence to distinguish reading span fromoperation span, because each of these tasksexplained significant, unique variance in single-word reading ability. It would be simple and

elegant to claim that working memory span taskscan be thought of as entirely domain specific orcompletely domain general. However, onceagain, the data do not support such an inter-

pretation. Instead, it seems necessary to con-clude that there are both domain-specific anddomain-general aspects to performance in work-ing memory span tasks. The analysis of individ-

ual differences also included various measuresof processing speed, taken either on-line in eachspan task or off-line from the Kit of Factor-referenced Tests (Ekstrom et al., 1976). As ex-pected, there was substantial overlap betweenspeed and span for both measures of scholasticattainment. However, in each case there wasvariance unique to span that was not attribut-able to speed. So once again, we have evidencethat span is indeed complex.

Longitudinal analysis in Hitch et al.’s (2001)study showed that a combined score for read-ing span and operation span explained signifi-cant unique variance in both number skills andword reading some 12 months later. This wasso even after taking out variance attributable tocombined span scores at the time of the scho-lastic assessment. This result is consistent withDaneman and Carpenter’s (1980) claim thatworking memory span is related to cognitive

ability. However, it goes further in demonstrat-ing a longitudinal relationship that is additionalto any concurrent relationship. One would ex-pect to see a longitudinal relationship if workingmemory is a factor in children’s acquisition of reading and arithmetical skills.

One interesting aspect of our research onchildren has been the absence of qualitativedevelopmental changes. Our initial view wasthat task switching might characterize youngerchildren’s performance, perhaps because theylacked sufficient capacity to share resourcesbetween processing operations and retention inworking memory, or had yet to acquire such astrategy (Towse & Hitch, 1995). With thisview, one would expect increasing evidence forresource sharing among older children. How-ever, this was not found. To check whether wemight have missed a crucial stage of develop-

ment by not testing children above the age of about 11, we investigated adult performance(Towse, Hitch, & Hutton, 2000).

Results once again showed significant effectsof card completion order in reading span andoperation span. Thus, for both tasks, spans werelarger when the card order was such that

5operation span

reading span

2

3

4

2 4 6

Time per card (sec)

S p a n

8753 109

Figure 5.3. Reading span plotted against readingtime and operation span plotted against the time toperform arithmetic operations. Each data pointrepresents data on span and speed for a different agegroup of children at one of the two waves of testingrun 1 year apart. (From Hitch et al., 2001.)




memory requirements started later rather thanearlierwithinatrial(asinTowseetal.,1998).Fur-thermore, there were no discernible within-trialchanges in the time taken to complete process-

ing operations in either task. Thus, there was noevidence that processing became less efficientas the memory load increased from the startto the end of a trial. The one major contrastbetween adults and children was related to pat-terns of individual differences. Among adults,unlike children, there were no significant cor-relations between the speed of processing oper-ations(whether measured on-line or off-line)andspan.

We interpret the adult data as showing thatthere is no clearly marked developmental tran-sition from task switching to resource sharing inworking memory. It seems that task switching isnot an immature strategy but one that persistsinto adulthood. Given that task switching alsoprovides a good account of adults’ errors in doingmental arithmetic (Hitch, 1978), it seems thatthis conclusion generalizes beyond laboratorytasks. The qualitative developmental change was

founded on the relative unimportance of indi-vidual differences in processing speed in relationto working memory span in adults compared tothat of children (see Chapter 6 for similar indi-cations of a shift away from the contribution of processing efficiency for span when moving fromchildren to adults). However, it is not entirelyclear what this means; it might reflect the greaterimportance of rehearsal in children’s workingmemory (but see Hutton & Towse, 2001) orsignal a difference in the balance of factors de-termining span, due, for example, to range ef-fects. Thus, whereas span in children includes acomponent that is independent of speed as wellas one that is correlated with speed, the first of these components may sweep up a much greaterproportion of the variance in adults.


What is your view on the critical source(s) of working memory variability within your targetpopulation(s) of study? Why do you focus onthe specific source(s) of variability in your re-

search? Working memory span trials involveasking participants repeatedly to engage insome nontrivial processing task and rememberpieces of information connected (semantically

or temporally) to the processing. Span measuresthe maximum number of pieces of informationthat can be retained under these circumstances.We argue against the simple view that span is ameasure of the resources available to workingmemory and nothing more. We argue insteadthat the temporal dynamics of trials are im-portant, because children differ in the rate atwhich processing can be executed. This in turnaffects the extent to which they are exposed to

the effects of forgetting transient information,and so influences their ability to complete thetask successfully.

We also suggest that there may be individualdifferences in the degree to which participantscan sustain information in accessible form whileengaged in a separate processing task. That is,overlaying the differences in the rate at whichprocessing can be completed are differences inhow destructive particular delays may be for

memories. We have recently attempted to de-velop tests of this aspect of memory, and wedescribe these below.

A further source of variability that we con-sider is age. Whereas a simple model mightsuggest that individual differences for childrenat a particular age are paralleled by differencesacross age, we conclude that these issues maybe dissociated. Explaining age differences onthe one hand, and individual differences withinan age group on the other, may require sepa-rate theoretical accounts. However, we also ac-knowledge that age as a variable is at best a proxymeasure for some underlying change, psycho-logical or biological, which takes place.

We discuss recent research that has exam-ined individual differences of response-timingprocesses. Different phases of the recall pro-cess are sensitive to the configuration of working

memory tests, and they relate to overall memoryscores and external ability measures. Responsetiming offers clues to some of the variation thatexists in working memory tests.

It is apparent, therefore, that we believe itis necessary to think about multiple sources of variability in working memory.




GOING BEYOND SPAN SCORESPER SE IN UNDERSTANDINGWORKING MEMORY

Recently, together with Nelson Cowan and othercolleagues, we have considered children’s work-ing memory span from an alternative and po-tentially complementary perspective, that of thechronometry of recall responses (Cowan et al.,2003; see also Towse & Cowan, 2005). As well asassessing the quality of memory recall responses,we also considered the timing of the successfulresponse sequences. Over repeated trials (being aspan procedure, this varied across participants),

we partitioned every correctly recalled responsesequence into separate segments. Past researchhas established distinct components in a recallstring in immediate-memory tasks (e.g., Cowanet al., 1998). These include the preparatory in-terval (the initial delay between the recall cueand the onset of the participants’ recall), worddurations (the time taken to articulate each re-called item), and the inter-word intervals (thepauses between each response). On the basis of

this previous research we measured the durationsof the corresponding segments in recall in vari-ous working memory span tasks.

Cowan et al. (2003) considered responsetiming in both children and adults on differentmeasures of memory. In two separate studies,participants were assessed on reading span,counting span, and listening span (participantslistened to a sentence, decided whether it wastrue or not, and remembered the last word inthe sentence). A digit span test was incorpo-rated within one experiment also, providing acontrast between short-term memory and thevarious working memory measures. Variouscognitive measures were also collected. Theseincluded reading- and numerical-skills attain-ment and high-school grade percentiles (forcounting span and listening span).

One of the most striking aspects of the re-

sults was that response timing did not clearlydifferentiate digit span from counting span, al-though it did differentiate listening span andreading span from counting span. The principaldifferences were in the inter-word pauses, whichwere almost an order of magnitude slower in thetwo working memory span tasks that involved

language comprehension. Thus, children weredoing something very different in listening spanand reading span compared with counting span.One suggestion is that children were using a

strategy of recalling information about the sen-tences and using this to identify their finalwords. No equivalent of this strategy would bepossible in the case of counting span insofar asthe processing, array counting, is not distinctive.Moreover, the different working memory spansalso produced different patterns of correlationswith scholastic measures, providing further ev-idence that they are not strictly equivalent. Thisview indicates that working memory span is

dependent on the content of processing and, toa different degree across different span tasks,raises further challenges for the theory that theresource cost of processing is critical to perfor-mance.

A second major finding was that responsetime measures could account for additional var-iance in scholastic measures over and abovevariance associated with working memory spans.Thus, for reading span, response time measures

correlated with standardized tests of readingand number skills and this was separable fromtheir relationship with memory performanceper se. These observations offer further confir-mation that response time measures afford adifferent and distinctive insight into memoryprocesses.

In summary, Cowan et al. (2003) providefurther evidence for the complexity of work-ing memory span, by pointing to some of theadditional variables that contribute to perfor-mance. The timing of recall is a valuable sourceof information, capable not only of revealingsubstantial differences in the way various spantasks are performed but also unique variationin scholastic attainment. These observationsmove us further away from the idea of work-ing memory span as a simple, domain-free mea-sure.

Additional Measuresof Working Memory

As we have just argued, the analysis of re-sponse timing can help to build a theoreticalpicture of working memory, revealing certain




commonalities with short-term memory (in thegeneral pattern of specific phases of recall),some striking differences from short-term mem-ory, and indeed differences among working

memory tasks themselves (in the duration of the recall phase). The approach is also rele-vant to individual-difference issues, especiallysince some but not other phases of recall areassociated with working memory span scores.Response-timing analysis serves another usefulpurpose, however, in demonstrating the value of taking measures other than just the number of recallable items as a way of informing us aboutworking memory.

This is particularly relevant in the context of the task-switching account of children’s workingmemory and its development. We have beenarguing that forgetting results from the require-ment to hold recent information while complet-ing processing activities. Processing operationsthat take a long time can be detrimental forworking memory performance because memo-ries can degrade. Moreover, there may be indi-vidual differences in the performance of pro-

cessing operations and, separately, there mayalso be individual differences in the resilience of memory representations. That is, even assumingthat a memory is initially formed at the samestrength in two individuals, the course of for-getting may be more rapid for one of them.Even if these ideas are only partly accurate, itfollows that there is more to working memorythan just the number of independent items thatcan be remembered.

There are several different ways of exploringwhether there are individual differences in thedurability of representations in working mem-ory. We have recently attempted to operation-alize this issue by examining whether the en-durance of memories can be measured in ameaningful way for children (Towse, Hitch,Hamilton, Peacock, & Hutton, 2005). We de-vised a task in which children completed a fixed

number of problems that involved ‘‘processingþ

memory,’’ as in complex span. However, insteadof increasing the number of problems oversuccessive sets of trials to determine span, weincreased the time required to complete theprocessing operations for the given number of

problems. We then derived a measure calledworking memory period, which represented thefurthest point in the test at which items couldbe successfully remembered. In particular, we

studied a reading-period task, in which childrencompleted three sentences and then remem-bered the words that they generated, and oper-ation period task, in which children calculatedthe answer to four arithmetic problems and thenremembered these solutions. As should be ap-parent from the description, the objective wasto attempt to assess the functional duration of working memory for a fixed number of items,rather than the limit on the number of items

that can be recalled.Theresults indicated that theperiod score had

reasonable test-retest stability, in that it was atleast comparable with span (and in one analysis,exceeded the reliability of span). Moreover, theperiod score had predictive power: operationperiod correlated with children’s reading andnumber skills. Reading period likewise correlatedwith these cognitive skills, but particularly withreading. Towse et al. (2005) argued that working

memory period was a potentially useful experi-mental device; the use of fixed length lists makesit amenable to addressing specific theoreticalquestions. That in itself makes it an interestingapproach. However, in the context of a chapteron individual differences, it is the overlap invariance between the working memory periodscore and cognitive ability that is of particularrelevance. Thus the length of the interval overwhich children could maintain a small numberof items was significantly related to the abilityto identify words or complete mathematical op-erations. We believe that this speaks to the valueof asking questions about the limits of workingmemory that go beyond the almost ubiquitous,‘‘How many?’’

It should also be noted that we do not assumethat the passage of time causes forgetting inworking memory. Such an interpretation would

be consistent with the data: as sentence lengthincreases, there is more forgetting caused bytemporal decay. However, it could equally bethe case that longer processing events (such assentences) put additional strain on the ability toinhibit non-target information and increase the




likelihood of inhibitory failure (see Chapter 9).It might also be the case that more protractedsentences generate greater representational over-lap with the target memory items, and that

representational distinctiveness is an impor-tant component of memory (Saito & Miyake,2004). Since we are not in a position to dis-criminate between these and other possi-bilities, our proposals are more modest: theendurance of representations is a relevant attri-bute for working memory, and the endurancecan be tested by varying the exact processingevents that accompany or produce the memorystimuli.

LIMITS OF THE TASK-SWITCHINGACCOUNT

We have described a series of experiments thatprovide a variety of evidence for the view thatchildren’s working memory performance is af-fected by the retention duration of the memorystimuli. The idea that memories are perishable,

and that delaying the point of recall can impairthe quality of recall, seems to us to have muchmerit, and forms the basis of the task-switchingmodel we advocate here. It is consistent witha very large research literature on immediatememory (i.e., short-term memory) and makesintuitive sense. It is also consistent with datafrom the experiments designed to test whether itis credible. It helps to explain experimentalphenomena and to account for individual dif-ferences. And it can be complemented by otherfindings—for example, with respect to the con-tribution of recall processes, and situations whereprocessing and memories are mutually support-ive, since even where processing scaffolds laterrecall, one can anticipate that that forgettingtakes place during the execution of processingactivity. Thus, the task-switching model avoidsthe need to draw a sharp distinction between

processing (for example, that which occurs whena sentence is read for comprehension) and mem-ory (that which enables certain items to be re-tained for subsequent recall). Such distinctionsare inherent in the notion of a trade-off betweenprocessing and memory.

At the same time, the task-switching modeldoes not offer a full or comprehensive accountof working memory span. We have repeatedlysought to note this. Towse and Hitch (1995)

considered conceptual limits. Making the pro-cessing task shorter might improve span scores,but of course eliminating a concurrent pro-cessing task, akin to a short-term memory task,does not eliminate capacity constraints. Clearly,therefore, there must be other factors to con-sider. Furthermore, Hitch et al. (2001) notedthat controlling for processing speed did noteliminate age differences in working mem-ory, which implies that additional factors influ-

ence developmental change. Towse et al. (2000)found that adults’ working memory was not re-lated to their processing speed (see also Engle,Cantor, & Carullo, 1992) and thus, while pro-cessing delays did affect performance, workingmemory variation was primarily influenced bysome other variable. Towse, Hitch, and Hutton(2002), using an interpolated-task paradigm,argued that the task-switching model did noteasily account for age differences in memory

performance, nor did it offer a clear explanationfor, in one experiment, memory performancebeing equivalent when the retention intervalswere (unexpectedly) different. Response timinganalyses from Cowan et al. (2003) also suggestthat working memory performance is shaped atrecall, and not only during ‘‘retention’’ (see alsoHaarmann, Davelaar, & Usher, 2003).

Thus, it is important to distinguish our po-sition that retention delay is important for work-ing memory performance (perhaps especiallyamong children) from any straw man that sug-gests working memory is purely determined bytask-processing time. Moreover, we regard boththe inadequacy of task switching as a completeaccount and the proposal that working mem-ory is multifaceted and multidetermined to beboth realistic and exciting. The conclusion is re-alistic because working memory span is a com-

plex task that may well resist being satisfactorilycaptured by a single mechanism. The conclu-sion is exciting because it provides a challengein determining how the account can best beaugmented, and also because the accountneed not be exclusive; other hypotheses may




comfortably coexist alongside the ideas we haveproposed.

Some recent work, published as this chapterwas first drafted, has called into question the

task-switching account of working memory.Barrouillet, Barnadin, and Camos (2004) de-vised span tasks in which the processing oper-ations were discrete and externally paced. Byaltering the speed of pacing they were able toseparate out effects of the total duration of processing from the number of processing op-erations performed. Among their findings wasthe observation that, for a given number of processing operations, spans were higher when

the duration of processing was longer. Bar-rouillet et al. (2004) point out that this observa-tion goes against a simple task-switching accountin which ‘‘there is no active maintenance of stored totals that competes with the executionof counting operations’’ (Hitch et al., 2001,p. 185). As we have seen, this latter accountpredicts that span should decline as the totalduration of processing increases. Barrouilletet al. argue for a more sophisticated form of task

switching in which participants switch attentionbetween processing and maintenance duringintervening activity. Thus, when the externalpacing of such switching is high, this reduces thetime available for active maintenance in the gapsbetween processing operations, resulting in lowerspans. Barrouillet and colleagues also show thatspans are lower in their paradigm when the dif-ficulty of the processing operations is increased(e.g., retrieving a digit from long-term memory asin ‘‘4þ 1¼ ?’’ as opposed to simply reading thedigit as in ‘‘4þ 1¼ 5’’). This leads them to sug-gest that the attention-switching process is re-source limited. To account for these findingsthey propose a hybrid ‘‘time-based resource-sharing model’’ as an alternative to simple taskswitching.

We have a number of comments to makeabout this interesting and ingenious work. The

first is that we of course agree on the need toelaborate on the simple task-switching model;we have been arguing here that it is not enough.Towse et al. (2005) have noted the importanceof considering the retention function of eachmemorandum, rather than just the entire mem-ory set, while Cowan et al. (2003) have pointed

to processes during recall that affect workingmemory performance. Towse et al. (2002) alsorecognize the possibility that memory ‘‘con-solidation’’ might occur with interpolated tasks,

that is, rehearsal between presentation and re-call of memory stimuli. In short, we have movedbeyond the simple task-switching account thatBarrouillet et al. argue against.

Second, we have concerns about the gen-erality of their results. Thus, if one paces aprocessing task slowly enough it is entirelyunsurprising that participants will use any un-filled gaps to maintain the memory items (lead-ing to issues grappled with by Jarrold & Bayliss

in Chapter 6). We suggest that participants areless likely to switch to memory maintenancewhen processing involves a continuous chain of interdependent operations (e.g., as in reading asentence) than when operations are discrete andsequentially independent. Third, we note thatBarrouillet et al. studied adults. It is not clearthat their results and interpretation would nec-essarily apply to children (nor that their em-phasis on the resource demand of recall

activity fits entirely comfortably with the focuson controlled attention in retention [Chapter2], or representational overlap at recall, seeSaito & Miyake, 2004 in adults). Children mayshow the same effects, yet we caution against anypresumption that children and adults must bealike.

Thus, in summary, Barrouillet and col-leagues show that simple task switching cannotexplain adult performance in particular ver-sions of the working memory span task whereprocessing is discrete and externally paced. Wehave no problem accepting this. However, wedo question the degree to which one wouldwant to generalize from such tasks to otherworking memory span tasks, to situations whereprocessing and storage may be integrated ratherthan in competition, and to children. More-over, we think it is important to recognize the

degree of alignment in the different views, andnot just focus on differences. There is agree-ment in suggesting some role for task switch-ing, that is, alternation between processing andmemory activities; similarly, there is agreementtoo that memories may not be continuouslymaintained throughout an experimental trial,




has no independent role; rather our view is thatat some of the properties of working memorytasks can be attributed to processing differencesinstead of just memory; see Hutton & Towse,

2001). Furthermore, we agree with Hale andcolleagues (Chapter 7) that processing speed isimportant for working memory, and we sharewith them as well as with Jarrold and Baylissthe view that working memory comprises bothdomain-general and domain- or task-specificcomponents. Our inventory of functions thatfall under these two headings differs some-what from that of these authors, but this re-flects as much the choice of issues to focus on

as it does an allegiance to a limited set of var-iables.

Another of our proposals is that responsetime processes (that is, examining the chro-nometry of recall) can be informative not onlyfor how children carry out working memorytasks, but also for revealing some of the strate-gies and skills that differentiate those childrenwho show particular strengths at the task. Weview response-timing variables partly as mark-

ers for other cognitive processes, and we discusswhat these processes might represent.

We argue that not all working memory spantasks are the same. Although we agree that mostspan tasks show impressive and sometimessimilar correlational profiles, and we recognizethat they may share some core cognitive prop-erties, we review several pieces of evidence thatindicate nontrivial differences between workingmemory span tasks. We interpret these differ-ences as evidence that working memory spanreflects a combination of domain-specific anddomain-general factors. We discuss how thesedifferences might have an impact on variationin working memory, in particular consideringage-related changes in span.

We also believe that accounts of long-termworking memory are potentially relevant tounderstanding working memory and develop-

mental change (see, e.g., Ericsson & Delaney,1999). The notion that networks of long-termknowledge representations can facilitate the ef-ficient structuring and integration of targetmemories is a view that appears consistent withthe ideas we analyze in this chapter. For ex-ample, it provides one way of conceptualizing

the domain specificity of different workingmemory span tasks.

An important question that we feel requiresinvestigation (i.e., remains unresolved) is the

overlap between working memory as measuredby span tests, and working memory as under-stood within concepts of mental control (anissue developed by Reuter-Lorenz and Jonidesin Chapter 10). We regard this question asrelevant within a developmental context be-cause notions of executive control extend toissues such as theory of mind and conflict res-olution (Towse & Cowan, 2005).

WHAT DOES A WORKING MEMORYSPAN TASK MEASURE?

In this chapter, we have drawn evidence from anumber of sources, principally our own, to il-lustrate some of the cognitive processes mea-sured by working memory span tests. What havewe learned? In working memory span tasks, itis self-evident that the participant cannot com-

pletely and continuously focus effort on theretention of information because there are pro-cessing requirements, such as the need to com-prehend a sentence. This processing task is in-fluential in a number of ways. It increases thelikelihood that information will be forgottensince memory representations may not be ac-tively maintained (Towse & Hitch, 1995; Towseet al., 1998). The processing task itself may re-quire information to be remembered, addingfurther stress of the system’s ability to retain theexperimentally identified memory items (Towseet al., 2002). In the case of reading span, theprocessing task provides a temporal and se-mantic context for the memory items, such thatrecall may involve consideration of the entireprocessing episode (Copeland & Radvansky,2001; Cowan et al., 2003; Osaka, Nishizaki,Komori, & Osaka, 2002). Of course, the con-

tribution from processing may vary substantiallywith the type of working memory test—outsideof language-based spans, processing events mayoffer primarily a temporal context for discrimi-nating memory items.

We believe these recent perspectives in-crease the attractiveness of the principle of an




episodic-buffer component of working memory(Baddeley, 2000), insofar as it establishes a venuefor bringing together different types of repre-sentation (including semantic, in the sense of

thematic, information). It also dissociates ex-ecutive control from memory functions as partof the central executive, and is thereby highlycompatible the task-switching account, as well asa perspective that incorporates domain-specificviews of memory. An episodic buffer offers astructural alternative to the long-term workingmemory framework (e.g., Ericsson & Delaney,1999).

A dominant assumption driving theoretical

research into working memory and variation inworking memory is that it is legitimate to askthe simple question of how people carry outworking memory tasks. At issue, then, is whichtheoretical account is the most satisfactory.Towse and Cowan (2005) have recently sug-gested that a more pluralist approach might beappropriate. That is, given the complexity of working memory, and the richness of the datafrom working memory studies, individuals may

differ not only in information-processing ca-pacities but also in the type of strategy that theycall upon. That is, perhaps there are severalqualitatively different ways in which a task canbe completed, just as there are different theo-ries of working memory represented in thisvolume. There are several sources of evidencefor this position. For example, recent analysishas examined recall timing in reading span asa function of task experience. Towse, Cowan,Horton & Whytock (2006) compared how chil-dren recalled correct sequences on two sepa-rate occasions and compared recall when thetrial format was either unfamiliar, or followedexposure to similar trials. Of relevance to thepresent argument, it was found that the timingof recall changed even in circumstances whereoverall span levels were equivalent. Further-more, there were changes in the patterns of

correlations between memory and ability aschildren were exposed to reading span trials.Thus, rather than performance just improving,it seems as though experience at the readingspan task led to changes in performance strat-egy. The results underline the possibility thatthere is more flexibility in the way that the task

is accomplished than many theoretical positionswould concede.

Furthermore, as we noted earlier, Cowanet al. (2003) report important differences among

working memory tasks. Preparatory intervals andinter-word pauses are longer for reading spanand listening span than for counting span. Oneexplanation for this is that the sentence framesin the reading and listening span tasks providemuch more in the way of a distinct representa-tion than for counting span (where the countingprocess for different arrays is really pretty muchalike), and these sentence representations areaccessed during recall. Therefore, some working

memory span tests may involve integration be-tween processing and memory instead of simplycompetition. If one accepts this perspective, itseems a small step to the argument that someindividualsmightapproachspantasksbyattempt-ing to integrate the different requirements, whileothers choose not to do so.

Finally, we have noted that both children andadults are affected by the retention duration of memorystimuli.However, individual differences

in adults’ working memory span are not corre-lated with on-line processing speed, although thecorrelation is reliable among primary-schoolchildren. Thus, there is a qualitative difference.This conclusion could be rephrased as adults’strategies on the working memory span tasks canbe differentiated from those of children, despitesome underlying continuities. Since adults andchildren differ in this way, it would not be sur-prising to discover that some adults have a per-formance profile that is indistinguishable fromthat of children, and some children have a pro-file that is indistinguishable from that of adults.In other words, there are qualitative differencesin the way that individuals approach the task.

QUESTION 4: CONTRIBUTIONSTO GENERAL WORKING

MEMORY THEORY

What does the variability within your targetpopulation of study tell us about the structure,

function, and/or organization of working mem-ory in general? Variability is of course inher-ent to a developmental perspective on working




memory. The challenge is to understand andexplain developmental change, and assess thecorrespondence between variation in workingmemory both within and across age groups.

Task performance shows clear changes withage, accompanied by profound changes in theskills and experience that older children andadults can bring to bear on cognitive tasks. Ourresearch identifies both qualitative and quanti-tative changes in working memory span perfor-mance across age, and we attempt to integratethese into a model of the task in which multipleprocesses are seen to be relevant.

Although research is still very much build-

ing up a picture of the nature and mechanismsunderlying differences between children andadults, it is already apparent that what is truefor adult working memory is not necessarilytrue in every case for children’s working mem-ory, though it may be in some. Certainly, oneneeds to be cautious in assuming that conclu-sions from children will apply to adults, or viceversa, and we attempt to provide illustrations of such developmental differences. We note that

some of the contemporary challenges to the ar-gumentswe put forward come from research withadults (Barrouillet et al., 2004; Saito & Miyake,2004). That is not cause to dismiss these criticalfindings, of course, but it does mean that theyneed to be evaluated in their context.

One of the aspects of many, though not all,complex working memory tasks is that thecompletion of processing is under the controlof the participant. Therefore, as children varyin their ability to carry out relevant processingoperations (within and between ages), their abil-ity to interleave these activities with memoryrequirements is affected.

With respect to the value of individual dif-ference, we argue that experimental and cor-relational approaches are best seen as havingan interactive and iterative relationship in thespecification of theory about working memory.

They are often suited to investigating differenttypes of issues, though they work best when theyprovide convergent evidence for a particularconclusion.

We use our research into working memoryspan to exemplify the gains from both research

methodologies. Our initial collaborative work oncounting span adopted an experimental ap-proach to assess the veracity of the dominanttheory of working memory mechanisms. Subse-

quent work (see also Barrouillet & Camos, 2001;Barrouillet et al., 2004; Saito & Miyake, 2004)has pursued an experimental approach in an at-tempt to understandthecognitive building blocksof the working memory span paradigm. At thesame time, the implications of our early findingsfeed into issues such as domain specificity of pro-cesses, the impact of processing speed on the de-termination of working memory ability, and therelationship between working memory and high-

level cognitive abilities. Individual-differenceanalyses offer a highly suitable vehicle for ad-dressingthesequestions.In turn, theseapproacheshave led to further predictions for experimentalwork, and indeed have encouraged the develop-ment of alternative and potentially complimen-tary measures of working memory (Towse et al.,2005).

TOWARD A CONCLUSION

Our studies of working memory span in childrenpoint to a number of general observations thatseem to apply across different working memoryspan tasks. One set of these observations relatesprincipally to what we have called the temporaldynamics of working memory. Thus, the timedurations for which temporary information mustbe held are important, such that the longer theinterval the lower the working memory span.This is evident both in experimental manipula-tions (where the sequence of processing opera-tions is varied so that different patterns of dura-tions are experienced) and in the finding thatchildren who process information faster tend tohave higher spans. The speed–span relationshipis found both across age groups and across indi-viduals when effects of age are partialed out

(Hitch et al., 2001; Towse et al. 1998), but it isnot found in adults (Towse et al., 2000). Theimportance of time intervals is consistent withtask switching, according to which one factorlimiting working memory span is the rapid for-getting of temporary information while per-




forming a sequence of operations. In contrast, wefound only very modest experimental support forthe common explanation of span as reflecting alimit on the capacity for resource sharing. Thus,

children’s spans were unaffected by a substantialdifference in the difficulty of processing opera-tions (Towse & Hitch, 1995), and there was onlylimited support for the prediction that fewer re-sources remain available to sustain processingoperations as memory load increases. Neverthe-less, there was some evidence for an interactionbetween memory load and processing efficiency,underlining our view that task switching doesnot give complete account of the limit on work-

ing memory span. At the same time as finding commonalities

across various working memory span tasks, sig-nificant and substantial differences were revealedin some analyses of individual differences. Thus,while different span tasks did tend to correlatewith one another, they also showed different pat-terns of correlation with children’s performanceon tests of reading and arithmetic and with theirscores on tests of processing speed. As we have

noted, these observations point to a role for do-main-specific processes in any particular test of working memory span. To develop a fuller the-oretical account of the complexities of span, wetake the view that it is necessary to tease apart thedomain-specific component of any particularworking memory span task from the domain-freecomponent common to all such tasks. We sug-gest that task switching is an important factor inthe domain-free component of working memoryspan. From our evidence it seems more impor-tant and pervasive than resource sharing, but wecannot rule out resource sharing or indeed otherfactors as components of the domain-free com-ponent of working memory span.

As regards the domain-specific componentof working memory span, we note that this hastwo orthogonal dimensions, corresponding tomodality of information and information con-

tent, as in Fodor’s (1983) horizontal vs. verticalclassification scheme. In the present context,modality refers chiefly to whether informationis verbal or visuospatial, whereas content refersto the knowledge domain to which the infor-mation relates, such as reading or arithmetic.

The span tasks we have studied in detail allinvolve a substantial verbal component. As aconsequence, we interpret our empirical evi-dence for domain specificity in terms of dif-

ferences between the knowledge domainstapped by each span task. One way of thinkingabout this type of specificity is in terms of theconcept of long-term working memory (Erics-son & Delaney, 1999), whereby the organiza-tion of knowledge structures within a domaininfluences the operation of working memory.We note, however, that other investigators havefound evidence for domain specificity in spantasks that seems more appropriately interpreted

in terms of modality. For example, in the studyof Bayliss et al. (2003), evidence for domainspecificity came from working memory spantasks that involved different combinations of visuospatial and verbal processing and storage.Modality specificity of this type fits more neatlywith the multicomponent model of workingmemory (Baddeley, 1986; Baddeley & Hitch,1974; see also Hale, Bronik, & Fry, 1997). Thenotion of an episodic buffer provides an addi-

tional domain-specific constraint that can helpaccount for differences between tasks, particu-larly complex memory tasks likely to involveintegrated or multiple representations incorpo-rating thematic information from processing.In turn, this suggests that working memoryspan or period tasks are not just dual-task para-digms. Moreover, the four-component work-ing memory model may allow for an explanationof both vertical and horizontal domain speci-ficity.

We find complementary evidencefromexper-imental and individual-differences approachesto determining mechanisms of working memoryspan in children. The two approaches convergeon our main conclusion, namely that workingmemory is constrained by a temporal dynamic. A second conclusion stems mostly from studies of individual differences, that working memory

span is multifaceted and involves domain-freeand domain-specific sources of variation. A morespecific example of this conclusion is the argu-ment we pursue that working memory can re-quire the integration of task requirements, andnot just resolution of the competition between




them. We attribute the sensitivity of workingmemory to temporal factors to the domain-freecomponent of span, given that we find consistenteffects over a variety of span tasks. We speculate

that the domain-specific component of span hastwo orthogonal dimensions, one of which re-flects the modality of information storage, as inBaddeley’s (1986) account of the structure of workingmemory,theother ofwhichisknowledgebased, as in Ericsson & Kintsch’s (1995) view of long-term working memory.

Our central goal is to develop a more satis-factory and complete account of working mem-ory and its development. In this regard, we value

the methodological control that is available fromexperimental research, and which provides afoundation from which an understanding of cognitive systems can be understood through

classical hypothesis testing. Insofar as workingmemory span tasks correlate well with complexcognitive skills, there are reasons to believe thatour interest in working memory has applications.

Yet, appreciating the forces that drive this andother individual differences relationships is for usa means toward our central goal, and not an endin itself. Individual differences provide a com-plementary perspective on working memory andprovide another way of testing theoretical ac-counts, sometimes with greater purchase than ispossible experimentally. Likewise, mapping outthe differences in the way that younger childrenand older children carry out working memory

tasks is a staging post on the road toward appre-ciating why the mature system takes the formit does, and how developmental change takesplace.



OF WORKING MEMORY

Working memory is a limited-capacity system

that maintains and operates on fragile repre-

sentations in complex thought. We regard

working memory as a multicomponent system

that comprises a central executive, a phono-

logical loop, a visuospatial sketchpad, and

possibly an episodic buffer. We view working

memory span as a truly complex measure, yet

our data lead us to believe that the interactionbetween processing and memory is an impor-

tant ingredient that distinguishes it from many

other tasks. This interaction occurs in different

and subtle ways, with elements of both com-

petition and cooperation.



We argue that variation in working memory in

normally developing children arises from

several factors. These include differences in

the rate at which information can be pro-

cessed, the ability to sustain information in

accessible form while engaged in a separate

processing task, response-timing processes,and the parameters of specific subsystems. A

further source of variability is age. However,

we suggest that separate theoretical accounts

may be required to explain age differences

on the one hand, and individual differences

within an age group on the other. We ac-

knowledge that age is at best a proxy measure

for some underlying psychological or biologi-

cal change.


MEMORY VARIATION

We anticipate that several additional sources of

working memory variation will be identified

besides those we have identified in our own re-

search. We recognize that processing time itself

may not be the material cause of forgetting in

working memory tasks and that differences

in susceptibility to interference or the ability toinhibit irrelevant material may also contribute.

We share with others the view that working

memory comprises both domain-general and

domain-specific components. We suggest that

far frommemory activities occurring in isolation,




Note

1. Close examination of this study reveals an am-

biguity. The plotted figure in Case et al.’s (1982)

article suggests that adults were counting objects

at a rate of 450 ms per item, yet the Results sec-

tion records a rate of 490 ms. The value used here

is taken from the text, although this may be

a typographical error, since it makes the adultdata appear more discrepant from the children’s

speed–span function than one might expect from

other aspects of the study.

References

Alloway, T. P., Gathercole, S. E., Willis, C., & Adams,

A. (2004). A structural analysis of working mem-

ory and related cognitive skills in young children.

Journal of Experimental Child Psychology, 87(2),

85–106.

Baddeley, A. D. (1986). Working memory. Oxford:

Clarendon Press.

Baddeley, A. D. (2000). The episodic buffer: a new

component of working memory. Trends in Cog-


Baddeley, A. D. (1994). Working memory: The

interface between memory and cognition. In

D. L. Schacter & E. Tulving (Eds.), Memorysystems 1994 (pp. 351–367). Cambridge, MA:

MIT Press.

Baddeley, A. D., & Hitch, G. J. (1974). Working mem-

ory.InG.H.Bower(Ed.), Thepsychology of learn-

ingandmotivation:Advancesinresearchandtheory

(Vol. 8, pp. 47–89). New York: Academic Press.

Baddeley, A. D., & Lieberman, K. (1980). Spatial

working memory. In R. Nickerson (Ed.), Atten-

tion and performance (pp. 521–539). Hillsdale,

NJ: Lawrence Erlbaum Associates.

Baddeley, A. D., Lewis, V. J., & Vallar, G. (1984).

Exploring the articulatory loop. Quarterly Jour-

nal of Experimental Psychology, 36, 233–252.

Barrouillet, P., Bernadin, S., & Camos, V. (2004).

Time constraints and resource sharing in adults’working memory span. Journal of Experimental


Baddeley, A. D., Thomson, N., & Buchanan, M.

(1975). Word length and the structure of short-

term memory. Journal of Verbal Learning and

Verbal Behavior , 9, 176–189.

Barrouillet, P., & Camos, V. (2001). Developmental

increase in working memory span: Resource

sharing or temporal decay? Journal of Memory

and Language, 45, 1–20.

Bayliss,D.M.,Jarrold,C.,Baddeley,A.D.,Gunn,D.M.,

& Leigh, E. (2005). Mapping the developmental

constraintsonworkingmemoryspanperformance.

Developmental Psychology, 41(4), 579–597.

Bayliss, D. M., Jarrold, C., Gunn, D. M., & Baddeley,

A. D. (2003). The complexities of complex

span: Explaining individual differences in work-

ing memory in children and adults. Journal of

Experimental Psychology: General, 131, 71–92.Case, R. (1985). Intellectual development: Birth to

adulthood. New York: Academic Press.

Case, R., Kurland, M., & Goldberg, J. (1982).




the processing context and long-term knowl-

edge contribute to working memory perfor-

mance.

4. CONTRIBUTIONS TO GENERALWORKING MEMORY THEORY

Our data lead us to suggest that information

processing and storage interact in working mem-

ory through task switching. This account iden-

tifies the fragility of representations as a limiting

factor in working memory, and is a counter-

weight to the usual assumption of a limit on

resource sharing. We propose that task switch-

ing is a general feature of working memory

but that the balance between task switching

and other limiting factors varies across individu-als and during development. Other data suggest

that working memory has the same multicom-

ponent structure in children as in adults. Overall,

working memory appears to serve a similar

function in adults and children, given that vari-

ation in working memory in children is related to

cognitive skills and their acquisition.




Copeland, D. E., & Radvansky, G. A. (2001).

Phonological similarity in working memory.

Memory & Cognition, 29, 774–776.

Cowan, N., Towse, J. N., Hamilton, Z., Saults, J. S.,

Elliott, E. M., Lacey, J. F., Moreno, M. V., &Hitch, G. J. (2003). Children’s working

memory processes: A response-timing analysis.


132, 113–132.

Cowan, N., Wood, N. L., Wood, P. K., Keller, T. A.,

Nugent, L. D., & Keller, C. V. (1998). Two

separate verbal processing rates contributing to

short-term memory span. Journal of Experimen-


Daneman, M., & Blennerhassett, A. (1984). How toassess the listening comprehension skills of pre-

readers. Journal of Educational Psychology, 76,

1372–1381.

Daneman, M., & Carpenter, P. A. (1980). Individ-

ual differences in working memory and read-


Behavior , 19, 450–466.

Ekstrom, R. B., French, J. W., Harman, H. H., &

Dermen, D. (1976). Manual for kit of factor-

referenced cognitive tests. Princeton, NJ: Edu-cational Testing Service.

Emerson, M. J., Miyake, A., & Rettinger, D. A.

(1999). Individual differences in integrating

and coordinating multiple sources of infor-

mation. Journal of Experimental Psychology:

Learning, Memory, and Cognition, 25, 1300–

1321.

Engle, R. W., Cantor, J., & Carullo, J. J. (1992).


comprehension: A test of four hypotheses. Jour-

nal of Experimental Psychology: Learning, Mem-

ory and Cognition, 18, 972–992.

Ericsson, K. A., & Delaney, P. F. (1999). Long-term




Models of working memory (pp. 257–297). New


Ericsson, K. A., & Kintsch, W. (1995). Long-termworking memory. Psychological Review, 102,

211–245.

Fodor, J. A. (1983). The modularity of mind.

Cambridge, MA: MIT Press.

Hale, S., Bronik, M. D., & Fry, A. F. (1997). Verbal

and spatial working memory in school-age chil-

dren: Developmental differences in susceptibility

to interference. Developmental Psychology, 33,

364–371.

Haarmann, H. J., Davelaar, E. J., & Usher, M.

(2003). Individual differences in semantic short-term memory capacity and reading compre-

hension. Journal of Memory and Language, 48,

320–345.

Halliday, M. S., & Hitch, G. J. (1988). Develop-

mental applications of working memory. In

G. Claxton (Ed.), New directions in cognition

(pp. 193–222). London: Routledge and Keegan

Paul.

Hitch, G. J. (1978). The role of short-term working

memory in mental arithmetic. Cognitive Psy-chology, 10, 302–323.

Hitch, G. J., & Halliday, M. S. (1983). Working

memory in children. Philosophical Transac-

tions of the Royal Society of London (B) 302,

325–340.

Hitch, G. J., Halliday, M. S., & Littler, J. E. (1989).

Item identification time and rehearsal rate as

predictors of memory span in children. Quar-

terly Journal of Experimental Psychology, 41A,

321–337.Hitch, G. J., & McAuley, E. (1991). Working

memory in children with specific arithmetical

learning difficulties. British Journal of Psychol-

ogy, 82, 375–386.

Hitch, G. J., Towse, J. N., & Hutton, U. M. Z.

(2001). What limits working memory span?

Theoretical accounts and applications for scho-

lastic development. Journal of Experimental


Hitch, G. J., Woodin, M. E., & Baker, S. (1989).

Visual and phonological components of work-

ing memory in children. Memory & Cognition,

17, 175–185.

Hulme, C., Thomson, N., Muir, C., & Lawrence, A.

(1984). Speech rate and the development of


tal Child Psychology, 38, 241–253.

Hutton, U. M. Z., & Towse, J. N. (2001). Short-term

memory and working memory as indicesof children’s cognitive skills. Memory, 9, 383–

394.

Klatzky, R. L. (1980). Human memory: Structures and

processes (2nd ed.). San Francisco: Freeman.

Leather, C. V., & Henry, L. A. (1994). Working

memory span and phonological awareness tasks




as predictors of early reading ability. Journal of

Experimental Child Psychology, 58, 88–111.

Osaka, M., Nishizaki, Y., Komori, M., & Osaka, N.

(2002). Effect of focus on verbal working mem-

ory: Critical role of the focus word in reading.Memory and Cognition, 30, 562–571.

Quinn, J. G., & McConnell, J. (1996). Irrelevant pic-

tures in visual working memory. Quarterly Jour-

nal of Experimental Psychology, 49A, 200–215.

Ransdell, S., & Hecht, S. A. (2003). Time and resource

limits on working memory: Cross age consistency

in counting span performance. Journal of Exper-

imental Child Psychology, 86, 303–313.

Saito, S., & Miyake, A. (2004). On the nature of

forgetting and the processing-storage relation-ship in reading span performance. Journal of


Siegel, L. S., & Ryan, E. B. (1989). The develop-

ment of working memory in normally achiev-

ing and subtypes of learning disabled children.

Child Development, 60, 973–980.

Towse, J. N., & Cowan, N. (2005). Working memory

and its relevance for cognitive development. In

W. Schneider, R. Schumann-Hengsteler &

B. Sodian (Eds.), Young children’s cognitivedevelopment: Interrelationships among execu-

tive functioning, working memory, verbal ability,

and theory of mind (pp. 9–37). Mahwah, NJ:


Towse, J. N., Cowan, N., Horton, N., & Whytock, S.

(2006). Task experience and children’s working

memory performance: A perspective from recall

timing. Manuscript submitted for publication.

Towse, J. N., & Hitch, G. J. (1995). Is there a

relationship between task demand and storage

space in tests of working memory capacity? Quar-


108–124.

Towse, J. N., Hitch, G. J., Hamilton, Z., Peacock, K.,

& Hutton, U. M. Z. (2005). Working memoryperiod: The endurance of mental representa-

tions. Quarterly Journal of Experimental Psychol-

ogy, 58A, 547–571.

Towse, J. N., Hitch, G. J., & Hutton, U. M. Z.

(1998). A reevaluation of working memory

capacity in children. Journal of Memory and

Language, 39, 195–217.


(2000). On the interpretation of working mem-

ory span in adults. Memory & Cognition, 28,341–348.


(2002). On the nature of the relationship

between processing activity and item retention

in children. Journal of Experimental Child


Towse, J. N., & Houston-Price, C. M. T. (2001).

Combining representations in working mem-

ory: A brief report. British Journal of Develop-

mental Psychology, 19, 319–324.Walker, P., Hitch, G. J., Dewhurst, S. A., Whiteley,

H. E., & Brandimonte, M. A. (1997). The repre-

sentation of nonstructural information in visual

memory: Evidence from image combination.

Memory & Cognition, 25, 484–491.

Yee, P. L., Hunt, E., & Pellegrino, J. W. (1991).

Coordinating cognitive information: Task ef-

fects and individual differences in integrating

information from several sources. Cognitive Psy-

chology, 23, 615–680.




6

Variation in Working Memory Due to

Typical and Atypical Development

CHRISTOPHER JARROLD and DONNA M. BAYLISS

Studies of working memory performance inadults have shown that the ability to hold in-formation in mind while manipulating or pro-cessing other material is a reliable predictor of arange of other skills. For example, in a seminalstudy, Daneman and Carpenter (1980) pre-sented adult undergraduate participants with areading span task in which they were requiredto read a series of sentences and remember thefinal words from each of these sentences forsubsequent serial recall. The maximum num-ber, or span, of final words that individualscould successfully hold in mind was found to berelated to verbal comprehension skills and to in-dividuals’ verbal Scholastic Aptitude Test (SAT)scores. Subsequent research has confirmed thatperformance on this kind of working memory

measure, namely a task that combines the needto process information with the apparently si-multaneous requirement to store the product of that processing for subsequent recall, can pre-dict adult participants’ verbal abilities (see Da-neman & Merikle, 1996) as well as their spatialskills (Shah & Miyake,1996),mathematics com-

petence (Daneman & Tardif, 1987), reasoningabilities (Kyllonen & Christal, 1990), and gen-eral fluid intelligence (Conway, Cowan, Bun-ting, Therriault, & Minkoff, 2002; Engle,Tuholski, Laughlin, & Conway, 1999).

Given these findings, a better grasp of thecauses of variation in working memory per-formance would clearly have important im-plications, both in terms of the theoreticalunderstanding of the processes involved inhigher-order cognitive abilities such as readingand mathematics, and in terms of potential ed-ucational benefits. It is possible, of course, thatcertain sources of variance in working memorytask performance are irrelevant to the predictionof other abilities (see Maybery & Do, 2003).However, the factors that mediate the link be-

tween working memory measures and othercognitive skills must be a subset of those factorsthat constrain performance on working memorytasks. Consequently, one can ask two specificresearch questions about variance in workingmemory that are linked to the general theoret-ical questions addressed in each chapter of this

134



volume. The first is: what are the critical sourcesof variation in working memory task perfor-mance? The second, which has clear implica-tions for theoretical accounts of general working

memory theory, is: what aspect of this variancemediates the relationship between task perfor-mance and abilities such as reading and math?

The aim of our research is to address these tworesearch questions by exploring the workingmemory abilities of typically and atypically de-veloping individuals. Both of these populationshave the potential to provide informative answersto these questions for two main reasons. First,these samples are associated with relatively

greater variation in ability and performance thanundergraduate participants, who are necessarilydrawn from a much narrower intellectual range.This larger range increases the power of anindividual-differences approach to highlight thedifferent constraints on task performance and thepatterns of association between constructs. Of course, a representative sample of typically de-veloping children will, by definition, vary in IQto the same extent as a similarly representative

sample of adult participants from the generalrather than student population. This is not thecase, however, for individuals undergoing atypi-cal development. A greater degree of variation inintelligence is possible among individuals withgeneralized learning difficulties, although insuch a sample the degree of covariation amongdifferent aspects of intellectual functioning (forexample, verbal and nonverbal ability) is likely tobe no greater than that seen in the typical pop-ulation. Work with individuals experiencing spe-cific learning difficulties provides the potentialfor teasing apart domains of functioning. Forexample, verbal and nonverbal abilities mightappeartoplayasimilarroleindeterminingperfor-mance on a range of working memory tests in thetypical population, but this might simply reflectthe degree of correlation between these abili-ties. Assessing individuals with specific verbal or

nonverbal learning difficulties might indicatethat verbal and nonverbal skills are in fact po-tentiallyindependentconstraintsonperformanceon certain tasks.

It could be argued that this kind of issuecould be addressed equally easily among atypical adult sample by impairing a domain of

functioning through the use of selective dual-task interference (cf., Baddeley, 1993; althoughsee Hegarty, Shah, & Miyake, 2000). The secondadvantage of employing developmental groups is

that one can explore causal relationships, in termsof either constraints on working memory taskperformance or associations between workingmemory measures and academic achievement,in a way that adult experimental methods do notallow, simply because children’s abilities changeso rapidly over time. Such developmental studiesdo not necessarily require the use of longitudi-nal designs, although these are particularly infor-mative (Bishop, 1997); rather, differences in age

or ability between individuals can be used to ex-plore the causal links between domains of abilityand working memory performance (e.g., Case,Kurland, & Goldberg, 1982; Fry & Hale, 1996).One can then determine the factors that lead toage-related change in working memory, whichneed not necessarily be the same as those factorsthat underpin individual differences within anage group (Conway et al., 2002; Jenkins, Myer-son, Hale, & Fry, 1999; Towse, Hitch, & Hutton,

1998), as well as the consequences of workingmemory development. Once again, work withindividuals experiencing atypical developmentcan provide a strong test of the causal relation-ships between domains of cognitive functioning,thus enabling the separation of factors that arecollinear in the typical case. For example, in arecent study, Jarrold, Baddeley, Hewes, Leeke,and Phillips (2004) examined verbal short-termmemory performance among children withlearning difficulties who had an equivalent levelof vocabulary knowledge but differed in rate of vocabulary attainment, to determine the likelycausalrelationbetweenthesemeasures.Moregen-erally, determinationof whether working memoryproblems are a cause or a consequence of anindividual’s learning difficulties is central to aproper understanding of the broader educationalimplications of such difficulties.

CRITICAL SOURCES OF VARIATIONIN WORKING MEMORY

In working with these populations, our ap-proach starts from a principled analysis of the

Variation in Working Memory Due to Typical and Atypical Development 135



construct of working memory and the typeof measures used to index this construct. If working memory is defined as the storage of to-be-remembered information during the

simultaneous processing or manipulation of other, perhaps related, information, then broadlyspeaking there are two obvious sources of po-tential variation in working memory ability. In-dividuals may vary in terms of their ability to holdinformation in mind (storage), and/or in theirability to simultaneously manipulate informa-tion (processing). In other words, individuals’performance on the type of complex span taskused by Daneman and Carpenter (1980) might

be expected to depend on their ability to re-member the list of to-be-recalled target words,and to read the sentences that provide thesewords to be remembered.

In fact, a focus on this type of measureshows how likely it is that storage constraintsoperate to limit working memory performance.Because the dependent variable in a complex-span procedure is typically the number of itemssuccessfully recalled, it is almost a truism to

suggest that performance will depend on stor-agecapacity.Indeed,theevidenceofwordlengthand phonological similarity effects in complexspan tasks (La Pointe & Engle, 1990; Tehan,Hendry, & Kocinski, 2001) indicates that short-term storage factors are operating in theseworking memory tests (cf., Baddeley, Thom-son, & Buchanan, 1975; Conrad, 1964). Atthe same time, working memory measures areseen by most investigators to capture somethingmore than short-term memory tasks. Clearly,the difference between complex span measuresand simple span tests such as digit span andword span is that the former involve a degree of active manipulation or processing of informa-tion not present in the latter; this, we wouldargue, is what distinguishes working from short-term memory. The fact that complex span taskstend to be stronger predictors of other skills than

corresponding simple span tasks (e.g., Conwayet al., 2002; Daneman & Carpenter, 1980;Engle, Tuholski, et al., 1999; Kail & Hall, 2001;Oberauer, Schulze, Wilhelm, & Suß, 2005)suggests that the processing requirements of thetask do affect working memory performance in ameaningful way.

The storage and processing requirements of the complex span task are therefore two plau-sible sources of variation in working memoryperformance and, in turn, potential mediators

of the link with higher-level cognitive abilities.However, a third possible source of variancecould arise from the need to combine, coordi-nate, or integrate these two component aspectsof the task. Duff and Logie (2001) com-pared the effect of combining the processingand storage operations of a complex span task,relative to performance on these componentsconsidered in isolation. They showed a smallbut reliable cost on span performance when

these two components were combined, whichthey attributed to the demands of coordinat-ing these operations. As noted above, this resultin itself does not imply that the link betweencomplex span performance and higher-levelabilities is mediated to any meaningful extentby individual differences in coordination ability,whatever that entails. However, Towse andHouston-Price (2001) showed that the abilityto coordinate two sets of representations in a

short-term memory paradigm was related to ed-ucational ability, even when performance on thetwo separate aspects of the combined task wasaccounted for.

In short, therefore, working memory spantasks involve both storage of to-be-rememberedinformation and the processing of other infor-mation, and possibly also the coordination orcombination of the two. Consequently, thereare at least three potential sources of variation inworking memory performance that we aim toconsider in our work.

OVERARCHING THEORY

These three sources of variation in workingmemory can be integrated within a theoreticalframework, namely Baddeley’s (1986) working

memory model. This model distinguishes be-tween the short-term storage of information andthe executive control of those systems involvedin this storage (see also Baddeley & Hitch,1974). As a result, it is consistent with the no-tion that storage demands can operate as a po-tentially independent constraint on working




memory performance. In addition, the modelposits functionally independent storage systems,the phonological loop and the visuospatialsketchpad, dedicated to the short-term mainte-

nance of verbal and visuospatial informationrespectively. The capacity of these subsystems isthought to be tapped by simple span tasks thatrequire the relatively passive maintenance of information without its transformation, such asdigit span and Corsi span tasks (Corsi, cited inMilner, 1971).

Evidence for the potential distinctiveness of these two systems comes from a number of sources. Verbal and visuospatial dual tasks have

been shown to have selective interference ef-fects on the short-term maintenance of ver-bal and visuospatial information (Hale, Bronik,& Fry, 1997; Hale, Myerson, Rhee, Weiss, &

Abrams, 1996; Logie, Zucco, & Baddeley, 1990;see Chapter 7, this volume), and functionalimaging studies have implicated different neu-roanatomical substrates for these systems (e.g.,Smith, Jonides, & Koeppe, 1996; see Chapter10, this volume). Consistent with these findings,

selective deficits of verbal and visuospatial short-term memory have been observed in neuro-psychological patients (see Vallar & Papagno,1995) and among individuals with specificlearning difficulties and/or developmental dis-orders. For example, verbal short-term memorydeficits have been documented among in-dividuals with specific language impairment(see Montgomery, 2003) and Down syndrome(see Jarrold, Baddeley, & Phillips, 1999),while specific impairments of visuospatial short-term memory have been reported among indi-viduals with nonverbal learning disabilities(Cornoldi, Della Vecchia, & Tressoldi, 1995)and Williams syndrome (Vicari, Carlesimo,Brizzolara, & Pezzini, 1996). Indeed, it hasbeen suggested that the contrast in short-termmemory skills of individuals with the geneticconditions of Down syndrome and Williams

syndrome represents a double dissociation inshort-term memory impairments (Wang &Bellugi, 1994; see also Jarrold, Baddeley, &Hewes, 1999).

From our standpoint, these data do notnecessarily indicate that verbal and visuospatialshort-term memory are subserved by entirely

distinct and encapsulated systems, as Badde-ley’s working memory model would suggest.There are likely to be a number of processesthat are common to both verbal and visuospa-

tial short-term memory tasks (Chuah & May-bery, 1999; Jones, Farrand, Stuart, & Morris,1995; Pickering, Gathercole, & Peaker, 1998;Smyth & Scholey, 1996a, 1996b). However, itis also the case that these tasks differ in waysthat allow them to be dissociated in certaincircumstances. One possible source of thisdifference is in terms of the representations heldin short-term memory; Baddeley’s distinctionbetween the phonological loop and visuospa-

tial sketchpad may reduce to a fundamentaldifference in the nature of verbal and visuo-spatial information and the ways in whichthese different representations can be encoded,maintained, and retrieved (see Chapter 10). If so, then evidence of short-term memory deficitsin developmental disorders may, in some casesat least, reflect a more fundamental difficultyin dealing with verbal or visuospatial repre-sentations in general (see Hulme & Roodenrys,

1995; Jarrold, 2001). Nevertheless, regardless of the source of these impairments, it does appearthat the nature of to-be-remembered informa-tion, or the content of working memory, will bea potentially important constraint on complexspan performance (cf., Oberauer, Suß, Schulze,Wilhelm, & Wittmann, 2000).

What is less clear is how one would map thenotion of processing demands as a constrainton complex span performance onto the work-ing memory model. In the early instantiation of this model, Baddeley and Hitch (1974) refer toa ‘‘central processor . . . [that] forms the core of working memory’’ (p. 81), which might suggestthat the ‘‘central executive’’ of the current work-ingmodelisresponsibleforwhatwehavetermedthe processing portion of a complex span task.Indeed, Engle and colleagues (e.g., Engle,Tuholski, et al., 1999) have suggested that what

differentiates complex and simple span tasks, andhence working memory from short-term mem-ory, is central-executive functioning. However,even in the original Baddeley and Hitch model,and more clearly in the more recent versions of this account (Baddeley, 1986, 2000), the cen-tral executive’s role is really one of ‘‘control




processing’’ of the kind required in monitoringand coordination of memory activities (e.g.,Baddeley, 1996). There is no obvious sense inwhich the processing component of a complex

task—that is, the task that is interleaved betweenstorage episodes—involves control of this form,and consequently we would argue that the pro-cessing aspect of a complex span task need not beexecutive in nature.

Evidence to support this view comes from astudy by Russell, Jarrold, and Henry (1996),who examined the working memory abilities of individuals with autism. Autism is thought tobe associated with executive difficulties (Pen-

nington & Ozonoff, 1996; Russell, 1997), andconsequently one might expect impaired work-ing memory in this condition (see also Ben-netto, Pennington, & Rogers, 1996). As a test of this account, Russell et al. employed threecomplex span tasks, each of which containedtwo conditions that varied the level of com-plexity of the processing involved, the predic-tion being that individuals with autism wouldbe particularly impaired when complex span

tasks involved more complex processing (cf.,Braver et al., 1997). However, on one task, in-dividuals with autism showed a significantlysmaller complexity effect than controls. In thiscounting span task (cf., Case et al., 1982) par-ticipants were required to count the number of dots on a series of screens and then recallthe series of totals. ‘‘Easy’’ processing involvedcounting dots arranged in a canonical pattern,as seen on dice, while ‘‘difficult’’ processinginvolved counting distributed patterns of dotsthat also contained distractor items to preventsubitizing.

Analysis of the time taken to complete thecounting component of these tasks showed thatindividuals with autism, unlike controls, werenot aided by the canonical representation of dots (Jarrold & Russell, 1997). Indeed, ratherthan simply reading off the total from this ca-

nonical representation, individuals with au-tism tended to count the dots one by one. Thisrelative difficulty in carrying out the processingcomponent of this condition of the countingspan task led to impaired working memory per-formance. However, Jarrold and Russell (1997)argued that this failure to perceive the global

form reflected a low-level bias toward localprocessing of visual stimuli known to be asso-ciated with autism (see Frith & Happe, 1994),rather than any form of higher-level executive

problem.These data, along with similar evidence fromstudies employing complex span tasks with otherindividuals with learning difficulties (e.g., Hitch& McAuley, 1991; Siegel & Ryan, 1989), show,unsurprisingly, that an individual’s ability toperform the processing component of a complexspan task will affect their overall performance.However, to the extent that that component it-self is non-executive in nature, processing con-

straints on performance need not be executive.Similarly, there is no reason to suspect thatsuch constraints on performance need necessar-ily be domain general in nature. The processingcomponents of different tasks may share com-mon aspects—they may, for example, be limitedby general speed of processing (Fry & Hale,1996; Salthouse, 1996; Salthouse & Babcock,1991)—but an individual might still struggle onone kind of processing task while finding an-

other relatively easier, as the Russell et al. (1996)data indicate.

This is not to say that Engle is incorrect tosuggest that there are executive aspects to com-plex span tasks that are not present in simplespan tests. The working memory model wouldsuggest that working memory differs fromshort-term memory in terms of the added needfor executive control. However, by definitionthis control must reflect the higher-level com-bination of the processing and storage compo-nents of the task, rather than the ability toperform these components in isolation. AsKyllonen and Christal (1990) argue, the execu-tive aspect of working memory reflects ‘‘con-current processing and storage efficiency, inde-pendent of both the concurrent operations andthe efficiency of nonconcurrent storage’’ (p.425). Consequently, if Baddeley’s (1986) cen-

tral executive is involved in mediating complexspan performance, it must be through someexecutive requirement associated with com-bining or coordinating the processing andstorage aspects of the task. Furthermore, for theexecutive to be a meaningful system it must bedomain general in its operation; it makes little




sense to have a central control system that canonly coordinate a limited set of tasks. Of course,it may be possible that certain combinations of storage and processing overlap or interact in

different ways (cf., Shah & Miyake, 1996), butif Baddeley’s model of the central executive iscorrect, then one would expect the executivecoordination of processing and storageto be com-mon to all complex span tasks (cf., Emerson,Miyake, & Rettinger, 1999).

In sum, the theoretical structure providedby Baddeley’s (1986) working memory modelis broadly consistent with the three potentialsources of variation in working memory ability

outline above. Performance on a complex spantask may depend on an individual’s ability tostore the memory content of the task, to carryout the process interleaved between storage ep-isodes, and to control the simultaneous opera-tion of these two aspects of the test. In addition,we would argue that constraints of content andof process are potentially domain specific, al-though there may be common aspects to each of these constraints across tasks. In contrast, con-

trol constraints must be domain general if theyare to be interpreted as evidence of any higher-level executive activity (Engle, Tuholski, et al.,1999).

CONSIDERATION OFOTHER ACCOUNTS

These points are not particularly contentious;most theorists would accept that complex spantasks differ from short-term memory tests bythe addition of a processing component thatinvolves manipulation of material, while atthe same time sharing the need to store to-be-remembered target information (Dixon, Le-Fevre,&Twilley,1988;Engle,Kane,&Tuholski,1999; Waters & Caplan, 1996). Where ac-counts differ is over the question of which of

these potential constraints mediates the linkbetween complex span performance and otherabilities, and of how these various constraintsinterrelate. Clearly the processing componentof the complex span task plays a role in de-termining its predictive power, because thisis what differentiates working memory from

short-term memory measures, but it is not ob-vious why the addition of processing has thiseffect.

Traditional resource-sharing accounts (Case

et al., 1982; Daneman & Carpenter, 1980) ar-gue that processing and storage demands drawon a common, limited-capacity pool of workingmemory resources (cf., Baddeley & Hitch,1974). Given this trade-off between storage andprocessing requirements, relatively efficientprocessing frees up relatively greater resourcesfor storage; indeed, individual differences instorage capacity and processing efficiency arecollinear in this account. According to this view,

‘‘processing efficiency not storage is the real lo-cus of individual differences in working memorycapacity’’ (Daneman & Carpenter, 1987, p.494). Similarly, the development of complexspan performance need not reflect a change inoverall working memory capacity, but rather achange in the efficiency with which one cancarry out the processing requirements of the task(Case et al., 1982).

Alternative, resource switching accounts (e.g.,

Hitch & Towse, 1995; Chapter 5, this volume)accept that processing plays a role in constrain-ing complex span performance, but argue thatthis role is less direct and is mediated instead bythe influence that processing time has on thestorage demands of a task. Towse and colleaguesemphasize the commonalities between short-term and working memory measures and notethat the storage requirements of any immedi-ate recall paradigm will vary with the totalduration of that test. In the context of complexspan measures, where the onset of succes-sive processing intervals is typically self-paced,relatively less efficient processing will lead tothe task taking relatively longer overall, withconsequently greater likelihood of forgetting to-be-remembered information. The key differ-ence between this account and resource-sharingtheories is that although it accepts that process-

ing time does affect storage demands, there is noinevitable relationship (or sharing of resources)between processing and storage constraints.Instead, the suggestion that the participantswitches between these phases of the task meansthat they are potentially independent. Indeed,in a situation where processing time is held




constant, processing difficulty should have noeffect on task performance (Towse, Hitch, &Hutton, 2002). This account accepts that de-velopmental improvements in complex span

performance follow from changes in processingefficiency, and individual differences in proces-sing efficiency play a role in mediating the linkbetween complex span performance and otherabilities (Hitch, Towse, & Hutton, 2001). Re-maining questions in this account are whetherchanges in storage capacity moderate age-related change in complex span performance,and whether variance in storage capacity isseen as a further, independent factor underly-

ing the relationship between complex span per-formance and higher-level abilities (see Chap-ter 5).

This relationship may, of course, by drivenby variation in the ability to combine the storageand processing phases of a complex span task.Engle and colleagues argue that complex spantasks tap executive ‘‘controlled attention,’’ andone reason for this may be that they requireparticipants to coordinate their component

operations. In line with this suggestion, whenEngle, Tuholski, et al. (1999) partialed outshort-term memory ability from an estimate of individuals’ working memory ability as mea-sured by complex span tasks, they found that theresidual variance correlated with an index of general fluid intelligence. As noted, this ap-proach does not conclusively demonstrate thatthis residual variance represents the executiveability to coordinate storage and processing,because it fails to control for basic variance inprocessing efficiency. However, Conway et al.(2002) showed that residual variance in workingmemory performance, having accounted forshort-term memory ability, was not related to anindex of general speed of processing. This isconsistent with the view that this residual vari-ance captures more than the ability to carry outthe processing aspects of complex span tasks.

Indeed, Kyllonen and Christal (1990) foundthat an estimate of working memory ability thatcontrolled for the ability to carry out both theprocessing and storage aspects of the compo-nent working memory tasks (Experiment 1,model 1x) was closely related to an index of individuals’ reasoning ability.

Our analysis of the possible constraints oncomplex span performance shares features withall three of the accounts outlined above, andtherefore differs to some extent from each of

them. We concur with the resource-switchingaccount’s emphasis on the storage constraintson complex span performance and would ar-gue, contrary to resource-sharing views, thatthese are potentially independent of the effectsof processing efficiency. Indeed, although com-plex span tasks tend to be stronger predictors of higher-level abilities than simple span mea-sures, there are occasions when simple storagemeasures are equally strong and reliable pre-

dictors of these other skills (Bayliss, Jarrold,Baddeley, & Gunn, 2005; Cowan et al., 2003;Hutton & Towse, 2001; Oakhill & Kyle, 2000;Shah & Miyake, 1996; see also Maybery & Do,2003). In some cases this might reflect the factthat the processing requirements of the com-plex span task are not sufficiently demanding toreliably reduce span levels relative to simplespan (Myerson, Jenkins, Hale, & Sliwinski,2000; although see Shah & Miyake, 1996).

Nevertheless, because simple span measures dopredict these other abilities in these instances,storage constraints may play some role in me-diating the relationship between complexspan performance and higher-level abilities (seealso Cantor, Engle, & Hamilton, 1991; Engle,Tuholski, et al., 1999). In addition, studies inwhich the domain of the content of simple andcomplex span tasks has been manipulated haveoften shown higher correlations between pairs of short-term and working memory tasks that sharecontent than those observed between pairs of working memory tasks that do not share con-tent (Kane, Hambrick, Tuholski, Wilhelm,Payne, & Engle, 2004; Maybery & Do, 2003;Oberauer & Suß, 2000; Shah & Miyake, 1996;although see Oberauer, Suß, Wilhelm, & Witt-man, 2003). In other words, tasks that sharestorage constraints but differ in processing re-

quirements tend to have more in common thantasks that differ in storage constraints yet shareprocessing requirements.

Like the resource-sharing account but some-what at odds with the resource-switching andcontrolled-attention views, we accept that theability to carry out the particular processing




component of any complex span task may in-fluence performance. Part of this ability mayreflect the effect of variation in processing timeon overall storage demands, as Towse and col-

leagues suggest, but we are open to the possi-bility that processing constraints affect per-formance even under fixed time conditions.Finally, we predict a potentially executive aspectto complex span tasks that arises from the need tocombine storage and processing operations and,in line with Kyllonen and Christal (1990),would argue that one should assess this by par-tialing out uncontaminated estimates of bothstorage and processing abilities from complex

span performance.

OUR METHODOLOGICALAPPROACH

Our approach to identifying the constraintson working memory has been to examine theextent to which variation in complex span per-formance can be attributed to individual differ-

ences in the ability to perform the processingand storage requirements of the complex spantask in isolation. This involves partitioning thevariance in complex span performance into theunique and shared contributions associatedwith each component of working memory. Al-though this technique for examining individualdifferences in working memory abilities is notuncommon (Kane et al., 2004; Salthouse &Babcock, 1991; Shah & Miyake, 1996), thenovelty of our approach is in the design of ourcomplex span tasks. Across a series of studies wehave used a common methodology in which wesystematically vary the domain of processingand storage involved in each task. To do this, wefirst developed two processing tasks that sharethe same basic characteristics but differ in thecontent of the information to be manipulated(verbal and visuospatial). We then integrated

these tasks with two types of storage (verbal andvisuospatial) to create four complex span tasksin which the processing and storage compo-nents are factorially combined across the verbaland visuospatial domains. Another importantaspect of our design is careful control of thetiming of processing and storage episodes so that

the duration of the trials within each task is al-ways the same.

The strength of this design is that it allowsus to directly compare performance across the

processing and storage domains of complex spantasks. It also allows us to examine whether theconstraints on complex span performance arelargely domain general in nature and show aconsistent pattern of results across all complexspan tasks, or are domain specific and thereforedependent on the particular combination of processing and storage involved. Perhaps themost important aspect of this design, however,is that it allows us to take independent measures

of the processing and storage involved in eachcomplex span task and account for the variancein performance that is associated with these.Theoretically, this leaves us with a relativelypure estimate of any potentially executive con-tribution to complex span performance that fol-lows from combining the processing and storagecomponents of the task. As a result, this ap-proach has allowed us to address the two ques-tions central to our work: what are the various

constraints on complex span performance, andhow do these constraints mediate the link be-tween complex span performance and higher-level cognition?

WHAT ARE THE CONSTRAINTSON COMPLEX SPANPERFORMANCE?

Two Initial Studies

We first explored this question in two studiesthat examined the nature of constraints under-lying complex span performance in both chil-dren and adults (Bayliss, Jarrold, Gunn, &Baddeley, 2003). In the first study, 75 childrenaged 8 and 9 years completed a battery of fourcomplex span tasks—two independent mea-

sures of processing efficiency and two indepen-dent measures of storage capacity. Consistentdisplay characteristics were used across all fourcomplex span tasks: nine different-colored cir-cles were presented in a random arrangement onthe screen, with one of the digits 1 to 9 shown inthe center of each circle. In addition, one of the




circles was presented with a beveled edge, thelocation of which varied on successive trials. Inthe tasks involving verbal processing, the chil-dren were presented with a verbal object name

(i.e., ‘‘milk’’) and were asked to find the circlethat corresponded to the color of the object asquickly as possible. This was thought to involveverbal processing in that the child was requiredto first associate the object name with the objectand then retrieve the color most often identifiedwith that object. Clearly efficiency of target de-tection could depend on a number of factorssuch as lexical semantic knowledge and visualimagery, a point we return to below. In the tasks

involving visuospatial processing, the childrenwere asked to find the circle with a beveled edgeas quickly as possible. Both processing tasksled the children to an appropriate target cir-cle, at which point they were asked to remembereither the number in the center of the targetcircle (verbal storage) or the location of the targetcircle (visuospatial storage) for later recall. Thetiming of each processing and storage intervalwas fixed so that if a child found the target circle

relatively quickly they did not move on to thenext processing episode of the trial until this setinterval had elapsed. In contrast, if a child failedto find the appropriate target within the availabletime, the correct storage item was shown to thembefore the presentation of the next processingepisode. The number of processing and storageepisodes in each list increased in a span proce-dure until the child could no longer recall thestorage items in correct serial order.

In addition to the four complex span tasksdescribed above, we also took independent mea-sures of the processing and storage requirementsof each task. The processing tasks correspondedto the processing involved in the complex spantasks but were presented without any associatedstorage requirements. They were designed assearch tasks, and required the participant to ei-ther decide what color a verbally presented ob-

ject was and touch the appropriately coloredcircle as quickly as possible (verbal), or scan thearray of circles for the circle with the bevelededge (visuospatial). By varying the number of distractor items present in these search displays,we were able to confirm that target detection inour visuospatial processing task depended on the

number of distractors present, a finding thatsuggests that participants were actively scanningthe visual display. In contrast, in the verbal pro-cessing task, performance was not affected by

number of distractors. This finding suggests thatregardless of the precise processes involved indetermining the typical color of a named object,these do not lead to visuospatial searching of thedisplay. Similarly, our independent measures of individuals’ storage capacity were taken by rep-licating the storage demands of the complex spantasks but without any concurrent processing re-quirements. Consequently, these correspondedto a digit span (verbal storage) and a Corsi span

(visuospatial storage) task. Analysis of the rela-tionship among these independent measures of processing and storage ability showed that theverbal and visuospatial processing measures werehighly correlated (r ¼ .75), suggesting that theremay be a common factor driving performance onthese measures. In contrast, the correlation be-tween the verbal and visuospatial storage mea-sures was not as strong (r ¼ .32), and was in factcomparable to the correlations between storage

and processing measures.These suggestions were borne out by an

exploratory factor analysis on the data from ourfour complex span tasks, our two processingtasks and our two storage tasks, which producedthree factors that together accounted for 72%of the total variance. The first factor corres-ponded to a general processing factor with strongloadings from both measures of processing ef-ficiency. The second factor showed loadingsfrom digit span and the two complex span tasksinvolving verbal storage, results suggesting thatit corresponded to a verbal storage factor. In con-trast, the third factor showed loadings from thetwo complex span tasks involving visuospatialstorage and Corsi span, which suggests that thisfactor was best interpreted as a visuospatial stor-age factor. Consistent with the Baddeley (1986)model, these results provide further support for

separation of the systems responsible for thestorage of verbal and visuospatial information.However, they also suggest the involvement of adomain-general processing component, whichis largely driven by both measures of pro-cessing efficiency. Given the simplicity of theseprocessing tasks, we would argue that this gen-




eral processing factor is non-executive in nature,and instead reflects the constraints imposed byindividual differences in the ability to performthe processing component of the complex span

task.To explore the relative importance of eachof these constraints, we examined the extent towhich individual variation in both processingefficiency and storage capacity determined per-formance on each complex span task. To dothis, we used a variance partitioning procedureto identify the unique and shared variance at-tributable to the processing and storage com-ponents (cf., Salthouse & Babcock, 1991). A

series of hierarchical multiple-regression anal-yses showed that the domain-appropriate mea-sures of storage capacity contributed uniquevariance to performance on each complex spantask (between 8.5% and 18.9% of the totalvariation). The independent measures of pro-cessing efficiency also accounted for significantunique variance in the two complex span tasksincorporating verbal processing (14.6% and19.3% when entered together) but not in the

two tasks involving visuospatial processing. Weargue that this result may be due to the visuospa-tial processing task not being demanding en-ough to influence performance on the complexspan tasks incorporating this type of processing.Indeed, the two complex span tasks involvingverbal processing showed a greater drop in per-formance relative to simple span performancethan the two complex span tasks involvingvisuospatial processing (cf., Oberauer & Suß,2000). In addition, there was very little sharedvariance between the processing and storagemeasures, indicating that most of the variationin span performance was accounted for byunique contributions. These results suggestthat, in children, complex span performance isconstrained by variation in both processing ef-ficiency and storage capacity, and furthermore,that these contributions are largely indepen-

dent of one another.In our second study, we were interested tosee if this pattern would replicate in an adultsample. The tasks from our first experimentwere modified to make them more appropriatefor adult participants; however, the basic char-acteristics of each task remained the same. The

one significant change that we did make was toconvert the simple visuospatial search task fromthe initial study into a conjunctive search task.This involved changing the display screen to an

array of big and small squares, half of whichwere presented with a beveled edge and half of which were not. The visuospatial processingtask involved searching the array to locatethe big square with a beveled edge as quickly aspossible. Only one of these squares was pre-sented in each display, thus making the searchtask more difficult, as the target could not bedistinguished from the distractors by the pres-ence or absence of a single feature. A sample

of 48 adults completed the four complex spantasks, the two independent measures of pro-cessing and the two independent measuresof storage. An exploratory factor analysis per-formed on the data from the complex spantasks and the independent measures of storageand processing revealed a three-factor structuresimilar to that found with the initial sample of children. The first factor corresponded to a vis-uospatial storage factor with strong load-

ings from Corsi span and the two complexspan tasks incorporating visuospatial storage.The second factor showed loadings from the twomeasures of processing efficiency, suggestingthat this was a general processing factor. Thethird factor showed high loadings of the twocomplex span tasks involving verbal storage andthe digit span task, results suggesting that thisfactor corresponded to a verbal storage factor.These results closely replicated the patternfound previously with children and providedfurther support for the existence of distinct ver-bal and visuospatial storage systems and a do-main-general processing component.

Also consistent with our initial results wasthe finding that the independent measures of verbal and visuospatial storage capacity eachcontributed significant unique variance to thecorrespondingcomplexspantasks(between12%

and 41%). The independent measures of pro-cessing efficiency uniquely accounted for 12%of the variance in the complex span task com-bining verbal processing with verbal storage;however, this was the only significant contri-bution of processing efficiency to complex spanperformance in the adult sample. Nonetheless,




the finding that processing efficiency was asignificant predictor of performance in one of the tasks indicates that an individual’s proces-sing ability can be, but is not necessarily, a

constraint on complex span performance evenunder conditions where the duration of eachtrial is held constant. On the basis of thesefindings, we argue that individuals could po-tentially vary in terms of both their ability toperform the processing activity of the complexspan task and their ability to maintain thestorage items in mind. Furthermore, we arguethat it is the balance of these constraints withinan individual that will determine their com-

plex span performance (cf., Hitch et al., 2001). A third source of potential variation that we

were interested in exploring was whether thecoordination of the processing and storage ep-isodes of the complex span task required anadditional executive ability, independent of theprocessing and storage abilities themselves.Borrowing from the approach of Kyllonen andChristal (1990; see also Engle, Tuholski, et al.,1999), we examined the residual variance in

complex span performance once variance as-sociated with the independent measures of processing efficiency and storage capacity wasremoved. In line with the arguments advancedabove, we maintain that if there is an executiveability involved in complex span performancethat is reflected in the residual variance, thenthis should be domain general and should beobserved on all tasks that involve the combi-nation of storage and processing requirements.In other words, residual variance should cor-relate across our different complex span tasks.The correlations between the residuals de-rived from our first two studies are shown in

Table 6.1. Although only two of these correla-tions were significant in the children’s data, allbut one of the residuals were significantlycorrelated in the adult data. This finding is

important because it indicates that these re-siduals do not simply reflect measurement er-ror but instead index an additional ability thatcontributes consistent variance to complex spanperformance (cf., Emerson et al., 1999). More-over, the fact that the residuals from all fourcomplex span tasks were correlated to a certaindegree in the adult sample suggests that thisability is domain general (cf., Oberauer et al.,2003), as one would expect if it is executive in

nature. Indeed, in the adult sample, the residualsfrom the two same-domain complex span tasks(verbal–verbal and visuospatial–visuospatial)shared 13% variance and the two cross-domaintasks (verbal–visuospatial and visuospatial–verbal) shared 11% variance, despite the fact thatthese pairs of tasks share none of their processingor storage elements.

A DEVELOPMENTAL STUDY

Consistent with our original hypotheses, theresults of these first two experiments provideevidence for three independent sources of work-ing memory variation: storage capacity (whetherit is verbal or visuospatial), processing efficiency,and executive ability. More recently, we haveturned our attention to exploring the factors thatdrive age-related increases in working memoryperformance. The results from the two experi-ments reported by Bayliss et al. (2003) highlightsome important similarities between childrenand adults, but also some important differences

TABLE 6.1. Correlational Analysis of Complex Span Residualsfrom Samples of Children (to the Bottom Left of the LeadingDiagonal) and Adults (to the Top Right of the Leading Diagonal).

Residuals 1. 2. 3. 4.

1. Verbal verbal — .296* .399** .364*2. Verbal visuospatial .126 — .329* .399**3. Visuospatial verbal .124 .233* — .2394. Visuospatial visuospatial .174 .293* .033 —

**p< .01; *p< .05. These data were not previously reported in Bayliss et al. (2003).




(i.e., processing efficiency seems to be a moreimportant predictor of complex span perfor-mance in children than in adults). It is possiblethat the constraints imposed on complex span

performance may follow different developmen-tal trajectories and, consequently, the relativeimportance of these constraints may changeacross different stages of development (cf.,Conway et al., 2002; Cowan et al., 2005). Thecapacity of working memory has been shown toincrease with development (Case et al., 1982;Hale et al., 1997; Towse et al., 1998) and de-cline during older adulthood (Salthouse &Babcock, 1991). A number of researchers have

argued that this age-related change in workingmemory capacity is directly related to changesin processing speed. For example, Fry and Hale(1996) found that developmental increases inprocessing speed accounted for most of the age-related increases in working memory capacity(cf., Salthouse & Babcock, 1991). However, thefindings from our previous work suggest thatdevelopmental increases in the capacity for short-term storage may also be important in driving

age-related improvements in complex span per-formance.

We addressed these issues in a develop-mentalstudy(Bayliss,Jarrold,Baddeley,Gunn,&Leigh, 2005) examining the complex span per-formance of 120 typically developing childrenaged 6, 8, and 10 years. To assess the claims of Fry and Hale (1996) and of Salthouse andcolleagues (Salthouse, 1994, 1996; Salthouse &Babcock, 1991), that speed of processing me-diates most of the age-related change in work-ing memory performance, we included anumber of verbal and visuospatial speed tasksdesigned to measure various levels of speededperformance. The most basic measures weredesigned as forced-choice reaction-time tasksthat were broadly linked to the verbal andvisuospatial domains. In both tasks, a fixationcross was displayed in the center of the screen,

with a picture of a bird on one side and a frogon the other. In the auditory speed task, thechildren were presented with either a short,high-pitch tone, which they were told corre-sponded to the bird, or a short, low-pitch tone,which corresponded to the frog. They were toldthat they had to listen to the tones and touch

the corresponding picture on the screen asquickly as they could without making mistakes.In the visual speed task, the participants werepresented with a picture of either the bird or

the frog in the center of the screen and weretold to touch the correct picture on the screenas quickly as possible. Reaction times for cor-rect touch-responses were taken as a measure of basic auditory and visual speed.

In addition, we also included tasks designedto measure verbal and visuospatial mainte-nance rate, which we thought might underliesome of the age-related variation in storagecapacity. In the verbal domain, it is commonly

assumed that storage items are maintained inthe phonological loop by means of a subvocalrehearsal process and that the rate at which onecan articulate a list of items provides a good ap-proximation of the rate of covert rehearsal(Baddeley, Lewis, & Vallar, 1984; Baddeleyet al., 1975). In line with this research, we tooka measure of each child’s articulation rate as asurrogate for the rate at which they were able tomaintain the verbal storage items. In the visuo-

spatial domain, Logie (1995) has suggested thatthe maintenance of spatial patterns may besupported by a spatially based system termedthe ‘‘inner scribe.’’ The most common methodof disrupting the spatial storage system in adual-task paradigm is spatial tapping (Logie,1995). Performance on this task is not generallytaken as an index of the ability to maintainvisuospatial material in short-term memory,however. So in contrast, we designed a mentalrotation task as our measure of visuospatialmaintenance rate. Various mental rotation taskshave been used as a measure of the speed withwhich a target figure can be spatially manip-ulated (Hegarty et al., 2000; Miyake, Fried-man, Rettinger, Shah, & Hegarty, 2001). Weargued that a mental rotation task would pro-vide a suitable visuospatial analogue of articu-lation rate. In line with our previous work, we

also took independent measures of the verbaland visuospatial processing involved in the com-plex span tasks, as well as independent mea-sures of each child’s verbal and visuospatialstorage capacity by means of a digit span anda Corsi span task, respectively. For practicalreasons, we chose to include only two of the




complex span tasks described above—the taskcombining verbal processing with verbal stor-age and the task combining visuospatial pro-cessing with visuospatial storage. As expected,

clear age trends were evident in each of themeasured variables.One of our interests was to evaluate whether

there was a common speed factor underlyingperformance on all of the processing and stor-age variables, or whether there were a numberof factors corresponding to the different typesof speeded tasks. To explore this issue, weperformed an exploratory factor analysis on thedata from the six processing speed tasks and

the two measures of storage capacity. Analysisof the results suggested a two-factor solution,which accounted for 70% of the total variance.The first factor showed strong loadings of thetwo basic speed measures, and moderate tostrong loadings of the two measures of proces-sing efficiency and the two measures of main-tenance rate, results suggesting a link to ageneral processing speed factor. The secondfactor can best be described as a storage factor

with high loadings of the two storage tasks anda moderate loading of articulation rate. Thisfinding suggests the presence of a general factorrelated to processing speed as well as a sepa-rate storage-related factor. In contrast to ourprevious findings, there was no evidence of domain-specific storage, as both the verbal andvisuospatial storage measures loaded on thesame factor. We argue that this may reflect theindependence of developmental and indivi-dual differences in working memory capacity.To a large extent, verbal and visuospatial storageabilities develop in parallel (Chuah & Maybery,1999; Gathercole, 1999) and so are both closelyassociated with age. Thus, across a large devel-opmental sample, dissociations between the twostorage systems are likely to be swamped bystrong correlations between the two factorsdriven by age-related improvements in both.

However, within a specific age group, when agedoes not have such an overriding influence, it ispossible to isolate the specific constraints thateach imposes on complex span performance.

To assess the adequacy of the two-factormodel, we performed a confirmatory factoranalysis. Digit span, Corsi span, and articulation

rate were linked to one latent storage variable,while the verbal and visuospatial processingmeasures, the auditory and visual speed vari-ables, rotation rate, and articulation rate were all

linked to a second latent variable named speed.Initially, this model did not provide a good fit tothe data. However, modification indices sug-gested that the errors from the two processingtasks and the two basic speed tasks should belinked, reflecting the fact that there was sometask-specific variance associated with these mea-sures. With this adjustment, the two-factor modelprovided an excellent fit to the data, as indicatedby all the measures of fit considered. Parameter

estimates for the final model are presented inFigure 6.1. The two factors were correlated, asindicated by the double-headed arrow connect-ing the two latent variables. However, a subse-quent analysis of a one-factor model, which isequivalent to assuming a perfect correlation be-tween the two factors, did not provide a satisfac-tory fit to the data. This discrepancy providedfurther support that two factors were needed todescribe the data. To explore the relationship

between each of these latent variables and age-related variance in complex span performance,we constructed a structural-equation modellinking each of the latent variables to a third la-tent variable drawn from the two complex spanvariables. On the basis of increasingly popularhypothesis that speed of processing mediatesmost of the age-related change in working mem-ory performance, we specified direct paths fromthe speed variable to the storage variable and thecomplex span variable. We also specified a directpath from the storage variable to the complexspan variable and direct paths from age to eachlatent variable. This initial model provided agood fit to the data; however, a number of thepaths were nonsignificant. These paths were de-leted and the model was reanalyzed. On the basisof the fit statistics, the revised model also pro-vided a good fit to the data. This final structural

equation model is presented in Figure 6.2. Forclarity, the observed variables contributing to thespeed and storage latent variables have beenomitted from the diagram.

As expected, the speed and storage vari-ables both have direct effects on complex spanperformance, and age has direct effects on the




speed and storage variables. Thus age-relatedchanges in both speed of processing and stor-

age capacity may contribute to age-related im-provements in complex span performance.Indeed, the absence of a direct link between ageand complex span indicates that these two ef-fects account for most age-related variance inworking memory performance. The key finding,however, is that there was no link between thespeed and storage variables. That is, the age-

related effect of storage capacity on complexspan performance was not entirely mediated by

age-related changes in speed of processing. Thisfinding is in contrast with the developmentalcascade hypothesis of Fry and Hale (1996) andsuggests that processing speed and storage-related factors provide separable constraints onage-related changes in complex span perfor-mance. The model presented in Figure 6.2,however, does appear to be consistent with our

Figure 6.1. Parameter estimates for the final two-factor model following confirmatory factoranalysis. RT¼ reaction time; V-S¼ visuospatial. (Adapted with permission from Bayliss, D. M.,Jarrold, C., Baddeley, A. D., Gunn, D. M., & Leigh, E., Mapping the developmental constraintson working memory span performance, Developmental Psychology, 41, p. 588, 2005, AmericanPsychological Association.)

Figure 6.2. Structural-equation model of relations among age and components of complex spanperformance. V-S¼ visuospatial. (Adapted with permission from Bayliss, D. M., Jarrold, C.,Baddeley, A. D., Gunn, D. M., & Leigh, E., Mapping the developmental constraints on workingmemory span performance, Developmental Psychology, 41, p. 590, 2005, American Psychological

Association.)




own finding that processing efficiency and stor-age capacity impose independent constraints oncomplex span performance.

Taken together, the evidence from the studies

reported here suggests that there are at least threesources of variation in working memory perfor-mance. We have clearly shown the importanceof individual differences in domain-specificstorage and domain-general processing to com-plex span performance in children and adults.Furthermore, there is evidence of a third sourceof potentially executive variation. Examinationof the factors driving age-related changes incomplex span performance also revealed two

sources of independent variation, processingspeed and storage capacity. The implication of these findings is that complex span performanceis multidetermined; one or all of these inde-pendent sources of variation may be importantin mediating the relation between complex spanperformance and cognitive skills such as readingand mathematics.

WHAT MEDIATES THE LINK BETWEENCOMPLEX SPAN PERFORMANCEAND HIGHER-LEVEL COGNITION?

Predictors of Higher-LevelCognition in TypicallyDeveloping Individuals

Having identified at least three independentsources of variation in complex span perfor-mance, we were interested in exploring the re-lationship between these factors and level of educational attainment. A subsequent analysisof the data from our initial work (Bayliss et al.,2003) showed that the scores from the generalprocessing factor accounted for unique vari-ance in children’s reading and mathematicsperformance (7% and 6%, respectively). In ad-dition, scores from the visuospatial storage fac-

tor uniquely accounted for an additional 13% of the variation in children’s mathematics perfor-mance. In the adult sample, scores from theverbal storage factor uniquely accounted forover 25% of the variance in reading perfor-mance. These findings suggest that variation inchildren’s general processing efficiency may be

an important mediator of the relationship be-tween their complex span performance andreading and mathematics ability. Moreover, itseems that variance related to visuospatial stor-

age abilities is particularly important for math-ematics performance in children (cf., McLean& Hitch, 1999; although see Bull, Johnston, &Roy, 1999), whereas in adults, verbal storageability may be an important mediator of readingperformance. Converging evidence for the im-portance of storage abilities to academic at-tainment comes from our developmental studyin which scores from the storage factor con-tributed unique variance to reading and math-

ematics performance across development (10%and 11%, respectively). Developmental differ-ences in general speed of processing also con-tributed significant unique variance to readingand math performance in this study (5% in bothcases).

Residual variation independent of indi-viduals’ ability to carry out the processing andstorage components of complex span tasks wasalso found to be an important predictor of read-

ing and math skills in both children and adultsin our initial studies (Bayliss et al., 2003). In ourfirst study, the residuals from the two complexspan tasks involving verbal storage were signifi-cantly correlated with children’s reading andmathematics performance. Furthermore, in theadult sample assessed in our second study, mostof the correlations between the residuals fromthe complex span tasks and the measures of reading, mathematics, and performance on theRaven’s Standard Progressive Matrices were ei-ther significant or approaching significance.These findings provide further evidence thatthere is a third, potentially executive, source of variation in complex span performance. If theseresiduals do reflect variation in some executiveability, then the fact that this variation is an im-portant predictor of academic abilities is con-sistent with the findings of Engle, Tuholski, et al.

(1999) and Conway et al. (2002), who arguedthat controlled attention was the component of working memory responsible for the predictivepower of the complex span task. However, whatour results suggest is that individual differencesin general processing speed and storage abilitymay also mediate the relationship between




complex span performance and reading andmathematics ability. To address this issue fur-ther, we have begun to examine how differentsources of working memory variation map onto

task variance in reading.

Predictors of Higher-LevelCognition in AtypicallyDeveloping Individuals

To gain a greater understanding of the re-lationship between the various components of working memory and reading skills, we assessedreading ability on three different levels: the

ability to recognize written letters, the ability todecode and recognize written words, and theability to make sense of sentences (cf., Badde-ley, Gathercole, & Spooner, 2003). To maxi-mize the variation in reading abilities assessedby these tests, we examined the performance of 80 children with moderate learning difficulties.By examining the working memory abilities of these children, we expected to gain a clearerpicture of how the various components of

working memory influence different aspects of reading skill. Each child completed a batteryof our working memory tasks: two complex spantasks combining verbal processing with verbalstorage in one case and visuospatial storage inthe other, two independent measures of storage(digit span and Corsi span), and an indepen-dent measure of verbal processing efficiency. Inaddition, we also included two measures of phonological awareness (rhyme detection andinitial phoneme deletion) taken from the Pho-nological Abilities Test (Muter, Hulme, &Snowling, 1997) (see Bayliss, Jarrold, Baddeley,& Leigh, 2005, for full details).

A series of hierarchical regression analyseswere performed on each reading test to exam-ine the unique contribution of these variousmeasures to the prediction of different aspectsof reading ability. The storage measures were

the only variables to contribute significant un-ique variance to the letter decision task, ex-plaining just over 10% of the total variancewhen entered together. In contrast, in the worddecision task, the two storage measures failedto independently account for any significantunique variance in performance, whereas the

two measures of phonological awareness to-gether accounted for approximately 10% of thevariance uniquely and the two complex spanmeasures uniquely accounted for approximately

7% of variance when entered together. The mea-sures of phonological awareness also accountedfor a significant amount of variance in perfor-mance on the sentence decision task (12%) in-dependently of the other variables. The storagemeasures again accounted for very little vari-ance when entered after the other variables,whereas the complex span measures uniquelyexplained 5% of the variance in performance onthe sentence decision task.

Consequently, across these three levels of reading, the relative importance of each aspectof working memory performance varied (cf.,Swanson & Berninger, 1995, 1996). Examina-tion of the differential contribution of variablesfrom the more basic letter recognition skillsthrough to the more comprehensive level of reading skill required in the word and sentencedecision tasks revealed a decrease in the con-tribution of basic storage ability and an in-

crease in the contribution of complex spanperformance. These findings suggest that com-plex span performance is more than just thesum of its parts. The complex span measurescontributed variance to the word and sentencedecision tasks over and above that accounted forby measures of storage ability, processing effi-ciency, and phonological skills. Thus, complexspan performance may tap an additional work-ing memory ability, possibly a controlled atten-tion or executive component, an ability involvedin key aspects of reading skill. This idea is con-sistent with numerous studies from the Englelab, which have emphasized the controlled-attention component as the driving force be-hind the predictive power of the complex spantask. However, the finding that other compo-nents of working memory are important at morebasic levels of reading ability suggests that if one

of these components of working memory fails todevelop normally, these early stages of learningto read may not be navigated successfully. A deficit of this form could have repercussionsnot only for later stages of reading but also nu-merous other areas of a child’s developmentthat are either directly dependent on or in some




quences of variation at this microlevel that mir-ror the resource-sharing and resource-switchingaccounts of complex span at the task level. Onthe one hand, engagement in processing may

affect the ease with which individuals can si-multaneously engage in other activities such asmaintenance operations. Alternatively, proces-sing may prevent individuals from carrying outany other activity at all.

Our data do not rule out the possibilitythat processing disrupts one’s ability to shareresources for the reactivation or rehearsal of to-be-remembered material. However, otherfindings indicate that an explanation of the pro-

cessing effect in terms of its blocking of maintenance activities may be preferable. Thecrucial support for this form of account comesfrom studies that have attempted to separateduration and difficulty of processing by askingindividuals to continuously perform processingtasks for a fixed interval. For example, Halford,Maybery, O’Hare, and Grant (1994, Experi-ment 4) compared the effects of a fixed dura-tion of either forward or backward counting on

the recall of a digit list, and found that whileboth processing tasks reduced recall relative toa baseline condition, the size of this effect wascomparable in each case. Similar results havebeen reported by Towse et al. (2002). Thesefindings are clearly inconsistent with the no-tion that individuals are sharing resourcesduring processing phases, and instead suggestthat any form of processing will block mainte-nance regardless of its cognitive demand. If oneextrapolates these findings to our complex-spansituation, then one would predict that individualdifferences in processing efficiency would notrelate to performance if individuals used upall of the available time for processing. If so,then any effect of processing that we do see inour tasks would be due to the duration of theprocessing within this fixed time window, ra-ther than its difficulty (cf., Conway & Engle,

1996).Other studies have provided somewhat dif-ferent results from those of Halford et al.(1994) and Towse et al. (2002), and have sug-gested that processing difficulty can have aneffect on item retention, independent of dura-tion effects (e.g., Barrouillet, Bernardin, & Ca-

mos, 2004; Barrouillet & Camos, 2001; Posner& Rossman, 1965). It seems possible, however,that in these cases where apparent processingdifficulty effects are observed the tasks involved

allowed space for individuals to carry out otheractivities while completing processing. In otherwords, a form of micro–resource switchingmight occur during these processing tasks(Barrouillet & Camos, 2001; Barrouillet et al.,2004; Gavens & Barrouillet, 2004). In fact, if this suggestion is correct, then it might not be somuch the demand of a processing task that de-termines whether switching to retention-basedactivities can be carried out alongside proces-

sing, but rather the rate at which that task ispresented (Barrouillet & Bernardin, 2002).Barrouillet and Bernardin (2002) reported datashowing that relatively undemanding proces-sing tasks can have a severe impact on complexspan performance if presented at a rapid pace(see also Barrouillet et al., 2004). If this is thecase, then provided that a processing task is ra-pid enough to keep participants busy it willprevent switching to retention-based activities,

and the duration but not the difficulty of thatprocessing will be what determines degree of forgetting. In contrast, if the task allows partic-ipants to slow the rate at which they performprocessing to the point at which it can be tem-porarily interrupted, then a reduced effect of processing difficulty may appear to occur, eventhough it may more accurately reflect an in-crease in the time spent switching away fromprocessing to maintenance activity. Towseet al.’s (2002) manipulation may have beensuccessful in preventing this form of switchingbecause of the way in which processing opera-tions were continuously repeated (although seeHalford et al., 1994, Experiment 1, for contra-dictory results).

A MODEL

These points lead us to propose a novel accountof complex span performance that is consistentwith Baddeley’s working memory model andTowse and colleagues’ resource-switching hy-pothesis. It also extends these approaches byconsidering the potentially independent effects




of storage and processing constraints, as well asthe possibly executive requirements of com-bining these operations. One of our mainclaims is that variation in the rate at which in-

dividuals can complete the processing opera-tions of our tasks leads to variation in the extentof forgetting to-be-remembered information, asmaintenance operations are prevented for theduration of processing. However, a key differ-ence between our complex span procedure andthe design employed by Towse et al. (2002) isthat our fixed-time window procedure leads notonly to periods when maintenance is preventedbut also to those when reactivation of the

to-be-remembered information can occur (cf.,Cowan, 1992; Cowan et al., 1998), presumablyby some form of rehearsal. This potential forreactivation, we would argue, is what accountsfor the importance of storage constraints in ourprocedures. Of course, our design is such thatvariation in time available for reactivation iscollinear with variation in processing duration.

As a result, one might expect that processingand storage influences could not be separated

in the way that we have demonstrated in ourstudies. However, while the extent of informa-tion loss from immediate memory in the ab-sence of maintenance operations depends ontime elapsed, the success of these maintenanceoperations depends on the rate at which indi-viduals can reactivate to-be-remembered infor-mation. Consequently, provided this rate of reactivation is not determined by the same fac-tors that limit speed of completion of processingoperations, these effects can be independent inour design. That is, although individuals whoprocess quickly will have more time availablefor maintenance activities, the extent to whichthey benefit from this time will also be deter-mined by the rate at which they can reactivatethe to-be-remembered information in that pe-riod (cf., Oberauer & Kliegl, 2001).

This explanation is strongly supported by

the results of our developmental study, whichprovide evidence that rate of reactivation of stored information can be dissociated from anindividual’s general speed of processing (seealso Cowan, 1999; Cowan et al., 1998; Jarrold,Hewes, & Baddeley, 2000). In modeling thesedata it was shown that although articulation rate

was associated with a general index of process-ing speed, it shared additional variance withmeasures of storage. This finding suggests thatthe storage construct that emerges in this model

captures variance in rate of reactivation of to-be-remembered information. One possibility isthat this variance represents speed of subvocalrehearsal. A central tenet of Baddeley’s (1986)model is that immediate serial recall of verbalinformation is determined by the rate at whichindividuals can articulate and thereby rehearsethis material (e.g., Baddeley et al., 1975). How-ever, the storage construct shown in Figure 6.1is drawn from tasks that involve either verbal or

visuospatial storage. Consequently, it seemsunlikely that this indexes verbal subvocal re-hearsal rate specifically. Instead, it may reflect acommon reactivation mechanism that operateson different domains of content in verbal andvisuospatial immediate-recall tasks (cf., Joneset al., 1995). Indeed, other studies have shownthat articulation rate is an equally strong pre-dictor of visuospatial and verbal short-termmemory performance (Chuah & Maybery,

1999; Smyth & Scholey, 1992, 1996a). This isnot to say that all variance in storage ability isdomain general in nature; as noted above, ourother studies show that the content of to-be-remembered information is an importantsource of variability in some instances. Takentogether, therefore, our data indicate that verbaland visuospatial storage abilities can dissociateand give rise to separable variance at the level of individual differences (see also Jarrold, Badde-ley, & Hewes, 1999; Smith et al., 1996), but thatthe factor driving development in verbal andvisuospatial storage is common to both domains(Chuah & Maybery, 1999).

The one source of variability left unex-plained by the above analysis is the residualvariance in complex span performance that weobserve when we partial out independent esti-mates of individuals’ storage and processing

abilities. This factor may reflect the executivecost of coordinating these aspects of the task,which in our terms would correspond to theswitch between phases of processing, duringwhich maintenance is blocked, and periods of reactivation. One way of testing this proposalwould be to examine the residuals that arise






OF WORKING MEMORY

We are guided by Baddeley’s (1986) model of

working memory that distinguishes between

domain-specific short-term storage of verbal and

visuospatial information and domain-general

executive-controlprocesses.Weargue,however,

that the distinction between short-term storage

systems may reflecta distinction in content rather

than process, and that age-related change in

verbal and visuospatial storage may be medi-

ated by a common mechanism. We also suggest

that the processing operations involved in tradi-

tional working memory tasks are typically not in

and of themselves executive in nature. Rather,

we argue, in line with Engle and colleagues (e.g.,

Engle, 2002), that executive control may well be

required in complex span tasks as a result of the

combination of storage and processing demands.

2. CRITICAL SOURCES OF WORKINGMEMORY VARIATION

Our principled analysis of the demands of

complex span tasks suggests that variation in

working memory performance might arise for

three main reasons. First, individuals might differ

in their storage capacity, whether in the verbal

or visuospatial domain. Second, variation might

arise due to differences in the efficiency with

which individuals carry out the processing op-erations of the task. Finally, the requirement

to combine storage and processing operations

gives rise to a third potential source of variance

in performance. Our data support this analysis

by providing evidence that all three of these

factors operate to constrain working memory

performance and can play a role in mediating

the relationship between working memory and

other higher-level abilities.


MEMORY VARIATION

Our approach is at odds with resource-sharing

accounts of working memory variation, in that

we emphasize and find support for separable

influences of processing efficiency and storage

capacity on complex span performance. Thisfinding might also appear problematic for a

basic resource-switching hypothesis, as varia-

tion in processing efficiency affects perfor-

mance without affecting the time-dependent

storage demands of our tasks. Our model of

complex span performance is in fact consistent

with a resource-switching account if one ac-

cepts that reactivation of memory items occurs

in pauses between processing. In addition, we

suggest that individuals may vary in the rate at

which they forget information during pro-

cessing activities. This may well reflect varia-

tion in individuals’ ability to resist interference,

as argued by a number of authors in this vol-

ume.

4. CONTRIBUTIONS TO

GENERAL WORKINGMEMORY THEORY

There are three main implications of our re-

search that follow, at least in part, from our use

of typically and atypically developing popula-

tions. First, our data support a model that sits

somewhere between previous accounts that

have emphasized either the processing or stor-

age aspects of complex span. Our data also

confirm that complex span performance ismore than the sum of its parts, and suggest

ways of testing what is involved in combining

storage and processing operations. Second,

they show that the factors leading to differ-

ences across individuals of different ages need

not be the same as those that cause variation

among individuals of the same age. Finally, in

addition to confirming that complex span is

multiply determined, our findings show that

the constraints on working memory perfor-mance and the ways in which these mediate

the relationship with other higher-level abili-

ties can vary among individuals of different

ages and levels of ability.

155



from complex span–type tasks in which theordering of processing and storage episodes wasvaried to manipulate the number of switchesbetween these phases of the task.

An alternative and intriguing suggestion isthat this residual variance reflects differences inthe rate at which individuals forget informationwhile engaged in processing (Barrouillet et al.,2004; Hitch et al., 2001; Lepine, Barrouillet, &Camos, 2005). In our analysis the effect of processing reflects the time during which re-activation is prevented. However, it is possiblethat individuals vary in the rate at which in-formation is lost during that period. Crucially,

this variation will only be observed when stor-age is combined with a processing requirement,and so would represent an emergent property of the complex span design. Indeed, variation inrate of forgetting during processing might ariseas a result of executive factors that deter-mine individuals’ resistance to interference (seeChapters 2 and 9). Alternatively, individualdifferences in forgetting-rates might simply re-flect variation in a relatively low-level decay

parameter (Oberauer & Kliegl, 2001), whichwould remove the need to view this emergentconstraint in executive terms. Future researchcould separate these possibilities by comparingthe residuals, once processing efficiency andstorage capacity are accounted for, from com-plex span tasks that vary in the extent to whichthey give rise to distracting interference (cf.,Lustig, May, & Hasher, 2001). Our work buildson previous research using complex span tasks(e.g., Engle, Tuholski et al., 1999; Kyllonen &Christal, 1990) to show that one can derivemeaningful and predictive residuals that cap-ture more than the ability to complete thecomponent aspects of the task alone. Havingdemonstrated this, the challenge for our futureresearch is to properly specify exactly what thisresidual variance really represents.

Acknowledgments

The research described in this chapter was sup-

ported by a cooperative group component grant

from the United Kingdom Medical Research Coun-

cil to Christopher Jarrold and Alan D. Baddeley

(Grant G0000258, within Cooperative Group Grant

G9901359). Deborah Riby (ne Gunn) and Eleanor

Leigh also contributed considerably to this work.

References

Baddeley, A. (1986). Working memory. New York:


Baddeley, A. (1993). Working memory or working

attention. In A. Baddeley & L. Weiskrantz

(Eds.), Attention: selection, awareness, and con-

trol. A tribute to Donald Broadbent (pp. 152–

170). Oxford: Clarendon Press.

Baddeley, A. (1996). Exploring the central execu-

tive. Quarterly Journal of Experimental Psychol-

ogy, 49A, 5–28.Baddeley, A. D., Gathercole, S. E., & Spooner, A.

(2003). The reading decision test. Hove, UK:

Psychology Press.




and theory (Vol. 8, pp. 47–89). New York:

Academic Press.

Baddeley, A. D., Lewis, V., & Vallar, G. (1984).

Exploring the articulatory loop. Quarterly Jour-nal of Experimental Psychology, 36A, 233–252.

Baddeley, A. D., Thomson, N., & Buchanan, M.

(1975). Word length and the structure of short-

term memory. Journal of Verbal Learning and

Verbal Behavior, 14, 575–589.

Barrouillet, P., & Bernardin, S. (2002). Working

memory spans in adults depend on the pace of

the secondary task. Presented at the First Euro-

pean Working Memory Symposium, Ghent,

September.

Barrouillet, P., Bernardin, S., & Camos, V. (2004).

Time constraints and resource sharing in adults’

working memory spans. Journal of Experimental


Barrouillet, P., & Camos, V. (2001). Developmental

increase in working memory span: Resource

sharing or temporal decay? Journal of Memory


Bayliss, D. M., Jarrold, C., Baddeley, A., & Gunn,D. (2005). The relationship between short-term

memory and working memory: Complex span

made simple? Memory, 13, 414–421.

Bayliss D. M., Jarrold, C., Baddeley, A. D., Gunn,

D. M., & Leigh, E. (2005). Mapping the

developmental constraints on working memory




span performance. Developmental Psychology,

41, 579–597.

Bayliss, D. M., Jarrold, C., Baddeley, A. D., &

Leigh, E. (2005). Differential constraints on the

working memory and reading abilities of indi-viduals with learning difficulties and typically

developing children. Journal of Experimental


Bayliss, D. M., Jarrold, C., Gunn, D. M., &

Baddeley, A. D. (2003). The complexities of

complex span: Explaining individual differ-

ences in working memory in children andadults.


132, 71–92.

Bennetto, L., Pennington, B. F., & Rogers, S. J.(1996). Intact and impaired memory functions

in autism. Child Development, 67, 1816–1835.

Bishop, D. V. M. (1997). Cognitive neuropsychology

and developmental disorders: Uncomfortable

bedfellows. Quarterly Journal of Experimental

Psychology, 50A, 899–923.

Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides,

J., Smith, E. E., & Noll, D. C. (1997). A

parametric study of prefrontal cortex involve-

ment in human working memory. Neuroimage,5, 49–62.

Bull, R., Johnston, R. S., & Roy, J. A. (1999).

Exploring the roles of the visual-spatial sketch pad

and central executive in children’s arithmetical

skills: Views from cognition and developmental

neuropsychology. Developmental Neuropsychol-

ogy, 15, 421–442.

Cantor, J., Engle, R. W., & Hamilton, G. (1991).

Short-term memory, working memory, and

verbal abilities: How do they relate? Intelli-

gence, 15, 229–246.

Case, R., Kurland, D. M., & Goldberg, J. (1982).




Chuah, Y. M. L., & Maybery, M. T. (1999). Verbal

and spatial short-termmemory: common sources

of developmental change? Journal of Experi-

mental Child Psychology, 73, 7–44.Conrad, R. (1964). Acoustic confusions in immedi-

ate memory. British Journal of Psychology, 55,

75–84.

Conway, A. R. A., Cowan, N., Bunting, M. F.,

Therriault, D. J., & Minkoff, S. R. B. (2002). A

latent variable analysis of working memory ca-

pacity, short-term memory capacity, processing

speed, and general fluid intelligence. Intelli-

gence, 30, 163–183.

Conway, A. R. A., & Engle, R. W. (1996). Individual

differences in working memory capacity: Moreevidence for a general capacity theory. Memory,

4, 577–590.

Cornoldi, C., Dalla Vecchia, R., & Tressoldi, P. E.

(1995). Visuo-spatial working memory limita-

tions in low visuo-spatial high verbal intel-

ligence children. Journal of Child Psychology

and Psychiatry, 36, 1053–1064.

Cowan, N. (1992). Verbal memory span and the

timing of spoken recall. Journal of Memory and

Language, 31, 668–684.Cowan, N. (1999). The differential maturation of

two processing rates related to digit span. Jour-

nal of Experimental Child Psychology, 72, 193–

209.

Cowan, N., Elliot, E. M., Saults, J. S., Morey, C.,

Mattox, S., Hismjatullina, A., et al. (2005). On

the capacity of attention: Its estimation and its

role in working memory and cognitive abilities.

Cognitive Psychology, 51, 42–100.

Cowan, N., Towse, J. N., Hamilton, Z., Saults,J. S., Elliott, E. M., Lacey, J. F., et al.

(2003). Children’s working-memory processes:

A response-timing analysis. Journal of Experi-


Cowan, N., Wood, N. L., Wood, P. K., Keller, T. A.,

Nugent, L. D., & Keller, C. V. (1998). Two







Behavior, 19, 450–466.




422–433.

Daneman, M., & Tardif, T. (1987). Working mem-

ory and reading re-examined. In M. Coltheart(Ed.), Attention and performance XII (pp. 491–

508). Hillsdale, NJ: Lawrence Erlbaum Asso-

ciates.

Dempster, F. N. (1981). Memory span: Sources of

individual and developmental differences. Psy-

chological Bulletin, 89, 63–100.




Jarrold, C., Baddeley, A. D., & Phillips, C. (1999).

Down syndrome and the phonological loop:

The evidence for, and importance of, a specific

verbal short-term memory deficit. Down Syn-

drome Research and Practice, 6, 61–75.Jarrold, C., Hewes, A. K., & Baddeley, A. D. (2000).

Two separate speech measures constrain ver-

bal short-term memory in children. Journal

of Experimental Psychology: Learning, Memory,


Jarrold, C., & Russell, J. (1997). Counting abilities

in autism: Possible implications for central

coherence theory. Journal of Autism and De-

velopmental Disorders, 27, 25–37.

Jenkins, L., Myerson, J., Hale, S., & Fry, A. F. (1999).Individual and developmental differences in

working memory span across the life span.


Jones, D., Farrand, P., Stuart, G., & Morris, N.

(1995). Functional equivalence of verbal and

spatial information in serial short-term mem-

ory. Journal of Experimental Psychology: Learn-

ing, Memory, and Cognition, 21, 1008–1018.

Kail, R., & Hall, L. K. (2001). Distinguishing short-

term memory from working memory. Memory& Cognition, 29, 1–9.


Wilhelm, O., Payne, T. W., & Engle, R. W.

(2004). The generality of working memory

capacity: A latent-variable approach to verbal

and visuospatial memory span and reasoning.


133, 189–217.


ability is (little more than) working memory


La Pointe, L. B., & Engle, R. W. (1990). Simple and

complex memory spans as measures of working

memory capacity. Journal of Experimental Psy-

chology: Learning, Memory and Cognition, 16,

1118–1133.

Lepine, R., Barrouillet, P., & Camos, V. (2005).

What makes working memory spans so predic-

tive of high level cognition? Psychonomic Bulle-tin & Review, 12, 165–170.

Logie, R. H. (1995). Visuo-spatial working memory.

Hillsdale, NJ: Lawrence Erlbaum Associates.

Logie, R. H., & Marchetti, C. (1991). Visuo-spatial

working memory: Visual, spatial or central

executive? In R. H. Logie & M. Denis (Eds.),

Mental images in human cognition (pp. 105–

115). Amsterdam: Elsevier.

Logie, R. H., Zucco, G. M., & Baddeley, A. D. (1990).

Interference with visual short-term memory. Acta

Psychologia, 75, 55–74.Lustig, C., May, C. P., & Hasher, L. (2001). Work-

ing memory span and the role of proactive

interference. Journal of Experimental Psycho-

logy: General, 130, 199–207.

Maybery, M. T., & Do, N. (2003). Relationships

between facets of working memory and perfor-

mance on a curriculum-based mathematics test

in children. Educational and Child Psychology,

20, 77–92.

McLean, J. F., & Hitch, G. J. (1999). Workingmemory impairments in children with specific

arithmetic learning difficulties. Journal of Ex-

perimental Child Psychology, 74, 240–260.

Milner, B. (1971). Interhemispheric differences in

the localisation of psychological processes in

man. British Medical Bulletin, 27, 272–277.


P., & Hegarty, M. (2001). How are visuospatial


spatial abilities related? A latent-variable anal-ysis. Journal of Experimental Psychology: Gen-

eral, 130, 621–640.

Montgomery, J. W. (2003). Working memory and

sentence comprehension in children with spe-

cific language impairment: What we know so

far. Journal of Communication Disorders, 36,

221–231.

Muter, V., Hulme, C., & Snowling, M. (1997). The

Phonological Abilities Test. San Antonio, TX:

The Psychological Corporation.

Myerson, J., Jenkins, L., Hale, S., & Sliwinski, M.

(2000). Stocks and losses, items and interfer-

ence: A reply to Oberauer and Suß (2000). Psy-

chonomic Bulletin & Review, 7, 734–740.

Oakhill, J., & Kyle, F. (2000). The relation between

phonological awareness and working memory.

Journal of Experimental Child Psychology, 75,

152–164.

Oberauer, K., & Kliegl, R. (2001). Beyond resources:Formal models of complexity effects and age

differences in working memory. European Jour-

nal of Cognitive Psychology, 13, 187–215.

Oberauer, K., Schulze, R.,Wilhelm, O., & Suß, H.-M.

(2005). Working memory and intelligence—

their correlation and their relation: Comment




on Ackerman, Beier, and Boyle (2005). Psycho-


Oberauer, K., & Suß, H.-M. (2000). Working mem-

ory and interference: A comment on Jenkins,

Myerson, Hale, and Fry (1999). PsychonomicBulletin & Review, 7, 727–733.

Oberauer, K., Suß, H.-M., Schulze, R., Wilhelm, O.,

& Wittmann, W. W. (2000). Working memory

capacity—facets of a cognitive ability construct.

Personality and Individual Differences, 29,

1017–1045.

Oberauer, K., Suß, H.-M., Wilhelm, O., & Witt-

mann, W. W. (2003). The multiple faces of

working memory: Storage, processing, supervi-

sion and coordination. Intelligence, 31, 167–193.Pennington, B. F., & Ozonoff, S. (1996). Executive

functions and developmental psychopathology.

Journal of Child Psychology and Psychiatry, 37,

51–87.

Pickering, S. J., Gathercole, S. E., & Peaker, S. M.

(1998). Verbal and visuo-spatialshort-term mem-

ory in children: Evidence for common and dis-

tinct mechanisms. Memory & Cognition, 26,

1117–1130.

Posner, M. I., & Rossman, E. (1965). Effect of sizeand location of informational transforms upon

short-term retention. Journal of Experimental


Russell, J. (1997). Autism as an executive disorder .

Oxford, UK: Oxford University Press.

Russell, J., Jarrold, C., & Henry, L. (1996). Working

memory in children with autism and with

moderate learning difficulties. Journal of Child

Psychology and Psychiatry, 37, 673–686.

Salthouse, T. A. (1994). How many causes are

there of aging-related decrements in cognitive

functioning? Developmental Review, 14, 413–

437.

Salthouse, T. A. (1996). The processing-speed theory

of adult age differences in cognition. Psycho-

logical Review, 103, 403–428.

Salthouse, T. A., & Babcock, R. L. (1991). Decom-

posing adult age differences in working mem-

ory. Developmental Psychology, 27, 763–776.Shah, P., & Miyake, A. (1996). The separability of




chology, 125, 4–27.

Siegel, L. S., & Ryan, E. B. (1989). The develop-

ment of working memory in normally achiev-

ing and subtypes of learning disabled children.

Child Development, 60, 973–980.

Smith, E. E., Jonides, J., & Koeppe, R. A.(1996). Dissociating verbal and spatial work-

ing memory using PET. Cerebral Cortex, 6,

11–20.

Smyth, M. M., & Scholey, K. A. (1992). Determin-

ing spatial span: The role of movement time

and articulation rate. Quarterly Journal of Ex-

perimental Psychology, 45A, 479–501.

Smyth, M. M., & Scholey, K. A. (1996a). The

relationship between articulation time and

memory performance in verbal and visuospatialtasks. British Journal of Psychology, 87, 179–191.

Smyth, M. M., & Scholey, K. A. (1996b). Serial

order in spatial immediate memory. Quarterly

Journal of Experimental Psychology, 49A, 159–

177.

Swanson, H. L., & Berninger, V. (1995). The role

of working memory in skilled and less skilled

reader’s comprehension. Intelligence, 21, 83–

108.

Swanson, H. L., & Berninger, V. W. (1996). Indivi-dual differences in children’s working memory

and writing skill. Journal of Experimental Child


Tehan, G., Hendry, L., & Kocinski, D. (2001).

Word length and phonological similarity effects

in simple, complex and delayed serial recall

tasks: Implications for working memory. Mem-

ory, 9, 333–348.


(1998). A reevaluation of working memory

capacity in children. Journal of Memory and

Language, 39, 195–217.


(2002). On the nature of the relationship

between processing activity and item retention

in children. Journal of Experimental Child


Towse, J. N., & Houston-Price, C. M. T. (2001).

Combining representations in working mem-ory: A brief report. British Journal of Develop-

mental Psychology, 19, 319–324.

Turner, M. L., & Engle, R. W. (1989). Is working

memory capacity task dependent? Journal of





Vallar, G., & Papagno, C. (1995). Neuropsycholog-

ical impairments of short-term memory. In A. D.

Baddeley, B. A. Wilson, & F. Watts (Eds.),

Handbook of memory disorders (pp. 135–165).

Chichester: John Wiley & Sons. Vicari, S., Carlesimo, G. A., Brizzolara, D., & Pezzini,

G. (1996). Short-term memory in children with

Williams syndrome: A reduced contribution of

lexical-semantic knowledge to word span. Neu-

ropsychologia, 34, 919–925.

Wang, P. P., & Bellugi, U. (1994). Evidence from two

genetic syndromes for a dissociation between

verbal and visual-spatial short-term memory.

Journal of Clinical and Experimental Neuropsy-

chology, 16, 317–322.Waters, G. S., & Caplan, D. (1996). The measure-

ment of verbal working memory capacity and

its relation to reading comprehension. Quar-


51–79.




7

Developmental and Computational

Approaches to Variation in

Working Memory

YUKO MUNAKATA, J. BRUCE MORTON,

and RANDALL C. O’REILLY

You’ve lost your keys and are searching aroundthe house for them. To find them, you mustkeep in mind the goal of searching for the keysas you wander from room to room, searchingpockets, hooks, and containers. And you want tokeep track of where you have already searchedand where you have yet to search. Although weare not perfect at these skills, we are reasonablygood at them (as evidenced by the fact thattypically keys are ultimately found!).

In this chapter, we consider the workingmemory processes that contribute to our abili-ties to handle these kinds of tasks, and differ-ences that lead to variations in these workingmemory processes. We focus on two processes,active maintenance and information updating,that are central to working memory abilities.

We do not mean to claim that these are theonly processes involved in working memory,but they are important processes that otherworking memory processes likely tap or buildon. We describe computational models andbehavioral studies designed to test the mecha-nisms subserving active maintenance and infor-

mation updating. We discuss how variations inthese processes may contribute to variations inworking memory observed across development,across diseases and disorders, and across typicaladults. Understanding such variation may helpus to understand working memory in general.We close with a discussion of the relation be-tween our approach to working memory andothers’ in this volume (see Box 7.1 for sum-mary).

Before turning to these substantive issues,we provide a brief overview of the particularcomputational approach that we take. Ourmodeling work fits within the growing field of computational cognitive neuroscience (O’Reilly& Munakata, 2000), which focuses on the useof computational models to help understand

the relation between the brain and behavior.We use artificial neural network models thatmathematically simulate biological neurons.These networks can be manipulated and testedin very specific and precise ways, to assesstheories of how biological neural networks func-tion to produce our thoughts and behaviors.

162



Such models can be crucial for helping us tounderstand complex, nonlinear interactions of the sort that characterize brain–behavior rela-tions. Moreover, such models assist in theory

comparison and evaluation by requiring theo-ries to be specific and plausible enough thatthey can lead to working models, and by gen-erating testable predictions. For these reasonsand others, many researchers argue that suchmodeling work is essential for advancing the-orizing about cognitive functioning (e.g., El-man et al., 1996; O’Reilly & Munakata, 2000;Rumelhart & McClelland, 1986; Seidenberg,1993).

In the domain of working memory, com-putational models can help us to understandthe specializations required for the functions of activation maintenance and information up-dating. This kind of approach can inform anunderstanding of how and why different neuralregions subserve different functions. For ex-ample, converging evidence from a number of methods indicates that the prefrontal cortexand basal ganglia play an important role in

working memory (e.g., Bell & Fox, 1992; Braveretal.,1997;Brown&Marsden,1990;Goldman&Rosvold, 1972; Miller, Erickson, & Desimone,1996; Petrides, 1994; Smith & Jonides, 1998;Stuss & Benson, 1984). What is it about thesebrain regions that supports these specializa-tions? Why, for example, doesn’t parietal cortexplay more of a role than prefrontal cortex? Whydoesn’t all of the neocortex contribute equallyto working memory? A computational approachcan help answer these kinds of questions, asdescribed in the next section.

OVERARCHING THEORYOF WORKING MEMORY

Our overarching theory of working memoryfocuses on the computational mechanisms un-

derlying the active maintenance and updatingof information. Active maintenance refers toholding information ‘‘in mind’’ in a robust form,for example, when it is no longer present in theenvironment, across delays, and in the face of distraction caused by ongoing processing. Forexample, to find your keys, you must actively

keep this goal in mind throughout the search,rather than being distracted by what you comeacross in the environment (e.g., a stack of billsthat need to be paid!). Active maintenance can

be subserved by the sustained firing of neurons,as observed in prefrontal cortex during workingmemory tasks (e.g., Fuster, 1989; Miller et al.,1996).

Updating of information in working mem-ory refers to the process of interrupting theactive maintenance of current information, sothat new information can be represented inworking memory. This updating process is crit-icalforflexiblebehavior.Forexample,insearch-

ing for your keys, if you are actively maintainingthe goal of searching yesterday’s pants pocketsbut the keys are not there, you want to interruptthe maintenance of that location and move onto other possibilities. Updating may be guidedby specialized systems that signal whether infor-mation should be maintained or interrupted(knowing ‘‘when to hold ’em and when to fold’em’’). Such signals might be mediated by sig-nals from the basal ganglia to the prefrontal

cortex (e.g., Frank, Loughry, & O’Reilly, 2001;O’Reilly & Frank, 2006).

We focus on these complementary processesof maintenance and updating in our investiga-tions of working memory for two related reasons.First, active maintenance and updating may berelatively amenable to computational investiga-tion, because theycanbe defined, implemented,and manipulated in computational models inrelatively straightforward ways. Other aspects of working memory, such as the manipulation of information (e.g., when mentally computing4217),aremorecomplicatedandmaybemoredifficult to link directly to underlying neuralmechanisms initially. Second, we believe thatactive maintenance and updating of informa-tion may form the bases for other processes re-lated to working memory. In some cases (e.g.,mental multiplication), additional mechanisms

may be required, building on maintenance andupdating. In other cases, a process that mayappear distinct from maintenance and updating(e.g., inhibition) may actually emerge fromthese basic processes without requiring addi-tional mechanisms. In either case, maintenanceand updating are likely to play an important role

Developmental and Computational Approaches 163



(e.g., Miyake & Shah, 1999), so we focus ourcurrent efforts on these processes.

We are particularly interested in the kinds of representations and learning that allow systems

to specialize in active maintenance and infor-mation updating. We use neural network mod-els to investigate the relation between relevantbrain areas (specifically prefrontal cortex andbasal ganglia) and working memory processes.Much of this work is guided by a considerationof computational trade-offs in different memorydemands, as described next.

Computational Trade-offsin Active Maintenance

A computational perspective can provide in-sight into how and why neural regions are spe-cialized for different functions (reviewed inO’Reilly & Munakata, 2000). Such specializa-tions can be understood in terms of computa-tional trade-offs, whereby two objectives cannotbe achieved simultaneously. That is, as a systemspecializes on its ability to achieve one objec-

tive, it must relinquish its ability to achieveanother objective. For example, there is a com-putational trade-off between fast learning andslow learning; a system that specializes in learn-ing rapidly is not well suited to learning grad-ually and vice versa. Thus, if there are demandson a system for both fast and slow learning, thesefunctions are likely to depend on distinct neuralregions with unique specializations. Similarly,there is a computational trade-off between rep-resentations that are highly overlapping andrepresentations with little overlap, so that if bothare desired, they too are likely to rely on spe-cialized neural systems.

These kinds of computational trade-offs,between distinct types of learning and repre-sentations, can provide insight into the spe-cializationsof neuralsystems subservingworkingmemory functions. We first explore such trade-

offs in very simple models, to see how speciali-zations are required to maintain information inan active form across time and to updateinformation (O’Reilly, Mozer, Munakata, &Miyake, 1999; O’Reilly & Munakata, 2000).We later consider the elaboration of suchmodels to simulate performance on specific

tasks and explore sources of working memoryvariability.

Consider the simple network in Figure 7.1(O’Reilly & Munakata, 2000). This network

contains input and hidden units that representa monitor, speakers, and keyboard. Weightsconnect hidden units that represent semanti-cally related information; in this case, eachhidden unit is connected to the other two.Such interactive representations confer seman-tic benefits, such as allowing a system to go fromincomplete information to activate relatedinformation. This kind of connectivity couldsubserve semantic networks of the sort observed

in posterior cortical regions (e.g., McClelland &Rogers, 2003; Lambon-Ralph, Patterson, Gar-rard, & Hodges, 2003).

However, such interactive representationsalso come with a price: loss of information whenit is supposed to be maintained across delays.When this network is presented with a monitorand speakers, the network correctly activates the

Input

Hidden

KeyboardSpeakersMonitor

Monitor Speakers Keyboard

Figure 7.1. Interconnected network. Weights con-nect hidden units that represent semantically relatedinformation. Such connectivity could subserve se-mantic networks of posterior cortical areas. (Adaptedfrom O’Reilly & Munakata [2000], Figure 9.18,p. 301. Copyright 2000 MIT Press.)




corresponding hidden units (top half of Fig. 7.2).However, when the input is removed, the acti-vation spreads across the hidden units via theweights connecting all of the units (bottom half

of Fig. 7.2). As a result, during the maintenanceperiod, it is no longer clear what the networkwas initially presented with; the network hasfailed to cleanly maintain this information.Such failures to reliably maintain activationare observed in posterior cortical areas (e.g.,Miller & Desimone, 1994; Steinmetz, Connor,

Constantinidis, & McLaughlin, 1994). Tran-sient maintenance may be observed in theseareas, but it is much less robust than thatin prefrontal cortex, for example, failing to sus-

tain activity in the face of interfering stimuli.ThesimplenetworkinFigure7.1canbeelab-orated to improve its active maintenance abil-ities. For example, higher-order representationscan be added, which connect with lower-levelfeatures that go together (Fig. 7.3). This im-proves the network’s ability to maintain infor-mation after it is removed. Instead of theactivation simply spreading from Monitor and

ut dtrial Eventcycle Inp Hid en

0 Input9

0 Input19

0 Input29

0 Input39

0 Input49

0 Input59

0 Input69

0 Input79

0 Input89

0 Input99

1 Maintain9

1 Maintain19

1 Maintain29

1 Maintain39

1 Maintain49

1 Maintain59

1 Maintain69

1 Maintain79

1 Maintain89

1 Maintain99

Figure 7.2. Input and hidden-unit activity as theinterconnected network in Figure 7.1 is presentedwith two inputs (top half of figure), and then thoseinputs are removed (bottom half of figure). Each rowcorresponds to one time-step of processing. Eachunit’s activity level is represented by the size of the

corresponding black square. The network correctlyactivates the corresponding hidden units when theinputs are present, but fails to maintain this infor-mation alone when the input is removed, because of interactive representations. (Adapted from O’Reilly& Munakata [2000], Figure 9.19, p. 301. Copyright2000 MIT Press.)

Input


Hidden

TV Synth

Terminal

Hidden2


Figure 7.3. Semantic network with higher-orderrepresentations. Weights connect hidden units thatrepresent semantically related information, with alayer of higher-order representations that connectwith lower-level features. (Adapted from O’Reilly &Munakata [2000], Figure9.20, p. 302. Copyright2000MIT Press, used with permission.)




Speakers to Keyboard, for example, the higher-order representation of TV is preferentially acti-vated in the second hidden layer, and this pref-erentially activates Monitor and Speakers.

However, because the system is still relativelyinterconnected (e.g., Monitor also connects toTerminal and Speakers also connects to Synth),this solution is not particularly robust. When asmall amount of noise is introduced intothe network processing (of the sort that ourbrains likely contend with on a regular basis),activation again spreads beyond the initial in-put, because of the connections with other units.

More isolated representations may be re-

quired for systems to maintain representationsover delays, in the absence of input, and in theface of noise (e.g., for working memory). Anextreme form of such isolated representations isshown in Figure 7.4. In this network, each in-put unit is connected to its corresponding hid-den unit, and each hidden unit is connectedonly to itself, rather than to the other semanti-cally related hidden units. When this network ispresented with a monitor and speakers, the

network correctly activates the correspondinghidden units (top half of Fig. 7.5). When the

input is removed, the activation is maintained inthese units (bottom half of Fig. 7.5), becausethere is no way for the activation to spreadfrom these units to any other units. As a result,

this network successfully maintains the previ-ously presented information during the main-tenance period. This solution is robust to noisein the network processing. This kind of isolatedconnectivity could subserve active maintenanceabilities of prefrontal cortical regions.

Dynamic Gating and InformationUpdating in the Basal Ganglia

What about the process of updating informationin working memory? We can again explore this

Input

Monitor

Hidden

Speakers Keyboard


Figure 7.4. Network with isolated representations.Each hidden unit connects to only itself, rather thanto other semantically related units. Such connec-tivity could subserve active maintenance abilities of prefrontal cortical areas.

trial Eventcycle Input Hidden

0 Input9

0 Input19

0 Input29

0 Input39

0 Input49

0 Input59

0 Input69

0 Input79

0 Input89

0 Input99

1 Maintain9

1 Maintain19

1 Maintain29

1 Maintain39

1 Maintain49

1 Maintain59

1 Maintain69

1 Maintain79

1 Maintain89

1 Maintain99

Figure 7.5. Input and hidden-unit activity as thenetwork in Figure 7.4 is presented with two inputs(top half of figure), and then those inputs are removed(bottom half of figure). The network activates thecorresponding hidden units when the inputs are pres-ent, and maintains this information when the input isremoved, because of isolated representations.




issue first through simple simulations. Considera revised set of inputs presented to the networkin Figure 7.4. As in Figure 7.5, these inputsstart with a stimulus that is presented and then

removed, but then a new input pattern is pre-sented and then removed. In some cases, thegoal might be to ignore the second stimulus andmaintain the first stimulus throughout the delayand the second input (e.g., when the first stim-ulus represents a location to search for keys andthe second stimulus represents a stack of bills tobe paid). In other cases, the goal might be toupdate with the second stimulus (e.g., when thefirst stimulus represents a location to search for

keys that turns out to be empty, and the secondstimulus represents a new location to search).

One parameter that affects whether thenetwork maintains or updates is the strength of its recurrent connections. When these are rel-ativelyweak,thenetworkupdatesinsteadofmain-taining. When these are relatively strong, thenetwork maintains instead of updating. How-ever, such static systems, set either to consistentlymaintain or to consistently update information,

would not be particularly useful for workingmemory. For example, a system that alwaysmaintained information would fail to evermove beyond whatever it represented first (e.g.,to search in pants pockets). On the flip side, asystem that always updated information wouldfail to withstand interference from distractors(e.g., the stack of bills to be paid). Thus, to beuseful, the working memory system insteadneeds to be dynamic, maintaining and updatinginformation as required across different situa-tions and with different inputs.

This dynamic switching between maintain-ing and updating can be achieved via a gatingmechanism (Fig. 7.6A) (e.g., Cohen, Braver, &O’Reilly, 1996; Hochreiter & Schmidhuber,1997; O’Reilly, Braver, & Cohen, 1999). Whenthe gate is open, working memory representa-tions can be updated (e.g., from incoming

sensory inputs), and when it is closed, workingmemory representations are protected frominterference from such inputs, leading to robustmaintenance. The basal ganglia have a num-ber of neural specializations that appear ideallysuited to serve as a gating mechanism to theprefrontal cortex (PFC) working memory sys-

tem; we have captured these mechanisms in anintegrated prefrontal–basal ganglia workingmemory (PBWM) model (Frank et al., 2001;O’Reilly & Frank, 2006). First, the basal gan-glia are strongly interconnected with the PFC.Second, the output of the basal ganglia is mod-

ulatory on PFC, in that it disinhibits prefrontalneurons, instead of directly exciting them. Thisdisinhibition interaction is one way that neu-ral systems can achieve gating, which is afundamentally modulatory interaction. Third,there are a number of parallel loops intercon-necting PFC and the basal ganglia (Alexander,

SensoryInput

WorkingMemory 4920054

8675309Jenny, Igot your # ...

Myphone # is4920054

4920054

Gating open closed

a) Update b) Maintain

... PFC

...

PosteriorCortex

...

...

thalamus

Striatum(matrix)

}GP (tonic act)excitatory

inhibitory{

disinhib

Figure 7.6. A: Illustration of dynamic gating. Whenthe gate is open, sensory input can rapidly updateworking memory, but when it is closed, it cannot,thereby preventing other distracting informationfrom interfering with the maintenance of previously

stored information. B: The basal ganglia (striatum,globus pallidus [GP], and thalamus) are inter-connected with frontal cortex through a series of parallel loops. Direct-pathway striatal neurons dis-inhibit prefrontal cortex (PFC) by inhibiting toni-cally active GP internal–segment (and substantianigra pars reticulata, not shown) neurons, releasingthalamic neurons from inhibition. This disinhibi-tion provides a modulatory or gating-like function(Go signal). There are also indirect-pathway neurons(not shown) that provide a counteracting inhibitory(NoGo) signal.




DeLong, & Strick, 1986), so that the basalganglia can provide multiple independentgating signals (i.e., selective gating), so thatsome areas of PFC can be maintaining while

others are updating.Figure 7.6B provides a more detailed (butstill simplified; see Frank et al., 2001; O’Reilly &Frank, 2006, for fuller details), anatomical pic-ture of the PFC–basal ganglia system, showingthe direct pathway through the dorsal striatum,globus pallidus (GP), thalamus, and back toPFC. This pathway produces disinhibitory mod-ulation of PFC. The GP neurons are tonicallyactive and thus tonically inhibit the thalamus.

When a striatal neuron fires (they are usuallyinactive), it inhibits the GP neurons to which itprojects, thus disinhibiting the thalamus, whichis reciprocally interconnected with the frontalcortex via excitatory connections. This thalamicdisinhibition thus enables, but does not directlycause (i.e., gates), a loop of excitation into thefrontal cortex. The effect of this excitation in thePBWM model is to toggle the state of bistablecurrents in the prefrontal neurons. Thus, when

prefrontal neurons are in the up state, they havea persistent excitatory current that helps themremain active over time, while other neurons inthe down state lack this extra excitation (Dur-stewitz, Kelc, & Gunturkun, 1999; Durstewitz,Seamans, & Sejnowski, 2000; Fellous, Wang, &Lisman, 1998; Wang, 1999). This bistablemaintenance is further supported by recurrentexcitatory connections among other such pre-frontal neurons, and the combination providesimportant computational advantages (Frank etal., 2001; O’Reilly & Frank, 2006).

In short, the firing of a direct-pathway neu-ron, which we refer to as a Go signal, togglesthe maintenance of information in PFC in thePBWM model. These Go neurons are activateddirectlybystimulusinputscontextualizedbydes-cending PFC projections, via learned weights. If a PFC neuron is not maintaining information,

and a Go signal is received, it will start main-taining its current representation. If it is alreadymaintaining something, then the Go signal willturn off this maintenance, allowing it to startmaintaining something else. To clear an exist-ing representation and store a different one (i.e.,an update), two Go signals are required. This

toggling pattern of behavior has been observed inprefrontal neurons in vitro (J. Seamans, personalcommunication, January 2002). There are alsostriatal neurons that project via an indirect path-

way, with the effect of increasing the level of inhibition on the thalamic pathway. We referto these as the NoGo neurons in the PBWMmodel—they compete with the Go neurons andenable the PFC to continue to maintain cur-rently stored information. Interestingly, theseNoGo neurons only have their effect by pre-venting (out-competing) Go neuron firing; theydo not result in any kind of direct effect onthe PFC.

In other work, we have shown that the ap-propriate patterns of Go and NoGo firing inthe basal ganglia gating system can be learnedvia powerful reinforcement-learning mecha-nisms that are widely thought to be supportedby other aspects of the basal ganglia system(Contreras-Vidal & Schultz, 1999; Houk, Ad-ams, & Barto, 1995; Joel, Niv, & Ruppin, 2002;O’Reilly & Frank, 2006; Schultz, Dayan, &Montague, 1997; Suri & Schultz, 2001). We

will describe these learning mechanisms ingreater detail later. The resulting model wasable to learn complex working memory tasks,including those requiring multiple levels of maintenance and updating of working memoryoperating in parallel, and thus serves as an initialplatform for developing biologically based cog-nitive models of working memory function.Work is currently under way in applying thismodel to a wide range of working memory tasksthat have previously been modeled with a vari-ety of other existing working memory models(e.g., AX-CPT, Stroop, Wisconsin Card Sortingtask, Erikson Flanker task), with the goal of de-veloping a ‘‘unified model’’ of working memoryfunction.

Summary

These simulations demonstrate computationaltrade-offs in active maintenance, which providea strong computational foundation for our overalltheory of working memory function. Interactiverepresentations can support semantic knowledgeand isolated representations can subserve activemaintenance of information across delays, of




the sort required for working memory. Both typesof representations are useful, but there is acomputational trade-off between them; a singlesystem cannot simultaneously specialize in inter-

connected and isolated representations. As a re-sult, one neural system (posterior cortex) mayspecialize in interconnected representations,while another system(prefrontal cortex) mayspe-cialized in isolated representations. This com-putational approach is consistent with (and mayhelp to make sense of) findings from neurosci-ence on the anatomy (Levitt, Lewis, Yoshioka, &Lund, 1993) and physiology (Rao, Williams, &Goldman-Rakic, 1999) of PFC, which may sug-

gest more isolated representations in this region.Further, the system that specializes in active

maintenance must also be able to dynamicallyswitch between robust maintenance and rapidupdating with new information. This switchingrequires a dynamic gating mechanism, and thebasal ganglia have appropriate specialized neu-ral mechanisms to achieve this dynamic gatingfunction through extensive disinhibitory con-nections with the PFC. Furthermore, the basal

ganglia also contain learning mechanisms cap-able of training the dynamic gating mechanismsin task-relevant ways. Thus, we think our inte-grated PBWM model has the elements in placefor a fully self-contained theory of both mainte-nance and control of working memory, withoutrelying on unexplained ‘‘homunculi’’ such as acentral executive.

We have investigated to varying degreesthese components of our overarching theory—isolated representations for active maintenance,and a gating system for information updating—and their relevance to variation in workingmemory. In what follows, we describe findingsfrom this research, and we consider avenues forfurther exploration within this framework.

CRITICAL SOURCES OF WORKING

MEMORY VARIATION

There are many potential sources of workingmemory variation within our framework, whichwe have investigated to varying degrees. Forexample, people could differ in the degree of isolation and strength of working memory rep-

resentations, the learning of working memoryrepresentations, and the efficiency of gating pro-cesses. After briefly presenting some of thesepossibilities, we will focus our discussion on

contributions from variations in the strengthof working memory representations, the mainsource of variation we have explored to date.

Gating and Learningof Representations

We expect that our models will provide a richsource of predictions regarding the specific con-tributions of the basal ganglia dynamic gating

system to individual variability. Differences ingating abilities should be important both in the‘‘mature’’ trained form of the networks and overthe developmental time course of learning tasks.By virtue of having numerous anatomicallyspecified mechanisms, our full PBWM modelhas the potential for exploring a wide range of sources of individual variability in workingmemory function, which could in principle beindependently assessed through various neuro-

science measurement techniques. This potentialhas yet to be realized, however, as we are justat the initial stages of exploring these models.Nevertheless, we can provide some examples of what these differences might look like.

Diseases of the basal ganglia provide one ex-treme source of variability. For example, Par-kinson patients can exhibit working memorydifficulties similar to those of frontal patients(e.g., Brown & Marsden, 1990). However, ourmodel would predict that a more careful anal-ysis would reveal some differences. Specifically,we would expect that Parkinson patients shouldnot be impaired on raw maintenance of work-ing memory information per se, but rather theyshould be impaired on updating working mem-ory. This can be manifest in cases where oneneedstochange(e.g.,reverse)priorpatternsofre-sponding. Consistent with this prediction, evi-

dence in both humans and animals suggests thatbasal ganglia damage produces selective deficitsin reversal learning (e.g., Brown & Marsden,1990; Ragozzino, Ragozzino, Mizumori, &Kesner, 2002). Another example of the potentialspectrum of phenomena that could be ad-dressed with our model comes from the detailed




analysis of a set of cognitive control tasks inpopulationswithschizophrenia,Sydenhamcho-rea, Tourette syndrome, and attention-deficithyperactivity disorder (ADHD) (Casey, Dur-

ston, & Fossella, 2001). Casey et al.’s (2001)framework for understanding the PFC and basalganglia system has much in common with ourPBWM model; they were able to interpret be-havioral patterns across tasks and populations interms of deficits in Go–NoGo dynamics con-tributed by the basal ganglia. Our PBWMmodel may provide increasingly detailed andmechanistically explicit explanations and pre-dictions of this sort.

Another source of variability that arises inour computational models is variability in thekinds of representations that develop over learn-ing. Even with fixed learning mechanisms andparameters, the complex interactions betweenthese mechanisms, random initial weights, andsimulated environmental experiences can pro-duce different representations after learning.These differences in representations, particu-larly in the PFC and basal ganglia systems,

could then substantially affect a whole range of working memory–related behaviors. We haverecently explored one aspect of representationaldevelopment in a PFC model, looking at howthe PFC component of the network was able todevelop more discrete, abstract, rule-like rep-resentations of stimulus dimensions than theposterior-cortex component. These more ab-stract representations then greatly facilitatedcross-task generalization, where experience withitems in one task context generalized to otherrelated task contexts (Rougier, Noelle, Braver,Cohen, & O’Reilly, 2005). The development of these representations interacted critically withthree factors: (1) the presence of neural spe-cializations associated with both the PFC andbasal ganglia; (2) the need to maintain stimulusdimensions over contiguous trials; and (3) thebreadth of experience across multiple differ-

ent task contexts. Thus, individual variability inany of these factors could lead to importantdifferences in the kinds of representations thatdevelop, and consequently in the ability to per-form more abstract forms of generalization ortransfer. Having explicit computational modelsenables us to explore complex interactions such

as these, which might be too difficult to managein purely verbal terms.

Variations in Strength of Working

Memory Representations

The main source of variation we have exploredto date is the strength of actively maintainedrepresentations. These explorations are basedon the idea that representations are graded innature rather than being all-or-nothing, presentor absent (reviewed in Munakata, 2001). Thatis, instead of simply remembering or knowingsomething or not knowing it, we remember and

know things to differing degrees. In the case of working memory representations, this graded-ness might be instantiated in terms of the num-ber of neurons contributing and the firing ratesof those neurons. This graded-representationsapproach has been applied to understandingvariation with development and following braindamage (e.g., Farah, Monheit, & Wallace,1991; Farah, O’Reilly, & Vecera, 1993; Joanisse& Seidenberg, 2003; Munakata, McClelland,

Johnson, & Siegler, 1997; Plaut & Booth, 2000).The developmental work has demonstratedhow the strengthening of working memory rep-resentations can lead to variations observedacross infants and children at different ages.Relatively weak working memory representa-tions might suffice for tasks that children carryout successfully early on, but stronger repre-sentations are required for tasks that childrenmaster later in development.

We have focused on the strength of workingmemory representations as a source of variabil-ity for two primary reasons. First, this seems tobe a highly plausible candidate source of vari-ability, given that representations can vary instrength in numerous ways and result in variabil-ity in performance. Second, this frameworkmay provide a parsimonious, unified account of variability. Much of the variability observed

across development has been explained in some-what piecemeal ways. For example, infants suc-ceed on looking measures of memory beforereaching measures because of problem-solvingdeficits specific to reaching tasks. They thensucceed on reaching measures with a single-target location before reaching measures with




multiple-target locations, because of inhibitiondeficits specific to multiple-location reachingtasks. And so on. Although the variety of factorsposited across such accounts may be relevant to

developmental change, there may also be im-portant contributions from single factors (such asthe strength of working memory representations)thatchangegraduallywithdevelopment.Insomecases, consideration of these single factors mayobviate the need for the variety of factors positedto explain variability. These ideas have been in-stantiated in neural-network models, and result-ing behavioral predictions have been tested andconfirmed.

As reviewed below, these models implementthe idea that computational trade-offs demandspecialized brain regions, and demonstrate howgraded changes in active maintenance cansimulate variability across development andtasks. Moreover, these models demonstrate howinhibitory control can arise as a functional con-sequence of active maintenance. The modelsprovide a framework for understanding varia-tion in cognitive control and attention, across

ages and tasks. Specifically, they provide in-sight into why infants and children often repeatold behaviors that are no longer appropriate,despite apparent knowledge of the correct re-sponse, and how this changes with development.We focus on three instances of such variation: incard sorting, object search, and tasks with visiblesolutions.

Card Sorting As described earlier, in card-sorting tasks (Ze-lazo, Frye, & Rapus, 1996), children first sortcards one way (e.g., by color) and are then askedto switch to sort the same cards in a differentway (e.g., by shape). Although children cor-rectly answer simple questions about the newsorting rule (e.g., ‘‘Where do trucks go in theshape game?’’), most 3-year-olds perseverate, by

inappropriately sorting the cards in the old way.By age 5 years, children perform at ceiling bothsorting cards and answering questions. We usedneural-network models to explore how variationin active maintenance mechanisms might leadto this sort of age- and task-related variation(Morton & Munakata, 2002a). In the models,

sorting cards one way leads to a bias for partic-ular features. Switching to a new sorting rulecreates a conflict between previously and cur-rently relevant features. Strong, active mainte-

nance of the new rule is required for resolvingthis conflict, whereas weak maintenance suf-fices for non-conflict tasks such as answeringsimple questions about the new rule.

The network consisted of three input layers(visual features, rule, and verbal features), twohiddenlayers(internalrepresentation andPFC),and an output layer (Fig. 7.7; see also Cohen &Servan-Schreiber, 1992). The visual-featureslayer encoded the shape (truck or flower) and

color (red or blue) of the cards, the verbal-features layer encoded verbal statements

Visual Features Verbal Features Rule

R B T F R B T F C S

InternalRepR B

C S

PFC

T F

Output

"In the color game, red ones go here...

and blue ones go here."

"Here’s a red one."

"Where does it go?"

"We’re playing the color game."

B TR F

Figure 7.7. Simplified version of the card-sortingnetwork and the elements of a trial. R ¼ red; B¼

blue; T¼ truck; F¼flower; C¼ color; S¼ shape;PFC¼ prefrontal cortex. In the inputs with ‘‘gohere,’’ the corresponding output unit was activatedfor the network to indicate where the card should go.(Adapted from Morton & Munakata [2002], Figure1, p. 259. Copyright 2002 Wiley Periodicals, Inc.,used with permission)




measures are equated for conflict (Munakata & Yerys, 2001).

The network’s ability to ultimately over-come bias depended critically on how strongly

the PFC units maintained the new rule. Whenrecurrent connections were weak and the shaperule was only weakly maintained, PFC unitsoffered little support to the shape units in theinternal-representations layer. Consequently,the color units tended to win the competitivestruggle and the network sorted by color. By con-trast, when recurrent connections were strongand the shape rule was strongly maintained,PFC units offered considerable support to the

shape units in the internal-representation layer.Under these circumstances, the shape unitstended to win and the network correctly sortedby shape. These results suggest that the asso-ciation between age-related advances in cog-nitive control and development of the frontallobes may be due to advances in active main-tenance mechanisms.

In addition, these results speak to the issueof active maintenance and inhibition discussed

earlier. In the model, what might appear to bechanges in inhibitory abilities arises as a func-tional by-product of changes in active mainte-nance mechanisms. There is no inhibitorysystem per se in the model, and the model’sperformance improves without any changesto the inhibitory connections throughout themodel.

A-not-BThis framework is general enough to accountfor age-related changes in cognitive control andattention across various tasks and ages. For ex-ample, an earlier version of the model (Muna-kata, 1998; see also Dehaene & Changeux,1989) simulated age- and task-related variabilityin the A-not-B task, in terms of the strength of active maintenance abilities. In this task (Pia-

get, 1954), infants watch and successfully searchfor a toy hidden at one location (called the A location) for several trials. The toy is then hid-den at an alternate B location. After a brief de-lay, most infants search perseveratively for thetoy at the A location, a phenomenon referred toas the A-not-B error. However, infants show

some sensitivity to the correct location of the toyin their looking behaviors. They occasionallygaze at the correct hiding location while reach-ing to the previous location (Diamond, 1985;

Hofstadter & Reznick, 1996; Piaget, 1954), andthey show earlier sensitivity to the correct hidinglocation in gaze-only versions of the task (Hof-stadter & Reznick, 1996) and in violation-of-expectation versions (Ahmed & Ruffman, 1998).In the model of the A-not-B task (Munakata,1998), weak active maintenance abilities suf-ficed for success on A trials, because there was noconflict from preceding trials (just as in the firstrule trials in the card sorting task). Further, weak

active maintenance abilities sufficed for successon gaze and violation-of-expectation versions of the task, but not for the standard reaching task.This looking–reaching dissociation resultedfrom differences between the two systems in thefrequency of updating. The looking system wasallowed to update continually throughout thetrials (as infants are typically able to do in the

A-not-B task), whereas the reaching system wasonly allowed to update at the end of each trial (as

infants are allowed to do when the hiding ap-paratus is pushed to within their reach). Themore frequently updating looking-system wasable to make better use of weak representationsof the object’s hiding location than the less fre-quently updating reaching system, yielding thesame looking–reaching dissociation observed ininfants. Ultimately, with stronger active main-tenance abilities, the model succeeded on bothlooking and reaching variants of the A-not-B task.

Again, apparent improvements in inhibitoryabilities emerged from the development of ac-tive maintenance abilities.

Visible Solutions

Some researchers have argued that active main-tenance abilities cannot account for problems of control in development, because perseveration

occurs even in tasks in which solutions are fullyvisible. One example is a task in which infantsare presented with two towels, one with a dis-tant toy on it and the other with a toy behind it(Aguiar & Baillargeon, 2000). As in the A-not-Btask, infants initially pull the correct towel toretrieve the toy (i.e., the one with the toy on it).




Then the towels are switched, for example, sothat the towel with the toy on it was on theleft and is now on the right. As in A-not-B, in-fants perseverate, pulling the towel in the old

location even though it does not yield the toy.However, unlike A-not-B, the infant can seewhich towel will yield the toy. How can activemaintenance be involved when the solution isfully visible?

To investigate this question, we applied thesame neural-network approach used for cardsorting and A-not-B to the towel-pulling task(Stedron, Sahni, & Munakata, 2005). Specifi-cally, we investigated the effects of changes

in active maintenance when solutions arefully visible. We discovered that the same in-creases to active maintenance abilities that im-proved the model’s performance on card sortingand the A-not-B task similarly improved per-formance on the towel-pulling task, by increas-ing the network’s attention to the correct towel.

The network (Fig. 7.8) comprised two inputlayers encoding the location, identity, andplacement of the objects (toy on or attached),

internal-representation and PFC layers re-presented the location of the objects (left orright), and an output layer indicated the net-work’s response. As in the previous models, ex-citatory connections linked correspondingunits,and these connections changed with experi-ence according to a Hebbian learning rule.Inhibitory connections were present within theinternal representation, PFC, and output layers.PFC units were linked to themselves throughexcitatory recurrent connections, allowing themto actively maintain a representation of thecurrent (and visible) location of the towel sup-porting the toy. Again, these recurrent connec-tions were varied continuously in strength toinvestigate the effects of graded changes in ac-tive maintenance with age.

The simulated task consisted of four A trialsin which a toy and a towel were present in both

locations (with one toy on its towel and theother toy not on its towel), followed by a B trialin which toy placement was reversed. Thenetwork performed well on A trials, which ledto a bias for the towel location that supportedthe toy on A trials. Reversing the toy place-ment on B trials therefore created a competi-

tion between this bias and the current inputindicating that the toy was now placed on the op-posite towel. Whether the network overcame thisbias and responded correctly on B trials thereforedependedon input from the PFC and the strengthof the recurrent connections. When recurrence

was low and PFC only weakly represented in-formation about the new towel supporting thetoy, input to the internal-representations layerwas weak and the network succumbed to itsinitial bias. However, when recurrence was highand the PFC formed strong representations,input to the internal-representations layer was

Presentation

Left

Toy Tow On Att Toy Tow On Att

Right

PFC

InternalRep

Output

Choice

Delay

L R

L R

L R

Figure 7.8. Towel-pulling network and the ele-

ments of an A trial (Stedron, Sahni, & Munakata,2005). The input units encode information about theidentity of objects (toy and towel) and their place-ment (on and attached). A toy sitting behind a towelwould activate the toy and towel units only (as onthe right side of the trial shown). A toy sitting on atowel would activate the toy, towel, and on units (ason the left side of the trial shown). A toy that at-tached to its supporting towel would activate the toy,towel, on, and attached units; this condition is not

be discussed here.




comparatively high and the network respondedcorrectly.

Thus, this model demonstrated how a sim-ilar mechanism—active maintenance—could

subserve performance in both tasks with obvi-ous working memory demands and tasks with-out such obvious demands. In tasks with fullyvisible solutions, active maintenance supportsattention to relevant information in the envi-ronment. This perspective shares much incommon with that of Braver, Gray, and Burgess(Chapter 4) and Kane, Conway, Hambrick, andEngle (Chapter 2).

In sum, the card-sorting, A-not-B, and towel-

pulling models implement the core assumptionsof our working memory theory and explorepossible consequences of graded changes inactive maintenance abilities. Together, theyprovide a unified framework for understandingsources of age- and task-related variability incognitive control.

CONTRIBUTIONS TO GENERAL


We next address the question of how our ap-proach informs the study of working memory ingeneral, in terms of what we have learned fromvariation across cognitive development andfromthe PBWM modeling work outlined above.

We believe that the variability observedacross cognitive development is very informa-tive for understanding cognition more gener-ally. First, general cognitive processes likelycontribute to the variability observed both acrossdevelopment and in the mature system. Chil-dren may simply reveal these processes in moreobvious ways (e.g., throughtheir errors), whereasadults show them in more subtle ways (e.g., intheir reaction times, and in their errors underdemanding conditions). Second, understand-ing developmental trajectories may be crucial

for understanding functioning of the maturesystem. We elaborate these points and addresspotential skepticism about the relevance of de-velopment.

Finally, the PBWM-modeling work out-lined above provides a direct implementationof many aspects of our general working mem-

ory theory. We elaborate this model in the finalpart of this section.

Clearer Windows into

Underlying Processes

One example of developmental studies provid-ing a clearer window into underlying processesis in the types of cues children and adults use toreorient themselves after becoming disoriented.Children tend to reorient themselves accordingto geometric information about the layout of the environment (e.g., that a corner of a roomhas a long wall to the left of a short wall), while

failing to use featural information about theroom (e.g., that one wall in the room is blue)(Hermer & Spelke, 1994). Although adults cangenerally reorient themselves using bothgeometric and featural information, children’sfailures to use featural information do not re-flect cognitive processing unique to children.

Adults show similar patterns of behavior underdemanding conditions, when they must carryout an unrelated secondary task while trying to

reorient themselves (Hermer-Vazquez, Spelke,& Katsnelson, 1999). It appears that commoncognitive processes are at work in children’s andadults’ reorienting; in both cases, the use of geometric information is more robust than theuse of featural information. Thus, the changesthat lead to variability across development (aschildren progress from reorienting on the basisof geometric information alone to reorientingthrough featural information as well) are likelyto be relevant to understanding the parallelvariability that adults show in reorienting underdifferent conditions.

Another example arises in children’s andadults’ perseveration, or the repeating of priorbehaviors when they are no longer relevant.Children often perseverate in very obvious ways,repeating incorrect behaviors again and again.For example, after 3-year-olds sort cards ac-

cording to one rule (e.g., by their shape), theyperseverate with this rule even after they areasked repeatedly to switch to sorting the cardsby a new rule (e.g., by their color) (Zelazo et al.,1996). Moreover, they perseverate despite be-ing able to answer questions about the new rulecorrectly (e.g., about where blue cards should




go in the color game). Again, adults seem tohave no difficulty with such simple tasks, butthey show similar patterns of behavior in moresubtle ways. Specifically, adults are slowed in

their responses under the same conditions inwhichchildrenperseverate(e.g.,whentheymustswitch to a new rule; Diamond & Kirkham,2005; Morton & Munakata, 2004). And, adultsshow this slowing despite being able to respondquickly to questions about the new rule(Morton & Munakata, 2004). Thus, it againappears that the processes contributing tovariability across development (as children prog-ress from perseverating despite answering ques-

tions correctly to flexibly switching) are likelyrelevant to understanding the parallel variabilitythat adults show in their response times for tasksof flexibility. We will return to this example inthe final section of the chapter, to consider im-plications for working memory processes.

Developmental Trajectory

An understanding of developmental trajectories

could also inform debates about neural spe-cializations in the mature system. For example,there is considerable debate about neural spe-cializations for face processing (e.g., Haxbyet al., 2001; Kanwisher, 2000; Tarr & Gauthier,2000). If there are neural regions that are spe-cialized for processing faces, does this reflectan inherent specialization tuned for facesper se? Or, does this reflect a more general sys-tem that gets tuned to frequently attendedstimuli through learning? If children showedincreasing use of these brain regions with thedevelopment of expertise for faces (or otherkinds of stimuli), this might support the moredomain-general account. In contrast, if infantsshowed high use of these brain regions withearly exposures to faces, this might support themore domain-specific account.

We will discuss in the final section some

initial attempts to understand the developmentof representations in PFC that can supportsystematic behavior in an ‘‘adult’’-like model.In this case, the developmental process is criticalbecause the adult system appears to be capableof ‘‘magical’’ powers of generalization—we can

almost instantly perform novel tasks without thekinds of extensive training procedures requiredby monkeys (and standard neural networkmodels). We think these ‘‘magical’’ powers ac-

tually reflect the extended development of alarge vocabulary of basic cognitive skills, whichcan then be flexibly deployed in the adult torapidly solve novel tasks. Thus, studying the de-velopment of this vocabulary is critical for un-derstanding the functioning of the adult system.

In these ways, we believe that variabilityacross development provides an important win-dow onto general cognitive processes, whichcan lead to the discovery of parallel, more sub-

tle, indicators of these processes in adults and anunderstanding of how the mature system func-tions.

But Is Development Really Relevant?

Some investigators have expressed skepticismabout the relevance of developmental variabil-ity for understanding variability in the maturesystem. The reasoning goes as follows: as chil-

dren develop, they progress from not having askill to having that skill; variability across de-velopment thus arises from the addition of newskills. In contrast, adults have all of the variouscandidate skills of interest; variability acrossadults thus must arise from factors other thanthe addition of new skills. Variability across de-velopment and across adults must come fromdifferent sources.

We believe this argument is flawed for many,if not most, cases of developmental variability.Children rarely progress from simply not havinga skill to having that skill. Development is in-stead generally more graded and variable (e.g.,Munakata et al., 1997; Siegler, 1996; Thelen &Smith, 1994), with children progressing fromless automatic, robust, or frequently used skills toskills that are more advanced on these dimen-sions. For example, in the domain of arithmetic,

a rough glance might suggest that childrenprogress from simply not knowing how to addto knowing how to add (e.g., around 7 years of age). A more careful analysis, however, showsthat children across an age span of several yearspossess a similar repertoire of adding strategies




(e.g., counting on fingers, starting with the lar-ger addend and counting up by the smaller ad-dend, and retrieving the answer from memory).The developmental variation comes from dif-

ferences in the weighting of these various strat-egies, rather than the addition of new strategies.This is not to say that children (or adults, for thatmatter) have all of the relevant cognitive skills,and change is always a matter of gradual chan-ges in weighting. However, much of the varia-tion across development can be understood interms of such gradual changes.

As a result, in many cases like this it is inac-curatetoviewdevelopmentintermsofastaircase

model, with children cleanly progressing fromone developmental stage with a particular set of skills to the next stage with a new set of skills.

A more appropriate model may be one of overlapping waves, with children progressingthrough graded changes in a variety of skills(Siegler, 1996). The latter model highlights thepotential relevance of developmental changesto understanding individual differences in themature system. That is, graded differences in

skills in their automaticity, robustness, and fre-quency of use likely contribute to variabilityacross development and in the mature system.

Prefrontal–Basal GangliaWorking Memory Gating andInformation-Updating Models

As discussed earlier, we believe that modelingwork is essential for informing and advancingtheory. We have described our models of activemaintenance and how they may contribute tounderstanding variation in working memory.Here, we discuss how such active maintenancemechanisms may be modulated to support be-haviors in more complex situations.

Active maintenance in PFC can be modu-lated by adaptive gating mechanisms that candynamically switch between robust mainte-

nance and rapid updating. As summarized ear-lier, we have identified neural mechanisms inthe basal ganglia that are well suited for thisadaptive gating role (Frank et al., 2001;O’Reilly & Frank, 2006). Specifically, the Go(direct-pathway) neurons in the dorsal striatum

can disinhibit the PFC, allowing it to rapidlyupdate what it is maintaining. The indirect-pathway NoGo neurons compete with these Goneurons to prevent this updating, enabling ro-

bust maintenance of currently active PFC rep-resentations. These gating mechanisms raise anumber of important questions, including twothat we address in this section. The first is, whatdetermines when these Go and NoGo neuronsfire? Without a clear mechanistic explanation of this, the gating mechanism would amount to ahomunculus. The second question is, whatimplications does the presence of an adaptivegating mechanism have for the development of

PFC representations?

Learning to Gate in the Basal Ganglia

Our general answer to the first question re-garding the firing of the Go and NoGo gatingneurons in the basal ganglia is that powerfullearning mechanisms shape the firing of theseneurons in response to task demands. Specifi-cally, our PBWM model uses a Pavlovian-style

reinforcement learning mechanism called per-ceived value and learned value (PVLV) thatrepresents a synergy between biological mecha-nisms and computational demands (O’Reilly &Frank, 2006). Biologically, such a mechanismis supported by the dopaminergic systems of theventral basal ganglia in much the same way asthe closely related temporal-differences learningmechanism (Contreras-Vidal & Schultz, 1999;Houk et al., 1995; Joel et al., 2002; Schultz et al.,1997; Suri & Schultz, 2001). Computationally,PVLV solves the temporal credit assignmentproblem, which is the critical problem in train-ing an adaptive gating system.

The temporal credit assignment problemarises when the consequences of an action aredelayed in time from the point when the actionneeds to be taken. For example, consider asimple working memory task like the spatial

working memory task, where a spatial locationmust be encoded, maintained over a delay, andthen a response to that location must be made.From the gating mechanism’s perspective, thistask requires a Go signal at the time of thestimulus to update working memory. However,




H i d

d e n ( 8

3 u

n i t s )

T a s k

H i d

d e n

( 1 6

u n i t s )

Left Stimulus Right Stimulus Task

NF MF SF LF A B C D E

A1 A2 A3 A4

B1 B2 B3 B4

C1 C2 C3 C4

D1 D2 D3 D4

E1 E2 E3 E4

A1 A2 A3 A4

B1 B2 B3 B4

C1 C2 C3 C4

D1 D2 D3 D4

E1 E2 E3 E4

A1 A2 A3 A4

B1 B2 B3 B4

C1 C2 C3 C4

D1 D2 D3 D4

E1 E2 E3 E4

No

AG

Dimension Cue

R e s p o n s e

C u e

H i d

d e n

( 1 6

u n i t s )

P F C

C o n t e

x t

( 3 0

u n i t s )

A

P o s t e

r i o r

P + R e

c

P + S e

l f S R

N

S R N P

F C N o

G a t e

F u l l

No PFC PFC

25

50

75

G e n e r a l i z a t i o n

% C

o r r e c t

Task PairsAll Tasks

CrossTask Generalization

Training Regimen

Network ConfigurationB

Figure 7.9. A: The full prefrontal cortex (PFC) model of Rougier et al. (2005, used with permission)Stimuli are presented in two possible locations (left, right). Rows represent different dimensions,labeled A–E, and columns represent different features (1–4). Other inputs include a task inputindicating current task to perform (NF, MF, SF, LF), and, optionally, a verbal cue as to the currentlyrelevant dimension (explicit cue conditions only). Output responses are generated over the verbal-response layer. The AC unit is the adaptive critic, providing a temporal differences–based dynamic

178



the benefits of having correctly activated this Gosignal do not come until after the delayed re-sponse based on the encoded stimulus. There-fore, the learning mechanism must somehow

learn from these subsequent rewards to performa Go gating action earlier in time. This is achallengingcomputationalproblem.ThePVLV mechanism solves it by perceiving rewardvalue (PV) associated with stimuli (maintainedin PFC) that were previously associated withlearned reward values (LV). So, by trial-and-error, the system maintains stimuli in PFC, andif these are associated with reward, then whenthese stimuli later appear again, the resulting

PV signal reinforces Go firing to these stimuli.Biologically, the firing of dopamine neurons inthe ventral tegmental area (VTA) and substantianigra pars compacta (SNc) (heavily inner-vated by the basal ganglia) reflect the firingpatterns of the PVLV model (O’Reilly & Frank,2006; Schultz et al., 1997). These dopamineneurons then modulate learning in the striatum,producing appropriate patterns of reinforce-ment for Go and NoGo firing in our PBWM

model.One of the most important features of the

PBWM model relative to earlier gating modelsis that it can provide selective gating signals todifferent regions of PFC, so that some PFCworking memory representations can be main-tained while others are updated. We refer tothese separately updatable PFC regions (andassociated parallel cite-loops through the basalganglia) as stripes, in reference to the anatom-ically isolated patterns of connectivity charac-terized by Levitt et al. (1993). By virtue of havingthese separate stripes, the learning mechanism

must also solve a structural credit assignmentproblem in determining which stripes are re-sponsible for maintaining different separableitems of task-relevant information. This addi-

tional mechanism is based on inhibitory pro-jections from the substantia nigra pars reticulata(SNr) to the dopamine neurons in the SNc,which produces a shunting inhibition thatmodulates dopamine firing in a stripe-specificmanner. The resulting stripe-specific dopaminesignals provide selective reinforcement foronly those stripes that are responsible for thecurrent perceived value (PV) reinforcementsignals.

Development of Rule-like Prefrontal Cortex Representations

The presence of an adaptive gating mechanismcan impose important constraints on the typesof representations that form in the PFC system,which in turn can impact the overall behaviorof the system in important ways. In particular,we have recently shown that a network having

an adaptive gating mechanism developed ab-stract, rule-like representations in its simulatedPFC, whereas models lacking this mechanismdid not (Rougier et al., 2005). Furthermore, thepresence of these rule-like representations re-sulted in greater flexibility of cognitive control,as measured by the ability to generalize knowl-edge learned in one task context to other tasks.These results may have important implicationsfor understanding how PFC can contribute totasks in ways that are not obviously related toworking memory function (e.g., by supportingmore regular, rule-like behavior).

gating signal to the PFC context layer. To evaluate the features of this architecture, many variantswere tested, from a single 145 hidden-unit layer between inputs and verbal response (with andwithout recurrent connectivity) to a simple recurrent network (SRN), with a context layer that is acopy of the hidden layer on the prior step. B: Cross-task generalization results (% correct on test set)

for the full PFC network and a variety of control networks, with either only two task (Task Pairs) or allfour tasks (All Tasks) used during training. Overall, the full PFC model generalizes substantiallybetter than the other models, and this interacts with the level of training such that performance on theall-tasks condition is substantially better than the task-pairs condition. With one feature left out of training for each of four dimensions, training represented only 31.6% (324) of the total possiblestimulus inputs (1024); The roughly 85% generalization performance on the remaining test itemstherefore represents good productive abilities.




Rougier et al. (2005) trained a range of dif-ferent models on a varying number of relatedtasks operating on simple visual stimuli (e.g.,name a ‘‘feature’’ of the stimulus along a given

‘‘dimension’’ such as its color, shape, or size;match twostimulialongoneofthesedimensions;compare the relative size of two stimuli). Thegeneralization test for the cognitive flexibility of the models involved training a given task on asmall percentage (e.g., 30%) of all the stimuli,and then testing that task on stimuli that weretrained in other tasks. To explore the impact of the adaptive gating mechanism and other ar-chitectural features, a range of models having

varying numbers of these features were tested. As shown in Figure 7.9, the model with the

full set of prefrontal working memory mecha-nisms achieved significantly higher levels of gen-eralization than those of otherwise comparablemodels that lacked these specialized mecha-nisms. Furthermore, this benefit of the prefron-tal mechanisms interacted with the breadth of experience the network had across a range of different tasks. The network trained on all four

tasks generalized significantly better than onetrained on only pairs of tasks, but this was onlytrue for the full PFC model. Thus, the modelexhibited an interesting interaction betweennature (the specialized prefrontal mechanisms)and nurture (the breadth of experience): bothwere required to achieve high levels of gener-alization. We consider the protracted period of development of the PFC (up through late ad-olescence; Casey et al., 2001; Diamond &Goldman-Rakic, 1986; Huttenlocher, 1990;Lewis, 1997; Morton & Munakata, 2002b) asthe time frame over which prefrontal repre-sentations are shaped, and the huge breadth of experience during that time then leads to whatsystematic reasoning abilities we have as adults.

The main reason why the prefrontal mech-anisms led to such good generalization in oursimple task domain is that they enabled the

network to develop discrete, abstract, rule-likerepresentations of the stimulus dimensions (Fig.7.10). Specifically, the network was trained suchthat a given stimulus dimension was relevantacross a series of individual training trials. Fur-thermore, in some cases, the network had to‘‘guess’’ what this relevant dimension was on the

basis of trial-and-error feedback, and maintain itover a sequence of trials in which the dimensionwas the same. Critically, the robust activation-based working memory functions of the prefron-

tal model enabled the network to maintain thesame representation over time, and thereforethese representations learned to abstract the di-mensional information that was common acrosstrials, while filtering out the irrelevant infor-mation that varied across these trials.

Other comparison networks that could main-tain information over time, but lacked a dy-namically gated working memory system (e.g.,a simple recurrent network or SRN; Elman,

1990), ended up using a variable set of repre-sentations over a given dimension and thus didnot develop the appropriate abstractions. Wethink this pattern of results reflects a generalrationale for the PFC developing more ab-stract representations than posterior cortex, andthus facilitating flexible generalization to novelenvironments: abstraction derives from themaintenance of stable representations over time,interacting with learning mechanisms that ex-

tract commonalities over varying inputs. Sup-porting this view are data showing that damageto PFC impairs abstraction abilities (e.g.,Dominey & Georgieff, 1997) and that PFC inmonkeys develops more abstract category repre-sentations than those in posterior cortex (Freed-man, Riesenhuber, Poggio, & Miller, 2002;Nieder, Freedman, & Miller, 2002; Wallis,

Anderson, & Miller, 2001).Importantly, the rule-like PFC representa-

tions that developed in the model differ fromthe symbolic representations typically assumedby traditional symbolic models of higher-levelcognition (e.g., Anderson & Lebiere, 1998;Newell, 1990). Unlike symbolic models, theserule-like PFC representations do not supportarbitrary symbol-binding operations, and are in-stead much more like standard neural-networkrepresentations, in that they obtain their mean-

ing through gradually adapting synaptic connec-tions with other representations in the system.Thus, they have considerably less intrinsicflexibility relative to arbitrary symbols. This isevident in that our model predicts moderatebut far from perfect levels of generalization ortransfer to new tasks (e.g., the 85% transfer




A

B

Figure 7.10. Representations that developed in four different network configurations tested byRougier et al. (2005). A: No prefrontal cortex (PFC) (posterior cortex) trained on all tasks. B: PFCwithout the adaptive gating mechanism (all tasks). C: Full PFC (all tasks). D: Full PFC trainedonly on task pairs (NF & MF in this case). Each panel shows the weights from the hidden units (A)or PFC (B–D) to the verbal-response layer. Larger squares correspond to units (all 30 in the PFC,and a random and representative subset of 30 from the 145 hidden units in the posterior model),smaller squares designate strength of the connection (lighter¼ stronger) from that unit to eachof the units in the verbal-response layer. Note that each row designates connections to verbal-

response units representing features in the same stimulus dimension (see Fig. 7.1). It is evident,therefore, that each of the PFC units in the full model (D) represents a single dimension and,conversely, that each dimension is represented by a distinct subset of PFC units. This pattern is lessevident in the model lacking an adaptive gating mechanism (B) and in the PFC model trainedonly on task pairs (D), and is almost entirely absent in the posterior model (A) in which the hiddenunits appear to encode arbitrary combinations of features across dimensions. Panel (E) shows thecorrelation of generalization performance in these cases with the extent to which the units dis-tinctly and orthogonally encode stimulus dimensions in a rule-like manner.



shown in Fig. 7.9B), which is consistent withthe empirical data (e.g., Gick & Holyoak,1987). In contrast, purely symbolic modelsincorrectly predict perfect generalization per-

formance, unless otherwise artificially handi-capped. Nevertheless, the rule-like representa-tions in our model can support flexible cognitivecontrol through a combination of other prop-erties. One is that the PFC representations aremore abstract and discrete in nature than rep-resentations that develop in posterior cortical

areas, which results in enhanced generalizationin novel environments or task contexts. An-other is that a large ‘‘vocabulary’’ of suchrepresentations can develop over sufficiently

broad experience, which then enables noveltasks to be performed by flexibly combiningexisting representations (generativity). Finally,specialized neural mechanisms support bothrapid updating and robust maintenance of PFCrepresentations, as well as the ability of theserepresentations to bias or influence processing

C

D

Figure 7.10. (continued)




in other cortical areas (Cohen, Dunbar, &McClelland, 1990).

For example, to support generalization in anovel environment, an appropriate abstract rep-resentation for the task at hand must be rapidlyactivated and then maintained in the face of novel distractions. This abstract representationthen imposes rule-like, top-down constraints onprocessing in other cortical areas, resulting inmore systematic, regular behavior. When de-mands change, the system must be able torapidly update to a new, more appropriate rep-resentation. To support generativity, appropri-ate novel combinations of existing representa-tions must be rapidly activated and maintainedor updated as necessary. In short, we argue thatflexible cognitive control emerges from mech-

anisms that control the dynamics of activation of existing representations, in contrast to traditionalsymbolic approaches where flexibility derivesfrom arbitrary symbol binding, or other ap-proaches suggesting that flexibility depends onrapidly learning entirely new representations(Duncan, 2001).

The extent to which development of theseabstract, rule-like PFC representations supportmore flexible cognitive control could be a veryimportant source of individual variability inworking memory function. In the extreme case,we argue that the vastly greater levels of cog-nitive flexibility exhibited by people relativeto that of other primate species may derivefrom the greatly extended time window overwhich human PFC representations develop(Casey et al., 2001; Diamond & Goldman-Rakic, 1986; Huttenlocher, 1990; Lewis, 1997;Morton & Munakata, 2002b), coupled, of course, with the expanded size of human PFC.In our model, the breadth of experience was acritical factor in shaping these PFC represen-tations. Thus, we would predict that individual

variability in exposure to a wide range of cog-nitive task demands would play an importantrole in determining subsequent cognitive flexi-bility. Increased levels of cognitive flexibilitycould potentially impact a wide range of work-ing memory tasks. For example, an individual’sability to more flexibly and efficiently perform a

FullPFCPair

0.25 0.50 0.75Rule Representation Measure

25

50

75

G e n e r a l i z a t i o n % C

o r r e c t

PosteriorAll

NoGateAll

FullPFCAll

RuleLikeness Predicts Generalization

r=.97

E

Figure 7.10. (continued)




novel task (few working memory tasks are rou-tinely performed in everyday life) may result in ahigher effective level of working memory span.

Also, the correlations between working memory

span and more direct measures of cognitiveflexibility (e.g., Raven’s progressive matrices)could be explained in terms of differences inbasic working memory span capacity producingcorresponding differences in the formation of abstract, rule-like PFC representations, whichare tapped by these tasks.

CONSIDERATION OF OTHER

SOURCES OF WORKINGMEMORY VARIATION

There are likely many similarities between ourtheory of working memory variation and othersources proposed in this volume, as well as pos-sible points of contrast. We view our frameworkas almost identical to that of Braver et al. (withwhom we collaborate; see Chapter 4), and sharemuch in common with those of Kane, Conway,

Hambrick, and Engle (Chapter 2) and Fried-man and Miyake (2004). For example, Braveret al. and Kane et al. emphasize the role of thePFC in providing a controlled attentional sys-tem for activating and supporting task-relevantinformation.Thisisalsocentraltoourapproach,as discussed in the context of models of variationin strength of working memory representations.We go further, however, in that we also em-phasize the dynamic interactions between PFCand hippocampus to support performance oncomplex span tasks (see O’Reilly, Braver, et al.,1999, for elaboration).

In general, we believe that a crucial step inidentifying similarities and differences (be-tween our theories and others, and among othertheories) will be to clarify the precise meaningof central theoretical constructs such as ‘‘work-ing memory,’’ ‘‘activation,’’ and ‘‘inhibition,’’

and to more directly map them to underlyingmechanisms. In what follows, we attempt torelate some of these constructs to the centralmechanisms in our framework. We also dis-cuss how domain-general and domain-specificindividual differences can arise from these un-derlying mechanisms. Finally, we explore the

roles of the PFC and posterior sensory areas inworking memory within our framework.

Mapping Constructs

and Mechanisms

To explore the issue of mapping theoreticalconstructs onto underlying neural mechanisms,we start with Chapter 9 (by Hasher, Lustig, andZacks) in this volume, which contrasts contri-butions from ‘‘activation’’ processes with thosefrom ‘‘inhibition’’ processes to working memorytasks. The authors find that priming or retrievalof semantic associates, tasks considered to tap

activation processes, is unrelated to individualdifferences in working memory (e.g., acrossaging). In contrast, prevention of interferencefrom stimuli presented on prior trials in order torespond correctly on the present trial, consid-ered an inhibitory process, does correlate withworking memory variation. They conclude thatinhibition processes arecentralto working mem-ory, and that variation in activation does notaccount for variation in working memory. As a

result, inhibition might not reflect the flip sideof activation, as others have argued (Cohen &Servan-Schreiber, 1992; Goldman-Rakic, 1987;Kimberg & Farah, 1993; Miller & Cohen,2001; Munakata, 1998; O’Reilly, Braver, et al.,1999; Roberts, Hager, & Heron, 1994). Thestudies by Hasher and colleagues are veryelegant, and we believe they point to importantaspects of working memory function and var-iability. However, we believe that their datasupport somewhat different conclusions aboutworking memory, given a consideration of themechanisms that might underlie their con-structs of activation and inhibition.

Specifically, their measures of activation(e.g., semantic retrieval and priming) do nottap sustained, prefrontally mediated activationof the sort that is crucial for working memory.Instead, the‘‘activation’’ tasks likely taptherichly

interconnected semantic networks of posteriorcortex. In support of this interpretation, fron-tal lesions do not affect fluency tasks that requiregeneration of instances of existing semantic ca-tegories (e.g., ‘‘name as many animals as youcan in 1 minute’’), but they do affect more ar-bitrary fluency tasks (e.g., the FAS task: ‘‘name




plan, Waters, and DeDe (Chapter 11) show anextreme case of domain specificity for syntaxprocessing, while other researchers have shownvarying levels of generality (e.g., accounting for

performance on the Raven’s progressive matri-ces or the antisaccade task; see for exampleChapters 2 and 3). As we elaborate below, theneural network models central to our frame-work can exhibit both domain-specific anddomain-general effects. Further, they may pro-vide important insights into what kinds of mechanisms lead to these effects. For example,we describe below how a domain-specific effectcan sometimes arise from a very general neural

parameter, and vice-versa. Therefore, we reit-erate our view that empirical findings of do-main specificity or generality cannot be trans-parently mapped onto underlying neuralmechanisms and that consideration of thesemechanisms can lead to different interpreta-tions of these kinds of results.

In general terms, neural-network modelscan explain domain specificity because allknowledge and processing in a network takes

place across dedicated sets of neurons andsynaptic connections. Thus, different domainsof knowledge will tap different sets of neuronsand synapses. The efficacy of one set of syn-apses depends on the complex history of ex-perience that has entrained these connections,thus introducing a strong element of cross-taskvariability. These general properties of networkshave been strongly confirmed in the strikingcontext- and task-specificity of human perfor-mance and, consequently, moderate levels of generalization or transfer (Glick & Holyoak,1987). Furthermore, the observed correlationsbetween working memory and other cognitivefunctions rarely exceed .5.

Against this backdrop of domain specificityare a number of factors that can lead to someamount of shared variability across tasks. Obvi-ously, the extent to which tasks share content

and processing demands will determine theirability to tap the same sets of connections, andthis can produce the observed shared variabilitywithin content domains, such as within verbalprocessing and spatial processing. Variability inglobal neural parameters can also give rise toshared variability across a range of tasks. For

example, variability in the production or releaseof a global neuromodulator like dopamine,which broadly impacts neocortical processingand learning, could clearly lead to correspond-

ing variability across a range of tasks.Nevertheless, variability in global parame-ters such as dopamine need not affect all tasksequally. An interesting example comes frompeople with phenylketonuria (PKU), who havea disturbance in the production of dopamine(Diamond, 2002). Specifically, they cannot con-vert phenylalanine to tyrosine, the precursor todopamine. Treatment takes the form of re-stricting dietary intake of phenylalanine, which

allows ingested tyrosine more opportunity tocompete with phenylalanine for transport intothe brain. Without any dietary remediationsevere mental retardation results, with impair-ments in performance across a broad range of cognitive tasks, presumably due to the broadeffects of dopamine across the cortex. Withmoderate dietary remediation (which leads toblood phenylalanine levels three to five timesthe normal level), performance on tasks that do

not depend on frontal cortex (e.g., spatial dis-crimination, visual paired comparison, linebisection) improves substantially (to normallevels),butfrontal-likedeficitsremain.Diamond(2002) interpreted this pattern as reflecting thehigher firing rate and higher rate of dopamineturnoverindopamineneuronsprojectingtoPFCthan in other dopamine neurons. As a result,residual deficits in producing dopamine withmoderate dietary remediation affected dopa-mine levels more in PFC than in other areas.Thus, variability in the global parameter of do-pamine production caused more variability insome tasks (those dependent on frontal cortex)than in others. More extreme dietary restrictionsare required to avert such deficits in fron-tal function (Diamond, Prevor, Callender, &Druin, 1997).

Another study in which global dopamine

efficacy was manipulated via dopamine receptoragonists also revealed an interesting and com-plex pattern of effects on frontal task perfor-mance (e.g., Kimberg, D’Esposito, & Farah,1997; Kimberg & D’Esposito, 2003). The au-thors found that the effects of dopaminergicmanipulations interacted with the working




sient maintenance in posterior areas is drivenby more robust maintenance in PFC. In theface of interfering inputs, maintenance wasuniquely supported by PFC neurons, not by

posterior areas. Thus, when precise and robustrepresentations are needed, as for working mem-ory, the unique specializations of the PFC–basal ganglia system may become essential.

Summary

The main theme that we have emphasized hereis that the grounding of constructs in neuralmechanisms can provide important constraints

on interpretations of behavioral findings of indi-vidual variability. What looks like variability ininhibitory function may actually reflect variabil-ity in active maintenance abilities. What lookslike domain-specific behavior may actually re-flect differential sensitivity to a global neuralparameter. What looks like an undifferentiatedcluster of brain areas supporting working mem-ory may actually reflect the interactions amongmore clearly specialized neural systems.

CONCLUSIONS

So, how do we manage to find our keys as wesearch around the house for them? We believethat the developmental and computational ap-proaches described in this chapter provideinsights into two processes—active maintenanceand information updating—that are central toworking memory abilities for such tasks.

First, we must keep in mind the goal of searching for the keys as we wander around thehouse. This kind of active maintenance requiresspecialized mechanisms, becausethere is a com-putational trade-off between interactive repre-sentations (that can support semantic networksin posterior cortex) and isolated representations(that can support active maintenance abilities in

prefrontal cortex). Further, we need to be able torevise our subgoals on the basis of where we havealready searched for the keys and where we haveyet to search. This kind of updating requires agating mechanism that may be implementedthrough the specialized neural circuitry of thebasal ganglia. Understanding these computa-

tional demands can help make sense of spe-cializedbrainregionsbeingrequiredfordifferentfunctions.Further, individual variation canthenbe understood in terms of computational differ-

ences in the abilities of these brain regions, suchas in the strength of active representations inPFC, or in the Go–NoGo dynamics of the basalganglia.

As discussed throughout the chapter, we be-lieve that developmental and computationalapproaches have been very important for un-derstanding such working memory processes.The developmental work on perseveration pro-vides a window into similar processes in adults.

The computational work has been critical forexploring complex interactions (e.g., betweenthe PFC and basal ganglia during development)and for demonstrating how basic mechanismscould subserve behaviors in nonintuitive ways(e.g., with variation in strength of working mem-ory representations alone leading to knowledge–action dissociations, and perseveration in the faceof visible solutions).

There are many other aspects of working

memory that we have not considered here (e.g.,Baddeley, 1986) that will be important to in-corporate into complete accounts. We believethat it will be essential to map theoretical con-structs onto underlying mechanisms in this pro-cess. Investigating the computational bases of different aspects of working memory shouldhelp to ground such theoretical constructs,which should in turn inform our understand-ing of how and why individuals vary in theirworking memory abilities.

Note

1. A more complete exploration of this idea would

implement greater interactivity among related

non-PFC units, but this has not yet been inves-

tigated.

Acknowledgments

The writing of this chapter was supported by NICHD

Grant 1R29 HD37163 and ONR grant N00014-03-1-

0428. We thank members of the Cognitive

Development Center and Computational Cognitive

Neuroscience Lab at the University of Colorado,




Boulder, and members of the Cognitive Develop-

ment Centre at the University of Western Ontario foruseful comments and discussion.

References

Aguiar, A., & Baillargeon, R. (2000). Perseveration

and problem solving in infancy. In H. Reese

(Ed.), Advances in child development and be-

havior (Vol. 27, pp. 135–180). New York: Aca-demic Press.

Ahmed, A., & Ruffman, T. (1998). Why do infants

make A-not-B errors in a search task, yet show

memory for the location of hidden objects in a

non-search task? Developmental Psychology, 34,

441–453.



OF WORKING MEMORY

Our research focuses on the computational

mechanisms underlying two components of

working memory: active maintenance of infor-

mation (e.g., maintaining driving directions

such as ‘‘Turn left at the light’’) and updating of

information (e.g., after turning left at the light,

replacing those directions with the next step).

We investigate the types of representations and

learning mechanisms that allow different brain

systems (particularly prefrontal cortex and basal

ganglia) to specialize in these components of

working memory. We believe that an under-

standing of such underlying mechanisms is es-

sential for informing and advancing theorizing

about working memory.

2. CRITICALSOURCESOFVARIATION

IN WORKING MEMORY

Possible sources of working memory variability

include (1) differences in the strength of actively

maintained representations (e.g., with repre-

sentations becoming more robust across delays

and in the face of interference with develop-

ment), and (2) differences in the tendency to

update or maintain information (e.g., with such

differences contributing to variations observed

in diseases and disorders such as Parkinson’s

disease and ADHD, as well as within the typicalpopulation). We focus on these types of sources

ofvariabilitybecausetheycanberelatedtocom-

putational mechanisms contributing to work-

ing memory and seem to be plausible candidate

sources of variability given biological, behav-

ioral, and computational considerations.


MEMORY VARIATION

We believe there are many other potential

sources of working memory variability, such as

those proposed in this volume. There are likely

many similarities between our approach and

others’, as well as points of contrast. For ex-

ample, we share with Braver et al. (Chapter 4)

and Kane et al. (Chapter 2) the view of working

memory representations providing a controlled

attentional system for activating task-relevant

information. We differ from Hasher et al. (Chap-

ter 9) in how we map theoretical constructs

onto underlying mechanisms; we believe that

a consideration of underlying mechanisms

leads to very different conclusions about con-

structs such as activation and inhibition.



Variability observed across development pro-

vides an important window into the basic

components of working memory. For example,

the same working memory tasks that children

robustly fail give adults difficulty in more

subtle ways (e.g., as evidenced in their reac-

tion times). This continuity across the life span

suggests that theories of working memory that

can explain variability across children of dif-

ferent ages may also explain working memoryin the mature system. For instance, a more

robust ability to actively maintain information

may explain developmental differences as well

as individual differences across people of the

same age, and serve as a critical component of

working memory in general.




Alexander, G. E., DeLong, M. R., & Strick, P. L.

(1986). Parallel organization of functionally

segregated circuits linking basal ganglia and cor-

tex. Annual Review of Neuroscience, 9, 357–

381. Anderson, J. R., & Lebiere, C. (1998). The atomic

components of thought. Mahwah, NJ: Lawrence

Erlbaum Associates.



Bell,M.A.,&Fox,N.A.(1992).Therelationsbetween

frontal brain electrical activity and cognitive

developmentduringinfancy. Child Development,

63, 1142–1163.

Benton, A. L. (1968). Differential effects of frontallobe disease. Neuropsychologia, 6, 53–60.

Braver, T. S., Cohen, J. D., Nystrom, L. E., Jonides, J.,

Smith, E. E., & Noll, D. C. (1997). A parametric

study of frontal cortex involvement in human

working memory. NeuroImage, 5, 49–62.

Brown, R. G., & Marsden, C. D. (1990). Cognitive

function in Parkinson’s disease: From description

to theory. Trends in Neurosciences, 13, 21–29.

Bunting, M. F. (2006). Proactive interference and

item similarity in working memory. Jour-nal of Experimental Psychology: Learning,

Memory and Cognition, 32, 183–196.

Butler, R. W., Rorsman, I., Hill, J. M., & Tuma, R.

(1993). The effects of frontal brain impairment

on fluency: Simple and complex paradigms.

Neuropsychology, 7, 519–529.

Casey, B. J., Durston, S., & Fossella, J. A. (2001).

Evidence for a mechanistic model of cogni-

tive control. Clinical Neuroscience Research, 1,

267–282.

Cohen, J. D., Braver, T. S., & O’Reilly, R. C.

(1996). A computational approach to prefrontal

cortex, cognitive control, and schizophrenia:

Recent developments and current challenges.

Philosophical Transactions of the Royal Society

(London) B, 351, 1515–1527.

Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990).

On the control of automatic processes: A parallel

distributed processing model of the Stroop effect.Psychological Review, 97, 332–361.

Cohen, J. D., & Servan-Schreiber, D. (1992).

Context, cortex, and dopamine: A connection-

ist approach to behavior and biology in schizo-

phrenia. Psychological Review, 99, 45–77.

Contreras-Vidal, J. L., & Schultz, W. (1999). A

predictive reinforcement model of dopamine

neurons for learning approach behavior. Jour-

nal of Computational Neuroscience, 6, 191–

214.Dehaene, S., & Changeux, J. P. (1989). A simple

model of prefrontal cortex function in delayed-

response tasks. Journal of Cognitive Neurosci-

ence, 1, 244–261.

Diamond, A. (1985). Development of the ability

to use recall to guide action, as indicated by

infants’ performance on A BB. Child Develop-

ment, 56, 868–883.

Diamond, A. (2002). A model system for studying

the role of dopamine in prefrontal cortexduring early development in humans. In M. H.

Johnson, Y. Munakata, & R. O. Gilmore (Eds.),

Brain development and cognition: A reader (pp.

441–493). Oxford: Blackwell.

Diamond, A., & Goldman-Rakic, P. S. (1986).

Comparative development in human infants

and infant rhesus monkeys of cognitive func-

tions that depend on prefrontal cortex. Society

for Neuroscience Abstracts, 12, 742.

Diamond, A., & Kirkham, N. (2005). Not quite asgrown-up as we like to think: Parallels between

cognition in childhood and adulthood. Psycho-


Diamond, A., Prevor, M., Callender, G., &

Druin, D. (1997). Prefrontal cortex cognitive

deficits in children treated early and continu-

ously for PKU. Monographs of the Society for

Research in Child Development, 62, Mono-

graph 252.

Dominey, P. F., & Georgieff, N. (1997). Schizophren-

ics learn surface but not abstract structure in a

serial reaction time task. Neuroreport, 8, 2877.

Duncan, J. (2001). An adaptive coding model of neural

function in prefrontal cortex. Nature Reviews


Durstewitz, D., Kelc, M., & Gunturkun, O. (1999).

A neurocomputational theory of the dopami-

nergic modulation of working memory func-

tions. Journal of Neuroscience, 19, 2807.Durstewitz, D., Seamans, J. K., & Sejnowski, T. J.

(2000). Dopamine-mediated stabilization of

delay-period activity in a network model of pre-

frontal cortex. Journal of Neurophysiology, 83,

1733–1750.




Elman, J. L. (1990). Finding structure in time.

Cognitive Science, 14, 179–211.

Elman, J. L., Bates, E. A., Johnson, M. H., Karmiloff-

Smith, A., Parisi, D., & Plunkett, K. (1996).

Rethinking innateness: A connectionist perspec-tive on development. Cambridge, MA: MIT Press.

Farah,M.J.,Monheit,M.A.,&Wallace,M.A.(1991).

Unconscious perception of ‘‘extinguished’’ visual

stimuli: Reassessing the evidence. Neuropsycho-

logia, 29, 949–958.

Farah, M. J., O’Reilly, R. C., & Vecera, S. P. (1993).

Dissociated overt and covert recognition as an

emergent property of a lesioned neural network.


Fellous, J. M., Wang, X. J., & Lisman, J. E. (1998). A role for NMDA-receptor channels in working

memory. Nature Neuroscience, 1, 273–275.

Frank, M. J., Loughry, B., & O’Reilly, R. C. (2001).

Interactions between the frontal cortex and

basal ganglia in working memory: A computa-

tional model. Cognitive, Affective, & Behavioral


Freedman, D. J., Riesenhuber, M., Poggio, T., &

Miller, E. K. (2002). Visual categorization and

the primate prefrontal cortex: Neurophysiologyand behavior. Journal of Neurophysiology, 88,

929–941.

Friedman, N. P., & Miyake, A. (2004). The relations

among inhibition and interference control func-

tions: A latent variable analysis. Journal of Exper-


Fuster, J. M. (1989). The prefrontal cortex: Anatomy,

physiology and neuropsychology of the frontal

lobe. New York: Raven Press.

Gick, M. L., & Holyoak, K. J. (1987). The cognitive

basis of knowledge transfer. In S. M. Cormier & J.

D. Hagman (Eds.), Transfer of learning: Contem-

porary research and applications (pp. 9–46).

Orlando, FL: Academic Press.

Goldman, P. S., & Rosvold, H. E. (1972). Effects of

selective caudate lesions in infant and juvenile

rhesus monkeys. Brain Research, 43, 53.

Goldman-Rakic, P. S. (1987). Circuitry of primate

prefrontal cortex and regulation of behavior byrepresentational memory. In F. Plum & V.

Mountcastle (Eds.), Handbook of physiology:

Vol. 5, The nervous system (pp. 373–417). New

York: Oxford University Press.

Haxby, J. V., Gobbini, M. I., Furey, M. L., Ishai, A.,

Schouten, J. L., & Pietrini, P. (2001). Distrib-

uted and overlapping representations of faces

and objects in ventral temporal cortex. Science,

293, 2425–2429.

Hermer, L., & Spelke, E. S. (1994). A geometric

process for spatial reorientation in young chil-dren. Nature, 370, 57–59.

Hermer-Vazquez, L., Spelke, E. S., & Katsnelson,

A. S. (1999). Sources of flexibility in human

cognition: Dual-task studies of space and lan-

guage. Cognitive Psychology, 39, 3.

Hochreiter, S., & Schmidhuber, J. (1997). Long

short-term memory. Neural Computation, 9,

1735–1780.

Hofstadter, M. C., & Reznick, J. S. (1996). Response

modality affects human infant delayed-responseperformance. Child Development, 67, 646–658.

Houk, J. C., Adams, J. L., & Barto, A. G. (1995). A

model of how the basal ganglia generate and

use neural signals that predict reinforcement.

In J. C. Houk, J. L. Davis, & D. G. Beiser

(Eds.), Models of information processing in the

basal ganglia (pp. 233–248). Cambridge, MA:

MIT Press.

Huttenlocher, P. R. (1990). Morphometric study of

human cerebral cortex development. Neuro-psychologia, 28, 517–527.

Joanisse, M. F., & Seidenberg, M. S. (2003). Phonol-

ogy and syntax in specific language impairment:

Evidence from a connectionist model. Brain and

Language, 86, 40–56.

Joel, D., Niv, Y., & Ruppin, E. (2002). Actor-critic

models of the basal ganglia: New anatomical

and computational perspectives. Neural Net-

works, 15, 535–547.

Kanwisher, N. (2000). Domain specificity in face

perception. Nature Neuroscience, 3, 759–763.

Kimberg, D. Y., & D’Esposito, M. (2003). Cognitive

effects of the dopamine receptor agonist pergo-

lide. Neuropsychologia, 41, 1020–1027.

Kimberg, D. Y., D’Esposito, M., & Farah, M. J.

(1997). Effects of bromocriptine on human

subjects depend on working memory capacity.

Neuroreport, 8, 3581–3585.

Kimberg, D. Y., & Farah, M. J. (1993). A unifiedaccount of cognitive impairments following

frontal lobe damage: The role of working mem-

ory in complex, organized behavior. Journal of


Lambon-Ralph, M. A., Patterson, K., Garrard, P., &

Hodges, J. R. (2003). Semantic dementia withcat-




for response and visual cue discrimination learn-

ing. Behavioral Neuroscience, 116, 105–115.

Rao, S. G., Williams, G. V., & Goldman-Rakic, P. S.

(1999). Isodirectional tuning of adjacent inter-

neuronsandpyramidalcellsduringworkingmem-ory: Evidence for microcolumnar organization in

PFC. Journal of Neurophysiology, 81, 1903.

Roberts, R. J., Hager, L. D., & Heron, C. (1994).

Prefrontal cognitive processes: Working memory

and inhibition in the antisaccade task. Journal of


Rougier, N. P., Noelle, D., Braver, T. S., Cohen,

J. D., & O’Reilly, R. C. (2005). Prefrontal cor-

tex and the flexibility of cognitive control:

Rules without symbols. Proceedings of the Na-tional Academy of Sciences USA, 102, 7338–

7343.

Rumelhart, D. E., & McClelland, J. L. (1986).

Parallel distributed processing: Explorations in

the microstructure of cognition, Vols. 1 and 2.

Cambridge, MA: MIT Press.

Schultz, W., Dayan, P., & Montague, P. R. (1997).

A neural substrate of prediction and reward.

Science, 275, 1593–1599.

Seidenberg, M. (1993). Connectionistmodelsandcog-nitive theory. Psychological Science, 4, 228–235.

Siegler, R. (1996). Emerging minds: The process of

change in children’s thinking. New York: Ox-

ford University Press.

Smith, E. E., & Jonides, J. (1998). Neuroimaging

analyses of human working memory. Proceedings

of the National Academy of Sciences USA, 95,

12061–12068.

Stedron, J., Sahni, S. D., & Munakata, Y. (2005).

Common mechanisms for working memory

and attention: The case of perseveration with

visible solutions. Journal of Cognitive Neurosci-

ence, 17, 623–631.Steinmetz, M., Connor, C., Constantinidis, C., &

McLaughlin, J. (1994). Covert attention sup-

presses neuronal responses in area 7a of the

posterior parietal cortex. Journal of Neurophys-

iology, 72, 1020–1023.

Stuss, D., & Benson, D. (1984). Neuropsychological

studies of the frontal lobes. Psychological Bul-

letin, 95, 3–28.

Suri, R. E., & Schultz, W. (2001). Temporal

difference model reproduces anticipatory neu-ral activity. Neural Computation, 13, 841–862.

Tarr, M. J., & Gauthier, I. (2000). FFA: a flexible

fusiform area for subordinate-level visual pro-

cessing automatized by expertise. Nature Neuro-

science, 3, 764–770.

Thelen, E., & Smith, L. B. (1994). A dynamic

systems approach to the development of cogni-

tion and action. Cambridge, MA: MIT Press.

Wallis, J. D., Anderson, K. C., & Miller, E. K. (2001).

Single neurons in prefrontal cortex encodeabstract rules. Nature, 411, 953–956.

Wang, X.-J. (1999). Synaptic basis of cortical persis-

tent activity: The importance of NMDA recep-

tors to working memory. Journal of Neuroscience,

19, 9587–9603.

Zelazo, P. D., Frye, D., & Rapus, T. (1996). An age-

related dissociation between knowing rules and

using them. Cognitive Development, 11, 37–63.




8

Variation in Working Memory

across the Life Span

SANDRA HALE, JOEL MYERSON, LISA J. EMERY,

BONNIE M. LAWRENCE, and CAROLYN DUFAULT

Our interest in working memory evolved out of our research on life span changes in cognitiveprocessing speed. Given the profound changesin response times (RTs) that are associated withcognitive development and aging (for a review,see Cerella & Hale, 1994), we wondered howage differences in speed might affect other as-pects of cognitive function. With respect toworking memory in particular, there seemed tobe multiple ways in which it might be affectedby changes in processing speed as well as mul-tiple ways in which working memory mightaffect higher-order cognitive processes such asreasoning. For example, when the time avail-able for processing is limited, slower processingof memory items might lead to poorer encod-ing, whereas slower processing of non-memory

information might take time away from re-hearsal, leading to more forgetting. The potentialeffects of speed on encoding are similar to thoseof Salthouse’s (1996) limited-time mechanism,whereas the effects on forgetting are similar tothose of Salthouse’s simultaneity mechanism andto Towse and Hitch’s (Chapter 5) task-switching

account of the development of working memory.Both poorer encoding and more forgetting leadto weaker memory traces, which would provide aless adequate basis for reasoning. Therefore, wespeculated that changes in cognitive processingspeed might precipitate a cascade in which in-creases or decreases in speed would lead tochanges in working memory function which, inturn, would directly affect higher levels of cog-nition (Fry & Hale, 1996, 2000; Kail & Salt-house, 1994). It should be noted that the ac-count proposed by Salthouse and that proposedby Towse and Hitch both emphasize how theconsequences of having to switch back and forthbetween tasks are affected by age-related changesin the time it takes to perform these tasks. Theseviews should be distinguished from accounts of

age-related changes in working memory thatemphasize changes in the ability to switch itself,a putative executive function (to be discussedlater).

When we began our research on workingmemory we were immediately drawn to theframework proposed by Baddeley (1986),

194



largely because Baddeley hypothesized that theworking memory system contains separate ver-bal and visuospatial subsystems. This hypothesiswas consistent with our findings on age-related

slowing of verbal and visuospatial processing,findings that also imply the existence of domain-specific processing systems. Specifically, olderadults are generally slower to process all infor-mation, but they are slowed to a substantiallygreater degree in the visuospatial domain and toa much lesser degree in the verbal domain (e.g.,Hale & Myerson, 1996; Lawrence, Myerson, &Hale, 1998; Lima, Hale & Myerson, 1991). Ac-cordingly, we hypothesized (correctly, as it turns

out) that working memory for visuospatial infor-mation would be more age sensitive than work-ing memory for verbal information (Myerson,Emery, White, & Hale, 2003; Myerson, Hale,Rhee, & Jenkins, 1999).

The methodology that we adopted for study-ing age differences in working memory wasinfluenced by Logie, Zucco, and Baddeley’s(1990) elegant study of domain-specificity inthe working memory function of young adults.

The basic design of this study, which involvedtesting memory spans for verbal and visuospa-tial information while participants performedverbal and visuospatial secondary tasks, seemedideally suited for examining age-related differ-ences in working memory. Logie et al. foundthat memory for verbal and visuospatial infor-mation was selectively affected by secondarytasks, so that a secondary task from the samedomain as that of the memory items producedmuch greater interference than a secondary taskfrom the other domain. Nevertheless, in bothcases a secondary task from one domain didproduce some interference with memory foritems from the other domain, and this cross-domain interference was interpreted as showingthe role of the central executive. That is, per-forming any two tasks concurrently, in this casethe primary memory task and the interfering

secondary task, was assumed to require atten-tional resources, leaving less resources for con-trol of the domain-specific slave systems directlyresponsible for maintaining the memory items.This would be true regardless of whether theprimary and secondary tasks were from the sameor different domains, but when they were from

the same domain, the secondary task also se-lectively interfered with the slave system for thatdomain, producing further interference withmemory performance.

The experimental paradigm pioneered byLogie et al. (1990) seemed to provide an excel-lent tool for dissecting the effects of age-relatedchanges in working memory. By developing ver-bal and visuospatial span tasks that producedequivalent performance in young adults whenthere was no secondary task (i.e., equivalent do-main-specific simple span tasks), we could thenadminister these tasks to other groups and assessthe relative rates of development or decline in

the efficiency of the two domain-specific slavesystems. Importantly, by going on to examineinterference by a secondary task from a differentdomain than that of the primary memory spantask we could assess age differences in executivefunction, and by measuring the extent to whicha secondary task from the same domain pro-duced additional interference, we could assessage differences in the sensitivity of the slavesystems. Such sensitivity could reflect either

decay of memory traces during performanceofthesecondarytask,whichpresumablyinterruptsrehearsal, or the actual displacement or degra-dation of memory traces by interference fromsecondary task information.

This experimental paradigm, involving com-bining span tasks and secondary tasks from bothverbal and visuospatial domains, also seemedwell suited to evaluating the proposals of otherdevelopmental researchers. A dominant per-spective in recent theorizing about changes inworking memory across the life span has beenone emphasizing changes in resistance to in-terference (e.g., Dempster & Brainerd, 1995;Richardson et al., 1996). Such changes havebeen described by a variety of terms, includingattentional capacity, inhibitory function, andexecutive control, by researchers advocating dif-ferent working memory models. Moreover, sim-

ilar viewpoints have been adopted by those whostudy children (e.g., Bjorklund & Harnishfeger,1990) and older adults (Hasher & Zacks; 1988),as well as by those who focus not on age differ-ences but on individual differences amongyoung adults (Conway & Engle, 1996). Thisperspective has considerable face validity, given

Variation in Working Memory across the Life Span 195



the age-related changes that occur in the frontallobes considered in the context of the effects of frontal lobe lesions (Dempster, 1992; Kane &Engle, 2002; Moscovitch & Winocur, 1995;

West, 2000). Although differences in terms (e.g.,attention, inhibition, capacity, resources, andcontrol) may reflect different conceptualiza-tions of the issue to some extent, there arenonetheless core similarities, and these simi-larities are sufficient to lead to similar pre-dictions regarding the fundamental issue: resis-tance to interference.

Thus, using the Logie et al. (1990) para-digm we were prepared to assess a number of

critical sources of working memory variation:potential differences in executive function andin the efficiency of the verbal and visuospa-tial slave systems, as well as in the sensitivityof these systems to interference. Of particularinterest was whether measures of these func-tions would rise and fall in concert, or whethera particular measure would follow a uniquetrajectory, lagging or leading through devel-opment and/or aging. In addition, having estab-

lished the often dramatic changes in processingspeed that occur at both ends of the life span(e.g., Hale, 1990; Hale, Lima, & Myerson,1991; for a review see Cerella & Hale, 1994),we wondered what role these changes had inother aspects of cognition (see also Kail &Salthouse, 1994). Our conjectures ultimatelyturned into the developmental cascade model,which posits that the changes in processingspeed that occur with age lead to correspondingchanges in working memory (i.e., faster pro-cessing leads to increases in working memorycapacity and function, and slower processingleads to decreases in working memory capacityand function). The cascade model also positsthat these speed-related changes in workingmemory have consequences for other, higher-order aspects of cognition, with reasoning abil-ity, as measured by tests of fluid intelligence,

being the paradigmatic example (Fry & Hale,1996).It has been said that individual differences

represent a crucible in which to test general(nomothetic)psychologicaltheory(Underwood,1975), and the same argument can be applied

to age differences (e.g., Kotary & Hoyer, 1995).For example, if Baddeley’s (1986) theoreticalframework is correct, then given an age groupknown to have an executive deficit, this defi-

cit should render the group’s performance of aprimary memory span task especially sensitiveto interference by a secondary task from thesame domain as the primary task. Similarly,if items are maintained in working memorythrough rehearsal, and faster rehearsal permitsmore items to be maintained, then given agroup with a deficit in the speed of processinginformation from one specific domain, thatgroup’s memory span for items from that do-

main should be lower than their span for itemsfrom another domain.

Of course, observing age differences consis-tent with the predictions of a general theoryprovides support, but not proof, of the theo-ry’s validity. Moreover, when predictions are notsupported, the fault may lie either with the theoryor with the additional assumptions involvedin the test. For example, a failure to observeenhanced sensitivity to interference by a cross-

domain secondary task in a group with a putativeexecutive deficit could occur because the theo-retical assumptions are incorrect about executiveinvolvement in performance of two concurrenttasks. Alternatively, the failure could occur be-cause the group does not have the hypothesizeddeficit, and often only further research can re-solve such questions. In either case, age differ-ences can play a critical role in testing theoreticalassumptions about the fundamental mecha-nisms underlying working memory performanceas well as those about the nature of differencesbetween age groups.

Over the past several years our research in-terests have expanded to include not only ques-tions regarding age differences but also morebasic questions, such as the nature of the dif-ferences between simple and complex spantasks, and how these differences play out in both

normal and pathological development and ag-ing. This focus on more basic issues arose nat-urally from our developmental research, in partbecause of the general theoretical implicationsof age differences in working memory. In thischapter, we will first describe what we have




observed in experiments with young adults, whoare assumed to be performing at peak efficiencyrelative to other groups. These findings will thenbe compared with what we have observed in

experimental studies of children and adoles-cents, followed by a comparison with healthy,normally aging older adults. In the presentcontext it is important to note that young adultsare not just a control group for studying agedifferences. Rather, young adulthood is justanother developmental stage, albeit one fromwhich it is often easiest to recruit experimentalsubjects. Thus, in our view, findings from youngadults are an important part of the life span

developmental story. After describing the resultsof studies that focused directly on age differ-ences, we will describe the results of two studieson individual differences in working memoryand fluid intelligence, one at each end of the lifespan. Finally, we will review the implications of our findings for the four focal questions onworking memory proposed by the editors of thisvolume.

WORKING MEMORY INYOUNG ADULTS

Baddeley (1986) proposed that the workingmemory system consists of at least three com-ponents. There are two domain-specific subsys-tems, the phonological loop for maintenance of verbal information and the visuospatial sketch-pad for maintenance of visuospatial information,and a central executive that exercises attentionalcontrol and performs other executive functions.Following the approach used by Baddeley andcolleagues (e.g., Logie et al., 1990), our initialgoal was to develop a set of memory span tasksthat would enable independent assessment of these three components in different age groupsacross the life span.

Accordingly, we developed two simple span

procedures to serve as primary tasks, which wehypothesized would each engage one of the twodomain-specific subsystems for tempoarily stor-ing the memory items, and two secondary tasks,one verbal and one visuospatial, which we hy-pothesized would interfere with the maintenance

of verbal and visuospatial information, respec-tively (Hale, Myerson, Rhee, Weiss, & Abrams,1996; for a review, see Jenkins, Myerson, Hale, &Fry, 1999). Of particular interest was the extent to

which a secondary task from one domain wouldaffect performance on memory span for itemsfrom the other domain (e.g., the effect of a visu-ospatial secondary task on verbal memory span),indicating executive involvement in the coordi-nationoftasksfromdifferentdomains(Haleetal.,1996; Hale, Bronik, & Fry, 1997; Myerson et al.,1999).

For the primary memory span tasks, a seriesof items (either digits or letters in the verbal

tasks or locations in a grid in the visuospatialtasks) were displayed one by one on a com-puter monitor, followed by a signal to recall theitems that were just presented. For the verbalsecondary task, participants had to say aloudthe color of each item as it appeared. For thevisuospatial secondary task, participants had toindicate the color of each item by touching thematching color in a palette located to the rightof the primary span stimuli. A schematic draw-

ing depicting these procedures is shown in Fig-ure 8.1. Although our early studies required vocalrecall of digits and manual recall of locations(using marks made directly on the computermonitor), our more recent studies have used atouch screen (with letters rather than digits forrecall of verbal items) or a computer mouse toindicate recall responses (e.g., Jenkins, Myerson,Joerding, & Hale, 2000; Lawrence, Myerson,Oonk, & Abrams, 2001).

Several features of these procedures should benoted. First, to better isolate the mechanisms thatcause verbal and visuospatial secondary tasks tohave different effects on working memory, wedesigned these tasks using an approach pio-neered by Brooks (1968). That is, our verbal andvisuospatial secondary tasks both depend on thesame simple color discriminations, and differprimarily in terms of how those discriminations

are reported (i.e., by naming the color aloud inthe verbal secondary task, and by pointing toa matching color in the visuospatial secondarytask). Thus, it is the response requirement thatmakes the secondary tasks either verbal or visu-ospatial in nature.




Second, we chose to interleave the primaryand secondary tasks, rather than requiring par-ticipants to perform them concurrently. That is,participants were not performing the secondarytask when a memory item was presented andpresumably processed by the participant. Onlyafter the item had been presented could the par-ticipant perform the secondary task. This proce-dure was intended to minimize the possibilitythat our secondary tasks would compete with the

encoding of memory items, although the sec-ondary tasks might compete with rehearsal of theitems.

In addition, the interleaving of primary andsecondary tasks was intended to make our com-plex span procedures formally analogous to otherworking memory procedures such as the reading

span and computation span tasks (Babcock &Salthouse, 1990; Daneman & Carpenter, 1980).Note, for example, that in the prototypical read-ing span task, participants read a series of sen-tences and then try to recall, in order, the lastword of each sentence. That is, participants arepresented with multi-attribute stimuli (i.e.,strings of words, each of which may be thought of as an attribute of the whole sentence) to be pro-cessed, although only one attribute (i.e., the last

word of each sentence) of these stimuli is rele-vant to the primary memory task.Similarly, the verbal and visuospatial items

used in these procedures can be thought of as stimuli with multiple attributes (e.g., color,identity, and location), at least two of which mustbe processed (color and identity in the verbal task

Figure 8.1. Schematicdrawings of the proce-dures used in the threeworking-memory taskconditions for the verbal

( A) and spatial (B) do-mains. Note that thecolor palette representsred as black and blue ashorizontal stripes, andthat red memory itemsare shown as black.(Although memory itemscould be blue, no blueitems are shown.) Also,the color palette, whichis shown once in thisillustration, actuallyappeared adjacent toeach memory item; thepositions of the colors inthe palette varied ran-domly from item to item.




and color and location in the visuospatial task),although only one attribute (identity or location)must be recalled later. In the reading span task,performance of the secondary task (correctly

reading a sentence) implies that the memoryitem (the last word) has been encoded. Similarly,Hale et al. (1996) argued that successful perfor-mance of the secondary task (correctly reportingthe color of the item) in these procedures impliesthat the memory item has been encoded, allow-ing us to focus on why and how the secondarytasks interfere with the maintenance of theseitems in memory.

Third, and finally, we measured memory span

as the length of the series of memory items thatcan be correctly recalled, because use of thismetric facilitates comparison of performance ondifferent memory span tasks. The advantage of measuring span in terms of series length maybe illustrated by comparing this approach withperformance measures based on the number orproportion of correct trials. With such measures,scores depend on both the range of series lengthstested and the number of trials at each length. In

the case of the working memory measures in-cluded in the Wechsler Adult IntelligenceScale–third edition (WAIS-III) and WechslerMemory Scale–third edition (WMS-III; Psy-chological Corporation, 1997), for example, it isdifficult to compare scores on the Digit Span,Spatial Span, and Letter-Number Sequencingsubtests, because all three subtests differ in termsof the total number of correct trials that arepossible. One may get around this problem, atleast in part, by using z scores (e.g., Myerson,Emery, et al., 2003). However, such an approachsacrifices information on the absolute level of performance, and thus may obscure the practicalimplications of any age differences observed. Indesigning Web sites that are friendly to users of all ages, for instance, it is not helpful to know thatthere are significant age differences if one doesnotalsoknowhowmanyitems,onaverage,young

and older adults can hold in working memory.In our first working memory study, the ex-perimental procedures depicted in Figure 8.1were administered to a group of undergradu-ates. We observed that a secondary task requir-ing a response in the same domain as theprimary memory span task led to a decrease in

span of approximately 1.5 items or more, rela-tive to when the primary memory task was per-formed without a secondary task (Hale et al.,1996, Experiment 1). Moreover, this was true

for both verbal and visuospatial spans (i.e., bothfor digit span combined with naming the colorof the memory items and for location spancombined with touching the matching colors).In contrast, memory spans were surprisinglyunaffected by a secondary task when it requireda response in the other domain (i.e., locationspan combined with naming the color of thememory items and digit span combined withtouching the matching colors). These results

represent a classic double dissociation and arestrongly supportive of Baddeley’s (1986) modelof a working memory system with separate ver-bal and visuospatial subsystems.

The interference with memory span pro-duced by a same-domain secondary task appearsto reflect the additional demands that the sec-ondary task places on the specific subsystem en-gaged in maintaining memory items. For ex-ample, naming colors aloud may interfere with

subvocal rehearsal of verbal items. Nevertheless,the failure of secondary tasks from one domainto cause at least a small decrement in memoryspan for items from the other domain is puz-zling (e.g., the lack of effect of pointing to amatching color on digit span). It suggests thattask switching, which is generally thought to bean executive function, does not always lead to adecrease in the efficiency of the verbal andvisuospatial memory subsystems. Or put anotherway, the attentional control involved in coordi-nating performance of the primary and sec-ondary tasks in this study does not appear to usethe same resources as those used in the main-tenance of memory information.

Our interest in understanding the way inwhich the visuospatial secondary task engagedthe visuospatial memory subsystem was piquedby these data. The effect of the verbal second-

ary task on verbal memory span was explainablein terms of the idea that the verbal secondarytask interrupted covert articulatory rehearsal of verbal memory items (e.g., Baddeley, Lewis, &

Vallar, 1984; Gathercole & Baddeley, 1993; cf.,Nairne, 2002). The mechanism underlying theeffect of the visuospatial secondary task on




visuospatial span was less apparent, in part be-cause the visuospatial sketchpad (and in par-ticular, the nature of visuospatial rehearsal) hasreceived far less investigation than the phonolog-

ical loop, and in part because the very existence,let alonethe nature, of visuospatial rehearsal iscontroversial (e.g., Postle, Awh, Jonides, Smith,& D’Esposito, 2004; Washburn & Astur, 1998;see Chapter 10, this volume).

Accordingly, we were interested in evaluat-ing two of Baddeley’s (1986) suggestions regard-ing the types of processes that might be engagedduring the maintenance of visuospatial memoryitems: mental imagery and covert eye move-

ments. We found that, contrary to the hypoth-esis of imagery-based rehearsal, a secondary taskrequiring mental rotation did not affect memoryspan for spatial locations (Hale et al., 1996,Experiment 2). However, simply requiring aneye movement away from the grid in which thememory items were presented did lead to a re-duction in memory span, and requiring a visu-ally guided pointing movement in addition tothe eye movement significantly added to the

interference (Hale et al., 1996, Experiment 3).Taken together, these findings suggest that theshifts in spatial attention that accompany eyemovements and direct limb movements disruptthe mechanism(s) used to temporarily maintainlocation information.

Lawrence et al. (2001) explored this issuefurther by using sophisticated eye-monitoringequipment to ensure that participants actuallymade accurate saccades in the eye-movementconditions and maintained fixation in the otherconditions, including those involving shifts of spatial attention. Again, visuospatial and verbalmemory span tasks were interleaved with sec-ondary tasks by means of an experimental par-adigm similar to that described above and shownin Figure 8.1. In Lawrence et al.’s first experi-ment, the secondary task required participantsto saccade to a sudden-onset peripheral target

presented following each memory item. Thesaccade target was randomly located to the rightor the left of where the memory items appeared.

As can be seen in the upper left panel of Figure8.2, having to execute saccades caused a de-crease in memory for locations of approximately

2.5 items but had relatively little effect onmemory span for letters.

Lawrence et al.’s (2001) second experimentused the same two types of primary memory

tasks (verbal and visuospatial) and three typesof secondary eye-movement tasks: ‘‘reflexive’’saccades (using the same procedure as in Ex-periment 1), antisaccades, and prosaccades.The antisaccade task required participants toexecute a saccade away from the peripheraltarget rather than towards the target as in thereflexive saccade condition. The prosaccadetaskrequired participants to execute an eye move-ment in the direction indicated by a central cue

(an arrow pointing right or left). As shown in theupper right panel of Figure 8.2, all three types of saccades produced decreases of approximatelythe same size in visuospatial memory span. Thisfinding is surprising from theoretical perspec-tives that emphasize the role of inhibition, con-trolled attention, or frontal lobe functions inworking memory, becausesuch perspectives pre-dict a special status for antisaccades, as these areassumed to require inhibition and controlled

attention, functions that have been attributed tothe frontal lobes (e.g., Kane, Bleckley, Conway,& Engle, 2001).

In a third experiment, Lawrence et al. (2001)demonstrated that directional limb movementsalso led to a decrease in visuospatial memoryspan approximately equivalent to that producedby saccadic eye movements. Importantly, thelimb movements were executed in the absenceof eye movements and without visual feedback.These findings suggest that the interference pro-duced by eye movements is not the result of theirvisual consequences (i.e., visual transients, sac-cadic suppression, or the resetting of retinalcoordinates), because the limb movements hadno visual consequences. Rather, interferencefrom eye movements is a property of both di-rected limb and eye movements, perhaps be-cause they both are accompanied by shifts in

spatial attention. It should be noted that theseresults do not show that attention shifts are usedto rehearse visuospatial memory items. They dosuggest, however, that such shifts disrupt theprocesses involved in active maintenance of lo-cation information.




In a follow-up study, Lawrence, Myerson,and Abrams (2004) conducted two experimentsdesigned to more directly determine whether

covert shifts of spatial attention differentiallydisrupt working memory function in the verbaland visuospatial domains. In Experiment 1,which again using the interleaved secondary-task paradigm, participants were required tomake a same–different discrimination about

two successive stimuli (two Xs, two þs, or one X and one þ) that were both presented eithercentrally or peripherally. For the peripheral

stimuli, one stimulus was presented approxi-mately 10 degrees to the left of fixation and theother was presented approximately 10 degreesto the right of fixation. Importantly, partici-pants maintained a central fixation (i.e., no eyemovements were allowed) while making this

8Verbal

N o n e

S a c c a

d e

N o n e

S a c c a

d e

Spatial

6

M e m o r y S p a n

4

Secondary Task Requirement

2

8Verbal

N o s h i

f t

A t t n

s h i

f t

N o s h i

f t

A t t n

s h i

f t

Spatial

6

M e m o r y

S p a n

4


2


8

N o s h i

f t

A t t n

s h i

f t

E y e s h i f t

Spatial

6

M e m o r y

S p a n

4


2

8Verbal

N o n e

S a c c a

d e

P r o - s a

c c

A n t i - s a

c c

N o n e

S a c c a

d e

P r o - s a

c c

A n t i - s a

c c

Spatial

6

M e m o r y S p a n

4

2

Figure 8.2. Memory spans plotted as a function of condition. Data in upper panels are fromLawrence et al. (2001); data from lower panels are from Lawrence et al. (2004). The spans for theNone conditions in the upper panels represent the average of two no-eye-movement controlconditions.




same–different discrimination, so that covertattention shifts were required to identify theperipheral stimuli. As shown in the lower leftpanel of Figure 8.2, visuospatial memory span

was much lower (approximately 1.5 items) whenthe secondary task stimuli were presented inthe periphery (and shifts of spatial attentionwere required) than when the stimuli werepresented centrally, and this decrease in mem-ory span due to spatial attention shifts was ob-served only in the visuospatial domain.

Experiment 2 of Lawrence et al. (2004) ad-dressed the question of whether overt and covertattention shifts differentially affect visuospatial

working memory. Three different secondarytasks involving discrimination of a target (an X

or aþ) were interleaved with a primary memoryspan task. One of these secondary tasks requiredthat the participant maintain fixation (as inExperiment 1) while discriminating centrallypresented targets, and another required them tomaintain fixation while discriminating targetsthat appeared either to the right or left of thefixation point. In the third condition, the target

also appeared to the right or the left of fixation,but participants were required to make an eyemovement to the target.

Relative to the condition in which partici-pants discriminated central targets, the sec-ondary tasks requiring attention shifts and eyemovements caused a decrease in visuospatialmemory span. The decrease, however, wasmuch greater (nearly twice as large) for the eye-movement condition than for the attention-shift condition (see lower right panel of Fig.8.2), even though both conditions involvedexactly the same stimuli. Whether actual eyemovements simply involve more spatial atten-tion than covert shifts of attention or whetherthere are actual differences in the quality of the attention involved is still unclear. On thebasis of this series of experiments, however, wewould stress that the highly selective conse-

quences of both eye movements and shifts of spatial attention (which exclusively affect mem-ory spans for location information) argue againstthe idea that spatial attention involves a general(i.e., domain-independent) resource.

Recently, we turned back to Baddeley’soriginal concept of the working memory sys-

tem, which he described in terms of the tem-porary holding and manipulation of information(Baddeley, 1986). Although a large number of working memory experiments have been con-

ducted using both interleaved and concurrentsecondary tasks that require processing infor-mation irrelevant to the primary memory task,far fewer studies have been conducted withworking memory tasks that actually requiremanipulation of the memory items.

For example, consider the reading span task,which is perhaps the paradigmatic working mem-ory task and the one that provided the model forthe interleaved secondary task procedures used

in all of our studies described thus far. The sec-ondary task in reading span (i.e., reading sen-tences) involves processing information, most of which (except for the final words) is irrelevant tothe memory task. Although such procedures areclearly appropriate for investigating attentionalaspects of working memory, they may be lessappropriate for investigating certain aspects of working memory that underlie performance onhigher-order cognitive tasks. Specifically, rea-

soning and problem solving often require thatsome subset of information be held for a shortperiod of time while it is manipulated or com-bined with other information to produce anacceptable solution to the problem being solved.Mental arithmetic problems, for example, clearlyrequire use of this aspect of working memory.

Emery, Myerson, and Hale (2002) havesuggested that this aspect of working memoryfunction (i.e., the temporary storage of itemsin order to manipulate them) may be betterstudied using what may be termed manipula-tion span tasks than by pairing primary spantasks with interleaved or concurrent second-ary tasks. Letter-number sequencing, a workingmemory task that has been added to the mostrecent version of the Wechsler Memory Scale(WMS-III; Psychological Corporation, 1997),is an example of a manipulation span task. In

letter-number sequencing, a series of alternat-ing letters and numbers (e.g., ‘‘K 2 G 8’’) arepresented, and individuals being tested are re-quired to sequence and then recall these items;numbers must be recalled first in ascendingorder followed by letters in alphabetical order(‘‘2 8, G K’’). Unlike backward span tasks,




which appear to involve off-line manipulationof memory items (i.e., rearrangement occursafter the whole series has been presented), letter-number sequencing, at least in the form that

we have adapted for computer administration,appears to involve on-line manipulation of theitems, and this manipulation may actually facil-itate recall.

Our interest in letter-number sequencingwas stimulated by the unexpected results of ourfirst experiment using this procedure (Emeryet al., 2002, Experiment 1). This experimentwas designed specifically to compare manipu-lation span procedures with those combining

primary and secondary tasks. As in many of ourprevious experiments, participants performedtwo different primary memory span tasks, onefrom the visuospatial domain (i.e., memory forlocations in a grid) and one from the verbaldomain (i.e., memory for letters and numbers).For each of these span tasks, there were threeconditions: (1) a simple span condition involv-ing forward recall, (2) a complex span condi-tion that combined the primary span task with

an (interleaved) secondary task from the samedomain, and (3) a manipulation span condi-tion requiring participants to reorganize thememory items prior to recall (see Fig. 8.3).Participants recalled verbal memory items bytouching the appropriate boxes in two grids,one containing digits and the other containingletters, and they recalled visuospatial memoryitems by touching the appropriate boxes in anempty grid. In all conditions, memory itemswere presented at a rate of one every 3 s to allowtime for secondary task responses or on-linemanipulation of the memory items, dependingon the condition.

In the complex-span conditions, the sec-ondary tasks involved color judgments (reportedeither by saying the color of each memory itemor by pointing to a matching color, dependingon the domain of the primary task) and were

similar to those used in Hale et al.’s (1996)study. For the manipulation span condition inthe visuospatial domain, participants were toldto imagine each memory item shifted one col-umn to the right and to recall these new lo-cations (n.b., if shifting an item resulted in alocation that was off the grid, participants were

instructed to wrap it around to the leftmostcolumn of the grid). For the manipulation spancondition in the verbal domain, we used a com-puterized version of the letter-number sequenc-

ing task described previously. Letters and digitswere presented alternately, and participantswere required to sort and sequence the items,recalling the digits first (in numerical order) andthe letters second (in alphabetical order).

The question of interest in this experimentwas whether secondary tasks and manipula-tion tasks would have similar negative effects onworking memory, relative to the correspondingsimple, forward span tasks in each domain. The

results revealed that both types of procedures(performing a visuospatial secondary task andspatiallymanipulatingthememoryitems)causedapproximately equivalent amounts of interfer-ence with visuospatial memory span. The re-sults for the verbal conditions, however, revealeda very different pattern: requiring a verbal sec-ondary task caused verbal memory span to de-crease significantly (replicating our previousstudies), but requiring participants to reorganize

the memory items actually resulted in betterperformance than simple forward recall of let-ters and numbers!

This puzzling finding may be best under-stood in the context of two follow-up experi-ments by Emery et al. (2002) that focused ex-clusively on letter-number sequencing. In thefirst follow-up study, we found that the benefi-cial effect of sequencing letters and numbersdepends on the rate at which they are presented.When the presentation rate was 2.25s per item,participants’ spans were approximately 1.0 itemlarger when they had to sequence the items thanwhen they had to recall them in the order pre-sented. When the rate was 1.25s per item,however, memory spans in the forward and se-quenced recall conditions were approximatelyequal. The fact that the benefit from sequenc-ing the items depends on the presentation rate

(whereas simple, forward spans are relativelyunaffected by rate) suggests that sequencing oc-curs on-line, contingent on the amount of timeavailable, and not just off-line, after all the itemshave been presented.

In the second follow-up study, Emery et al.(2002) added a condition in which the stimuli




87 9

54 6

21 3

RQ S

KJ L

FB G

T

M

H

87 9

54 6

21 3

RQ S

KJ L

FB G

T

M

H

87 9

54 6

21 3

RQ S

KJ L

FB G

T

M

H

Primary Only

A

Letter-Number Span Tasks

Secondary:Color Naming

Transform:Sequencing

K

2

G

8

K

8

K

2

G

8

K 2 G 8 K 2 G 8 2 8 G KTouch Touch Touch

“red”

“red”

“white”

“white”

X

X

X

X

X

X

X

Touch

X

X

Touch

touch

red

touchwhite

X

X

Touch

Primary Only

B

Location Span Tasks

Secondary:Color Touching

Transform:Move Right

Figure 8.3. Schematic drawings of the procedures used in the three conditions of the letter-number span tasks ( A) and the three conditions of the location span tasks (B). As in Figure 8.1, redmemory items are depicted as black, and no blue items are shown.

204



were presented in a presorted order (i.e., as-cending numbers first, alphabetically orderedletters second), rather than being sequenced bythe participants. Memory span for items pre-

sented in order was higher than forward recallof alternately presented (and randomly ordered)letters and numbers. When participants had tosequence the items themselves, memory spanswere intermediate between those for the othertwo conditions. Thus, there appear to be bothcosts and benefits associated with manipulatingmemory items: ordered items can be recalledmore readily than unordered items, but mem-ory items may be lost during the reorganization

process that produces an ordered list.The costs and benefits observed with letter-

numbering sequencing stands in contrast to theresults for backward spans (e.g., Myerson, Em-ery, et al., 2003), in which only the costs of manipulation are seen. The sources of thesecosts are probably not hard to explain: rearrang-ing memory items not only takes time, it mayresult in proactive interference from the originalsequence. As Emery et al. (2002) have shown,

however, the costs of reorganization may some-times be outweighed by the benefits, and anal-ysis of the difference between backward spanand letter-number sequencing procedures mayshed light on the source of these benefits.

Span tasks deliberately test participants at thelimits of their abilities, which implies that therepresentations of at least some memory itemswill be degraded by the time participants mustattempt to recall them. Reorganization of itemsinto a more familiar, predictable series, as inletter-number sequencing, may facilitate thereconstruction of such degraded representa-tions (a process sometimes termed redintegra-tion; Schweickert, 1993). In contrast, rearrangingmemory items in reverse order probably fails tobenefit recall because the backward version of the list is typically no more familiar or predict-able than the original, forward version of the list.

Of course, degradation of memory repre-sentations (whether through decay or interfer-ence) occurs continuously, and the sooner thatitems can be organized the less likely it is thattheir representations will become too degradedto be recoverable. This would explain whyslower presentation rates, which permit more

on-line reorganization, result in greater bene-fits to recall performance. Interestingly, the pro-cesses involved in reorganizing memory itemson-line, like those involved in their active main-

tenance, appear to be highly domain specific.This is evidenced by the fact that, as Emeryet al. (2002) observed in a final experiment, aninterleaved visuospatial secondary task had rel-atively little effect on letter-number sequenc-ing performance, whereas an interleaved verbalsecondary task completely eliminated the ben-efits of sequencing. These results suggest thatsequencing uses relatively little in the way of general attentional resources.

Taken together, the series of memory spanexperiments with young adult participants de-scribed above provides strong support for Bad-deley’s (1986) hypothesis of separate subsystemsfor the temporary maintenance of verbal andvisuospatial information. A particularly robustfinding, observed in a number of experimentsand with a variety of procedures, is that inter-leaved secondary tasks produce predominantlydomain-specific interference with performance

on primary memory span tasks. For example,color and shape discriminations only interferewith memory for numbers and letters if the re-sponse is vocal rather than manual. In contrast,verbal memory spans are relatively unaffectedby having to point to colors that match those of the memory items, by eye movements or di-rected limb movements (regardless of whetherthey are or are not visually guided), or by spatial-attention shifts, all of whichdo affect visuo-spatial spans. Importantly, the antisaccade taskproduces no more interference with visuospa-tial working memory than other eye-movementtasks.

These findings suggest that the executivefunctions involved in inhibiting responses, con-trolling attention, and switching back and forthbetween primary and secondary tasks producerelatively little interference with the temporary

maintenance of information in working mem-ory. Indeed, the presence or absence of an ad-ditional processing requirement consistentlyplayed a much smaller role in determining thenumber of items that could be recalled thanwhen the additional processing and the infor-mation to be maintained involved the same




domain (i.e., verbal vs. visuospatial). Manipu-lation span experiments suggest that the pro-cesses involved in manipulating the contents of working memory may also be relatively domain

specific. Taken together, these insights into whatdoes and does not affect memory span in youngadults have helped guide the design and inter-pretation of our studies of age-related differ-ences in working memory across the life span.

WORKING MEMORY ACROSSTHE LIFE SPAN

A major difficulty in comparing the effects of ex-perimental manipulations on the performanceof different age groups is the lack of equivalentbaseline performance. Perhaps the most famil-iar examples of this come from studies of agedifferences in RTs, and the problem of com-paring semantic priming effects is typical (e.g.,Chapman, Chapman, Curran, & Miller, 1994;Hale & Myerson, 1995; Myerson, Ferraro, Hale,& Lima, 1992). Children and older adults both

show larger semantic priming effects thanyoung adults. That is, their RTs decrease morethan those of young adults when a lexical de-cision is primed by the preceding semantic con-text. The difficulty in interpreting this reliablefinding, which on the face of it would seemto imply that children and older adults are bet-ter at using context, is that children and olderadults have longer RTs than those of youngadults to begin with (i.e., in the absence of anyrelevant context). Thus, children and olderadults may benefit more from context simplybecause they find decisions more difficult in theabsence of context, and not because they usecontext more effectively.

Various analytical tools (e.g., regression anal-yses based on Brinley plots and state-tracegraphs) have been developed to deal with theproblem of unequal baselines with respect to

RTs (e.g., Myerson, Adams, Hale, & Jenkins,2003; Verhaeghen & Cerella, 2002). The prob-lem of unequal baselines also occurs, however,with respect to memory spans (e.g., Jenkins et al.,1999), for which graphical and regression ap-proaches similar to those used with RTs are ar-

guably less appropriate (Oberauer & Suß, 2000;cf., Myerson, Jenkins, Hale, & Sliwinski, 2000).One approach to dealing with this problem inthe case of memory span data is to do what

amounts to an ‘‘end run’’ around it. If two tasksare matched in difficulty for one of the groupsbeing compared, then any difference in perfor-mance of the two tasks by another group clearlyindicates that the latter group is more (dis)ad-vantaged with respect to one of the tasks. Morespecifically, if a verbal and a visuospatial spantask are equally difficult for young adults, then if older adults have lower visuospatial spans thanverbal spans (and they do), clearly age affects

visuospatial working memory more than it af-fects verbal working memory. In this case, in-terpretation of the result is straightforward eventhough older adults may have lower spans thanthose of young adults on both verbal and visuo-spatial tasks.

How does one create verbal and visuospatialtasks of equal difficulty? One approach takesadvantage of the fact that memory spans arelarger for items drawn from a smaller set. Thus,

digit span is larger than letter span, which islarger than word span. Similarly, the number of locations in a matrix that can be remembereddecreases as the size of the matrix increases. As itturns out, young adults’ digit spans are approx-imately the same size as their visuospatial spanswhen they are required to remember locationsi n a 44 grid (e.g., Hale et al., 1996; see Fig. 8.1above), thereby facilitating comparisons withother age groups on these tasks. Alternatively,one can match spans in a child or older-adultgroup and then compare spans on the sametasks in young adults ( Jenkins et al., 2000).Other approaches are possible as well, but forthe most part we have relied on matching youngadults’ verbal and visuospatial spans, and thisstrategy has been relatively successful, as we willshow in the following sections on children andolder adults.

Children

To examine how development affects workingmemory, we tested children with the same basicexperimental paradigm that we had previously




in 8-, 10-, 12-, and 19-year-olds. (Importantly,Fry and Hale also administered processing-speed and fluid-intelligence tests, the results forwhich will be considered below in the section

Individual Differences.) In both studies, simpleverbal and visuospatial memory spans wereroughly equivalent in young adults, but in chil-dren, verbal spans were higher than visuospatialspans, with the difference between the two do-mains decreasing with the age of the child group(see Fig. 8.5). The finding that visuospatialmemory span approaches adult levels moreslowly than verbal memory span is an intriguingone, and may be related to the finding that

visuospatial memory spans decline more rapidlyin old age, as discussed in the following section.

Older Adults

To understand how normal aging affects work-ing memory, we tested healthy younger andolder adults (mean ages, 20 and 67 years, re-spectively) using the experimental paradigm in-troduced by Hale et al. (1996), with the same

two modifications (fewer colors and slowerpresentation rates) used by Hale et al. (1997)in testing children. As reported by Myerson et al.(1999), and shown here in Figure 8.6, thepattern of the data from both the young- andolder-adult groups conformed to the double

dissociation observed in our original study of younger adults (Hale et al., 1996). That is, forboth groups, only same-domain secondary tasksnegatively affected memory span performance.

As can be seen, however, there was a strikingdifference between the two age groups: on thevisuospatial tasks, older adults showed deficitsthat were considerably larger (i.e., nearly 2.5items) than those on the three verbal tasks (i.e.,approximately 1 item).

From several theoretical perspectives, includ-ing Baddeley’s (1986) working memory model,frontal-lobe aging theories (Moscovitch &Winocur, 1995; West, 2000), and the inhibi-

tion-deficit framework (Hasher & Zacks, 1988;Zacks, Radvansky, & Hasher, 1996), it is inter-esting to note that normal aging does not appearto be associated with a decline in executivefunctions, at least as measured in terms of theeffects of having to switch back and forth be-tween tasks. That is, different-domain secondarytasks had little or no effect on memory span inolder adults, and although same-domain sec-ondary tasks did interfere with performance of

primary memory span tasks, the observed inter-ference effects were no larger than those ob-served in young adults. In contrast to this pres-ervation of function, the marked decrease invisuospatial memory spans with age appears tobe part of a more general decline in the effi-ciency of visuospatial processing, as evidencedby greater visuospatial than verbal slowing andgreater age-related deficits in visuospatial thanverbal learning (Jenkins et al., 2000).

This decline in visuospatial processing effi-ciency without an accompanying breakdown inthe independence of the two domain-specificworking memory subsystems may be contrastedwiththeresultsobtainedinchildren.Recallthat,like older adults, children show lower visuo-spatial than verbal spans (see Fig. 8.4 above),but whereas secondary tasks from one domainproduce interference with memory spans in the

other domain in children (at least before 10years of age), older adults’ spans are relativelyunaffected by such procedures (Hale et al.,1997; Myerson et al., 1999). These findingssuggest that differences in verbal and visuo-spatial working memory on the one hand, and

8

6

4

2

M e m o r y S p a n

8 10

Fry & Hale (1996)

Verbal Spatial

Hale et al. (1997)

12

Age (years)

14 16 18

Figure 8.5. Verbal and spatial memory span plottedas a function of age. Data are from Fry and Hale(1996) and Hale, Bronik, and Fry (1997).




differences in the functional independence of the two domains on the other may reflect two

separate developmental phenomena. As documented by Jenkins et al. (1999), the

differential decline in older adults’ verbal andvisuospatial working memory has been repli-cated in our laboratory several times. Althoughthis finding has been somewhat controversial,its robustness is evidenced by the results of ouranalysis of the normative data from the mostrecent Wechsler Memory Scale (WMS-III). Inparticular, our analyses revealed more rapidadult age-related decline in visuospatial spanthan in digit (i.e., verbal) span (Myerson, Em-ery, et al., 2003). Confidence in this findingstems not only from its consistency with previ-ous results in our laboratory, but more impor-tantly from the fact that the verbal–visuospatialdifference in the WMS-III data is based on anormative sample of more than 1000 partici-pants between the ages of 20 and 90 years.

The WMS-III data also reveal two other sig-nificant findings. Perhaps surprisingly, the dif-ference between forward and backward spans isnot affected by age in either domain. The effectof age on the difference between forward andbackward spans, like the difference between

verbal and visuospatial spans, has been contro-versial; indeed, the WAIS-III technical manual

specifically states that backward digit span ismore affected by aging than forward digit span.Nevertheless, as depicted in the left panel of Figure 8.7, the normative sample data clearlyshow that the difference between forward andbackward spans remains constant as people age(Myerson, Emery, et al., 2003).

The second important observation based onthe WMS-III data is that performance on theletter-number sequencing subtest shows a un-ique curvilinear pattern of decline (see Fig.8.7, right panel), decreasing relatively slowlyuntil around age 65, after which the decline ismore precipitous. In contrast, digit and spatialspans (both forward and backward) showedrelatively more linear declines (Myerson, Em-ery, et al., 2003). These findings, taken togetherwith the results of our previous study of letter-number sequencing in young adults (Emery

et al., 2002), suggest that the distinction be-tween off-line and on-line manipulation maybe important for understanding age differenceson these working memory tasks.

Two recent experiments by Emery, Myerson,and Hale (2003) addressed the question of the

Young Adult Group

8

N o

n e

V e r b a

l

S p a t i a

l

N o

n e

V e r b a

l

S p a t i a

l

Spatial

6

M e m o r y S p a n

4


2

Verbal

Older Adult Group

N o

n e

V e r b a

l

S p a t i a

l

N o

n e

V e r b a

l

S p a t i a

l

Spatial


Verbal

Figure 8.6. Verbal and spatial memory span plotted as a function of secondary task requirementfor the young-adult group (left panel) and the older-adult group (right panel). Data are fromMyerson, Hale, Rhee, and Jenkins (1999).




role of on-line manipulation of memory itemsin the effects of age on letter-number sequenc-ing. In Experiment 1, Emery et al. (2003) testedyounger and older adults (mean ages of 20 and76 years) on the letter-number sequencing tasksdescribed earlier along with an additional con-dition in which the series of memory items wasfollowed by a cue indicating the order in whichthe items were to be recalled (see Fig. 8.8).Focusing first on the three conditions previouslystudied by Emery et al. (2002), the results foryoung adults replicated those of the earlier study(see the three right-most bars in the left panel of

Fig. 8.9): Memory spans were lowest for forwardrecall of alternating letters and numbers andhighest for forward recall of presorted memory

items, whereas sequenced recall produced in-termediate-sized spans. As shown in the rightpanel of Figure 8.9, although the memory spansof older adults in these three conditions werenot as large as those of young adults, they oth-erwise had a similar pattern.

Of particular interest in this experiment wasthe condition with the sequencing cue. Thiscondition encouraged off-line processing be-cause participants had to maintain the memoryitems in the order of presentation (alternatingletters and numbers) until the end of the series,at which time the cue informed them whether

or not to sequence the memory items (ascend-ing digits followed by ascending letters). Mem-ory spans on off-line processing trials (repre-

8.0

6.0

4.0

2.0

0.020 40

Digits Forward

Digits Backward

60 80

Age (years)

M e a n L o n g e s t S e r i e s

0.0

–0.5

–1.0

–1.5

–2.020 40

Digit SpanSpatial SpanLetter-Number Sequencing

60 80

Age (years)

S p a n ( z - s c o r e )

Figure 8.7. Performance on memory span tests plotted as a function of age. Mean longest series of digits forward and backward are shown in the left panel. Z scores for digit span, spatial span, andletter-number sequencing are shown in the right panel. Data are from Myerson, Emery, White,and Hale (2003).

Figure 8.8. Schematicdrawings of the proce-dures for the letter-number span tasks used inEmery, Myerson,and Hale (2003).




sented by the left-most bar in each of the twopanels of Fig. 8.9) were lower than spans in thesequencing condition (third bar from the leftin each panel), in which participants could re-arrange the items on-line (i.e., during presen-tation of the items). Although both groupsbenefited from the opportunity to rearrangeitems on-line (as compared with off-line), theyounger adults benefited to a greater extent thanthe older adults.

One possible explanation for this greaterbenefit from on-line sequencing observed inthe younger adults is that age-related slowingmay have made it more difficult for the olderadults to manipulate memory items on-line. InExperiment 2, Emery et al. (2003) directlytested this hypothesis by presenting memoryitems (alternating letters and numbers) at threedifferent rates (1.5 s per item, 2.5 s per item, and3.5 s per item), including one faster and oneslower than that in the preceding experiment.Presentation rate was crossed with recall order

(forward vs. sequenced) to yield six conditions.The data, shown in Figure 8.10, show thatpresentation rate did not affect performance byeither young or older adults (mean ages of 20and 77 years) in the forward span conditions. Inthe sequenced conditions, however, presenta-tion rate did affect memory spans. Rather thaneliminating age differences, slower presentationrates actually caused them to increase. Pre-sumably, young adults’ faster processing speed

means that they can reorder more memory itemsthan older adults given the same amount of time, and thus young adults benefit more thanolder adults from increases in the amount of time available for on-line sequencing.

Overall, our studies of working memory inyounger and older adults have led us to thefollowing conclusions. Regardless of the task,older adults show smaller memory spans thanthose of younger adults, and this is particularlytrue for visuospatial working memory tasks,which show reliably larger age differences than

Figure 8.9. Memory span plottedas a function of recall condition foryoung- and older-adult groups.Data are from Emery, Myerson,and Hale (2003).

Figure 8.10. Memory span plottedas a function of rate of item pre-sentation for the young-adult andolder-adult groups in the forwardand sequenced recall conditions.Data are from Emery, Myerson,and Hale 2003).




verbal working memory tasks. Within the ver-bal and visuospatial domains, however, the pat-terns of memory spans on different tasks areremarkably similar in the two groups and ap-

pear to provide little evidence that executivedeficitsunderlie age differences in working mem-ory. Recently, research in our lab has focused onmanipulation span tasks and revealed an impor-tant distinction between tasks that involve on-lineand off-line reorganization of memory items. Onthe one hand, the effects of requiring off-linemanipulation, as reflected in the difference be-tween forward and backward spans, appear to berelatively age invariant, contrary to common

belief. On the other hand, the effects of on-linemanipulation, as evidenced by the differencebetween forward spans and letter-number se-quencing spans, appear to depend on both theamount of time allowed to accomplish the re-ordering and the difference in processing speedbetween the groups being compared.

INDIVIDUAL DIFFERENCES

IN WORKING MEMORY

In addition to studying age-related differencesin working memory through an experimentalapproach, we have also conducted researchusing an individual-differences approach. Theadvantage of this approach is that it permitsone to test causal models by means of multiplevariables and to evaluate all of the hypothesizedrelationships simultaneously. Importantly, theindividual-differences approach allows us totest hypotheses about the linkage between age,working memory, and higher cognitive pro-cesses, such as the processes measured by tests of fluid intelligence. In particular, we have exam-ined age and individual differences in cognitiveabilities during childhood (Fry & Hale, 1996)as well as during middle and late adulthood(Hale, 2003). Thus, our models focus on the

relationship between age and fluid intelligence,and evaluate the role of potential mediating,hypothetical constructs (and the connectionsamong these constructs).

We first used the individual-differences ap-proach in a cross-sectional study of children,

adolescents, and young adults (Fry & Hale,1996). This study was designed to determinethe extent to which age-related improvementsin fluid ability (as measured by the Raven’s

matrices) are the result of age-related improve-ment in working memory ability that are, inturn, the result of age-related improvement incognitive processing speed. We termed thishypothetical causal account the developmentalcascade model. We contrasted the developmen-tal cascade model with a full developmentalmodel (shown in the upper panel of Fig. 8.11)in which age, cognitive processing speed, andworking memory function are all potential

predictors of fluid ability (paths 1, 2, and 3,respectively). In the full model, speed is also apotential predictor of working memory func-tion (path 4), and age serves as a predictor of both speed and working memory (paths 5 and6, respectively). According to the develop-mental cascade hypothesis, only paths 5, 4, and3 should be significant. (Note that the use of path analysis in lieu of a structural-equationmodel was dictated by the low number of vari-

ables associated with the each of the constructs.)To test the developmental cascade model,

Fry and Hale (1996) assessed 219 participantsbetween the ages of 7 and 19 years using abattery of tests that included four measures of cognitive processing speed similar to those usedpreviously in our laboratory (Hale, 1990; Hale,Fry, & Jessie, 1993). Four working memorytests, selected from the six original tasks (shownin Fig. 8.1, above), were also included in thebattery: verbal without a secondary task, verbalplus a verbal secondary task, spatial without asecondary task, and spatial plus a visuospatialsecondary task. Fluid ability was measured us-ing Raven’s Standard Progressive Matrices.

The results of the Fry and Hale (1996) studyprovide considerable support for the develop-mental cascade hypothesis in terms of how ageaffects speed and working memory, and how

these variables in turn affect fluid ability. As seenin the lower panel of Figure 8.11, the relation-ship between age and working memory was, aspredicted, mediated by processing speed, andthe relationship between speed and fluid abilitywas mediated by working memory. Counter to




the developmental cascade hypothesis, how-ever, paths 1 and 6 were statistically signifi-cant. That is, age had a direct effect on workingmemory, even with speed statistically con-trolled, and age also had a direct effect on fluidability, even with both speed and workingmemory statistically controlled.

Recently, we revisited the Fry and Hale(1996) data set to examine the unique andshared variances among age, speed, workingmemory, and fluid ability (Fry & Hale, 2000).Two separate hierarchical multiple-regressionanalyses revealed that nearly all of the variance(i.e., 97%) in working memory that could beaccounted for by age was shared variance, andmost of the variance (i.e., 80%) in fluid abilitythat could be accounted for by age was also

shared variance. That is, all but 3% of the totalage contribution to working memory was me-diated by speed, and all but 20% of the totalage contribution to fluid ability was mediatedby working memory. Thus, although the de-velopmental cascade model did not provide a

perfect fit to the data, it does tell much of thestory. Given that chronological age is only aproxy for various maturational changes and en-vironmental events, future research is neededto determine what can account for the re-maining 20% of the age-related variance influid ability.

It may be noted that our analyses used asingle working memory construct based on ver-bal and visuospatial tests of simple and com-plex memory spans. In contrast, Jarrold andBayliss (Chapter 6) distinguished two workingmemory constructs, simple span (or storage)and complex span, in their analysis of the de-terminants of working memory in childrenbetween the ages of 6 and 11. Jarrold andBayliss reported that although both speed and

simple span contributed to complex memoryspan, speed did not contribute to simple span.From the perspective of the developmentalcascade hypothesis, this latter finding is some-what surprising, and appears to contradict theresults of previous studies (e.g., Kail, 1992; Kail

Figure 8.11. Path analysis of data fromFry and Hale (1996). The full devel-

opmental model of relations amongage, processing speed, working mem-ory, and fluid ability is shown in theupper panel, and the path coefficientsfor the full model are shown in thelower panel. The path between speedand fluid ability (shown by broken line)failed the Wald test. Retesting with thatpath omitted resulted in negligiblechanges in the path coefficients.




testing, and all had visual acuity corrected to20/40 or better.

The participants were administered a batteryof three verbal and three visuospatial working

memory tasks similar to those used in our pre-vious studies but adapted for the touch screen(two of the visuospatial conditions are shown inthe first two conditions depicted in the lowerpanel of Fig. 8.3), and two tests of fluid ability,Raven’s Progressive Matrices and the Wood-cock Johnson Concept Formation subtest. TheWoodcock Johnson Concept Formation subtestconsists of a series of simple shapes shown in avariety of colors and one of two sizes (small or

large). In each problem, a subset of adjacentobjects appears inside a rectangle, and partici-pants are instructed to state the rule that governswhich shapes are inside the rectangle and whichare outside. For example, if a small red triangleand a large yellow square were inside the rect-angle, and a large blue triangle, a large red cir-cle, and a large blue square were outside therectangle, then the correct response would be‘‘small or yellow.’’

Examination of the zero-order correlationsbetween the working memory measures and thefluid-ability tests revealed that the single bestpredictor of both the Raven’s test and Wood-cock Johnson subtest was the simple visuospa-tial working memory measure (i.e., visuospatialmemory span when there was no secondarytask requirement). The correlations betweensimple visuospatial span and the Raven’s testand Woodcock Johnson subtest were .63 and.55, respectively. With regard to the verbalspan tasks, the complex verbal working-memorymeasure (verbal span with a verbal secondarytask) was the best predictor of the two fluidmeasures, with r s of .38 and .40, respectively.Nevertheless, it should be noted that all three of the visuospatial working memory tasks hadhigher correlations with the fluid-ability mea-sures than the best verbal predictor.

Our finding of a high correlation of sim-ple visuospatial span with fluid ability in olderadults is consistent with recent findings inyoung adults (see Chapters 2 and 3). The highcorrelations observed among the visuospatialspan tasks, and between both the simple and

complex spans and the tests of fluid ability, arealso consistent with the finding that simplevisuospatial span predicts performance tests of visuospatial abilities as well as complex visuo-

spatial span and that the two types of visuospatialspan tasks are not easily differentiated psycho-metrically (Miyake, Friedman, Rettinger, Shah,& Hegarty, 2001; Shah & Miyake, 1996).

Next, the data from these working memorytasks and tests of fluid ability were subjected toconfirmatory factor analysis. We hypothesizedthat our working memory tasks would load ontwo factors, one visuospatial and one verbal,and that the fluid-ability tests would load on a

third factor. This model provided an extra-ordinarily good fit to the data (w2¼ 9.80, p¼.912, RMSEA < .0001; see Fig. 8.13, leftpanel). It should be noted that two of the con-structs (spatial working memory and fluid abil-ity) were themselves highly correlated (r ¼ .99),an issue we will return to later. We also con-ducted a confirmatory factor analysis of datafrom the working memory tasks in which therewere two hypothesized factors based on the

distinction between simple and complex spantasks, as suggested by Engle, Tuholski, Laugh-lin, and Conway (1999). This analysis, whichcontrasted span tasks with and without sec-ondary tasks, revealed that a model based onthe simple–complex distinction provided aninadequate fit to the data, although it is im-portant to recall that our sample consisted of older adults whereas Engle et al.’s sampleconsisted of young adults.

Given that our hypothesized three-factorstructure (consisting of verbal working mem-ory, spatial working memory, and fluid ability)was confirmed, we proceeded to use structuralequation modeling to address the followingquestion: are age-related deficits in visuospatial(but not verbal) working memory the cause of age-related deficits in fluid ability? Our struc-tural equation model (see Fig. 8.13, right

panel) provided an excellent fit to the data,w2¼ 23.91, p¼ .409, RMSEA ¼ .024). The

only statistically significant predictor of fluidability was visuospatial (not verbal) workingmemory, so the effect of age on fluid ability wasentirely indirect, via the influence of age on




visuospatial working memory. Although agealso affected verbal working memory, this effectwas not nearly as pronounced as the effect of age on visuospatial working memory. Age ex-plained more than twice as much of the

variance in older adults’ visuospatial workingmemory, compared with their verbal workingmemory.

Finally, the high correlation between the spa-tial working-memory and fluid-ability factorsobserved in our first confirmatory factor analysissuggested that a two-factor (rather than three-factor) solution might provide a good fit tothe data. Therefore, we conducted another con-firmatory analysis to test a factor model con-sisting of a verbal-ability (or verbal workingmemory) factor and a visuospatial-ability factor(based on both visuospatial working memoryand fluid ability). The results of this analysisshowed that the fit to the data was slightly im-proved (w2¼ 10.46, p¼ .941, RMSEA < .0001).This small improvement aside, application of Ockham’s razor would suggest a preference forthe simpler (two-factor) model given that both

provide adequate fits. This finding implies that,at least for older adults, visuospatial workingmemory (including both simple and complexspans) and fluid ability represent facets of aunitary visuospatial factor (namely, visuospatialability) that shows a general decline throughoutlate adulthood.

DISCUSSION OF THE FOURQUESTIONS

Overarching Theory

of Working MemoryBaddeley’s (1986) model has served as a frame-work for much of our research on workingmemory. From our perspective, the attractionof Baddeley’s model is that it includes sepa-rate, content-specific subsystems. As we notedpreviously, a model of working memory withseparate verbal and visuospatial subsystems wasparticularly appealing because our work oncognitive aging has shown that older adults areless efficient at processing visuospatial infor-mation than verbal information (e.g., Hale &Myerson, 1996; Lawrence et al., 1998; Limaet al., 1991). Moreover, it seemed to us that datafrom different age groups (from childhood tolate adulthood) might prove to be a fertiletesting ground for Baddeley’s model. For themost part, our results have been consistent withthis model, although our findings suggest that

secondary tasks interfere with memory spanwhen they engage the same domain-specificsystems as the primary memory task, rather thanwhen they engage domain-independent execu-tive functions.

Many studies of working memory (includ-ing our own) have examined the effect of

Figure 8.13. Results of confirmatory factor analysis (left panel) and structural equations modeling(right panel) of data from Hale (2003). Note that for the model shown in the right panel, neitherthe path between age and fluid ability nor the path between verbal working memory (WM) andfluid ability reached statistical significance. WJ¼Woodcock Johnson subtest.




requiring people to maintain informationwhile performing secondary tasks that requireadditional processing. However, the additionalprocessing typically involves information irrel-

evant to the primary memory task (e.g., Dane-man & Carpenter, 1980; Engle et al., 1999;Hale et al., 1996; Lawrence et al., 2001). Mostrecently, we have been motivated by Baddeley’s(1986) original definition of the working mem-ory system to explore tasks such as letter-numbersequencing that explicitly involve manipulat-ing memory items (Emery et al., 2002, 2003;Myerson, Emery, et al., 2003). We believe thatsuch tasks are in some ways closer to the original

concept of working memory and provide anecessary supplement to studies using primaryand secondary tasks.

Critical Sources of WorkingMemory Variation

The major source of variation in workingmemory ability among the participants in ourstudies is age. Of course, chronological age is not

itself a causal variable. Rather, age is a proxy forat least two distinct sets of neurobiological pro-cesses, with the term maturation used to dis-tinguish the neurobiological changes that oc-cur during childhood from the changes duringthis period that are due to learning, and theterm senescence used to refer to neurobiologicalchanges during adulthood. The specific physio-logical and anatomical changes that occur withage are presently not well understood, but withchanges in the number of cortical neurons nowruled out, researchers have increasingly focusedtheir attention on changes in white matter. Thenature of these changes appears to differ qual-itatively at the two ends of the life span (forreviews, see Paus et al., 2001; Peters, 2002).In school-age children and adolescents, in-creases in white-matter volume are accompaniedby decreases in gray-matter volume, whereas

both white- and gray-matter volumes decreasethroughout adulthood and there is an increase inthe incidence of white-matter lesions. The routeby which these neurobiological changes affectworking memory is not yet clear, but the fact thatwhite-matter changes are involved is consistentwith the hypothesis that changes in processing

speed contribute to changes in working memoryat both ends of the life span.

The maturation of the working memory sys-tem from 8 to 20 years is itself a complex process,

involving both increased myelination and syn-aptic pruning, which differentially affects exec-utive, verbal, and visuospatial functions. Perhapssurprisingly, the ability to switch back and forthbetween tasks, an executive function, appears tomature most quickly—at least when switchinginvolves alternating between a primary memorytask and a simple secondary task (Hale et al.,1997)—with visuospatial memory function be-ing the last to reach adult levels (Fry & Hale,

1996; Hale et al., 1997). As might be expected from the neurobio-

logical evidence, the process of senescent de-clineinworkingmemoryfunctioninindividuals20 to 80 years of age is not a simple mirror imageof the system’s maturation during childhood.That is, although visuospatial working memoryappears to decline at a faster rate than verbalworking memory during adulthood (e.g., Cor-noldi & Vecchi, 2003; Myerson et al., 1999;

Myerson, Emery, et al., 2003), there appears tobe no decline in the ability to switch betweentasks, at least as measured by its effects on mem-ory span. The substantial decline in memory forlocations appears to be a specific instance of amore general age-related decline that affects avariety of visuospatial functions, including pro-cessing speed and learning ability as well asworking memory (e.g., Hale & Myerson, 1996;Jenkins et al., 2000; Lawrence et al., 1998). Thisdifferential visuospatial decline may itself be aninstance of a more general principle in whichthe later a function is acquired, the more sen-sitive it is to disruption, including disruption byage-related declines in basic neural processes.

The kind of information to be rememberedalso plays an important role in individual dif-ferences in working memory ability, just as itdoes in age differences. That is, the functional

distinction between verbal and visuospatialworking memory observed in experimentalstudies (e.g., Hale et al., 1996; Lawrence et al.,2001; Logie et al., 1990) appears to be importantin studies of individual differences in ability aswell (e.g., Hale, 2003; Shah & Miyake, 1996).Interestingly, working memory for numerical




items appears to be highly related to that forverbal items (e.g., Oberauer, Suß, Schulze,Wilhelm, & Wittman, 2000), a finding sug-gesting that people remember the names of

numbers rather than remembering more ab-stract concepts of quantities. The finding thatworking memory for syntactic information isdistinct from that for verbal items (Caplan &Waters, 1999; DeDe, Caplan, Kemtes, & Waters,2004; see Chapter 11) provides further evidenceof the important role played by content in thestructure of working memory.

Consideration of Other Sourcesof Variation in Working Memory

In addition to age and domain, processing speedis another important source of variation inworking memory. Indeed, age-related variationin speed underlies much of the age-relatedvariation in working memory. Correlations be-tween speed and working memory factors areparticularly strong in cross-sectional data fromsamples representing a large age range (e.g.,

Fry & Hale, 1996; Salthouse, 1996). The cor-relations observed in same-age samples are typ-ically not quite as strong (e.g., Ackerman, Beier,& Boyle, 2002), although the difference may bedue to restriction of range. The marked atten-uation of correlations between age and workingmemory when speed is statistically controlled isconsistent with developmental cascade modelsin which age-related changes in speed lead di-rectly to changes in working memory capacity(Fry & Hale, 1996; Salthouse, 1996). Jarroldand Bayliss (Chapter 6) found that speed wasrelated to complex but not simple spans in el-ementary-school children, but our data from asimilar age group indicate that speed accountsfor age differences in both types of spans. Im-portantly, there is general agreement that to-gether, speed and simple span explain much of the age-related variance in complex span, and

this conclusion appears to hold for the wholedevelopmental period from the age at whichchildren start elementary school through toyoung adulthood.

As Fry and Hale (1996) showed, however,age-related differences in speed do not com-pletely explain age-related differences in work-

ing memory in school-age children and youngadults. Similarly, the gradual slowing with ageof cognitive processing in adults does not com-pletely explain the age-related decline in adults’

working memory (Verhaeghen & Salthouse,1997). Similar views are expressed by Towseand Hitch (Chapter 5), who also found strongcontributions of processing time on workingmemory spans, but acknowledged that they donot completely explain either age or individualdifferences in memory span.

In general, our findings on the effects of same-domain and different-domain secondarytasks provide little support for the hypothesis

that executive processes draw on a general (i.e.,domain-independent) attentional resource (e.g.,Baddeley, 1986; Kane & Engle, 2002; Chapter2). Indeed, Oberauer et al. (Chapter 3) arguethat different executive processes share littlecommon variance. Moreover, neither of theexecutive processes that have arguably the mostface validity (i.e., switching between tasks andthe ability to resist interference) is strongly re-lated to working memory capacity (Oberauer,

Lange, & Engle, 2004; Chapter 3). Rather thansupporting the role of executive processes, ourfindings are more consistent with the hypoth-esis that interference results when primary andsecondary tasks engage the same neural sys-tems, as when secondary tasks requiring vi-sual orienting are paired with visuospatial spantasks (Corbetta, Kincade, & Shulman, 2002;Smith & Jonides, 1999). This is not to say thatwhen interference is observed, the neural sys-tems engaged by the primary and secondarytasks overlap completely, merely that there issome overlap, and that it is this overlap thatproduces the interference.

Another possible explanation couched inmore cognitive terms that might account for ourresults is that, as Nairne (2002) has suggested,interference results from overwriting due tosimilarity between primary and secondary task

cues, rather than from decay of memory tracesduring performance of the secondary task or fromdepletion of domain-independent resources byhaving to coordinate primary- and secondary-taskoperations. As Baddeley (1996) has suggested,however, what is termed the central executivemay not be a single, unified system but a set of




separate, interacting control functions, and thusit may be premature to draw conclusions aboutexecutive functions in general.

Contributions to General WorkingMemory Theory

Our developmental research has taught us is tobe wary of assuming that one has a good un-derstanding of how young adults perform anytask or battery of tasks, and this is true evenwhen, at first blush, the nature of young-adultperformance seems obvious. When we exam-ined our first data from young adults perform-

ing two different types of primary memory spantasks (verbal and visuospatial) under three dif-ferent secondary-task requirement conditions(i.e., none, same-domain, different-domain),we were delighted to obtain clear evidence of domain-specific interference. The fact that wedid not obtain any different-domain interfer-ence (i.e., verbal interference with visuospatialspan, or visuospatial interference with verbalspan) seemed fortuitous, because it meant that if

we did observe different-domain interference inother groups (e.g., older adults), we would beable to clearly interpret the findings as reflectingexecutive deficits.

Over the years, with systematic replicationafter replication in which not only young adultsbut also most school-age children and healthyolder adults have failed to show different-domain interference, we have come to believethat the domain-specificity of secondary-taskinterference has a deeper meaning. Ratherthan making it possible to find central executivedeficits easily, the relative lack of different-domain interference in healthy individuals be-tween 10 and 80 years of age indicates that trulygeneral (i.e., domain-independent) executivefunctions such as task switching simply may notplay a major role in determining memory span.Thus, although task switching and attention

switching may take time (e.g., Garavan, 1998;Rogers & Monsell, 1995), they do not neces-sarily interfere with working memory, unlessthe shifts are within a domain. For example,shifts of spatial attention affect memory for lo-cations, but not memory for letters (Lawrenceet al., 2004).

Inhibition is another putatively general-executive function that appears to play little rolein determining memory span, at least with ourprocedures. Antisaccades (often assumed to be a

good index of central-executive function becausethey require the inhibition of a pre-potent re-sponse) lead primarily to domain-specific inter-ference. That is, such saccades appear to interfereprimarily with the performance of visuospatialtasks, and the interference they generate is nogreater than the interference produced by othertypes of eye movements (Lawrence et al., 2001).

Again, however, it is important to note thatcentral-executive functions may not be a single,

unified system but a set of independent, inter-acting control functions (Baddeley, 1996). Cau-tion may be especially warranted in the case of inhibition, where empirical studies have shownthat putative measures of inhibition are oftenpoorly correlated (Kramer, Humphrey, Larish, &Logan, 1994; Shilling, Chetwynd, & Rabbitt,2002). Thus our findings on the antisaccade taskmay not necessarily be generalized to other sit-uations involving inhibition.

There is one final implication that maybe drawn from our findings on the domain-specificity of interference by secondary tasks thatapplies particularly to cognitive neuroscience,although as our knowledge in this area advancesit will obviously have implications for generalcognitive theory (e.g., Kane & Engle, 2002). If we are correct in thinking that interference re-sults when primary and secondary tasks engagethe same neural systems, then determining theoverlap in activation in neural imaging studiesmay be as important as determining which areasare uniquely activated by specific tasks (e.g., Awh& Jonides, 2001; Corbetta et al., 2002; Smith &Jonides, 1999). The same implication applies toneurophysiologic research on neural activity atthe single-unit level, where studying the neuronsin a particular brain area whose activity is mod-ulated by the same stimulus characteristics as

those in other areas may be as important as study-ing those neurons whose activity is modulated byunique stimulus characteristics (e.g., Snyder,Batista, & Andersen, 2000). Indeed, Oberaueret al. (2004) have pointed to the need for theo-ries of interference that indicate more preciselythe conditions under which task similarity affects




working memory. The present suggestion re-garding overlap in neural activation answers theircall. Of particular interest for both cognitiveneuroscience and general cognitive theory will

be whether the degree of overlap in activation by

different tasks predicts the extent to which thesetasks will interfere with each other, and whetherage and individual differences in the degree of overlap can predict age and individual differ-

ences in interference.



OF WORKING MEMORY

Baddeley’s (1986) model of working memory

with separate verbal and visuospatial subsys-

tems has served us well in our research. Gener-

ally, our results have been consistent with this

model, although we find that secondary tasks

interfere with memory span only when they

engage the same domain-specific subsystem as

that of the primary memory task, rather than

whenever they engage domain-independent ex-

ecutive functions. Recently, we have used pro-

cedures that explicitly involve manipulating

memory items, consistent with Baddeley’s orig-inal definition of working memory. These pro-

cedures (e.g., letter-number sequencing) provide

an important complement to procedures in

which secondary tasks require processing irrel-

evant information (e.g., operation span).



We have observed primarily quantitative dif-ferences across the life span from age 10 years

onwards. Prior to age 10, the verbal and visuo-

spatial subsystems are not yet completely in-

dependent. Following general improvements in

late childhood and adolescence, older adults

show a general decline that is exacerbated in

the visuospatial domain. Chronological age is

only a proxy, however, for neurobiological

changes, especially those involving white mat-

ter because of its role in processing speed.Changes in processing speed affect working

memory function at both ends of the life span,

but the changes during senescence do not

appear to be a simple mirror image of those

during maturation.


MEMORY VARIATION

Processing speed is an important source of

variation that appears to underlie much of

the age-related variation in working mem-

ory. We have shown, however, that processing

speed and storage do not completely explain

all of the age-related variance in children’s

complex memory span. We have considered,

but rejected, roles for two important execu-

tive processes: the ability to inhibit irrele-

vant information and the ability to switch

back and forth between tasks. Instead, our

findings are more consistent with the hypoth-esis that, across most of the life span, interfer-

ence results when primary and secondary tasks

engage the same domain-specific neural sys-

tems.



Perhaps our strongest contribution to generalworking memory theory stems from a finding

that we first observed in young adults: Only

secondary tasks involving the same domain

interfere with performance of a primary mem-

ory task. The relative lack of different-domain

interference in cognitively healthy individuals

between 10 and 80 years of age indicates that

truly general (i.e., domain-independent) ex-

ecutive functions such as task switching do

not play a major role in determining memoryspan. Thus, although task switching and at-

tention switching take up processing time,

they do not necessarily interfere with working

memory except when the shifts are within a

domain.




Hale, S. (1990). A global developmental trend in

cognitive processing speed. Child Development,

61, 653–663.

Hale, S. (May 2003). Short-term memory, working

memory, and fluid intelligence in older adults.Presented at the American Psychological Soci-

ety, Atlanta, GA.

Hale, S., Bronik, M. D., & Fry, A. F. (1997). Verbal

and spatial working memory in school-age

children: Developmental differences in suscep-

tibility to interference. Developmental Psychol-

ogy, 33, 364–371.

Hale, S., Fry, A. F, & Jessie, K. A. (1993). Effects

of practice on speed of information processing

in children and adults: Age-sensitivity and age-invariance. Developmental Psychology, 29, 880–

892.

Hale, S., Lima, S. D., & Myerson, J. (1991). General

cognitive slowing in the nonlexical domain: An

experimental validation. Psychology and Aging,

6, 512–521.

Hale, S., & Myerson, J. (1995). Fifty years older, fifty

percent slower? Meta-analytic regression mod-

els and semantic context effects. Aging and Cog-

nition, 2, 132–145.Hale, S., & Myerson, J. (1996). Experimental

evidence for differential slowing in the lexical

and nonlexical domains. Aging, Neuropsychol-

ogy, and Cognition, 3, 154–165.

Hale, S., Myerson, J., Rhee, S. H., Weiss, C. S., &

Abrams, R. A. (1996). Selective interference with

the maintenance of location information in work-

ing memory. Neuropsychology, 10, 228–240.




psychology of learning and motivation: Advances

in research and theory (Vol. 22, pp. 193–225).

San Diego: Academic Press.

Jenkins, L., Myerson, J., Hale, S., & Fry, A. F.

(1999). Individual and developmental differ-

ences in working memory across the life span.


Jenkins, L., Myerson, J., Joerding, J. A., & Hale, S.(2000). Converging evidence that visuospatial

cognition is more age-sensitive than verbal cog-

nition. Psychology and Aging, 15, 157–175.

Kail, R. (1992). Processing speed, speech rate, and

memory. Developmental Psychology, 28, 899–

904.

Kail, R., & Park, Y.-S. (1994). Processing time, artic-

ulation time, and memory span. Journal of Ex-

perimental Child Psychology, 57, 281–291.

Kail, R., & Salthouse, T. A. (1994). Processing speed

as a mental capacity. Acta Psychologica, 86,199–225.



of working-memory capacity. Journal of Experi-



prefrontal cortex in working-memory capacity,

executive attention, and general fluid intelli-

gence: An individual differences perspective.

Psychonomic Bulletin & Review, 9, 637–671.Kotary, L., & Hoyer, W. J. (1995). Age and the

ability to inhibit distractor information in visual

selectiveattention. ExperimentalAgingResearch,

21, 159–171.

Kramer, A. F., Humphrey, D. G., Larish, J. F., &

Logan, G. D. (1994). Aging and inhibition:

Beyond a unitary view of inhibitory processing

in attention. Psychology and Aging, 9, 491–512.

Lawrence, B., Myerson, J., & Hale, S. (1998).

Differential decline of verbal and visuospatialprocessing speed across the adult life span.

Aging, Neuropsychology, and Cognition, 5, 129–

146.

Lawrence, B. M., Myerson, J., & Abrams, R. A.

(2004). Interference with spatial working mem-

ory: An eye movement is more than a shift of

attention. Psychonomic Bulletin & Review, 11,

488–494.

Lawrence, B. M., Myerson, J., Oonk, H. M., &

Abrams, R. A. (2001). The effects of eye and

limb movements on working memory. Memory,

9, 433–444.

Lima, S. D., Hale, S., & Myerson, J. (1991). How

general is general slowing? Evidence from the

lexical domain. Psychology and Aging, 6, 416–

425.

Logie, R. H., Zucco, G. M., & Baddeley, A. D.

(1990). Interference with visual short-term mem-

ory. Acta Psychologica, 75, 55–74.Miyake, A., Friedman, N. P., Rettinger, D. A., Shah,

P., & Hegarty, M. (2001). How are visuospa-

tial working memory, executive function, and



eral, 130, 621–640.




Moscovitch, M., & Winocur, G. (1995). Frontal

lobes, memory, and aging. In J. Grafman, K. J.

Holyoak, & F. Boller (Eds.), Structure and

functions of the human prefrontal cortex (pp.

119–150). New York: New York Academy of Sciences.

Myerson, J., Adams, D. R., Hale, S., & Jenkins, L.

(2003). Analysis of group differences in proces-

sing speed: Brinley plots, Q-Q plots, and other

conspiracies. Psychonomic Bulletin & Review,

10, 224–237.

Myerson, J., Emery, L., White, D. A., & Hale, S.

(2003). Effects of age, domain, and processing

demands on memory span: Evidence for dif-

ferential decline. Aging, Neuropsychology, andCognition, 10, 20–27.

Myerson, J., Ferraro, F. R., Hale, S., & Lima, S. D.

(1992). General slowing in semantic priming

and word recognition. Psychology and Aging, 7,

257–270.

Myerson, J., Hale, S., Rhee, S. H., & Jenkins, L.

(1999). Selective interference with verbal and

spatial working memory in young and older

adults. Journal of Gerontology: Psychological

Sciences and Social Sciences, 54B, P161–P164.Myerson, J., Jenkins, L., Hale, S., & Sliwinski, M.

(2000). Stocks and losses, items and interfer-

ence: A reply to Oberauer and Suß (2000).


Nairne, J. S. (2002). Remembering over the short-

term: The case against the standard model.

Annual Review of Psychology, 53, 53–81.

Oberauer, K., Lange, E., & Engle, R. W. (2004).

Working memory capacity and resistance to

interference. Journal of Memory and Language,

51, 80–96.

Oberauer, K., & Suß, H. M. (2000). Working

memory and interference: a comment on Jen-

kins, Myerson, Hale, and Fry (1999). Psycho-

nomic Bulletin & Review, 7, 727–733.

Oberauer, K., Suß, H. M., Schulze, R., Wilhelm, O.,

& Wittman, W. W. (2000). Working memory

capacity—facets of a cognitive ability construct.

Personality and Individual Differences, 29, 1017–1045.

Paus, T., Collins, D. L., Evans, A. C., Leonard, G.,

Pike, B., & Zijdenbos, A. (2001). Maturation of

white matter in the human brain: A review of

magnetic resonance studies. Brain Research

Bulletin, 54, 255–266.

Peters, A. (2002). The effects of normal aging on

myelin and nerve fibers: A review. Journal of

Neurocytology, 31, 581–593.

Postle, B. R., Awh, E., Jonides, J., Smith, E. E., &

D’Esposito, M. (2004). The where and how of attention-based rehearsal in spatial work-

ing memory. Cognitive Brain Research, 20,

194–205.

Psychological Corporation (1997). WMS-III admin-

istration and scoring manual. San Antonio, TX:

Harcourt Brace & Company.

Richardson, J. T., Engle, R. W., Hasher, L., Logie,

R. H., Stolzfus, E. R., & Zacks, R. T. (1996).

Working memory and human cognition. New

York: Oxford University Press.Rogers, R. D., & Monsell, S. (1995). Costs of a pre-

dictable switch between simple cognitive tasks.


124, 207–231.




Schweickert, R. (1993). A multinomial processing

tree model for degradation and reintegration in

immediate recall. Memory & Cognition, 21,168–175.






Shilling, V. M., Chetwynd, A., Rabbitt, P. M. A.

(2002). Individual inconsistency across mea-

sures of inhibition: An investigation of the con-

struct validity of inhibition in older adults.

Neuropsychologia, 40, 605–619.

Smith, E. E., & Jonides, J. (1999). Storage and

executive processes in the frontal lobes. Science,

283, 1657–1661.

Snyder, L. H., Batista, A. P., & Andersen, R. A.

(2000). Intention-related activity in the poste-

rior parietal cortex: A review. Vision Research,

40, 1433–1441.

Underwood, W. D. (1975). Individual differencesas a crucible in theory construction. American

Psychologist, 30, 128–134.

Verhaeghen, P., & Cerella, J. (2002). Aging,

executive control, and attention: A review of

meta-analyses. Neuroscience and Biobehavioral

Reviews, 26, 849–857.




Verhaeghen, P., & Salthouse, T. A. (1997). Meta-

analyses of age-cognition relations in adulthood:

Estimates of linear and nonlinear age effects and

structural models. Psychological Bulletin, 122,

231–249.Washburn, D. A., & Astur, R. S. (1998). Nonverbal

working memory of humans and monkeys:

Rehearsal in the sketchpad? Memory & Cogni-

tion, 26, 277–286.

West, R. (2000). In defense of the frontal lobe hy-

pothesis of cognitive aging. Journal of the Inter-

national Neuropsychological Society, 6, 727–

729.

Zacks, R. T., Radvansky, G., & Hasher, L. (1996).Studies of directed forgetting in older adults.

Journal of Experimental Psychology: Learn-

ing, Memory, and Cognition, 22, 143–

156.




III

Working Memory Variation Due

to Normal and Pathological Aging



9

Inhibitory Mechanisms and

the Control of Attention

LYNN HASHER, CINDY LUSTIG, and ROSE ZACKS

‘‘Bigger is better.’’ So goes the message of manytheoretical perspectives on working memory,views that emphasize working memory as a‘‘mental workspace’’ which houses the represen-tations and processes that, at any given moment,are in the focus of attention. The intuition of such views is that the larger this workspace, orthe more representations one can have active atany given time, the better performance will beon most cognitive and social tasks.1 Virtually allviews of working memory share this perspective,including Baddeley’s (1986, 1992, 2000) andJust and Carpenter’s (1992). We describe an al-ternative view, one that could be described as‘‘good things come in small packages.’’ Our workas well as that of our collaborators focuses on theexecutive control processes that keep the mental

representation ‘‘packages’’ small and goal rele-vant. This, we have argued, enables a maximallyefficient information-processing system (e.g.,Hasher & Zacks, 1988; Hasher, Zacks, & May,1999).

Our focus is on a set of attentional or execu-tive control processes, all inhibitory, that operate

in the service of an individual’s goals to narrowand constrain the contents of consciousness tobe goal relevant. An uncluttered or narrowlyfocused ‘‘working memory,’’ rather than a largeone, is the ideal processing system: it will befaster to achieve a goal than will a more broadlydispersed system because it will not be slowed byirrelevant stimuli that occur in the task context,or by environmentally triggered thoughts, or byself-generated distraction. The narrow focusmaximizes the speed and accuracy of on-lineprocessing because it reduces the likelihood of switching attention to goal-irrelevant represen-tations such as those connected to a previoustask, an upcoming task, environmental distrac-tion, or subsidiary goals.

A narrowly focused processing system is also

ideal because it has the downstream benefitof ensuring accurate and rapid retrieval of theinformation it once focused on (Anderson &Bower, 1973). This claim follows from a richliterature pointing to substantial costs for retrievalof having entertained irrelevant informationduring encoding. Sometimes the irrelevant

227



information is explicitly part of the task environ-ment, as is the case, for example, when highlysimilar or overlapping information is learned(Anderson & Bower, 1973; Watkins & Watkins,

1976), when encoding takes places under divid-ed-attention conditions (Craik, Govoni, Naveh-Benjamin, & Anderson, 1996; Fernandes &Moscovitch, 2000), or when actually presentedinformation triggers activation of related infor-mation (e.g., Deese, 1959; Roediger & McDer-mott, 1995; Underwood, 1965). Whatever thesource, any additional information activated dur-ing encoding ‘‘enriches’’ the memory represen-tation of presented items andforms thebasis from

which intrusions are drawn and memory lapsesoccur, the latter due to fundamental interferenceprocesses. We note that all taskswhich depend onrapid and accurate retrieval of information thatwas once attended to will suffer to the degree towhich the processing system was initially broadly,rather than narrowly, tuned at encoding.

The detrimental effects of an ‘‘embarrass-ment of riches’’—i.e., of having too much infor-mation activated and in the focus of attention—

have been the primary interest of our researchprogram, rather than working memory per se.To that end, we have explored the nature of theinhibitory attentional-control processes thatlimit the momentary consideration of irrelevantinformation. We have also explored the impor-tance of these attentional-regulation processes toa wide variety of cognitive tasks, including (a)traditional working memory tasks (e.g., simplespan, verbal and visuospatial working memoryspan; Lustig & Hasher, 2002; Lustig, May, &Hasher, 2001; May, Hasher, & Kane, 1999;Rowe, Turcotte & Hasher, 2006); (b) basic-levelperceptual speed tasks used in the intelligence,developmental, and aging literatures (Lustig,Hasher, & Tonev, in press); (c) more conceptualtasks such as reading speed and reading com-prehension (Carlson, Hasher, Connelly, &Zacks, 1995; Connelly, Hasher, & Zacks, 1991;

Li, Hasher, Jonas, Rahhal, & May, 1998),problem solving and decision making (May,1999; Tentori, Osherson, Hasher, & May,2001); (d) attentional regulation (May, Kane, &Hasher, 1995) and control of primed or prepo-tent (but task-irrelevant) responses (Butler,Zacks, & Henderson, 1999; May & Hasher,

1998); and (e) long-term explicit and implicitmemory (Gerard, Zacks, Hasher, & Radvansky,1991; Kim, Hasher & Zacks, in press; Lustig &Hasher, 2001; Rowe, Valderrama, Lenartowicz

& Hasher, in press; Zacks, Radvansky, &Hasher, 1996).Our particular focus on these inhibitory-

based executive control processes differs frommuch of the early work on working memory,which centered on capacity for simultaneousmental operations and storage. Our emphasison executive processes fits well, however, withthe recent explosion of work on ‘‘executive con-trol’’ across the cognitive and cognitive neuro-

science literatures, including evidence that thecontrol processes involved in attention, work-ing memory, and long-term memory share com-mon neural substrates (Cabeza et al., 2003;Ranganath, Johnson, & D’Esposito, 2003). Re-cent work by Engle and colleagues has a similarperspective to our own, as their emphasis hasshifted toward working memory as an executiveattention system rather than as a ‘‘memory’’ sys-tem (e.g., Engle, 2002; Kane, Bleckley, Con-

way, & Engle, 2001; see also Chapter 2, thisvolume). Finally, our work also fits well with thegeneral effort to understand the processes in-volved in Baddeley’s construct of the ‘‘centralexecutive’’ (e.g., Baddeley, 2003).

In sum, our work is similar to that of manyother investigators in its focus on executive pro-cesses as a critical source of working memoryvariation as well as variation in many cognitivedomains. There is broad agreement that for in-dividuals sharing common goals, it is the effi-ciency of executive processes that is a majorsource of variation in the contents of conscious-ness and in many of the mental and physicalprocesses (e.g., memory and motor control)that are subsequently determined by the initialbreadth of focus (e.g., see Chapters 2 and 4, butsee Chapter 10 for a somewhat different view).

Our work differs from many others’ in that we

emphasize the role of inhibitory processes, orthose processes that keep consciousness free of irrelevant information that can impede the suc-cessful and efficient completion of a currentgoal. Our assumption is that the initial stageof activating representations is largely auto-matic and is driven by environmental and social

228 Variation Due to Normal and Pathological Aging



have taken two approaches: (1) the study of age-related inhibitory control deficits (i.e., groupdifferences), and (2) the study of inhibitorycontrol across the day (i.e., intra-individual dif-

ferences). The latter line of work is quite un-usual in mainstream cognition, but, as will beseen, it leads nicely to the conclusion that in-hibitory control can vary not just among groupsand individuals within a group but within indi-viduals themselves.

Because the intra-individual-differences ap-proach we take is an unusual one, we describe itbriefly before returning to elaborate on inhibi-tion. Our studies typically compare participants

with a particular type of circadian arousal pat-tern (Evening types and Morning types; Horne& Ostberg, 1976) who are tested early in themorning and late in the afternoon to providea snapshot of fundamental cognitive processesfunctioning across the day (see Winocur &Hasher,2002, for a brief review of related animalmodel evidence). Despite the folk nomencla-ture of these two ‘‘types,’’ they are well substan-tiated in physiology (e.g., Kerkhof & Lancel,

1991), including recent evidence of geneticcontributions to extremes in arousal patterns(e.g., Cermakian & Boivin, 2003; Hur, Bou-chard, & Lykken, 1998; see also the final sectionof this chapter). We also note that there are lifespan differences in overall arousal patterns, withmore than 70% of older adults (and many youngchildren) more likely to be at a peak in themorning than later in the day. This same time islikely a trough for many young adults, under10% of whom have Morning-type arousal pat-terns (Kim, Dueker, Hasher, & Goldstein, 2002;

Yoon, May, & Hasher, 2000; Yoon, May, Gold-stein & Hasher, in press).3

Of particular importance for present pur-poses, the data suggest that regardless of whetherone is a Morning- or Evening-type person, it isinhibitory processes that differ most across thecircadian cycle; excitatory-based processes seem

to show little variation across the day (e.g., Yoonet al., 2000). Our evidence suggests that inhibi-tory efficiency follows the arousal cycle andour assumption is that studying groups and in-dividuals with varying degrees of inhibitoryfunction (or single individuals at different pointsin their circadian arousal function) will help il-

luminate inhibitory processes and the roles theyplay in cognition, including in determiningapparent differences in working memorycapacity.

The decision to highlight inhibitory pro-cesses as critical determinants of both on-lineand downstream efficiency places us within asmall group of investigators who early on andquite independently looked at cognition from asimilar point of view and across many differentgroups and individuals (initially, Gernsbacher& Faust, 1991, and Dempster, 1991). Many in-vestigators subsequently arrived at similar orpartially overlapping views (see Duchek, Balota,

& Thessing, 1998; Harnishfeger & Bjorklund,1994; Nigg, 2001; Wenzlaff & Wegner, 2000;Chapters 2, 3, 4, and 10, this volume). By fo-cusing specifically on the role of inhibitoryprocesses, we differ somewhat from other inves-tigators who deal with executive or controlledattention processes in a generalized fashion,or who tie them to particular tasks such as setswitching.

As more researchers focus on executive con-

trol processes, it has become increasingly clearthat ‘‘executive control’’ is not a unitary con-struct and that the nature of the specific pro-cesses remains to be understood (e.g., Friedman& Miyake, 2004; Miyake et al., 2000; Sylvesteret al., 2003). Although in its early stages, recentwork suggests that different aspects of executivecontrol (which we view largely as different as-pects of inhibitory function) may be dissociableacross individuals, brain regions, and times of day (e.g., Friedman & Miyake, 2004; Lustig &Meck, 2001; May, Hasher, & Foong, 2005;Sylvester et al., 2003; West, Murphy, Armilio,Craik, & Stuss, 2002).

Our own attempts to understand the natureof inhibitory processes and their contributionsto performance led us to a framework that drawsdistinctions between three separate functions of inhibition, all of which serve to keep working

memory (i.e., the focus of attention) free of ir-relevant information (e.g., Hasher et al., 1999;Zacks & Hasher, 1994). Inhibitory processes actin the service of goals to (1) prevent irrelevantinformation from gaining access to the focus of attention, (2) delete no-longer relevant itemsfrom consideration, and (3) restrain prepotent




responses so that other, initially weaker responsecandidates can be evaluated and influence be-havior as appropriate for current goals.

In the next section of the chapter, we briefly

describe inhibitory functions and the effects thatvariations in their efficiency can have on per-formance on a number of tasks with an empha-sis on speed of processing and working mem-ory. We suggest that a view of executive controlthat focuses on inhibitory processes can offer acompeting account for group, individual, andintra-individual differences in speed and work-ing memory (among other cognitive phenome-non) without appealing to notions of ‘‘capacity’’

that in the attention literature have been sharplycriticized (e.g., Navon, 1984). Indeed, our evi-dence raises the possibility that what mostworking memory span tasks measure is inhibi-tory control, not something like the size of op-erating capacity (e.g., Just & Carpenter, 1992).In the final section, we discuss the potentialneurobiological underpinnings of the age andcircadian changes that have profound behavioraleffects on inhibitory regulation.

INHIBITORY PROCESSES

We have posited three inhibitory functions:access, deletion, and restraint (Hasher et al.,1999; Hasher, Tonev, Lustig, & Zacks, 2001;Hasher & Zacks, 1988; Zacks & Hasher, 1994).Each is a powerful player in determining thespeed and success of on-line processing. Two of them (access and deletion) are also major de-terminants of the speed and success of explicitretrieval while the third (restraint) can influ-ence successes, for example, when strong orprepotent responses are correct (e.g., stoppingat a traffic signal when it is red), and failures,when strong responses are wrong (cf. Radvansky& Curiel, 1998). Over the past 20 years ourwork has focused on exploring the nature of

these inhibitory functions and showing thatthey (a) operate across a wide range of tasks, (b)diminish with age over adulthood, and (c) varyacross the day with an individual’s circadianarousal pattern. It is important to note thatalthough our work takes a group- and intra-individual-differences approach, the theory be-

hind the work is a general theory of cognitionand, as such, applies to individual differences.

Access

The initial activation of representations is pre-sumed to be broad and virtually automatic. Theaccess function of inhibition is engaged in theservice of goals to determine which activatedrepresentations enter the focus of attention (e.g.,Cowan, 1993). When efficient, all irrelevantrepresentations are suppressed and the contentsof consciousness will be narrowly tied to goals. A dramatic example of narrow focus of attention is

the ‘‘inattentional blindness’’ effect, in whichunattended items in the center of the visual fieldare literally not ‘‘seen’’ (Mack & Rock, 1998;Most et al., 2001). Another is the state of ‘‘flow’’by which intense concentration enables indi-viduals to ignore the external world and passingof time (e.g., Csikszentmihalyi, Rathunde, &Whalen, 1993).

Our original work on the inferences gener-ated while reading suggested age differences

in the amount of information that gains access tothe focus of attention; as noted above, in an am-biguous context in which young adults gener-ated only one interpretation, older adults gen-erated more (Hasher & Zacks, 1988, Hamm &Hasher, 1992, Kim, Hasher, & Zacks, in press).Our recent work on the access function has fo-cused on its role in determining the speed withwhich tasks can be performed. To this end, wemanipulated the extent and nature of extrane-ous information present in a task environment.

For example, for older adults, the speed atwhich a decision is made about two letterstrings (e.g., XPFGN and XPFCN) being thesame or different is at least partially determinedby whether there are other letter strings si-multaneously present and competing for accessto attention (Lustig et al., in press). For youngadults, the presence of other letter-string prob-

lems has no effect on the speed at whichproblems are solved. These findings are critical,because letter comparison is one of a numberof tasks used to assess the notion of ‘‘percep-tual speed,’’ a concept that in the life-span andintelligence literatures is thought of as a cog-nitive primitive that establishes limits to an

Inhibitory Mechanisms and the Control of Attention 231



individual’s performance across a range of high-level cognitive tasks, including reasoning(Kail, 1993; Salthouse, 1996). As it happens,most tasks that assess ‘‘perceptual speed’’ use

highly cluttered displays (with many similarproblems on a page), an arrangement likely todisrupt the performance of some participants(those with reduced inhibitory function), butnot others. Our work suggests that the source of disruption (and the underlying cognitive prim-itive) is the access function which determinesthe ability to constrain task focus to just themomentarily relevant item.

Inefficient control over access can also slow

even highly practiced skills such as reading. Forexample, interspersing irrelevant words (in adistinctive font) amidst target text differentiallyslows reading for older adults (Carlson et al.,1995; Connelly et al., 1991; Duchek, et al.,1998; Dywan & Murphy, 1996; Li et al., 1998;Phillips & Lesperance, 2003). There are com-parable data showing age differences in dis-ruption effects when the distraction is in theauditory rather than visual mode (Tun, O’Kane,

& Wingfield, 2002). The selective-attentionliterature shows a similar phenomenon: undermany circumstances, older adults are differen-tially slowed to find a target amidst distraction(e.g., Plude & Hoyer, 1986; Zacks & Zacks,1993). From our perspective, all of these effectsare consistent with the idea that the accessfunction of inhibition is not as efficient for olderadults as it is for younger adults. These findings,and particularly those using simple perceptual-speed tasks, pose a challenge to views of pro-cessing speed as a cognitive primitive that un-derlies intelligence and the developmentaltrajectory of cognition across the life span (e.g.,Kail, 1993; Salthouse, 1996). Such findingssuggest that attentional regulation and particu-larly the access function of inhibition are part of theunderlyingmechanisms critical tocognition.

We note that the efficiency of the access

function also varies across the day in patternsconsistent with morning vs. evening arousalschedules. In one relevant study (May, 1999),participants were given a variant of the classicRemote Associates Task in which three veryloosely related words (e.g., rat, blue, and cot-

tage) were presented and the task was to gen-erate a word (cheese) that connects them. Thetarget words were presented alone on controltrials and with distraction on experimental tri-

als; participants were warned to ignore thedistraction that had been normed to either leadtoward the solution word or away from it.

Evening-type younger and Morning-typeolder adults were tested early in the morningor late in the afternoon. Performance on thecontrol (or distraction-free) sets did not differwith age or time of testing; however, the impactof distraction differed for both ages and timesof testing. Young adults were completely able

to ignore the distraction when tested in theafternoon, an effect similar to that seen in theinattentional blindness phenomenon (Mack &Rock, 1998; Most et al., 2001) and in percep-tual speed tasks completed in the presence orabsence of distraction (Lustig et al., in press). Inthemorning, however, young adults showed reli-able costs and benefits as the distraction ‘‘leaked’’in to influence performance. Thus, for youngadults, control over the access function is more

efficient in the afternoon than in the morning, apattern consistent with their Evening-arousaltypology.

For older adults, distraction is not ignored, ithelps or hurts performance at both testing times,but more so in the afternoon than in the morn-ing. The results for the older adults (and foryoung adults tested in the morning) cannot eas-ily be written off as ‘‘general performance defi-cits,’’ since baselines were equivalent when nodistraction was present. Instead, the extraneousinformation sometimes led to greater costs butalso to greater benefits, depending on the type of distraction that was present. What remainedconstant was the older adults’ relative failure torestrict attention away from the distractor items,consistent with the assumption that control overthe access function diminishes with age. Thisfailure also held for everyone tested at subopti-

mal times of day.More recent work shows that the conse-quences of a failure to control distractionare not just immediate, but can also impacton ‘‘downstream’’ performance 15 or 20 min-utes after initial exposure to the distraction




(Kim et al., in press; Rowe, Valderrama, Le-nartowicz & Hasher, in press). Furthermore,these experiments show ‘‘far transfer’’ effects,such that distraction in the context of one task

can influence performance on very differentsubsequent tasks. Indeed, in these two uniquecircumstances, the data show greater benefits forolder as compared to younger adults, rather thanthe typically seen greater costs. As well, thebenefits are greater at nonoptimal times thanat optimal times (Rowe et al., in press).

As an aside, what the data also show is thatfailing to attend to the time at which youngerand older adults are tested is probably a major

mistake, since more than 70% of older adultsare Morning types and a third or more of youngadults in university settings are Evening types(see, e.g., May, Hasher, & Stoltzfus, 1993; Yoonet al., 2000; Yoon et al., in press). If early-morning testing times are not used, and mostparticipants are tested later in the day (see Mayet al., 1993), age differences in access controlwill be exaggerated. As subsequent data show,this argument can likely be extended to the two

other inhibitory functions (deletion and re-straint) and, critically, the argument can alsobe extended to other cognitive tasks that haveinhibitory components.

In sum, our work and that of others suggeststhat across many situations, the ability to keepattention focused away from irrelevant infor-mation aids the fast and accurate processing of goal-relevant information. The access controlfunction influences performance on tests of processing speed, a construct often used alongor in competition with working memory capac-ity as an explanation for performance variationacross the life span (Park et al., 1996; Salthouse,1996), and on tests of reading and problem solv-ing, tasks often used as outcome measures instudies examining the predictive power of work-ing memory tasks (see review by Daneman &Merikle, 1996). From a theoretical perspective,

efficient inhibitory function is critical for con-trolling which pieces of information gain ac-cess to attention and, on the assumption thatco-occurrence is a major determinant of asso-ciation formation, how large the initial mem-ory bundles are. This in turn determines how

fast and accurate subsequent retrieval can be(Anderson & Bower, 1973). The impact of clut-tered or large memory bundles will be discussedfollowing the next section.

Deletion

Inhibition also serves to delete irrelevant infor-mation from the focus of attention. Irrelevantinformation may be active in the first instancebecause of the failure of the access function tocontrol ‘‘leakage’’ tied to subsidiary goals or to amismatch between the goals of an individualand those set by an experimenter or situation.

Deletion is critical for removing irrelevantrepresentations from the focus of attention soas to enable efficient processing of goal-drivenrepresentations. Deletion also removes once-relevant information that has become irrelevantbecause of a change in goals, context, task, orsituational demands, as can occur in a conver-sation when a topic changes, or in a task (whe-ther attention, memory, or problem solving)when one set of materials (or procedures) ends

and another begins. As noted earlier, the stimulus for this aspect of

our theoretical framework comes from the ob-servation that older adults not only allow alter-native interpretations of a passage to gain accessto their attention but also fail to delete those al-ternatives from consideration, even when it be-comes clear that they were incorrect (Hamm &Hasher, 1992; Hasher & Zacks, 1988). To es-tablish the generality of these initial findings, wecreated garden path sentences that ended with ahighly predictable but missing word that theparticipant generated and that was replaced, afew seconds later, by a less predictable word pro-vided by the experimenter. We then used animplicit task to measure the accessibility of theinitially generated word (the highly predictableending)—a word that became irrelevant in thecontext of the task as soon as the experimenter

provided an alternative ending to the sentence.We measured access to the no-longer relevantwords (and for other control items) for botholder and younger adults.

Across a series of studies, the ability to deletea no-longer relevant inference from memory




varied as a function of adult age and time of testing (e.g., Hartman & Hasher, 1991; May &Hasher, 1998; May, Zacks, Hasher, & Mul-thaup, 1999). For Evening-type young adults

tested in the afternoon (see May & Hasher,1998), deletion actually suppressed the no-longer relevant word to such a degree that sub-sequent use of those words to end new sentenceswas actually below baseline levels. Early in themorning, however, the availability of the no-longer relevant term was reliably above baselinelevels, showing time-of-day differences in theefficiency of the deletion function for youngadults that are consistent with their arousal pat-

tern. Older adults also showed time-of-day dif-ferences in deletion regulation, with worseperformance in the afternoon, consistent withtheir circadian arousal type. Overall, there wereprofound age and time-of-day differences in in-hibitory control over deletion.

Vulnerability to the effects of no-longer rel-evant information has been shown to vary acrossgroups and individuals who differ in readingability, in span scores, and on intelligence tests

(e.g., Chiappe et al., 2000, 2002; Dempster,1991; Gernsbacher & Faust, 1991; Kane &Engle, 2000). Note that if deletion is inefficient,the memory bundle representing a given eventor moment will consist of (at least) both relevantinformation and irrelevant information that re-mained active in consciousness, thus enabling an‘‘enriched’’ or cluttered memory bundle duringencoding. These larger bundles in turn result indifferentially poor retrieval (e.g., Anderson &Bower, 1973). In the next section, we considerthe impact of the deletion function on tasks in-tended to measure working memory.

Deletion and Working Memory Span

Working memory span tasks, including the by-now classic reading span task of Daneman andCarpenter (1980), typically present the partici-

pant with a series of ‘‘study’’ and recall test trials,each of which consists of a set of sentences tounderstand while preparing to recall the finalword of each, followed by an immediate recallof those final words. These sets vary in size (e.g.,from two to six sentences), and by convention(i.e., at least since the earliest IQ tests developed

by Binet) they are presented in an ‘‘ascending’’order so that the smallest sets are presented first.The largest set-size that a participant can reli-ably understand and for which all items in the

set can be recalled is a commonly used index of working memory capacity.The ascending administration requires dele-

tion to be efficient so that at any point in theseries of study trials consideration is narrowlyfocused on only the currently relevant set. If deletion is inefficient, items from prior sets will‘‘enrich’’ the memory representations of the cur-rent set, reducing the ability of participants torecall the current set accurately. The failure to

suppress no-longer relevant words enables pro-active interference (PI) to build up across trialsand to have its most detrimental effects on thelarge set-size trials that are last in the series yetcriticaltoattainingahighworkingmemoryscore.

On the basis of these observations of thetypical operations involved in assessing work-ing memory span, May, Hasher & Kane (1999)reversed the order of administration so that thelargest trials occurred first, before PI had a

chance to accumulate. This simple manipula-tion should have no effect on the measurementof working memory capacity per se, at least if capacity simply reflects the amount of infor-mation an individual can store and process intheir ‘‘mental workspace.’’ However, if deletion(and attendant PI) is involved in standard spantasks, the sequence manipulation should affecthow much irrelevant information is availableto that workspace from previous trials whenparticipants are attempting to recall the currentitems. Indeed, the reversed administration dra-matically improved the performance of olderadults on the reading span task, so that, ratherstartlingly, their performance no longer differedfrom that of young adults. (A more extreme ma-nipulation designed to reduce PI also improvedthe scores of young adults.) These findingssuggest that variation in deletion function

(or, inversely, in proactive interference causedby failures of the deletion function) plays amajor role in producing variation in workingmemory span (see also Bowles & Salthouse,2003).

Recent work suggests that this conclusionextends beyond the limits of the various ver-




sions of Daneman and Carpenter’s reading andlistening span tasks. The reversed-order ma-nipulation also increased the span scores of older adults on a Corsi-block version of a vi-

suospatial working memory span task (Roweet al., 2006). Bunting (2006) has shown thatthe operation span task introduced by Engleand colleagues (e.g., Engle, Cantor, & Carullo,1992), which is based on verifying the accuracyof equations while remembering words, is alsovulnerable to PI (see also Rowe et al., 2006).We have also found that circadian influencescan affect PI-heavy measures of working mem-ory span (Hasher et al., 2005; Yoon et al.,

2000), consistent with the conclusion that theefficiency of the deletion function varies acrossthe day. Taken together, these data suggest thatseveral of the most widely used versions of work-ing memory span are likely measuring some-thing other than capacity. We think it likelythat they index the efficiency of inhibitory as-pects of attention regulation.

Deletion may play a critical role not onlyin variability on working memory span tasks

per se, but also in those tasks’ ability to predictperformance on other measures. Lustig et al.(2001) replicated the May, Hasher & Kane(1999) results by showing that delivering a spantask in reverse order (so reducing PI) eliminatedage differences in working memory span per-formance, and further showed that the deletion-demanding aspects of the span task were criticalfor its ability to predict performance on proserecall (a standard outcome measure in the in-dividual difference tradition). For both youngerand older adults, manipulations that reduced PIand improved span scores also reduced theability of individual differences in span scores topredict individual differences in prose recall. Bybroad generalization, these data raise the possi-bility that whenever span tasks are used to selectparticipants to perform on other tasks andwhenever reliable correlations are obtained, the

mediating variable may well be inhibitory con-trol over nonrelevant information, not workingmemory capacity.

Further evidence that working memory spantasks do not measure capacity but instead some-thing like interference proneness comes from astudy demonstrating that prior experience with

other memory tasks can reduce estimates of thesize of an individual’s working memory span(Lustig & Hasher, 2002). Performance on otherretrieval tasks (e.g., paired associates and serial

learning) has long been known to be disruptedby prior laboratory experience (Greenberg &Underwood, 1950; Keppel, Postman, & Za-vortink, 1968; Underwood, 1957; Zechmeister& Nyberg, 1982). As with these classic memorytasks, the Lustig and Hasher (2002) findingsuggests that working memory span tasks mayalso be influenced by across-task proactive in-terference. Indeed, recent neuroimaging worksuggests that the same brain areas may mediate

both short- and long-term interference effects(Brush & Postle, 2003; Postle, Berger, Gold-stein, Curtis, & D’Esposito, 2001), a findingconsistent with the behavioral data.

The deletion function is critical not just forimmediate performance and working memorytasks (with their immediate recall trials), it is alsocritical for longer-term retrieval, since a broadfocus at encoding results in poorer retrieval(Anderson & Bower, 1973; Watkins & Watkins,

1976).4 It is not surprising, then, that older adultstypically show differentially poorretrieval relativeto that of younger adults (see Kane & Hasher,1995; Zacks, Hasher, & Li, 2000, for reviews).Consistent with this pattern of findings is evi-dence that retrieval is better, for both younger andolder adults, at peak than at off-peak times of day.This conclusion stems from a series of studiesusing materials ranging from prose to word listsand test tasks ranging from free recall to recog-nition (see Winocur & Hasher, 2002; Yoon et al.,2000, in press, for reviews).

Restraint

Restraint is the inhibitory mechanism that con-trols strong responses. It is probably also the mostwidely studied inhibitory mechanism and is ac-tually the mechanismthat many simply refer to as

‘‘inhibition’’ (e.g., Miyake et al., 2000, amongothers). Restraint has been studied using a varietyof tasks, including inhibition of return, Strooptasks of various sorts, and the stop-signal task. Itcan also be studied by looking at slips of thoughtand action as well as at schema-driven errors atretrieval, on the assumption that schemas are




strong responses to memory cues and so need tobe restrained for more detailed memories to beretrieved (see Alba & Hasher, 1983).

Direct evidence showing age and time-of-day

effects on control over strong responses comesfrom a variant of the stop-signal task, in which anoccasional signal occurs informing people of theneed to withhold a response that they otherwisemake quickly and accurately. A critical depen-dent measure is the proportion of stop trials onwhich errors are made (i.e., a ‘‘go’’ response ismade). In one study, older adults made moreerrors overall than young adults and everyonemade more errors at a nonoptimal time of day

(afternoon for older adults and morning foryoung adults) than at an optimal time (morningfor older adults and afternoon for young adults).The ability to withhold a strong response is re-duced with age as well as with performance at anoff-peak time of day (May & Hasher, 1998).

Comparable evidence with respect to agedifferences comes from the antisaccade task, inwhich people are instructed to respond to aperipheral stimulus (a brief onset) by looking

in the opposite direction to detect a limited-duration discrimination target. Because a pe-ripheral onset elicits a reflex response of look-ing toward the cue, restraint is required to lookin the correct direction (away from the onsetlocation), and older adults have greater diffi-culty than younger adults deploying the re-quired restraint. In particular, older adultsmake more looking-direction errors in the an-tisaccade task (Butler et al., 1999). Given therole inhibition plays in determining span size,it is not surprising that young adults show arelationship between span and performance onthe antisaccade task (Kane et al., 2001).

The ability to control strong responses canalso play a role in tasks requiring retrieval of detailed information when a strong response istriggered by a cue or context. A classic exampleof such errors occurs in the ‘‘Moses illusion’’

effect (Reder & Kusbit, 1991). Here people areasked to answer general-knowledge questionssuch as, ‘‘Who did Clark Kent turn into when hewent into a telephone booth?’’ Embedded in themidst of sensible questions are some that arenonsense, such as ‘‘How many animals of each

type did Moses take on the ark?’’ Yoon et al.(2000) reported that errors driven by strong re-sponses (e.g., to the biblical theme in the sen-tence) are more likely to occur at nonpeak times

ofday,andaremorelikelytooccurforolderthanfor younger adults.Other work shows that at nonoptimal times,

people are more likely to use easily accessiblestereotypes to judge individuals than they are atoptimal times (Bodenhausen, 1990). These er-rors of thought can be termed ‘‘slips’’ of thought,relating them to the ‘‘slips’’ of action litera-ture. This literature shows that strong motor re-sponses are less controllable at nonoptimal times

(Manley, Lewis, Robertson, Watson, & Datta,2002; May & Hasher, 1998), just as thoughtsare.

Attentional regulation of strong responses,like attentional regulation over distraction oraccess and deletion, appears to vary with cir-cadian arousal, and those variations are alsoseen in old rats. Winocur and Hasher (1999)found a similar pattern for old rats tested in aclassic Go–NoGo task at the beginning and at

the end of their activity cycle (Winocur &Hasher, 1999). Go responses did not changeacross the day, although the ability to withholda strong response was diminished at the end of the day for the old rats. Old rats also had moredifficulty performing a delayed matching-to-sample test (on which they have to reverse aprevious response) at the end of their activitycycle (Winocur & Hasher, 2004).

From a view emphasizing inhibitory func-tion, restraint processes are likely involved insituations conceived of by others as tapping‘‘task set’’ or ‘‘goal maintenance.’’ For example, aseries of Stroop experiments by Kane and Engle(2003) manipulated the ratio of congruent (sothat the ink color matched the color named bythe word) to incongruent (so that the ink colorconflicted with the color named by the word)trials. When there were many congruent trials,

participants with low working memory spanswere error-prone on those few trials that wereincongruent, and they were faster on congruenttrials. This was the case even though low-spanparticipants understood the goals of the task,and even when they received feedback after




every trial. These data can be seen as reflecting afailure of ‘‘task set’’ or ‘‘goal maintenance,’’ atleast at the level of having a goal control be-havior (e.g., Kane & Engle, 2003).

An inhibitory-based alternative explanationis at least equally possible. Like others (e.g., Ar-buckle & Gold, 1993), we consider the Strooptask to be an inhibitory control task that requirescontrol over strong responses (naming the word)in order to carry out a less dominant response(naming the color), thus primarily tapping intothe restraint function. Since working memorytasks have an inhibitory component that includescontrol over deletion and, very likely, given de-

letion failures, control over strong responsesfromprevious sets, it would not be surprising thatcontrol lapses in the Stroop task would be asso-ciated with poor performance on a span task.This might particularly be the case when theneed to control the nondominant response is notregularly reinforced.

Thus strong responses can seize control of both action and thought, and both patterns canbe seen for older adults and participants tested

at nonoptimal times of day. These effects canbe seen across a range of tasks, including atten-tion, memory, and language comprehension. Itis important to note that when strong responsesare correct, no time-of-day differences are ex-pected, since it is inhibition, not excitation, thatvaries with the arousal cycle and other importantindividual differences. As an example, the timeit takes to classify a word (e.g., chair ) as a mem-ber of a familiar category (furniture) does notdiffer across the day (e.g., May & Hasher, 1998;see Yoon et al., 2000).

WORKING MEMORY THEORY,CAPACITY, AND INHIBITION

In the previous sections we outlined our currentunderstanding and some of the relevant evi-

dence for the inhibitory control processes thatin our view are responsible for much of thevariation in working memory and cognitionmore generally. The relations between our viewsand those of other authors in this volume, andworking memory theory in general, have been

touched on throughout this discussion, but herewe focus more specifically on them.

The working memory model of Baddeleyand colleagues (Baddeley, 1986, 1992, 2000,

2003; Baddeley & Hitch, 1974) provides acommon heritage for most of the chapters inthis volume and for the vast majority of theworking memory literature more generally. Ourown work can be seen as focusing on the centralexecutive component of Baddeley’s system. LikeBaddeley, we were initially influenced by All-port’s (1989) and Shallice’s (Shallice & Burgess,1993) conceptions of control. Furthermore, weconsider the executive processes important for

‘‘working memory’’ to be domain general, andimportant across many areas of cognition, par-ticularly attention and memory, in close agree-ment with most of the contributors to thisvolume (see Chapters 2, 3, 4, 5, and 10). Fi-nally, we note that several contributors addressthe potential relations between group-level var-iation and individual differences–level variation(see Chapters 2, 4, 8, 10, and 11). Like severalother contributors (Chapters 8, 10, and 11), we

are especially concerned with variation due toaging.

Our approach differs from most other viewsin emphasizing inhibitory processes as sourcesof attentional regulation and thus of workingmemory variation. Although inhibitory pro-cesses are included in other views (see especiallyChapters 2, 3, and 10), we differ somewhat fromthese views by giving inhibition primary im-portance, and in so doing turning away fromnotions of capacity. We have avoided using thisterm in a loose sense, as we find it too easilyconfused with the idea that the ability to havemore information activated and at the focus of attention is always beneficial. Thus, we also turnaway from the metaphor of a large desk orworkspace as the best working memory, andconsider something more similar to a (truly ef-fective) spam blocker, allowing into the system

only information that is relevant to one’s goalsand concerns. As we have argued throughoutthis chapter, a mental workspace narrowly fo-cused on current concerns will be fast and ac-curate at on-line processing, in part because it isonly doing one task. Such a workspace is not




cluttered with previous tasks, upcoming tasks,social obligations, and short- and long-termpersonal concerns; it is simply doing the currenttask. Simply doing the current task also happens

to result subsequently in fast and accurate re-trieval of the information within that task. Whata narrow focus probably does not do is fostercreativity (Carson, Peterson, & Higgins, 2003).

Where we differ from others in the life-spandevelopmental literature and in the intelli-gence literature is in the notion that aspects of inhibitory regulation are central to determin-ing individual and time-of-day differences inboth perceptual speed and apparent working

memory capacity. In our view, the cognitiveprimitives upon which higher-order tasks buildare neither speed nor capacity, but instead areinhibitory regulation that occurs in the serviceof goals. It is important to note that individualsdiffer in their long-term goals (e.g., Kahneman,1973) and values (e.g., Rokeach, 1976). Whenresearchers are doing work across the adult lifespan (and probably with all non-universitystudents), it is particularly important to recog-

nize that younger and older adults differ onthese important dimensions (e.g., Carstensen &Lockenhoff, 2003). If information matchesgoals and values, the use of inhibitory processesshould be maximally efficient. Encoding willthen be narrow and retention levels high. In-deed, recent evidence suggests that age differ-ences in memory can be entirely eliminatedwhen the materials to be remembered matchthe goals, values and interests of older adults(May, Rahhal, Berry, & Leighton, 2005; Rah-hal,Hasher,&Colcombe,2001;Rahhal,May,&Hasher, 2002).

BIOLOGICAL BASES FORINHIBITORY VARIATION

The effects of age and time of day on inhibitory

function described above strongly suggest thatbiological influences play a major role in vari-ations in inhibitory efficiency. The field is innear-unanimous agreement that individual andgroup differences in frontal lobe structure andfunction contribute to individual and groupdifferences in executive processes such as in-

hibition (e.g., Engle, Tuholski, Laughlin, &Conway, 1999; Miyake et al., 2000; Moscovitch& Winocur, 1995; Park, Polk, Mikels, Taylor, &Marschuetz, 2001; Persad et al., 2002; West,

1996, 2000). The evidence for frontal lobe in-volvement in individual and group differencesin inhibition and other executive attention pro-cesses has been reviewed extensively elsewhereand is covered in more depth in other chaptersin this volume (see especially Chapters 2, 4, 7,10, and 11). Here we focus specifically on bio-logical evidence for variability in inhibitoryfunction, especially that due to age and circa-dian arousal.

Adult age differences in the structure andfunction of the frontal lobe structures most of-ten associated with working memory and inhi-bition are a major focus of work in the cognitiveneuroscience of aging (see reviews by Cabeza,2002; Grady & Craik, 2000; Raz, 2000; Reuter-Lorenz, 2002; Reuter-Lorenz et al., 2001). Therelevant neuroimaging findings are discussed inmore detail in Chapter 10; we will summarizesome of the more ubiquitous patterns here.

Prefrontal cortex structures, including thosemost often associated with working memory andinhibition, typically show the largest effects of age in structural brain studies (Raz, 2000). Infunctional imaging studies, older adults oftendiffer from young adults in showing either lessactivation in the brain regions typically associ-ated with task performance in young adults orshowing more activation, often in regions notassociated with task performance in youngadults (Cabeza, 2002; Grady & Craik, 2000;Reuter-Lorenz, 2002; Reuter-Lorenz et al.,2001). This additional activation is frequentlyinterpreted as a form of compensation for age-related increases in task difficulty or damage tostructures more typically associated with thetask. Other investigators have raised the possi-bility that it may represent a failure to createdistinct representations or a lack of functional

inhibition (e.g., Logan, Sanders, Snyder, Mor-ris, & Buckner, 2002; see Reuter-Lorenz &Lustig, 2005, for discussion of the functionalimplications of additional activations).

Although the neuroimaging literature mostoften focuses on changes in prefrontal cortex,there are also large changes in the subcortical




structures and neurotransmitter systems that in-teract with prefrontal cortex to modulate itsfunction. The size of age effects on the caudateand putamen, basal ganglia structures involved

in dopamine function, and on the locus coer-uleus, a brain structure involved in norepineph-rine function, is a close second to the amountfound for prefrontal cortex (Raz, 2000). Thesetwo catecholamine neurotransmitters, dopamineand norepinephrine, play important roles in at-tention and working memory. Changes in thesesystems may play an important but underratedrole in age changes in cognition (see discussionsby Braver & Barch, 2002; Li & Sikstrom, 2002;

Rubin, 1999).Of particular interest to the current discus-

sion is that dopamine and norepinephrinefunction appears to be essential to the ‘‘gating’’of information—that is, maintaining targetinformation and preventing irrelevant, non-target information from becoming activated(see reviews by Arnsten, 1998; Aston-Jones,Rajkowski, & Cohen,1999; Berridge, Arnsten, &Foote, 1993; Braver & Barch, 2002; see Chap-

ter 4, this volume). For example, neural-recording studies in rodents and primates showthat the phasic (stimulus-related) firing of cer-tain basal ganglia and locus coeruleus neuronsis largely target specific under normal condi-tions, with little or no firing to distractors (seereview by Arnsten, 1998). However, disruptionsin the tonic (state-related) levels of either do-pamine or norepinephrine lead to a loss of discriminability; both phasic firing to distractorsand behavioral false alarms increase (Arnsten,1998; Aston-Jones et al., 1999). In humans andother mammals, dopamine and norepinephrinefunction shows variation with both age (Arnsten,1998; Volkow et al., 1998) and circadian cycle(Aston-Jones, Chen, Zhu, & Oshinksy, 2001;Karlsson, Farde, & Halldin; 2000; Wirz-Justice,1984,1987).Further,thereisaninteractionsuchthat increased age is associated with shorter,

flatter, and often more irregular circadian cycles(Edgar, 1994; Hofman, 2000; Monk & Kupfer,2000; Weinert, 2000). These systems are thusprime candidates for the source of age- andcircadian-related variation.

Event-related potentials (ERPs) also provideevidence for age- and circadian-related changes

in the brain functions associated with workingmemory. In particular, P300, an ERP compo-nent strongly associated with the detection of target or unusual stimuli against a background

of distractors, shows significant variation in bothamplitude and latency over the course of theday (Geisler & Polich, 1990, 1992; Higuchi,Lui, Yuasa, Maeda, & Motohashi, 2000; Polich& Kok, 1995). P300 also shows differences as aresult of aging (see review by Polich, 1996),and animal studies provide compelling evi-dence for its link to locus coeruleus activity(Foote,Berridge,Adams,&Pineda,1991;Swick,Pineda, Schacher, & Foote, 1994). Thus far

there has been very little functional neuro-imaging (PET or fMRI) evidence of the influ-ence of circadian or age–circadian interactionson brain function. However, the evidence fromneurotransmitter and ERP studies suggests thatthese interactions are very promising areas forfuture investigation.

With regard to the possible relationshipsamong different functions of inhibition, we notea recent fMRI study that compared the brain

regions involved in switching with those involvedin the restraint of a prepotent response (Sylvesteret al., 2003; see Nelson, Reuter-Lorenz, Sylvs-eter, Jonides, & Smith, 2003, for a similar studyby this group). On each trial, participants werepresented with an arrow facing either the right orleft. For the switching task, participants had tocount the number of times each type of arrow(right or left) appeared during a block of trials;the arrow’s direction switched unpredictablyduring the block. For the restraint task, partici-pants had to press a button either correspondingto the direction in which the arrow was pointing(i.e., press the right button if the arrow is point-ing right; low-restraint condition) or one oppositethis direction (i.e., press the left button if thearrow is pointing right; high-restraint condition).

Although each task undoubtedly tapped mul-tiple processes, the switching task might be

thought of as preferentially requiring the deletionof one task set from working memory (e.g., countright arrows) to allow concentration on another(e.g., count left arrows). In contrast, the restrainttask likely preferentially required the restraint orsuppression of a natural inclination to press thebutton corresponding to the direction in which




the arrow was pointing. An intriguing question isthe degree to which these two tasks elicited thesame patterns of brain activity, thus suggesting ageneral executive function involved in both, or

distinct patterns specific to each task, implyingdifferent functions of executive control or inhi-bition.

There was a good deal of overlap in the brainregions activated by the two tasks: regions insuperior parietal cortex, medial frontal cortex,and left dorsolateral prefrontal cortex. Therewere also several interesting differences. Theswitching task activated several posterior re-gions more than the restraint task did—that is,

switching differentially activated bilateral ex-trastriate cortex and left posterior parietal cor-tex. The restraint task preferentially activatedregions in right parietal cortex, premotor cortex,frontopolar cortex, and bilateral basal gangliaregions including caudate and putamen. Theseresults on young-adult participants provide in-triguing evidence for the possibility that differ-ent functions of inhibition (or executive controlmore generally) may be mediated by different

brain structures (Sylvester et al., 2003). Thesedifferent structures may vary in their sensitivityto factors such as age and time of day, and thisdifference should manifest itself behaviorally.

In short, there is extensive evidence thatthe brain structures associated with workingmemory show a great deal of change with age,and that the functioning of those structures mayshow further variation across different times of day. In addition, recent brain imaging datasupport the idea of a distinction between differ-ent functions of inhibition or executive control,by suggesting that different functions may bedistinguished by the regions of the brain mostinvolved in their implementation (Sylvester etal., 2003). Behavioral evidence (e.g., Friedman& Miyake, 2004) is also suggestive in this regard.

Attempts to make direct connections betweenage-, circadian-, and function-related variations

in working memory performance are relativelynew, but represent a rich and exciting area forfuture research.

Our central view is that working memorycapacity is not the main issue for understand-ing higher-order cognition (nor is speed, as hasbeen argued in the literature on aging), rather,

inhibition and possibly other executive func-tionsare.Thecontentsofconsciousness,orwork-ing memory, are controlled by executive func-tions operating in the service of goals. These

executive functions are largely inhibitory in na-ture. There is a good deal of evidence to supportthis view, both in the present volume and else-where. Indeed, many views in this book havesome overlap with those we propose here, theideas of Kane and colleagues (Chapter 2) beingthe closest. For example, Kane et al. now de-scribe executive attention as the critical resourcethat determines both working memory capacityand inhibition. Of course, approaches favoring

inhibitory processes have not gone uncriticized(e.g., MacLeod, Dodd, Sheard, Wilson, & Bibi,2003: Miller & Cohen, 2002). The aging andtime-of-day differences reviewed here, however,will ultimately require some accommodation bythese alternative views.

CONCLUSIONS

Our emphasis on inhibitory processes, ratherthan on constructs such as capacity or resources,may be the characteristic that most differenti-ates our view from that of others. We haveconsistently maintained that inhibitory controlprocesses are the most likely sources of indi-vidual, group, and intra-individual variation inmeasures of working memory. We have pro-posed the existence of three inhibitory pro-cesses, access, deletion, and restraint (e.g.,Hasher et al., 1999), that together and possiblyindependently operate to regulate the contentsof consciousness. Our age and time-of-day workwith individuals at different points in theirarousal cycle suggests that all three processeschange with age and across the day, such thatregulation is better at peak times of day thanat off-peak times. The findings from animalmodels overlap, albeit not precisely, both the

age and time-of-day effects we have seen forpeople (Winocur & Hasher, 1999, 2002, 2004),suggesting a biological basis for these mecha-nisms. We have also reported evidence that in-hibitory processes underlie age differences inspeed of processing and underlie most tasks thatmeasure working memory capacity, as well as






OF WORKING MEMORY

We proposed a general theory of cognition

whose central view is that the best perfor-

mance on a variety of tasks occurs when the

contents of consciousness are narrowly fo-

cused on goal-relevant information (e.g.,

Hasher et al., 1999). Narrowing occurs in the

face of an individual’s internal and external

context—a world in which there is massive

activation triggered by the environment, the

recent past, near-future tasks, and subsidiary

goals. To tune this massive activation, we sug-

gest that inhibitory control is required, through

at least three control processes: access, dele-

tion, and restraint. Together with goals, these

processes determine the contents of con-

sciousness, or working memory. Our hypothe-

sized attentional mechanisms can be thought

of as at least partially fulfilling the functions of

the executive system of Baddeley’s workingmemory model (1986, 1992, 2003).


MEMORY VARIATION

Our work presumes that activation processes

vary minimally within and among individuals,

thus the critical source of individual and group

differences is the efficiency of inhibitory mech-

anisms and their underlying biology. Inhibitorycontrol appears to vary with age, and the syn-

chrony between an individual’s circadian

arousal pattern and the time of testing. There

are also substantial individual differences

within any age group. Inhibitory control is

particularly critical in situations to which there

are strong but erroneous response tendencies

and where there are salient sources of distrac-

tion, whether in thought or in the environment.

What we do not yet know is the degree towhich the three proposed attentional mecha-

nisms (access, deletion, and restraint) are fully

independent or partially overlapping mecha-

nisms. Also unclear is whether the pattern of

interdependence or independence remains the

same or changes across the adult life span and

within circadian arousal patterns at different

times of testing. The work of Friedman andMiyake (2004) suggests that access and dele-

tion may be the same for younger adults

whereas restraint is a separable process (see

Chapter 8). Our ongoing work addresses these

issues.


MEMORY VARIATION

A major alternative view proposes that the

capacity of working memory is the critical

determinant of individual differences on a

wide range of tasks. Our view stands in sharp

contrast to this and suggests instead that indi-

vidual differences in measures of capacity

(e.g., operation span, sentence span) and in the

ability of those measures to predict other

cognitive functions are actually due to varia-

tion in inhibitory control processes.We agree with Kane et al. (Chapter 2) that

the critical aspect of working memory mea-

sures is not that they measure capacity but that

they measure executive (or, in our view, at-

tentional) control processes. Indeed, we be-

lieve they best measure the ability to deal with

distraction (past, present, and future). We

agree that ‘‘executive attention’’ capabilities

are the major source of variation among indi-

viduals and that these capabilities are gen-eral and critical for a variety of intellectual

functions, including controlling interference,

memory, problem solving, and fluid intelli-

gence. In our view, however, the central as-

pects of control are inhibitory in nature; Kane

et al.’s view includes excitatory mechanisms

for maintaining the activation of representa-

tions, including goals. We have reviewed evi-

dence of equivalent activation across the day

in younger and older adults and so do notsee the need for assuming significant varia-

tion in activation processes. In this regard our

view differs from that of Munakata et al.

(Chapter 7).

(continued )

241



evidence to support the role of inhibition (and

circadian patterns) in determining long-termmemory performance. There are also clearfindings that excitatory processes do not changeacross the day (e.g., Yoon et al., 2000); we andothers believe these processes do not changewith age (e.g., Duchek et al., 1998). Thus, wesee these inhibitory processes, which we arguework together with an individual’s goals to de-termine the contents of consciousness, to be atthe heart of what many call working memory.

These mechanisms can be thought of as at leastpartially fulfilling the functions of the executivesystem of Baddeley’s working memory model(1986, 1992, 2003).

Thus, the critical source of working memoryvariability among (and within) people is inhi-bition. At the very least, we also know thatcircadian arousal patterns (and individual dif-ferences therein) influence the efficiency of inhibitory control. What we do not know is(a) the degree to which the three proposed in-hibitory executive functions (access, deletion,and restraint or suppression) are fully indepen-dent or partially overlapping mechanisms, and(b) whether with circadian arousal the patternof interdependence or independence remainsthe same or changes across the adult life spanand within an age group. Although the researchreviewed above in the section Biological Bases

for Inhibition Variation indicates that relevantfindings are beginning to appear in the litera-ture, we do not know a great deal about theunderlying biology of inhibitory control.

We would simply add that the richest ex-plications of the problems of mental controlwill come from research on a very broad range

of participants studied through a broad range of

approaches, including self-regulation of moti-vated behavior (e.g., Muraven & Baumeister,2000). From our analysis of the overall litera-ture, we suggest that for cognitive efficiency, anarrow, goal-driven focus is ideal for both on-line performance and subsequent retention of details. To achieve a narrow focus (or to reg-ulate attention effectively), inhibitory processesare required. We argue that there are threesuch processes (access, deletion, and restraint)

and that they vary within an individual, amongindividuals, and across the life span. Our viewsare not particularly tied to aging or circadianrhythms, but instead represent a general theoryof cognition that suggests that fundamen-tal regulatory mechanisms are inhibitory innature.

Notes

1. In this chapter we do not discuss the relevant lit-

erature on social issues and personality, but note

that there is empirical and theoretical overlap

among the domains in both tasks and the mech-

anisms that regulate them (see, e.g., Eysenck,

1995; Muraven & Baumeister, 2000). For example,

schizophrenic, creative, and low-span young adults

are all more likely to pick up information on the

unattended track in a dichotic listening experiment

(Conway, Cowan, & Bunting, 2001; Dykes & Mc-Ghie, 1976). Psychosis-prone, creative, and older

adults all show less habituation to repeated stimuli

(e.g., McDowd & Filion, 1992; Raine, Benishay,

Lencz & Scarpa, 1997).

2. In the lab, experimenters typically set goals for

participants, whereas in life people set goals for



Our approach speaks directly to the nature of

working memory. It suggests that working

memory capacity is not the cognitive primitive

it once appeared to be, and instead suggests

that the cognitive primitive (if there is one) is

inhibitory attentional control. Our studies have

included younger and older adults, as well as

individuals of this age range who differ in their

circadian arousal rhythms. We have also donesome work with animal models. All of these

studies point toward attentional regulation as

a critical determinant of intellectual perfor-

mance. This conclusion might remind readers

of Navon’s classic article (1977) on capacity as

a theoretical soup stone.

BOX 9.1. (continued )




themselves, sometimes a short-term one (finding

a cup of coffee in an unfamiliar city) and some-

times long-term ones (finding first editions of

classic psychology texts). Lab and life goals can

conflict, setting the stage for poor performance.Sometimes, participants may not adopt the goals

set by the experimenters.

3. Although there is a rich literature exploring per-

formance across the day, most of these studies are

done without reference to differences in circa-

dian arousal patterns. Although important for the

study of younger adults, the failure to take arousal

differences into account when comparing across

ages is particularly worrisome, given the substan-

tial differences in arousal patterns (see Goldsteinet al., 2006).

4. We note that access, too, plays a role in deter-

mining both short- and long-term memory per-

formance, including that on working memory

tasks, because this process also influences the size

of the memory bundles created during encoding.

These will be small or large, to the degree that

access is or is not efficient, respectively.

5. We note that deletion failures also set the stage

for the need for source monitoring, that is, theneed to distinguish whether an item or set of

items came from the current trial or a previous

trial. If items from a previous trial are successfully

deleted initially, few source decisions would be

required. Further, if items from a previous trial

were successfully deleted when that trial was over

and the new trial started, source decisions would

be easier to make.

References

Alba, J. W., & Hasher, L. (1983). Is memory

schematic? Psychological Bulletin, 93, 203–231.

Allport, A. (1989). Visual attention. In M. I. Posner

(Ed.), Foundations of cognitive science, pp 631–

682. Cambridge, MA: MIT Press.

Anderson, J. R., & Bower, G. H. (1973). Human

associative memory. Oxford: V.H. Winston &

Sons. Arbuckle,T. Y., & Gold, D. P. (1993) Aging, inhibi-

tion and verbosity. Journal of Gerontology:

Psychological Science, 48, P225–P232.

Arnsten, A.F.T. (1998). Catecholamine modulation

of prefrontal cortical cognitive function. Trends

in Cognitive Sciences, 2, 436–437.

Aston-Jones, G., Chen, S., Zhu, Y., & Oshinsky, M.L.

(2001). A neural circuit for circadian regulation

of arousal. Nature Neuroscience, 4, 732–738.

Aston-Jones, G., Rajkowski, J., & Cohen, J. (1999).

Role of locus coeruleus in attention and behav-ioral flexibility. Biological Psychiatry, 46, 1309–

1320.



Baddeley, A. D. (1992). Working memory. Science,

255, 556–559.

Baddeley, A. D. (2000). The episodic buffer: a new

component of working memory? Trends in Cog-


Baddeley, A. D. (2003). Working memory: Lookingforward and looking backward. Nature Reviews:






Academic Press.

Berridge, C. W., Arnsten, A. F .T., & Foote,

S. L. (1993). Noradrenergic modulation of cog-

nitive function: Clinical implications of ana-tomical, electrophysiological, and behavioural

studies in animal models. Psychological Medi-

cine, 23, 557–564.

Bodenhausen, G. V. (1990). Stereotypes as judg-

mental heuristics: Evidence of circadian varia-

tions in discrimination. Psychological Science,

1, 319–322.

Bowles, R. P., & Salthouse, T. A. (2003). Assessing the

age-related effects of proactive interference on

working memory tasks using the Rasch model.


Braver, T. S., & Barch, D. M. (2002). A theory of

cognitive control, aging cognition, and neuro-

modulation. Neuroscience and Biobehavioral

Reviews, 26, 809–817.

Brush, L. N., & Postle, B. R. (2003). Neural mecha-

nisms underlying the control of item-specific and

item-nonspecific proactive interference in human

working memory. Presented at the Society forNeuroscience 33rd Annual Meeting, New Or-

leans, November 8–12.

Bunting, M. F. (2006). Proactive interference and

item similarity in working memory. Journal of

Experimental Psychology: Learning, Memory &

cognition, 32, 183–196.




Butler, K. M, Zacks, R. T., & Henderson, J. M.

(1999). Suppression of reflexive saccades in

younger and older adults: Age comparisons on

antisaccade task. Memory & Cognition, 27, 584–

591.Cabeza, R. (2002). Hemispheric asymmetry in older

adults: The HAROLD model. Psychology and

Aging, 17, 85–100.

Cabeza, R., Dolcos, F., Prince, S. E., Rice, H. J.,

Weissman, D. H.,& Nyberg,L. (2003). Attention-

related activity during episodic memory retrieval:

A cross-function fMRI study. Neuropsychologia,

41, 390–399.

Cameli, L., & Phillips, N. A. (2000). Age-related

differences in semantic priming: Evidencefrom event-related brain potentials. Brain and

Cognition, 43, 69–73.

Carlson, M. C., Hasher, L., Connelly, S. L., &

Zacks, R. T. (1995). Aging, distraction, and the

benefits of predictable location. Psychology and

Aging, 10, 427–436.

Carson, S. H., Peterson, J. B., & Higgins, D. M.

(2003). Decreased latent inhibition is associ-

ated with increased creative achievement in

high-functioning individuals. Journal of Person-ality and Social Psychology, 85, 499–506.

Carstensen, L. L., & Lockenhoff, C. E. (2003).

Aging, emotion, and evolution. Annals of

the New York Academy of Science, 1000, 152–

179.

Cermakian, N., & Boivin, D. B. (2003). A molecular

perspective of human circadian rhythm disor-

ders. Brain Research Reviews, 42, 204–220.

Chiappe, P., Hasher, L., & Siegel, L. S. (2000).

Working memory, inhibitory control and read-

ing disability. Memory & Cognition, 28, 8–17.

Chiappe, P., Siegel, L. S., & Hasher, L. (2002).

Working memory, inhibition, and reading skill.

In S. P. Shohov (Ed.), Advances in psychology

research (Vol. 9, pp. 30–51). Huntington, NY:

Nova Science Publishers.

Connelly, S. L., Hasher, L., & Zacks, R. T. (1991)

Age and reading: The impact of distraction.

Psychology and Aging, 6, 533–541.Conway, A. P. A., Cowan, N., & Bunting, M. F.

(2001). The cocktail party phenomenon revis-

ited: The importance of working memory

capacity. Psychonomic Bulletin & Review, 8,

331–335.

Cowan, N. (1993). Activation, attention, and short-

termmemory.Memory&Cognition,21, 162–167.

Craik, F. I. M., Govoni, R., Naveh-Benjamin, M., &

Anderson, N. D. (1996). The effects of divided

attention on encoding and retrieval processes inhuman memory. Journal of Experimental Psy-


Csikszentmihalyi, M., Rathunde, K., & Whalen, S.

(1993). Talented teenagers: The roots of success

and failure. New York: Cambridge University

Press.

Daneman, M., & Carpenter, P. A. (1980). Individual

differences in working memory and reading.

Journal of Verbal Learning and Verbal Behavior,

19, 450–466.Daneman, M., & Merikle, P. M. (1996). Working



422–433.

Deese, J. (1959). On the prediction of occurrence of

particular verbal intrusions in immediate recall.

Journal of Experimental Psychology, 58, 17–22.

Demspter, F. N. (1991). Inhibitory processes: A ne-

glected dimension of intelligence. Intelligence,

15, 157–173.Duchek, J. M., Balota, D. A., & Thessing, V. C.

(1998). Inhibition of visual and conceptual

information during reading in healthy aging

and Alzheimer’s disease. Aging, Neuropsychol-

ogy, and Cognition, 5, 169–181.

Dykes, M., & McGhie, A. (1976). A comparative

study of attentional strategies of schizophrenic

and highly creative normal subjects. British

Journal of Psychiatry, 128, 50–56.

Dywan, J., & Murphy, W. E. (1996). Aging and

inhibitory control in text comprehension. Psy-

chology and Aging, 11, 199–206.

Edgar, D. M. (1994). Sleep–wake circadian cycles

and aging: Potential etiologies and relevance to

age-related changes in integrated physiological

systems. Neurobiology of Aging, 15, 499–501.

Engle, R. W. (2002). Working memory capacity as

executive attention. Current Directions in Psy-

chological Science, 11, 19–23.Engle, R. W., Cantor, J., & Carullo, J. J. (1992).


comprehension: A test of four hypotheses.


Memory and Cognition, 18, 972–992.




Hofman, M. A. (2000). The human circadian clock

and aging. Chronobiology International, 17,

245–259.

Horne, J. A., & Ostberg, O. (1976). A self-assessment

questionnaire to determine morningness-eveningness in human circadian rhythms. Inter-

national Journal of Chronobiology, 4, 97–110.

Hur, Y., Bouchard, T. J., & Lykken, D. T. (1998).

Genetic and environmental influence on

morningness-eveningness. Personality and Indi-

vidual Differences, 25, 917–925.

Just, M. A., & Carpenter, P. A. (1992). A capacity

theory of comprehension: Individual differences

in working memory. Psychological Review, 99,

122–149.Kahneman, D. (1973). Attention and effort. Engle-

wood Cliffs, NJ: Prentice-Hall.

Kail, R. (1993). Processing time decreases globally at

an exponential rate during childhood and ado-

lescence. Journal of Experimental Child Psychol-

ogy, 56, 254–265.


Engle, R. W. (2001). A controlled-attention

view of working-memory capacity. Journal of


Kane, M. J., & Engle, R. W. (2000). Working memory

capacity, proactive interference, and divided

attention: Limits on long-term memory retrieval.



Kane, M. J., & Engle, R. W. (2003). Working


The contributions of goal neglect, response

competition, and task set to Stroop interfer-


eral, 132, 47–70.

Kane, M. J., & Hasher, L. (1995). Interference. In G.

Maddox (Ed.), The encyclopedia of aging (2nd

ed., pp. 514–516). New York: Springer-Verlag.

Karlsson, P., Farde, L., & Halldin, C. (2000).

Circadian rhythm in central D1-like dopamine

receptors examined by PET. Presented at the

Third International Symposium on FunctionalNeuroreceptor Mapping, New York, June.

Keppel, G., Postman, L., & Zavortink, B. (1968).

Studies of learning to learn: VIII. The influ-

ence of massive amounts of training upon the

learning and retention of paired-associate lists.


ior, 7, 790–796.

Kerkhof, G. A., & Lancel, M. (1991). EEG slow-

wave activity, REM-sleep, and rectal tempera-

ture during night and day sleep in morning-typeand evening-type subjects. Psychophysiology, 28,

678–688.

Kim, S., Dueker, G. L., Hasher, L., & Goldstein, D.

(2002). Children’s time of day preference: Age,

gender, and ethnic differences. Personality and

Individual Differences, 33, 1083–1090.

Kim, S., Hasher, L., & Zacks, R. T. (in press). A benefit

of heightened susceptibility to distraction: An

aging study. Psychonomic Bulletin & Review.

Laver, G. D. (2000). A speed-accuracy analysis of word recognition in young and older adults.


Li, K. Z. H., Hasher, L., Jonas, D., Rahhal, T., &

May, C. P. (1998). Distractability, circadian

arousal, and aging: A boundary condition? Psy-

chology and Aging, 13, 574–583

Li, S., & Sikstrom, S. (2002). Integrative neurocom-

putational perspectives on cognitive aging, neuro-

modulation, and representation. Neuroscience

and Biobehavioral Reviews, 26, 795–808.Logan, J. M., Sanders, A. L., Snyder, A. Z., Morris,

J. C., & Buckner, R. L. (2002). Under-

recruitment and nonselective recruitment:

Dissociable neural mechanisms associated with

aging. Neuron, 33, 827–840

Lustig, C., & Hasher, L. (2001). Implicit memory is

not immune to interference. Psychological Bul-

letin, 127, 618–628.

Lustig, C., & Hasher, L. (2002). Working memory

span: The effect of prior learning. American

Journal of Psychology, 115, 89–101.

Lustig, C., Hasher, L., & Tonev, S. (in press).

Distraction as a determinant of processing speed.

Psychonomic Bulletin & Review.


Working memory span and the role of proactive

interference. Journal of Experimental Psychol-

ogy: General, 130, 199–207.

Lustig, C. & Meck, W. H. (2001). Paying attentionto time as one gets older. Psychological Science,

12, 478–484.

Mack, A., & Rock, I. (1998). Inattentional blindness:

Perception without attention. In R. D. Wright

(Ed.), Visual attention: Vancouver studies in




cognitive science (Vol. 8, pp. 55–76). New York:


MacLeod, C. M., Dodd, M. D., Sheard, E. D.,

Wilson, D. E., & Bibi, U. (2003). In opposition

to inhibition. In B. H. Ross (Ed.), The psychol-ogy of learning and motivation (Vol. 43, pp.

163–214). San Diego: Academic Press.

Manly, T., Lewis, G. H., Robertson, I. H.,

Watson, P. C., & Datta, A. K. (2002). Coffee

in the cornflakes: Time-of-day as a modulator

of executive response control. Neuropsychologia,

40, 1–6.

May, C. P. (1999). Synchrony effects in cognition:

The costs and a benefit. Psychonomic Bulletin

& Review, 6, 142–147.May, C. P., & Hasher, L. (1998). Synchrony effects

in inhibitory control over thought and action.

Journal of Experimental Psychology: Human

Perception and Performance, 24, 363–379.

May, C. P., Hasher, L., & Foong, N. (2005).

Implicit memory, age and time of day: Para-

doxical priming effects. Psychological Science,

16, 96–100.

May, C. P., Hasher, L., & Kane, M. J. (1999). The

role of interference in memory span. Memory &Cognition, 27, 759–767.

May, C. P., Hasher, L., & Stoltzfus, E. R. (1993).

Optimal time of day and the magnitude of age

differences in memory. Psychological Science,

4, 326–330.

May, C. P., Kane, M. J., & Hasher, L. (1995).

Determinants of negative priming. Psychologi-

cal Bulletin, 118, 35–54.

May, C. P., Rahhal, T., Berry, E. M., & Leighton,

E. A. (2005). Aging, source memory, and emo-

tion. Psychology and Aging, 20, 571–578.

May, C. P., Zacks, R. T., Hasher, L., & Multhaup,

K. S. (1999). Inhibition in the processing of

garden-path sentences. Psychology and Aging,

14, 304–313.

McDowd, J. M., & Filion, D. L. (1992). Aging,

selective attention, and inhibitory processes: A

psychophysiological approach. Psychology and

Aging, 7, 65–71.Miller, E. K., & Cohen, J. D. (2001). An integrative

theory of prefrontal cortex function. Annual

Review of Neuroscience, 24, 167–202.

Miyake, A., Friedman, N. P., Emerson, M. J., Witzki,

A. H., Howerter, A., & Wager, T. D. (2000). The

unity and diversity of executive functions and

their contributions to complex ‘‘frontal lobe’’

tasks: A latent variable analysis. Cognitive Psy-

chology, 41, 49–100.

Monk, T. H., & Kupfer, D. J. (2000). Circadianrhythms in healthy aging: Effects downstream

from the pacemaker. Chronobiology Interna-

tional, 17, 355–368.

Moscovitch, M.,& Winocur, G. (1995). Frontal lobes,

memory, and aging. In J. Gragman,K. J. Holyoak,

& B. Boller (Eds.) Structure and functions of the

human prefrontal cortex (pp. 119–150). New

York: New York Academy of Sciences.

Most, S. B., Simons, D. J., Scholl, B. J., Jimenez, R.,

Clifford, E., & Chabris, C. F. (2001). How notto be seen: The contribution of similarity and se-

lective ignoring to sustained inattentional blind-

ness. Psychological Science, 12, 9–17.

Muraven, M., & Baumeister, R. F. (2000). Self-

regulation and depletion of limited resources:

Does self-control resemble a muscle? Psycho-


Navon, D. (1984). Resources—A theoretical soup-

stone? Psychological Review, 91, 216–234.

Nelson, J. K.,Reuter-Lorenz, P. A., Sylvester, C. Y. C.,Jonides, J., & Smith, E. E. (2003). Dissociable

neural mechanisms underlying response-based

and familiarity-based conflict in working mem-

ory. Proceedings of the National Academy of

Sciences USA, 19, 11171–11175.

Nigg, J. T. (2001). Is ADHD a disinhibitory

disorder? Psychological Bulletin, 127, 571–598.

Park, D. C., Polk, T. A., Mikels, J. A., Taylor, S. F.,

& Marshuetz, C. (2001). Cerebral aging:

Integration of brain and behavioral models of

cognitive function. Dialogues in Clinical Neuro-

science, 3, 151–166.

Park, D. C., Smith, A. D., Lautenschlager, G., Earles,

J. L., Frieske, D., Zwahr, M., & Gaines, C.

(1996). Mediators of long-term memory perfor-

mance across the life span. Psychology and

Aging, 11, 621–637.

Persad, C. C. (2001). The role of inhibition in

working memory performance associatedwith age. Dissertation Abstracts International:

Section B: The Sciences and Engineering, 61,

4440.

Persad, C. C., Abeles, N., Zacks, R. T., & Denburg,

N. L. (2002). Inhibitory changes after age 60




and the relationship to measures of attention

and memory. Journals of Gerontology: Series B:

Psychological Sciences and Social Sciences,

57B, P223–P232.

Phillips, N. A., & Lesperance, D. (2003) Breakingthe waves: Age differences in electrical brain

activity when reading text with distractors.


Plude, D. J., & Hoyer, W. J. (1986). Age and the

selectivity of visual information processing.


Polich, J. (1996). Meta-analysis of P300 normative

aging studies. Psychophysiology, 33, 334–353.

Polich, J., & Kok, A. (1995). Cognitive and biological

determinants of P300: An integrative review.Biological Psychology, 41, 103–146.

Postle, B. R., Berger, J. S., Goldstein, J. H., Curtis, C.

E., & D’Esposito, M. (2001). Behavioral and

neurophysiological correlates of episodic cod-

ing, proactive interference, and list length effects

in a running span verbal working memory task.

Cognitive, Affective, & Behavioral Neuroscience,

1, 10–21.

Radvansky, G. A., & Curiel, J. M. (1998). Narrative

comprehension and aging: The fate of com-pleted goal information. Psychology and Aging,

13, 69–79.

Rahhal, T. A., Hasher, L., & Colcombe, S. (2001).

Instructional differences and age differences in

memory: Now you see them, now you don’t.


Rahhal, T. A., May, C. P., & Hasher, L. (2002). Truth

and character: Sources that older adults can

remember. Psychological Science, 13, 101–105.

Raine, A., Benishay, D., Lencz, T., & Scarpa, A.

(1997). Abnormal orienting in schizotypal per-

sonality disorder. Schizophrenia Bulletin, 23,

101–105.

Ranganath, C., Johnson, M. K., & D’Esposito, M.

(2003). Prefrontal activity associated with work-

ing memory and episodic long-term memory.

Neuropsychologia, 41, 378–389.

Raz, N. (2000). Aging of the brain and its impact on

cognitive performance: Integration of structuraland functional findings. In F. I. M. Craik &

T. A. Salthouse (Eds.), The handbook of aging

and cognition (2nd ed., pp. 1–90). Mahwah,

NJ: Lawrence Erlbaum Associates.

Reder, L., & Kusbit, G. W. (1991). Locus of the

Moses illusion: Imperfect encoding, retrieval, or

match. Journal of Memory and Language, 30,

385–406.

Reuter-Lorenz, P. A. (2002). New visions of the

aging mind and brain. Trends in Cognitive Sci-

ences, 6, 394–400.Reuter-Lorenz, P. A., & Lustig, C. (2005). Brain

aging: Reorganizing discoveries about the aging

mind. Current Opinion in Neurobiology, 15,

245–251.

Reuter-Lorenz, P. A., Marschuetz, C., Jonides, J.,

Hartley, A., & Koeppe, R. (2001). Neurocogni-

tive aging of storage and executive processes.

European Journal of Cognitive Psychology, 13,

257–278.

Roediger, H. L., III, & McDermott, K. B. (1995).Creating false memories: Remembering words

not presented in lists. Journal of Experimental


21, 803–814.

Rokeach, M. (1976). Beliefs, attitudes and values: A

theory of organization and change. San Fran-

cisco: Jossey-Bass.

Rowe, G., Turcotte, J., & Hasher, L. (2006). Proactive

interference in a working memory span tasks.

Manuscript in preparation.Rowe, G., Valderrama, S., Lenartowicz, A., &

Hasher, L. (in press). Attentional disregulation:

A benefit for implicit memory. Psychology and

Aging.

Rubin, D. C. (1999). Frontal-striatal circuits in

cognitive aging: Evidence for caudate involve-

ment. Aging, Neuropsychology, and Cognition,

6, 241–259.




Shallice, T., & Burgess, P. (1993). Supervisory

control of action and thought selection. In

A. D. Baddeley & L. Weiskrantz (Eds.), Atten-

tion: Selection, awareness, and control: A tribute

to Donald Broadbent (pp. 171–187). New York:


Swick, D., Pineda, J. A., Schacher, S., & Foote, S. L.

(1994). Locus-coeruleus neuronal-activity inawake monkeys: Relationship to auditory P300-

like potentials and spontaneous EEG. Experi-

mental Brain Research, 101, 86–92.

Sylvester, C. Y., Wager, T. D., Lacey, S. C.,

Hernandez, L., Nichols, T. E., Smith, E. E., &

Jonides, J. (2003). Switching attention and




resolving interference: fMRI measures of execu-

tive functions. Neuropsychologia, 41, 357–370.

Tentori, K., Osherson, D., Hasher, L., & May, C.

(2001). Wisdom and aging: Irrational prefer-

ences in college students but not older adults.Cognition, 81, B87–B96.

Tun, P. A., O’Kane, G., & Wingfield, A. (2002).

Distraction by competing speech in young and

older adult listeners. Psychology and Aging, 17,

453–467.

Underwood, B. J. (1957). Interference and forget-

ting. Psychological Review, 64, 49–60.

Underwood, B. J. (1965). False recognition pro-

duced by implicit verbal responses. Journal of

Experimental Psychology, 70, 122–129. Volkow, N. D., Gur, R. C., Wang, G. J., Fowler,

J. S., Moberg, P. J., Ding, Y. S., Hitzeman, R.,

Smith, G., & Logan, J. (1998). Association

between decline in dopamine activity with age

and cognitive and motor impairment in healthy

individuals. American Journal of Psychiatry,

155, 344–349.

Watkins, M. J., & Watkins, O. C. (1976). Cue-overload

theory and the method of interpolated attributes.

Bulletin of the Psychonomic Society, 7, 289–291.Weinert, D. (2000). Age-dependent changes of the

circadian system. Chronobiology International,

17, 261–283.

Wenzlaff, R. M., & Wegner, D. M. (2000). Thought

suppression. Annual Review of Psychology, 51,

59–91.

West, R. L. (1996). An application of prefrontal

cortex function theory to cognitive aging. Psy-

chological Bulletin, 120, 272–292.

West, R. L. (2000). In defense of the frontal

lobe hypothesis of cognitive aging. Journal of

the International Neuropsychological Society, 6,

727–729.

West, R., Murphy, K. J., Armilio, M. L., Craik, F. I.

M., & Stuss, D. T. (2002). Effects of time of day

on age differences in working memory. Journals

of Gerontology: Series B: Psychological Sciences

and Social Sciences, 57B, P3–P10.

Winocur, G., & Hasher, L. (1999). Aging and time-of-day on cognition in rats. Behavioral Neuro-

science, 113, 991–997.

Winocur, G., & Hasher, L. (2002). Circadian

rhythms and memory in aged humans and

animals. In L. R. Squire & D. L . Schacter

(Eds.), Neuropsychology of memory (3rd ed., pp.

273–285). New York: Guilford Press.Winocur, G., & Hasher, L. (2004). Age and time-of-

day effects on learning and memory in a non-

matching-to-sample test. Neurobiology of Aging,

25, 1107–1115.

Wirz-Justice, A. (1984). Dopamine receptor

rhythms. Biological Psychiatry, 19, 1274–1276.

Wirz-Justice, A. (1987). Circadian rhythms in mam-

malian neurotransmitter receptors. Progress in

Neurobiology, 29, 219–259.

Yoon, C., May, C. P., & Hasher, L. (2000). Aging,circadian arousal patterns, and cognition. In

D. C. Park & N. Schwarz (Eds.), Cognitive

aging: A primer (pp. 151–171). Philadelphia:

Psychology Press.

Yoon, C., May, C. P., Goldstein, D., & Hasher, L.

(in press). Aging, circadian arousal patterns,

and cognition. In D. C. Park & N. Schwarz

(Eds.), Cognitive aging: A primer (2nd ed.).

Philadelphia: Psychology Press.

Zacks, R. T., & Hasher, L. (1994). Directed ignoring:Inhibitory regulation of working memory. In D.

Dagenbach & T. H. Carr (Eds.), Inhibitory

processes in attention, memory, and language

(pp. 241–264). San Diego: Academic Press.

Zacks, R. T., Hasher, L., & Li, K. Z. H. (2000).

Human memory. In F. I. M. Craik & T. A.

Salthouse (Eds), The handbook of aging and

cognition (2nd ed., pp. 293–357). Mahwah, NJ:


Zacks, R. T., Radvansky, G. A., & Hasher, L. (1996).

Studies of directed forgetting in older adults.



Zacks, J. L., & Zacks, R. T. (1993). Visual search

times assessed without reaction times: A new

method and an application to aging. Journal of

Experimental Psychology: Human Perception

and Performance, 19, 798–813.

Zechmeister, E. G., & Nyberg, S. E. (1982). Humanmemory: An introduction to research and theory.

Monterey, CA: Brooks/Cole.




10

The Executive Is Central to Working

Memory: Insights from Age, Performance,

and Task Variations

PATRICIA A. REUTER-LORENZ and JOHN JONIDES

The model of working memory developed by Alan Baddeley and his colleagues has domi-nated the landscape of memory research forsome time (Baddeley, 1986; Baddeley & Hitch,1974). According to this model, information isstored in and retrieved from a set of buffers,each specialized for a different kind of infor-mation. According to many interpretations of this model, some simple tasks demand onlyencoding, storage, and retrieval of informationfrom the relevant buffer. An example mightbe a simple span task in which a set of lettersis presented and several seconds later the par-ticipant has to recall the letters in the orderof presentation. Other tasks, by assumption,require executive processes if the informationin a buffer needs to be manipulated prior to the

participant’s giving a response. An examplemight be an alphabetic span task in which aset of letters is presented, and several secondslater the participant has to recall the letters inalphabetical order. Executive processes, then,are an added component of the model, andthese processes are assumed to be drawn into

play when stored information must be trans-formed.

We maintain that this view of executiveprocessing is too limited. Our argument is thatexecutive processing is required in any workingmemory task just so long as information mustbe stored longer than a passive trace is retained.Indeed, the delayed-response task, one of themost widely used tasks to investigate the neuralcircuitry of working memory in animals, haslong pointed to the frontal cortex, the site of the central executive, as crucial to successfulworking memory performance (see Goldman-Rakic, 1987, for a review). What is debated iswhether prefrontal circuitry is the site of work-ing memory storage per se, or whether its con-tribution is primarily in the form of attentional

control (see, e.g., Postle, Druzgal & D’Espo-sito, 2003, for a review). Attentional control overinternal representations figures prominently inour view of the central executive. Although aconsensual definition of what constitutes anexecutive process is lacking, many prominentexamples in the literature include filtering

250



irrelevant information, inhibiting competingresponses, switching between representations,monitoring ongoing performance, and manag-ing multiple tasks, all of which entail atten-

tional control. A frontal-parietal executivesystem that includes lateral prefrontal cortex,superior and inferior parietal cortex, and me-dial frontal cortex, including anterior cingulateand perhaps supplementary motor cortex, isimplicated in these forms of attentional control(e.g., Corbetta, Kincade, & Shulman, 2002;Posner & Dehaene, 1994; Posner & Petersen,1990; Ravizza, Delgado, Chein, Becker, &Fiez, 2004).

By our argument, any attentional processrequired during working memory engages agood part of the same neural machinery in-volved when information needs to be manip-ulated. And it is this engagement of attentionalcontrol that is the heart of executive processing,not necessarily the specific manipulations thatmay be necessary in one task or another (e.g.,alphabetizing). Our argument is motivated bydata on individual differences in success at

working memory tasks, and data on changes inworking memory with age, as revealed by neuro-imaging experiments. We have found thatolder adults (and in some cases poor-perform-ing younger adults) recruit executive processesin the service of performance in a workingmemory task that putatively involves onlyencoding, storage, and retrieval. We take thisfinding to indicate that even simple storagetasks have an executive component to themthat is stressed in poor-performing young adultsand by neural declines that accompany normalaging. When these same participants are thenchallenged with a task that requires yet furtherapplication of attentional control, they are de-ficient in their performance because their at-tentional resources are already partly taken upby the basic encoding, storage, and retrievalprocesses. In short, our view has its genesis in

evidence suggesting that working memory maybe under the control of a single resource, at-tentional allocation. This resource can be taxedby high-task demands, by individual differ-ences in performance level, by normal aging,and perhaps by still more individual-differencevariables that have yet to be explored. Thus,

rather than viewing executive processing as amodule of working memory separate fromstorage, we view attentional allocation in vary-ing degrees as a necessary component of all

working memory storage.These data have led us to a different view of working memory than that of the canon. Theestablished view is that working memory taskscan be divided into two basic categories: thosethat require simple storage (e.g., serial span),also referred to as short-term memory tasks, andstorage tasks that also require executive pro-cessing (e.g., alphabetic span) (see, e.g., Engle,Tuholski, Laughlin, & Conway, 1999). In the

present volume this distinction is captured bythe dichotomy between ‘‘simple’’ and ‘‘com-plex’’ span tasks (Chapter 6), with the formerplacing minimal if any demands on executiveprocessing. Likewise, according to the viewdeveloped by Kane, Conway, Hambrick, andEngle (Chapter 2), complex span tasks mea-sure working memory capacity and entail ‘‘short-term memory’’ plus executive (attentional)processes. Oberauer, Suß, Wilhelm, and San-

der (Chapter 3) also distinguish tasks that in-volve simultaneous storage and processing.Engle et al. (1999) provide the strongest theo-retical and empirical elaboration of this view.They are concerned in particular with the ex-tent to which short-term memory and workingmemory are different constructs that aremeasurable by different tasks. Using the latent-variable approach, they conclude that these areindeed separate constructs and that it is the‘‘differential reliance on controlled attentionalprocessing that makes these two constructsdifferent theoretically and empirically’’ (p. 326).The authors go on to agree with Cowan (1995,1998), who posits that while closely related,short-term and working memory are repre-sented by separate factors and are differentiallyrelated to higher-cognitive processes. Theseconclusions have been echoed by researchers

who have taken a life-span approach and de-termined that across an age range from 20 to 92years, there is a measurable distinction betweenshort-term memory and working memory (Parket al., 2002). By contrast, our view is that allworking memory tasks, including so-called short-termmemoryor ‘‘storage-only’’ tasks, recruit some

Insights from Age, Performance, and Task Variations 251



degree of attentional control (and thereforeexecutive processing), and so this classificationis artificial and misleading. The amount of at-tentional control called upon by a task alleged

to require only storage will depend on theprocessing demands of that task and on the ageand level of working memory skill of the in-dividual. The simplest storage task may recruitattentional control in the hands of a poorperformer, and a good performer will need at-tentional control in some task contexts (i.e.,depending on load and retrieval demands).Overall, though, all working memory tasks re-quire attentional control to some degree, much

as perceptual processing on incoming sensoryinformation requires attentional control.

While most researchers explicitly acknowl-edge that no task is pure in its measurement of only storage, or only executive processes, thereis a temptation to equate tasks themselves withthe underlying constructs. In this chapter, wehope to persuade the reader that this equationshould be resisted, if not discarded. Tasks thatare described as primarily storage and mini-

mally executive in their demands, activatebrain areas associated with executive, and es-pecially attentional, control. Moreover, thesame task performed by an older adult or a poorperformer can activate control areas to an evengreater degree than one finds in the average-performing younger adult.

In the discussion below, we lay out our viewof the representational characteristics of work-ing memory, and then review how attentionalcontrol is implemented in working memorytasks widely viewed as requiring only storage.With respect to representation, our fundamen-tal view is that there are many cases in whichworking memory harnesses representationsused for perception in the service of storage. Inaddition, there are also frontal mechanismsthat play a role in storage. With respect to ex-ecutive processing in working memory, we ar-

gue that there are varieties of attentionalcontrol: those that are used in the service of continued activation of stored information(what is typically called rehearsal), and thoseused in the service of selecting and manipu-lating information. In both cases, we presentthe possibility that executive control is simply a

form of attentional allocation, but on mentalrepresentations, not on perceptual ones. Weare particularly mindful of the role thatneuroimaging data have played in furthering

our understanding of these characteristics of working memory. After this overall review, wethen describe evidence about the recruitmentof executive processes in the item recognitiontask, drawing largely from data comparingolder and younger adults.

REPRESENTATION INWORKING MEMORY

Suppose you need to align two visual objects ina slide of a Powerpoint presentation. You mightget the following instructions: click on oneof the objects; depress the shift key and clickon the other object; go to the menu bar and findthe ‘‘draw’’ functions; scroll down to the onelabeled ‘‘alignment,’’ and then choose the di-rection in which you want the objects aligned.This is one of a huge number of tasks that re-

quire working memory. The memory systemthat serves these tasks has a small storage ca-pacity, and the information stored there is re-tained for only a brief duration, measured inseconds. This short duration can be overcomeby enlisting rehearsal processes that maintainthe strength of the memory trace by refreshingand recycling it. The information in workingmemory is subject to frequent turnover as well.If, in the above example, you are interrupted incarrying out the directions by a telephone call,you are likely to have lost some or all of theinformation needed to carry out the alignment.One of the features of working memory thatmakes it useful is that the information that isstored is rapidly accessible. Another feature isthat the information can be manipulated andtransformed to meet the needs of various cog-nitive tasks, from preparing a Powerpoint talk to

solving mental arithmetic problems to reason-ing deductively. Working memory, in short, isindispensable for normal cognition.

We know a good deal about the psycholog-ical processes that underlie working mem-ory. While there are several models of theseprocesses that have been proposed (e.g.,




Baddeley, 1986; Cowan, 1995; Engle, 2002),the model from Baddeley and colleagues hasmotivated most research. According to thismodel, memoranda are stored in a set of buff-

ers that differ from one another in the type of information each stores. The most frequentlydiscussed of these buffers includes one forverbal information, one for spatial information,one for visual information that is not spatial,and one for episodic information that is nottied to the modality of coding (Baddeley,2001). By some accounts, each buffer has arehearsal process associated with it that allowsattention to be paid to the representations stored

in the buffer. Information in the buffers issubject to manipulation by executive processes,which in turn allow attention to be paid torelevant representations and shifted from onerepresentation to another as needed for sometask. For example, some of the contents of working memory may be critical to some task(e.g., remembering that the ‘‘Draw’’ item in thetool bar is the one that contains the ‘‘align-ment’’ tool) whereas others may be irrelevant

(e.g., the text tool that allows you to add text toa slide); it is the allocation of attention by ex-ecutive processes that allows us to focus on therelevant information and gate out what is ir-relevant.

There is much to recommend this model of the psychological processes underlying workingmemory. Even our view of the more ubiquitousrole of executive processes shares the assump-tion that attentional allocation is fundamentalto working memory, as the model of Baddeleyand colleagues argues. Moreover, the emphasison attentional control is common to othertheorists as well. Engle and colleagues (Engle,2002; see also Chapter 2, this volume) arguethat variation in working memory capacity canbe explained by variation in executive attention.

Along similar lines, Hasher and colleagues(Chapter 9) emphasize executive processes as

a significant source of individual variation,although they stress the inhibitory or gatingfunction of attentional control. Whichever viewone takes, it is timely to try to understand theneural implementation of the model’s compo-nents. What are the sites of storage? What arethe mechanisms of rehearsal and is rehearsal

applicable to all of the buffers? How are exec-utive processes represented in the brain? Is therea single executive processor, or are there severalthat mediate multiple types of executive func-

tions (e.g., inhibiting irrelevant information vs.shifting attention from one representation toanother)? In recent years, a good deal of exper-imental attention has been paid to these ques-tions with good productivity, bothfrom noninva-sive studies of humans and from invasive studiesof animals other than humans. Consider, forexample, some of what we know about the sitesof storage (Fig. 10.1, see color insert). By now,there is very good evidence that implicates a

network of brain areas that includes sites in oc-cipital, parietal, and frontal cortex in workingmemory for spatial information. Strong hintsabout parts of this circuit have come fromstudies of monkeys storing spatial informationin working memory tasks. We can trace the or-igins of these studies to the classic article byJacobsen (1935), who found that lesions todorsolateral prefrontal cortex in monkeys im-paired their performance in a spatial delayed-

response task. This work has led to studies of single-cell performance in spatial workingmemory tasks that have uncovered the impor-tance of cells in dorsolateral prefrontal cortexand superior posterior parietal cortex (e.g.,Chafee & Goldman-Rakic, 2000). In humans,the network may be even more extensive thanthis. There are several studies that show activa-tion in superior prefrontal cortex near the su-perior frontal sulcus tied to the retentioninterval of a spatial delayed-response task (e.g.,Courtney, Petit, Maisog, Ungerleider, & Haxby,1998; Jonides et al., 1993). We also know frommany studies that parietal cortex in the area of the superior parietal lobule and intraparietalsulcus is activated by spatial storage (e.g., Jo-nides et al., 1993). Furthermore, Awh and Jo-nides (2001) have shown that there is upwardmodulation of extrastriate occipital cortex when

participants have to store spatial information incontrast to nonspatial storage. This network of frontal, parietal, and occipital sites for the stor-age of spatial information is reminiscent of thedorsal stream of processing that has been im-plicated in the perception of spatial features of visual displays (discussed below).




We also have good information about thecircuit involved in the storage of visual infor-mation that is not spatial in character. Again,some of the footing for our understandingof this circuitry comes from invasive studiesof monkeys. Miller and Desimone (1994), forexample, have shown that inferotemporal cor-tex contains cells that are responsive to partic-ular visual stimuli not only when those stimuliare present in the visual environment, but alsowhen the experimental animal is required tostore a memory of a stimulus in preparation fora delayed match-to-sample task. These cellsseem to be subject to interference from othervisual stimulation, however. By contrast, cellsin inferior prefrontal cortex in these animalsare also selectively responsive to visual stimuliduring the delay period of a working memorytask, and these cells are less subject to inter-

ference effects. Evidence from human studiescorroborates the involvement of inferior tem-poral and inferior frontal sites in memory forvisual nonspatial stimuli. Smith et al. (1995)showed this in a PET experiment in whichgeometric objects were the memoranda. And

several studies have shown that when humanfaces are the memoranda, there is also activa-tion of inferior temporal regions. Indeed, itappears likely that the very same region in-volved in the perception of faces in the fusi-form gyrus is activated during the retentioninterval of a task in which faces must be stored(Druzgal and D’Esposito, 2003). So, again thereappears to be a parallel between the brain re-gions responsible for perception of visual in-formation and those responsible for its storage.

Although more complex, the same principlemay be true for the storage of verbal material.Postle, Berger, and D’Esposito (1999), forexample, have shown that there is activation of superior temporal and inferior parietal corticesduring the storage of verbal material in workingmemory. These are areas that are clearly in-volved in verbal processing for incoming in-

formation. An interesting synthesis of various lines of research on the parallel between perceptionand storage can be seen in a recent meta-analysis of the working memory literature(Wager & Smith, 2003). In this meta-analysis,

Figure 10.1. Lateral viewof the left hemisphere il-lustrating the location of Brodmann’s areas (nu-merical codes) known toparticipate in aspects of working memory. Colorcoding is used to identifythe neuroanatomical re-

gions in which theseBrodmann’s areas are si-tuated. SMA ¼ supple-mentary motor cortex;PFC¼ prefrontal cortex;DLPFC¼ dorsolateralPFC; VLPFC¼ ventro-lateral PFC.




the investigators collected activations from over60 imaging studies of working memory usingPET and fMRI. They then conducted a clusteranalysis on the peaks of activation that were

reported in these various studies to examine anumber of issues. One had to do with whetherthere were differences in activation sites as afunction of the type of material stored. Inparticular, the meta-analysis had a substan-tial corpus of activation sites for spatial storageand for storage of object information. The clearoutcome of this analysis was that spatial storageled to activation of superior posterior siteswhereas object storage led to activation of in-

ferior posterior sites. This result summarizesthe general point we made above, that there isa seeming parallel between storage and per-ception. Viewed from an evolutionary point of view, this parallel makes architectural sense.Evolution led to the development of special-ized brain mechanisms for the processing of sensory information in an exquisite way. Hav-ing evolved, these mechanisms may then havebeen harnessed for memory purposes as well as

perceptual purposes. In principle, what is re-quired for this to happen is the development of hysteresis in the perceptual processing streams,so that the activation of neural structures by asensory event might then outlast that event. Inthis way, the very same structures that respondto a sensory event then continue to respond inthe absence of that event, at least for a shortperiod of time. This would be an efficient wayto construct a system that had both selectiveperception and selective working memory forenvironmental events.

Even in the face of this obvious evolutionaryadvantage, there remains the indisputable factthat not only do posterior regions of cortexshow activation in working memory tasks, butso do frontal regions. What might be the func-tions of these two systems? One possibility issuggested by a classic result documented by

Malmo (1942). He showed that the deficit inspatial delayed responding caused by lesions infrontal cortex of monkeys (the result first docu-mented by Jacobsen, 1935) could be eliminatedby placing the animals in a darkened environ-ment. That is, in the light, damage to dorso-

lateral prefrontal cortex impairs performancein spatial delayed responding, but in the dark,it does not. A ready interpretation of this effectis that the animals have sources of potential

visual distraction in a lighted environment thatare not present in the dark. So, perhaps thefunction of the frontal mechanisms of workingmemory hinge on heightening resistance todistraction. In single-cell studies, Miller andDesimone (1994) confirmed this position byshowing that inferotemporal neuronal activa-tion in a working memory task for objects wasinterrupted by interfering visual events whereasfrontal activation was not. Likewise, studies by

Constantinidis and Steinmetz (1996) and diPelligrino and Wise (1993) make the samepoint about spatial working memory. That is,posterior representations of spatial informa-tion are disrupted by interfering informa-tion, whereas frontal representations are not.So, the frontal mechanism of working memorymay become engaged by resistance to inter-ference while the posterior mechanism may besensitive to such interference.

This hypothesis has received support from ameta-analysis of patients with deficits in work-ing memory conducted by D’Esposito andPostle(1999).Theysurveyedtheliteraturetoiso-late 11 studies of patients who showed im-paired working memory in spatial and verbalspan tasks. They found that damage to thefrontal cortex did not appear to predict workingmemory impairments for either kind of mate-rial, whereas damage to temporal cortex did. Bycontrast, when patients were confronted withdelayed-response tasks in which there was dis-tracting information presented during the re-tention interval, frontal damage did predictworking memory declines. A recent fMRI studyby Sakai, Rowe, and Passingham (2002) makesa similar point. When interference was intro-duced during the retention interval of a work-ing memory task, activation of Brodmann’s

area 46 differentiated trials in which memorywas successful from those in which it was not.Thus studies of monkeys and humans appear toconverge on the idea that frontally mediatedprocesses are critical for establishing more re-silient representations in working memory.




This view is examined in more detail in the fol-lowing section.

ATTENTIONAL CONTROLIN THE SERVICE OF SELECTIONAND ACTIVATION

If regions of posterior cortex participate in stor-age and if they are sensitive to the effects of in-terference, there must be a mechanism thatworks against this interference to create a longer-lasting representation, perhaps the one medi-ated by frontal cortical sites of working memory.

For verbal material, we have known for a longtime that this mechanism involves the recyclingof a phonological or articulatory code during aretention interval to keep it fresh (see, e.g.,Naveh-Benjamin & Jonides, 1984). The brainmechanisms that underlie this recycling appearto depend on a system that includes parietal andfrontal components (e.g., Awh et al., 1996;Corbetta et al., 2002; Ravizza et al., 2004). Theparietal components are dependent principally

on cells in intraparietal sulcus and superiorparietal lobule. We know from much work onthe processing of spatial displays that this systemis a key player in allocating and re-allocatingattention to places in space (see Kastner &Ungerleider, 2000). We also know that this samesystem is not restricted in its attention con-trol function to spatial stimuli; rather, it appearsto play a much more general role in allowing usto switch attention from one object to another,whether those objects are perceptual or memo-rial and whether they are spatial locations, rules,tasks, attributes, or visual objects (Yantis, 2003).

With this evidence in hand, it seemsreasonable to hypothesize that the parietalcomponent of verbal rehearsal may be involvedin controlling whether our attention is placedon one verbal object or another. It is thiscontrol that would allow us to cycle through

the individual items in our verbal storage buf-fer to keep them fresh. The other componentsof the system are frontal and may include re-gions that mediate either domain-general ordomain-specific processes. One anterior com-ponent of the verbal rehearsal circuit appearsto be located in left inferior frontal gyrus and

possibly left insula cortex. This region is thesubject of a good deal of discussion. The classicinterpretation of its function includes controlof articulation and internal speech (e.g., Pau-

lesu, Frith, & Frackowiak, 1993). By this ac-count, it seems sensible to hypothesize that it isthis site that represents verbal information andover which attention has its control as focus isshifted from one item in working memory toanother. Another possibility, however, is that acritical function of left inferior frontal gyrus isto control the selection of a piece of informa-tion among many (Kan & Thompson-Schill,2004). By this account, the role of this region

in rehearsal may be to select items in turn forattention among the several items held in mem-ory. That is, if the function of this left inferiorfrontal region is to allow selection of one repre-sentation among several alternatives, thiswould fit the function of rehearsal quite well.By this account, rehearsal amounts to cyclingthrough several representations in turn, re-quiring selection of which representation is tobe the focus of attention in each time epoch.

Regardless of whether this site is important forverbal representation or selection, this regionunquestionably plays an important role inrecycling information in the service of verbalmaintenance.

Rehearsal is not restricted to verbal infor-mation. Awh and Jonides (2001) have shownthat the concept of rehearsal is equally appli-cable to spatial information. Consider the spa-tial delayed-response task we discussed above.Several visual locations are marked on a screen,and then a delay interval of a few seconds en-sues, during which the locations must bestored. Then a single location is marked, andthe observer must decide whether this locationis one of those stored in memory. If rehearsal isinvolved in this task, what exactly is rehearsed?There are data showing that extrastriate andsuperior parietal cortices play a role in this task

(e.g., Jonides et al., 1993). We know that atleast a portion of these regions of cortex have atopographical organization such that adjacentlocations in the visual world are represented byadjacent locations in the brain. Thus, the spa-tial topography of a set of locations in spacecan be preserved by the spatial topography of




working memory. Under this model, rehearsalmay involve the repeated activation of the re-presented positions in space by activation of their respective neural representations. As with

verbal information, control over this cycling of activation of the extrastriate sites may be ex-erted by superior parietal cortex. This link hasyet to be firmly established, but what is estab-lished is that there is modulation of extrastriatecortex when spatial information must be stored.

Awh and Jonides (2001) have reviewed the rel-evant evidence for this case, and it consists of both behavioral and imaging findings. Behav-iorally, Awh, Jonides, and Reuter-Lorenz (1998)

showed that if an observer has to store a locationfor several seconds, that storage produces abenefit in performing a visual discriminationtask at the same location, compared to per-forming one at some other location. Further-more, diverting attention away from the storedlocation (by having an observer perform a dis-crimination at some other location) reducesperformance in the spatial working memorytask. Awh et al. (1999) followed up these find-

ings with an imaging experiment. They hadparticipants store three locations in workingmemory for a retention interval during which avisual stimulus was presented. This produced anup-regulation of extrastriate cortex, compared toa control condition in which verbal informationhad to be stored. A ready interpretation of theseresults is that rehearsal processes (possibly con-trolled by parietal mechanisms) exert an in-fluence on the activation of the relevant extra-striate regions, keeping them active during theretention interval.

These characterizations of rehearsal forverbal and spatial information share the ideathat there is a source of the rehearsal signal anda site for its action. The source may be acommon mechanism in parietal cortex thatdirects attention to relevant memorial repre-sentations. The parietal mechanism is likely to

act in concert with prefrontal mechanisms (i.e.,in inferior frontal gyrus and superior prefrontalcortex, see below) to select and shift betweenthe relevant representations. The site of acti-vation may vary with the type of informationbeing stored. In the case of verbal information,the source may control an articulatory or

phonological representation in inferior frontalcortex; in the case of spatial information, itmay control a spatial representation in extra-striate cortex. There may be other sites over

which parietal cortex can exert control as well(see research by Yantis, 2003), but our researchthus far does not go much beyond verbal andspatial storage.

One important source of information aboutthe hypothesis that attentional mechanismsplay a role in rehearsal comes from brain evi-dence on the mechanisms used to control at-tention to spatial locations and the mechanismsthat control spatial rehearsal. Kastner and

Ungerleider (2000) and Wager, Jonides, andReading (2004) have surveyed the literatureconcerned with the allocation of attention. Bothgroups have found a network of regions thatinclude parietal cortex, superior frontal cortex,and medial frontal cortex that are recruitedwhen subjects face the task of allocating theirattention to regions of space and when they haveto shift attention from one region to another.

Awh and Jonides (1998) examined the literature

and discovered that many of the same regionsinvolved in the allocation of attention are alsoinvolved in spatial working memory. Thisparallel allows us to draw the conclusion thatspatial working memory has harnessed mecha-nisms of spatial attention to help retain material,presumably by controlling rehearsal processes.What is even more interesting about this argu-ment is that Wager et al. (2004) found that thevery same mechanisms, by and large, participatein many shifts of attention, whether to spatiallocations, objects, attributes, rules, or tasks. Thisfinding leads to the further hypothesis, as yetuntested, that the attention allocation mecha-nism involved in verbal and spatial workingmemory may also be one involved widely in allsorts of rehearsal.

This review of storage and rehearsal mech-anisms leads us to the following caricature:

much of storage involves the very same mech-anisms that are involved in perception, andmuch of rehearsal involves the mechanismsthat are involved in shifts of attention in thevisual world. Although probably too broada con-clusion, it leads one to suppose that what hashappened in the evolution of working memory




is a co-opting of mechanisms that developedfor perceptual purposes. Of course, there areholes in this argument, not the least of which isthe importance of frontal mechanisms of stor-

age, especially in the face of interference. Also,we have a good many gaps in our knowledge of storage for information that is not presentedvisually. Indeed, even auditory presentation hasbeen neglected in imaging studies of storageand rehearsal. But the hypothesis does presenta guiding principle that may lead to productiveacquisition of more data to fill in the gaps.

WORKING MEMORY CIRCUITRYAND TASK DEMANDS

As we discussed at the start of this chapter,within the working memory literature, and es-pecially in the field of cognitive aging, workingmemory tasks are considered dichotomously toeither emphasize or downplay executive pro-cessing. A distinction is often made betweenrote short-term memory tasks, the so-called

storage-only tasks, and working memory tasksthat require ‘‘storage plus processing.’’ Thisdistinction may be useful as a description of thetask design; however, it is often taken to meanmore than that, and to refer to neural proces-sing operations that are presumed to underliethe tasks.

Consider, for example, the item recognitiontask, a canonical storage-only task, in which anumber of items are presented to be held inworking memory for several seconds. After thecompletion of the retention interval, a probeitem is presented and the participant mustdecide whether it matches one of the itemsin memory (Sternberg, 1966). Although theencoding, maintenance, and retrieval requiredto perform this task are not typically assumedto tax executive processes, the neuroimagingevidence reviewed above indicates that the

frontal–parietal executive circuit is actively re-cruited when item recognition tasks are per-formed. No matter how seemingly simple thetask, working memory tasks involve the selectiveactivation and inhibition of stored information,even if only in the service of maintaining thatinformation. Attentional control over internal

representations is entailed in rehearsal. So, onthis count alone, item recognition must involvesome degree of executive control. As we reviewbelow, there is now ample evidence that a

storage task combined with other factors, suchas larger memory load, particular task contexts,and the age and performance level of the in-dividual, can further modulate the recruitmentof brain regions that mediate executive pro-cesses.

Of course, building explicit processing re-quirements into a task will recruit mental op-erations to meet such requirements. Considerthe more complex 2-back task, which makes

obvious demands on storage-plus-processingoperations. In this task, a series of single itemsis presented. For each item, the participantmust decide whether it matches the item thatappeared two items back in the series. This taskalso involves rehearsal in that the participantmust rehearse the set of items held in memory.However, successful performance on the taskrequires in addition that the participantdrop no-longer relevant items from memory

(e.g., the one that is now three items back), addnew items as each is presented, and assign theproper ‘‘back’’ tag to the items in memory. Byany definition of executive processing, this taskinvolves more than the item recognition task.These examples make clear how differentprocessing requirements can be built into thetask to invoke different demands on executiveprocessing.

Nonetheless, it is well documented thatseemingly minor variations of a ‘‘storage-only’’task can recruit executive processing regions.Consider the study by Barch and colleagues(1997), in which the neural consequences of different variations of task difficulty were ex-amined. In their continuous-performance taskthe subject had to respond to a pre-specifiedletter sequence, and refrain from respondingto non-target sequences. Perceptual difficulty

was varied by degrading the visual quality of the letters. Increased memory demands wereinduced by increasing the delay between con-secutive letters. Relative to the short delay, thelong-delay condition was associated with greateractivation in left dorsolateral prefrontal cortexand left parietal areas 40/7. By contrast, the




perceptual degradation did not affect executiverecruitment, despite reducing accuracy and re-sponse speed. The increased difficulty of per-ceptual degradation did lead to greater activa-

tion of the anterior cingulate. These resultsindicate that executive recruitment is not auniversal response to increased task difficulty,but is a more specific response to increasedworking memory demands.

A number of studies have manipulatedmemory load, or the number of items to beretained in working memory in a ‘‘storage-only’’item recognition task. Without introducing ex-plicit changes in the executive processing de-

mands (e.g., manipulation of the items, ordual-task demands), activity in prefrontal re-gions associated with executive control increaseswith increasing set size. For example, Rypma,Prabhakaran, Desmond, Glover, and Gabrieli(1999) compared three- with six-letter memorysets and found greater dorsolateral prefrontalactivation during encoding of the larger set. Ina 2002 report by Rypma, Berger, and D’Espo-sito, dorsolateral prefrontal cortex activation

was found to be load sensitive during the re-tention interval. Ventrolateral prefrontal cortexwas also affected by load, but only for high-performing subjects. Likewise, for tasks requir-ing the short-term retention of spatial locations,load effects have been found in dorsolateral,ventrolateral, and medial prefrontal sites, andin parietal cortex (Glahn et al., 2002; Leung,Gore, & Goldman-Rakic, 2002). At least someparietal–frontal components of this executivecircuit may be domain general in that theycontribute to item recognition of both verbaland nonverbal materials (Ravizza et al., 2004).Speer, Jacoby, and Braver (2003) have furthershown that the context in which a particularmemory load is experienced can influence thecircuitry recruited to perform the task. Theypresented participants with an item recognitiontask in which the size of the set of items to be

held in memory varied from trial to trial. Inone condition, the memory set varied in sizefrom 3 to 6 items, and in the other, from 6 to11 items. Both conditions included a memoryload of six items, and the investigators com-pared brain activations for this identical mem-ory load depending on the context in which it

occurred. The hypothesis was that executivecontrol strategies would be more preparatory(proactive) in nature when participants ex-pected smaller loads because they would an-

ticipate being able to manage the number of items being presented. However, when theyexpected larger loads that typically exceed work-ing memory capacity, their strategies would bemore reactive. Indeed, the authors found tem-poral variations in the pattern of prefrontalactivation. When the load of six was embeddedamong smaller loads, these areas were re-cruited early in the trial, whereas these regionswere recruited later in the trial when the six-item

load was embedded in trials with longer lists.Taken together these results demonstrate thatexecutive control processes are recruited andplay an integral role in the tasks that explicitlyrequire only storage.

By manipulating the context of individualtrials in a working memory storage task, our labhas induced different forms of conflict andobserved different patterns of prefrontal re-cruitment in response to these demands. Con-

flict is created by varying the overlap of itemson consecutive trials. Familiarity conflict oc-curs when a negative probe letter on trial Nþ 1was a member of the memory set on trial N.Under these circumstances, the probe requiresthe subject to respond ‘‘no’’ because it is not amember of the current set; however, becausethe probe was a member of the previous set, itis highly familiar and likely to promote a ‘‘yes’’response. Indeed, it takes longer to respond‘‘no’’ to these familiar probes than to probesthat are unfamiliar (Jonides, Marshuetz, Smith,Koeppe,& Reuter-Lorenz, 2000; Jonides,Smith,Marshuetz, Koeppe, and Reuter-Lorenz, 1998).Moreover, this additional processing time isassociated with the increased activation of ventral prefrontal cortex, Brodmann’s areas 44/ 45 along the inferior frontal gyrus. Althoughdistinct from the more dorsolateral sites in

Brodmann’s areas 9/46 that are widely viewedas the frontal sites of executive control (e.g.,D’Esposito, Postle, Ballard, & Lease 1999;Petrides, 1994), this ventral region is clearlyresponsive to the increased demand for selec-tion and interference resolution that is evokedby overlapping probes. We find that when the




magnitude of familiarity conflict is variedparametrically, for example, when the letter r appears in the target set of trial N and trialNþ 1, and then appears as a negative probe in

trial Nþ 2, response time and left inferiorfrontal gyrus activation increase linearly. De-spite these performance and neural signatures,the cognitive demand produced by overlappingprobes goes unnoticed by the research partici-pants according to their self-reports. Moreover,this form of conflict does not activate the an-terior cingulate gyrus, an area that has beenwidely associated with increased task difficulty(Nelson, Reuter-Lorenz, Sylvester, Jonides, &

Smith, 2003) as well as cognitive control (Bot-vinick, Braver, Barch, Carter, & Cohen, 2001;Gehring, Goss, Coles, Meyer, & Donchin,1993). Nonetheless, contextual manipulationsof the trials within a task that otherwise demandsstorage, rehearsal, and retrieval operationsclearly recruit executive control processes tomediate the selection among competing codes.

Using a simple variation of the overlap de-sign, a different form of conflict can be en-

gendered that does recruit anterior cingulate. Again, the probe on trial Nþ 1 is a member of the target set on trial N, but in addition thisitem was a positive probe on that precedingtrial and thus is associated with a recent‘‘yes’’ response. The behavioral effects of thisresponse conflict are indistinguishable fromthe effects of highly familiar probes. However,these two forms of conflict produce dissociableneural signatures, in that response conflict ac-tivates anterior cingulate but does not produceany further increases in the activation of ventralprefrontal cortex. Particularly relevant for thepresent discussion is the fact that these controlprocesses are recruited merely by the contex-tual manipulation of a task widely viewed asone that involves only storage.

AGING, WORKING MEMORY, ANDATTENTIONAL CONTROL

The importance of executive processes in ca-nonical storage-only tasks is especially apparentin the neuroimaging evidence from studies of older adults. This evidence challenges a pro-

minent view that emerged from several decadesof behavioral research that claimed minimaleffects of aging on the performance of workingmemory tasks, so long as they required simple

storage (e.g., Craik, 1977; Craik & Jennings,1992; Craik, Morris & Gick, 1990; Dobbs &Rule, 1989). This relative preservation of per-formance on simple storage tasks was in contrastto marked performance declines on executiveprocessing tasks. For example, in a well-citedstudy by Dobbs and Rule (1989), there were noage differences on the Brown-Peterson short-term memory task, but working memory tasks,similar to the N-back task described above,

produced reliable age differences in perfor-mance. Data such as these have been taken toimply that storage capacity is less affected by agethan the capacity or speed with which executiveprocessing operations can be performed. In-deed, solid support for the conclusion thatstorage was relatively spared came from a num-ber of empirical reports and a meta-analysis (seeReuter-Lorenz & Sylvester, 2004, for a review;Babcock & Salthouse, 1990). These behavioral

results lead naturally to the simple-mindedneuroimaging prediction that older and youn-ger adults should show similar activation pat-terns when performing tasks that emphasizestorage, whereas marked age differences shouldemerge when explicit executive processing de-mands are entailed in the tasks. Neuroimag-ing studies comparing older and younger adultsreveal a very different picture, however, andargue further for the pervasive nature of exec-utive control in working memory tasks, in-cluding those that emphasize storage.

Since 1998, there have been at least 10published reports describing the effects of ageon the activation patterns evoked by workingmemory tasks that emphasize storage. Althoughthe detailed pictures differ somewhat from onestudy to the next, in all cases marked age dif-ferences in brain activity have emerged even

when performance differences have not. Oneof the most consistent findings is that olderadults activate regions of prefrontal cortexthat are not significantly activated in the youn-ger group. In some cases this age-relatedoveractivation is accompanied by age-relatedunderactivation in other regions, and in some




cases it is not. For example, in one study of spatial working memory from our lab, subjectswere required to remember the location of threedots appearing briefly on a computer screen.

Younger adults showed lateralized right frontalactivation in premotor and supplementary mo-tor cortex, whereas older adults, who performedas accurately as the young adults, activated thesesame regions bilaterally (Reuter-Lorenz, 2002;Reuter-Lorenz et al., 2000). Older adults alsoshowed activation of left dorsolateral prefrontalcortex,whereasyoungeradultsdidnot.Likewise,in a study requiring the visual maintenance of gabor patches, which are non-nameable ele-

mentary visual stimuli, McIntosh and col-leagues (1999) found that older adults showedmore activation than younger adults in leftdorsolateral prefrontal cortex and less activationin ventral prefrontal regions of the right hemi-sphere.

Similar patterns are also evident in the verbaldomain. We have found that like young adults,older adults activate verbal working memoryareas in the left-hemisphere. However, older

adults also activate additional regions of ventro-lateral and dorsolateral prefrontal cortex in theright hemisphere that are not activated by youn-ger adults performing the same task (Reuter-Lorenz et al., 2000; see also Cabeza et al.,2004). This pattern is evident in Fig 10.2 (seecolor insert) Although age-related overactiva-tion is not always found for verbal storage tasks(Rypma & D’Esposito, 2001), greater activationof dorsolateral prefrontal cortex is consistentlyassociated with higher performance in the oldergroup, even when young adults show an inversecorrelation. A positive correlation between dor-solateral prefrontal cortex activation and per-formance has also been found during verbalmaintenance for older adults (Reuter-Lorenzet al., 2000, Reuter-Lorenz, Marshuetz, Hartley,Jonides, & Smith, 2001). This positive rela-tionship suggests that prefrontal recruitment is

beneficial to seniors’ performance of storagetasks.What functions might be subserved by such

regions of overactivation in older adults? Onepossibility that we favor is that older adults relyincreasingly on attentional control and otherexecutive processes to manage the memory

load on storage tasks. Even in the absence of explicit executive processing task demands,executive recruitment could enhance attentionto context, inhibit distraction, and increase

cognitive control. We submit that in the agingbrain, additional executive resources are neededsimply to maintain items in memory. This in-terpretation makes sense in view of the fact thateven in younger adults an increase in workingmemory demand leads to greater activation of the prefrontal–parietal executive circuitry.

But if the age-related increases in executiveprocessing serve a compensatory function, whatare they compensating for? We believe that the

answer to this question lies in the more wide-spread alterations in neural functioning thataccompany normal aging (see Reuter-Lorenz& Lustig, 2005 for a review). Aging not onlydecreases synaptic efficiency (e.g., Kemper,1994) but also appears to be associated with anincrease in neural ‘‘noise’’ (e.g., Li, Linden-berger, & Sikstrom, 2001). Single-unit record-ings in rodents and monkeys indicate that thereis a decline in the selectivity of receptive

field properties in primary sensory cortices(see Godde, Berkefeld, David-Jurgens, & Dinse,2002, for a review; Schmolesky, Wang, Pu, &Leventhal, 2000). A breakdown in selectivitymeans a corresponding decline in the distinc-tiveness of the neural response to sensory input.Similar effects have recently been documentedin humans as well. Through use of fMRI, Parkand colleagues (2004) have shown that thespecific and localized activity associated withthe encoding of such stimuli as faces, places,and letters breaks down in older adults. Theyfound that whereas young adults have discreteand separable regions of localized activity inextrastriate cortex in response to these differentclasses of stimuli, the activation patterns in olderadults lack this selectivity, so that the brainregion that responds to places also responds tofaces, and vice versa. Thus the older brain is less

able to generate distinctive representations of in-coming stimuli, and may be less able to reac-tivate distinct representations stored in memoryas well. We submit that this compromise inrepresentational processes leads to an increasedreliance on attentional control and other ex-ecutive processes that normally play an integral




part in working memory but bear an increasingburden in the older brain. That is, age-relatedincreases in the recruitment of attentionalprocesses compensate for declines in thequality of the stored information and in thestorage operations themselves (cf. Gutchesset al., 2005).

It is important to point out, however, that an

age-related tendency to ‘‘over-recruit’’ executiveprocesses need not imply that these processesare spared the effects of aging. On the contrary,there is strong evidence that prefrontal regionsin particular suffer the consequences of normalaging. By measuring gray-matter volumes ob-tained from structural MRI scans, Raz and

colleagues (e.g., Raz, 2000) have documentedthat lateral prefrontal cortex is one of the re-gions that shows the greatest loss of volume inolder adults. Although this structural evidencesuggests the view that prefrontal overactivationin seniors is a sign of dysfunction in these re-gions, we find this interpretation unsatisfactory.If age differences in activation were caused by

differences in the sheer amount of neural tis-sue, then we would expect less activation inolder than in young adults, rather than more.While it is conceivable that the remaining ex-ecutive circuitry in these areas has to ‘‘workharder’’ to achieve the same output, this stilldoesn’t explain why the net activation levels

Figure 10.2. Surface-rendered images (left lateral, superior, and right lateral views) of PET acti-vations obtained from younger and older adults performing a verbal working memory task re-quiring the maintenance only of four letters. The predominance of left-sided activation in theregions of younger adults is evident in contrast to the left- and right-sided activations evident in theolder group. For more details see Reuter-Lorenz et al. (2000).




would be greater in seniors than in youngeradults. Although shrinkage of these areas iscorrelated with poorer performance on theWisconsin Card-Sorting Task (Gunning-Dixon

& Raz, 2003), a task known to rely on execu-tive processes mediated by lateral prefrontalcortex, it does not predict poorer performanceon measures of working memory. We takethese results to mean that despite age-related shrinkage, prefrontal regions can con-tribute to the executive demands of workingmemory and even do so in a compensatorymanner. But what happens in the older brainwhen executive processes are an explicit part of

the task? If older adults rely more on executiveprocesses at lower levels of task demand thanyoung adults, will older adults also reach theirresource limit at lower levels of task difficultycompared to young adults? We think this is ex-actly what happens (see also Reuter-Lorenz &Mikels, 2006).

Consider the age differences that have beenfound on several tasks that entail the explicitexecutive processing components required to

switch between two tasks. In one study by Di-Giralamo colleagues (2001) the performanceof younger and older adults was compared ontwo classification tasks that were performedseparately or intermixed. One task requiredsubjects to decide whether the number of digitsappearing in a string was greater or less thanfive, and the other task required them to decidewhether the value of the digits in the string wasgreater or less than five. The time it takes tomake each of these decisions is considerablylonger when the tasks are randomly intermixedthan when each task is done repeatedly in ablock of trials without the requirement toswitch back and forth between rules. Indeed,the switching cost, which is a measure of thedifference between pure-block and switch-block performance, is greater for older than foryounger adults. Neuroimaging measures ob-

tained during pure and switch blocks indicatedthat younger adults activated dorsolateral pre-frontal cortex and medial prefrontal (includinganterior cingulate) areas when switching wasrequired. By contrast, older adults activatedthese regions even in the pure-block conditions,and consequently showed less of an increment

in prefrontal activation in response to the actualswitching demand.

We found a comparable outcome in aworking memory task that required subjects to

encode and retain a short list of unrelatedwords while verifying a series of mathematicalequations (Smith et al., 2001). The best-performing young adults did not need to re-cruit dorsolateral prefrontal cortex to managethese dual task requirements, but poorer-performing young adults and older adults didrecruit this region when performing the task.Older adults also showed significant activationin premotor and parietal regions that was not

evident in the younger groups, as would beexpected if the attentional circuitry is increas-ingly taxed by age-related decline. The picturethat is emerging, then, is one of older adultsbeing more likely to recruit additional brainregions that are not activated by younger adultsperforming the same task. Moreover, the addi-tional regions recruited can arguably be linkedto the mediation of executive processes, namelyattentional control, which may be recruited by

this age group even when the task itself does notexplicitly make this demand.

Thus far our discussion of aging has focusedlargely on executive processing generally andattentional control in particular. However, wehave said little about inhibitory processing inthe aging brain. Hasher, Zacks, and colleagues(see Chapter 9) propose that declines in in-hibitory control play a central role in age-related changes in cognition. Their view wouldpredict that older adults should show less ac-tivation than young adults in brain regionsthought to mediate inhibitory control. As wementioned previously, a region of left inferiorfrontal gyrus is activated by younger adultswhen they confront the need to resolve inter-ference between the tendency to respond ‘‘yes’’to an item because it is highly familiar and thetendency to respond ‘‘no’’ to the item because

it is not part of the current target set (Jonideset al., 1998). Some form of inhibitory control islikely to be critical in resolving the interferenceproduced by these conflicting codes. We haveshown that older adults are not only dis-proportionately challenged when respondingduring these high-familiarity trials but also fail




to activate the region of inferior frontal gyrusrecruited by younger adults. This outcome isconsistent with the kind of inhibitory deficitsproposed by Hasher and Zacks (see also Gaz-

zaley et al, 2005). At the same time, however,these older adults display increased recruit-ment of what we have referred to as the executiveattentional circuitry mediated by prefrontal andparietal regions, and this pattern is associatedwith better performance overall (Reuter-Lorenzet al., 2001). Thus there appears to be a disso-ciation between older adults’ ability to recruitthe mechanisms mediating the inhibitory con-trol required to adjudicate between competing

codes and their recruitment of attentional con-trol processes engaged during rehearsal andencoding. We are currently investigating whe-ther there are individual differences withinolder or younger age groups that influence therelationship between an individual’s ability toengage interference resolution processes in in-ferior frontal gyrus and his or her tendency toactivate attentional control processes moregenerally.

We should also note that one of the mostprominent ‘‘signatures’’ of neurocognitive agingthat we have observed is an increase in bilateralactivity under task conditions that evoke later-alized activity in the younger adult. For verbalworking memory tasks that typically produceleft dominant activation, and for spatial loca-tion memory tasks that evoke right dominantactivation, older adults show both left and rightactivity in frontal and parietal regions. Thisincreased bilateral activity suggests that there isa breakdown of domain specificity as we age,such that older adults recruit neural resourcesfrom both verbal and nonverbal domains tocompensate for more general neural declines.This neural pattern predicts that older adultswill show greater interference between verbaland spatial tasks in a dual-task situation. How-ever, as Hale, Myerson, Emery, Lawrence, and

DuFault report (Chapter 8), domain specific-ity is preserved in older adults. Although themeaning of this discrepancy between the be-havioral and imaging results is unclear, it isclear that resolving it will require a carefulmapping of the brain areas engaged by primary

and secondary tasks when they are performedseparately in each age group, to examinethe degree of neural overlap. This overlap canthen be used to predict the amount of inter-

ference when the two tasks are performedconcurrently.To summarize then, studies of aging have

revealed the integrality of attentional and ex-ecutive processing operations to tasks that areotherwise viewed as requiring only storage.Compensatory recruitment of attentional pro-cesses may be especially critical to the extentthat representational mechanisms are compro-mised with age (Li et al., 2001; Park et al.,

2004). Our studies of aging bring home themessages conveyed throughout this chapter.Clearly, brain imaging shows that there is nosimple one-to-one mapping of tasks onto braincircuits. Indeed, equivalent performance levelscan recruit different circuitry, and the criti-cal dimension along which variation occursis in recruitment of resources for attentionalcontrol.

CONCLUSIONS AND DISCUSSIONOF FOUR THEMATIC QUESTIONS

QUESTION 1: OVERARCHINGTHEORY OF WORKING MEMORY

What we have reviewed is the basis for asomewhat different view of working memorythan the traditional one attributed to Baddeleyand colleagues and widely held in the cognitiveand aging fields. Their work is largely guidedby a distinction between working memorytasks that engage storage mechanisms alone andthose that engage storage mechanisms plus ex-ecutive processing. We argue instead that allworking memory is best conceived as requiringstorage and the sort of attentional control pro-

cesses that define the central executive. Ourview is based on both animal neurophysiologyand human neuroimaging evidence indicatingthe recruitment of a frontal–parietal executiveattention system in tasks presumed to empha-size only storage.




Specifically, we base our argument on threemajor sources of evidence. First, neuroimag-ing studies reveal a clear overlap in the brainregions required for attentional selection of sen-

sory percepts and the maintenance of the cor-responding representations in working memory.Put boldly, it seems possible that the very samemechanisms that control attention to incomingsensory information control attention to piecesof information in working memory. Prefrontalsites, in particular, that are linked to executivecontrol play an essential role in establishingrepresentational stability in the face of distrac-tion. Second, executive processes are recruited

by increases in load or retention interval, orminor modifications in the context of a task thatotherwise requires only storage. Thus alterationsin functional circuitry can occur without anyexplicit changes in the nature of what getsstored, and they require no knowledge or ex-plicit strategy changes on the part of the subject.Third, comparison of adults of different agesand performance levels indicates that executiveprocesses are readily recruited in tasks that are

nominally only storage tasks. Executive controloperations are thus integral to the circuitry of most working memory tasks, and the explicitstructure of the task itself is but one of the factorsthat determines their recruitment.

Of course, we see the possibility that exec-utive control operations are not all of a type.For example, they may be differentiable by thetype of information on which they operate. Evi-dence about the mutual non-interference be-tween visual and verbal working memory tasksbring this issue to the fore (e.g., Duncan,Martens, & Ward, 1997; Luck & Vogel, 1997).If this is the case, then executive processeswould be mischaracterized by calling them allexamples of attentional allocation. At present,there is simply too little evidence to draw aconclusion about the differentiability of exec-utive processes by type of material or by any

other taxonomy. On the one hand, it does ap-pear that working memory for different types of material may be less mutually interfering thanworking memory for the same type of material,a pattern suggesting at the very least a frac-tionation of executive processing by type of

material. On the other hand, if one looks to therecent literature on attentional processes inperceptual cases, it appears that there may be asingle set of mechanisms that control attention,

regardless of whether the attention is allocatedto spatial locations, objects, or auditory mate-rial (see Yantis, 2003). By analogy to this lit-erature, one might draw a parallel to attentionaloperations on working memory to propose thatthey are also mediated by a single mechanism.This issue is by no means settled, and it cer-tainly warrants further investigation.

One might question whether the tasks thathave appeared prominently in the literature

have simply been too taxing to avoid executiveoperations, even as conceived in the canonicalmodel of Baddeley and colleagues. That is,even according to this model, when the ca-pacity of working memory is exceeded, partic-ipants must employ strategies that will recruitexecutive processing to satisfy task require-ments. For example, they might temporallygroup items to increase their capacity. Withthis in mind, one might ask whether neu-

roimaging studies of working memory have all,by happenstance, stretched the limits of work-ing memory too much to see evidence of a‘‘pure’’ storage process with no executive com-ponent. While possible, we find this view un-likely. Even the most elementary workingmemory paradigm, an item recognition taskwith four or fewer items, yields activation insuperior parietal and/or medial frontal regionsthat are emblematic of attentional engagement(e.g., Smith, Jonides, & Koeppe, 1996). Thus,there appears to be evidence that even the sim-plest cases of working memory require somedegree of attentional control.


We have focused on aging as a critical sourceof variation in working memory. However, wehave departed from the major thrust of thisvolume in that our focus has been less on ex-plaining performance differences and more onthe underlying neural mechanisms and how




their recruitment varies with performance leveland age. Neuroimaging reveals that the samelevel of performance can be achieved in differ-ent ways: older and younger adults can perform

an item recognition task with equal levels of accuracy while relying on different neural cir-cuitry. Much of our discussion about aging hasbeen focused on the source and function of thisneural variation and how it might serve tominimize differences in performance. In fact,the neuroimaging data turn the variance to beexplained on its head, so to speak. Here, wehave aimed to understand the basis for thevariations in neural activation patterns and

what these patterns tell us about how the task isbeing performed.

QUESTION 3: CONSIDERATION OFOTHER SOURCES OF WORKINGMEMORY VARIATION

Although the data we have to support theseclaims are limited, we can speculate that the

source of performance variations differs inyounger and older adults, even though execu-tive processing recruitment figures prominentlyin both age groups. In younger adults it is mostadvantageous to rely minimally on executivecontrol, whereas for older adults, greater reli-ance on executive processes is beneficial toperformance. While we cannot say for certainwhat causes this interaction of performance andage, we believe that different factors are at workacross these age ranges (cf. Rypma, Berger, &D’Esposito, 2002; Rypma & D’Esposito, 2001;Rypma, Prabhakaran, Desmond, & Gabrieli,2001). At the very least we can suggest that forolder adults the increased reliance on executiveprocesses essentially compensates for declin-ing storage and representational processes by‘‘cleaning up’’ and ensuring the durability of thestored representations. In younger adults who

are low performers, greater recruitment of ex-ecutive resources may reflect lower efficiencythan that of their higher-performing counter-parts. Thus in both age groups the efficiencyof executive processing resources is likely toprove an important source of performance var-iation.

Others are now beginning to recognize theimportance of individual-difference variables inaccounting for working memory performance(cf., the other chapters in the other chapters

in this volume). For example, Braver and col-leagues (Chapter 4) conducted a set of studiesshowing that variation in fluid intelligence, asmeasured by the Raven’s Test of ProgressiveMatrices, indicates some important perfor-mance differences in working memory para-digms. By their account, the critical conceptualdimension captured by fluid intelligence iswhether a person is more proactive or morereactive in dealing with potential sources of in-

terference from unwanted information. Proac-tive types can prepare themselves more readilyfor upcoming interference and fend off some of its consequences; reactive types wait for the in-terference to be upon them before dealing withit. Of course, in situations where interferencecannot be predicted, both types must be reac-tive. In one set of studies, Gray, et al., (2003)showed that a difference in fluid intelligencepredicts whether individuals will prepare

themselves for the proactive interference likethat which invades the item recognition taskwe have investigated. Thus, this approachblends nicely with the data we have accumu-lated to suggest a further understanding of thedifferential recruitment of attentional controlby different populations and the other cogni-tive skills that may be correlated with this. Wemight expect, for example, that the greater theexecutive recruitment in older adults, themore likely they would be to exert proactivecontrol.

Our view also interleaves well with the em-phasis on inhibitory declines in aging, expressedby Hasher and colleagues (Chapter 9). To theextent that the deletion of irrelevant items fromworking memory is compromised, there will bean increased need for attentional control tomanage a larger memory load and increased ef-

fects of proactive interference. We haveshown that some aspects of inhibitory controland interference resolution may be com-promised with age, leading perhaps to an in-creased need for compensatory recruitment of other executive processes, including attentionalcontrol.




QUESTION 4: CONTRIBUTIONSTO GENERAL WORKINGMEMORY THEORY

There is little doubt about the groundbreakingconceptual contribution of the differentiationbetween storage and executive processing. Thisdistinction has motivated a great deal of valu-

able research on basic characteristics of workingmemory. We hope we have added to this con-ceptual contribution by suggesting that whathas been called executive processing (what we

have called attentional control) may be evenmore ubiquitous than previously imagined.There may be precious few situations in whichsome level of attentional control is not needed



OF WORKING MEMORY

We agree with Baddeley’s original model of

working memory that postulates storage and

executive components. However, we take is-

sue with the tendency in the field to minimize

the importance of executive operations, spe-

cifically attentional control, in the perfor-

mance of tasks classified as ‘‘short-term’’ rather

than ‘‘working memory’’ tasks. These so-called

storage-only tasks reveal pervasive recruitmentof executive processes that is accentuated

in older adults and poorer-performing young

adults, and when there are minor changes in

the tasks that do not explicitly require item

manipulation or other processes that typically

characterize working memory tasks.


MEMORY VARIATION

While we see aging as a critical source of

variation in working memory performance, our

focus has been less on explaining performance

differences per se and more on the underlying

neural mechanisms and how their recruitment

varies with performance level and age. In

particular, our data and data from other labs

have shown that the same level of task per-

formance in younger and older adults is asso-

ciated with recruitment of different neural cir-cuitry. Our efforts have been directed toward

trying to understand what accounts for this

variation in neural activation patterns and

what these variations tell us about the cogni-

tive operations underlying these tasks.


MEMORY VARIATION

In a general sense, we see the reliance on at-

tentional control as an important source of

variation in working memory performance.

Our working hypothesis is that, to the extent

that attentional control is recruited at lower

levels of task demands, the availability of

control processes will be limited when higher

levels of demand are imposed. When this limit

is reached, performance will suffer. Theindividual-difference variables that are critical

for predicting the availability of attentional

control include age, as mentioned above, and,

as others in the volume have noted, Gf.



By challenging the tendency to reify the dis-

tinction between storage and processing tasks,we hope to emphasize the utility of neural

evidence for clarifying the nature of psycho-

logical constructs. By revealing the involve-

ment of executive brain regions in ‘‘storage’’

tasks, neuroimaging evidence argues against a

simple equation of tasks with underlying pro-

cesses. Such evidence can guide a more pre-

cise taxonomy of the representational and

cognitive operations that mediate working

memory performance. Ultimately, it will becrucial to arrive at a fine-grained character-

ization of the central executive, which will in

turn provide a more accurate vocabulary for

specifying the cognitive demands of the tasks

we employ.




when working memory is called onto the field.Thus, rather than distinguishing between tasksneeding attentional control from tasks that donot, we think it is more productive to ask how

attentional control may play a role in any work-ing memory task. This perspective opens up thestudy of working memory tasks to a finer-grainedanalysis of the role of attentional control as itparticipates with representational processes tomediate what is a vital human skill.

Acknowledgments

Preparation of this chapter was supported by grants

AG18286 and MH60655. The authors thank LauraZahodne for her editorial assistance with this chapter.

References

Awh, E., and Jonides, J. (1998). Spatial selective

attention and spatial working memory. In R.

Parasuraman (Ed.), The attentive brain. (pp.

353–380). Cambridge, MA: MIT Press, 1998.

Awh, E., & Jonides, J. (2001). Overlapping mecha-

nisms of attention and spatial working memory.Trends in Cognitive Sciences 5, 119–126.

Awh, E., Jonides, J., & Reuter-Lorenz, P. A. (1998).

Rehearsal in spatial working memory. Journal

of Experimental Psychology: Human Perception

and Performance 24, 780–790.

Awh, E., Jonides, J., Smith, E. E., Buxton, R. B.,

Frank,L.R.,Love,T.,Wong,E.C.,&Gmeindl,L.

(1999). Rehearsal in spatial working memory:

Evidence from neuroimaging. Psychological Sci-

ence 10, 433–437.

Awh, E., Jonides, J., Smith, E. E., Schumacher, E.

H., Koeppe, R. A., & Katz, S. (1996). Dissoci-

ation of storage and rehearsal in verbal working

memory: Evidence from PET. Psychological

Science 7, 25–31.

Babcock, R. L., & Salthouse, T. A. (1990). Effects of

increased processing demands on age differences

in working memory. Psychology and Aging, 5,

421–428.Baddeley, A. D. (1986). Working memory. Oxford:

Clarendon Press.





Academic Press.

Baddeley, A. D. (2001). Is working memory still

working? American Psychologist, 56, 851–864.

Barch, D. M., Braver, T. S., Nystrom, L. E., Forman,S. D., Noll, D. C., & Cohen, J. D. (1997).

Dissociating working memory from task diffi-

culty in human prefrontal cortex. Neuropsycho-

logia, 35, 1373–1380.

Botvinick, M.M., Braver, T.S., Barch, D.M., Carter,

C.S., Cohen, J.D. (2001). Conflict monitoring

and cognitive control. Psychological Review,

108, 624–652.

Cabeza, R., Daselaar, S. M., Dulcos, F., Prince,

S. E., Budde, M., & Nyberg, L. (2004). Task-independent and task-specific age effects on

brain activation during working memory visual

attention and episodic retrieval. Cerebral Cortex,

14, 364–375.

Chafee, M. V., & Goldman-Rakic, P. S. (2000).

Inactivation of parietal and prefrontal cortex

reveals interdependence of neural activity dur-

ing memory-guided saccades. Journal of Neuro-

physiology, 83, 1550–1566.

Constantinidis, C., & Steinmetz, M. A. (1996).Neuronal activity in posterior parietal area 7a

during the delay periods of a spatial memory

task. Journal of Neurophysiology, 76, 1352–1355.

Corbetta, M., Kincade, J. M., & Shulman, G. L.

(2002). Neural systems for visual orienting and

their relationships to spatial working mem-

ory. Journal of Cognitive Neuroscience 14, 508–

523.

Courtney, S. M., Petit, L., Maisog, J. M., Ungerlei-

der, L. G., & Haxby, J. V. (1998). An area

specialized for spatial working memory in hu-

man frontal cortex. Science, 279, 1347–1351.

Cowan, N. (1995). Attention and memory: An inte-

grated framework. Oxford: Oxford University

Press.

Cowan, N., Wood, N.L., Wood, P.K., Keller, T.A.,

Nugent, L.D., & Keller, C.V. (1998). Two


short-term memory span. Journal of Experimen-tal Psychology: General, 127, 141–160.

Craik, F. I. M. (1977). Age differences in human

memory. In J. E. Birren & K. W. Schaie (Eds.),

Handbook of the psychology of aging (pp. 384–

420). Englewood Cliffs, NJ: Prentice-Hall.




Craik, F. I. M., & Jennings, J. M. (1992). Human

memory. In F. I. M. Craik & T. A. Salthouse

(Eds.), Handbook of aging and cognition (pp.

51–109). Hillsdale, NJ: Lawrence Erlbaum

Associates.Craik, F. I. M., Morris, R. G., & Gick, M. L. (1990).

Adult age differences in working memory. In

G. Vallar & T. Shallice (Eds.) Neuropsycholog-

ical impairments of short-term memory (pp. 247–

267). New York: Cambridge University Press.

D’Esposito, M., & Postle, B. R. (1999). The

dependence of span and delayed-response per-

formance on prefrontal cortex. Neuropsycholo-

gia, 37, 1303–1315.

D’Esposito, M., Postle, B. R., Ballard, D., & Lease, J.(1999). Maintenance versus manipulation of

information held in working memory: An

event-related fMRI study. Brain and Cognition,

41, 66–86.

DiGirolamo, G. J., Kramer, A. F., Barad, V., Cepeda,

N. J., Weissman, D. H., Milham, M. P.,

Wszalek, T. M., Cohen, N. J., Banich, M. T.,

Webb, A., Belopolsky, A. V., & McAuley, E.

(2001). General and task-specific frontal lobe

recruitment in older adults during executive pro-cesses: A fMRI investigation of task-switching.

Neuroreport, 12, 2065–2071.

di Pellegrino, G., & Wise, S. P. (1993). Effects of

attention on visuomotor activity in premotor

and prefrontal cortex of a primate. Sensorimotor

and Motor Research, 10, 245–262.

Dobbs, A. R., & Rule, B. G. (1989). Adult age

differences in working memory. Psychology and

Aging, 4, 500–503.

Druzgal, T. J., & D’Esposito, M. (2003). Dissecting

contribution of prefrontal cortex and fusiform

face area to face working memory. Journal of

Cognitive Neuroscience, 15, 771–784.

Duncan, J., Martens, S., & Ward, R. (1997).

Restricted attentional capacity within but not

between modalities. Nature, 387, 808–810.

Engle, R. W. (2002). Working memory capacity as

executive attention. Current Directions in Psy-

chological Science, 11, 19–24.Engle, R. W., Tuholski, S. W., Laughlin, J. E., &

Conway, A. R. A. (1999). Working memory,

short-term memory, and general fluid intelli-

gence: A latent-variable approach. Journal of

Experimental Psychology: General, 128, 309–

331.

Gazzaley, A., Cooney, J.W., Rissman, J. & D’Esposito,

M.D. (2005). Top-down suppression deficit un-

derlies working memory impairment in normal

aging. Nature Neuroscience, 8 (10), 1298–1302.

Gehring, W. J., Goss, B., Coles, M. G. H., Meyer,D. E., & Donchin, E. (1993). A neural system

of error detection and compensation. Psycho-


Glahn, D. C., Kim, J., Cohen, M. S., Poutanen,

V-P., Therman, S., Bava, S., Van Erp, T. G. M.,

Manninen, M., Huttunen, M., Lonnqvist, J.,

Standertskjold-Nordenstam, C. G., & Cannon,

T. D. (2002). Maintenance and manipulation in

spatial working memory: Dissociations in the

prefrontal cortex. NeuroImage, 17, 201–213.Godde,B.,Berkefeld, T.,David-Jurgens,M.,& Dinse,

H. R. (2002). Age-related changes in primary

somatosensory cortex of rats: Evidence for par-

allel degenerative and plastic-adaptive pro-

cesses. Neuroscience and Biobehavioral Reviews,

26, 743–752.

Goldman-Rakic, P. S. (1987). Circuitry of the

primate prefrontal cortex and the regulation

of behavior by representational memory. In

F. Plum (Ed.), Handbook of physiology, thenervous systems, higher functions of the brain.

Section I, Vol. V (pp. 373–417). Bethesda, MD:

American Physiological Society.

Gray, J.R., Chabris, C.F., Braver, T.S. (2003).


gence. Nature Neuroscience, 6, 316-322.

Gunning-Dixon, F. M., & Raz, N. (2003). Neu-

roanatomical correlates of selected executive

functions in middle-aged and older adults: A

prospective MRI study. Neuropsychologia, 41,

1929–1941.

Gutchess, A. H., Welsh, R.C., Hedden, T., Bangert, A.,

Minear, M., Liu, L.L., & Park, D.C. (2005).

Aging and the neural correlates of successful

picture encoding: Frontal activations for de-

creased medial temporal activity. Journal of Cog-

nitive Neuroscience, 17, 84–96.

Jacobsen, C. F. (1931). The functions of frontal

association areas in monkeys. Comparative Psy-chology Monographs, 13, 1–60.

Jacobsen, C. F. (1935). Functions of the frontal

association area in primates. Archives of Neu-

rology and Psychiatry, 33, 558–569.

Jonides, J., Marshuetz, C., Smith, E. E., Reuter-

Lorenz, P. A., & Koeppe, R. A. (2000). Age




differences in behavior and PET activation reveal

differences in interference resolution in verbal

working memory. Journal of Cognitive Neurosci-

ence, 12, 188–196.

Jonides, J., Smith, E. E., Koeppe, R. A., Awh, E.,Minoshima, S., & Mintun, M. A. (1993). Spa-

tial working memory in humans as revealed by

PET. Nature 363, 623–625.

Jonides, J., Smith, E. E., Marshuetz, C., Koeppe,

R. A., & Reuter-Lorenz, P.A. (1998). Inhibition

in verbal working memory revealed by brain

activation. Proceedings of the National Academy

of Sciences USA 95, 8410–8413.

Kan, I. P., & Thompson-Schill, S. L. (2004). Selec-

tion from perceptual and conceptual repre-sentations. Cognitive, Affective, and Behavioral


Kastner, S., & Ungerleider, L. G. (2000). Mecha-

nisms of visual attention in the human cortex.

Annual Review of Neuroscience, 23, 315–341.

Kemper, T. L. (1994). Neuroanatomical and neuro-

pathological changes during aging and demen-

tia. In M. L. Albert & J. E. Knoefel (Eds.),

Clinical neurology of aging (2nd ed.) (pp. 3–

67). New York: Oxford University Press.Leung, H. C., Gore, J. C., & Goldman-Rakic, P. S.

(2002). Sustained mnemonic response in the

human middle frontal gyrus during on-line

storage of spatial memoranda. Journal of Cog-

nitive Neuroscience, 14, 659–671.

Li, S., Lindenberger, U., & Sikstrom, S. (2001).

Aging cognition: From neuromodulation to

representation. Trends in Cognitive Sciences, 5,

479–486.

Luck, S. J., & Vogel, E. K. (1997). The capacity of

visual working memory for features and con-

junctions. Nature, 390, 279–281.

Malmo, R. (1942). Interference factors in delayed

response in monkeys after removal of frontal

lobes. Journal of Neurophysiology, 5, 295–308.

McIntosh, A. R., Sekuler, A. B., Penpeci, C., Rajah,

M. N., Grady, C. L., Sekuler, R., & Bennett,

P. J. (1999). Recruitment of unique neural sys-

tems to support visual memory in normal aging.Current Biology, 9, 1275–1278.

Miller, E. K., & Desimone, R. (1994). Parallel

neuronal mechanisms for short-term memory.

Science, 263, 520–522.

Naveh-Benjamin, M., & Jonides, J. (1984). Mainte-

nance rehearsal: A two-component analysis.


Memory, and Cognition 10, 369–385.

Nelson, J. K., Reuter-Lorenz, P. A., Sylvester, C. Y.,

Jonides, J., & Smith, E. (2003). Dissociable

neural mechanisms underlying response-basedand familiarity-based conflict in working mem-

ory. Proceedings of the National Academy of

Science USA, 100, 11171–11175.

Park, D. C., Lautenschlager, G., Hedden, T.,

Davidson, N. S., Smith, A. D., & Smith, P. K.

(2002). Models of visuospatial and verbal

memory across the adult life span. Psychology

and Aging, 17, 299–320.

Park, D. C, Polk, T. A., Park, R., Minear, M.,

Savage, A., & Smith, M. R. (2004). Aging re-duces neural specialization in ventral visual cor-

tex. Proceedings of the National Academy of

Sciences USA, 101, 13091–13095.

Paulesu, E., Frith, C. D., & Frackowiak, R. S. J.

(1993). The neural correlates of the verbal compo-

nent of working memory. Nature, 362, 342–343.

Petrides, M. (1994). Frontal lobes and working

memory: Evidence from investigations of the

effects of cortical excisions in nonhuman pri-

mates. In F. Boller & J. Grafman (Eds.), Hand-book of neuropsychology (Vol. 9, pp 59–82).

Amsterdam: Elsevier.

Postle, B. R., Berger, J. S., and D’Esposito, M.

(1999). Neuroanatomical double dissociation

of mnemonic and executive control processes

contributing to working memory performance.

Proceedings of the National Academy of Science

USA, 96, 12959–12964.

Postle, B. R., Druzgal, T. J., & D’Esposito, M.

(2003) Seeking the neural substrate of working

memory storage. Cortex, 39, 927–946.

Posner, M. I., & Dehaene, S. (1994). Attentional net-

works. Trends in Neuroscience, 17, 75–79.

Posner, M. I., & Petersen, S. E. (1990). The attention

systems of the human brain. Annual Review of


Ravizza, S. M., Delgado, M. R., Chein, J. M.,

Becker, J. T., & Fiez, J. T. (2004). Functional

dissociations within the inferior parietal cortexin verbal working memory. NeuroImage, 22,

562–573.

Raz, N. (2000). Aging of the brain and its impact on

cognitive performance: Integration of structural

and functional findings. In F. I. M. Craik and

T. A. Salthouse (Eds.) Handbook of aging and




cognition–II. (pp. 1–90) Mahwah, NJ: Lawr-

ence Erlbaum Associates.

Reuter-Lorenz, P. A. (2002). New visions of the

aging mind and brain. Trends in Cognitive

Sciences, 6, 394–400.Reuter-Lorenz, P. A., Jonides, J., Smith, E. E., Hartley,

A., Miller, A., Marshuetz, C., & Koeppe, R. A.

(2000). Age differences in the frontal lateraliza-

tion of verbal and spatial working memory

revealed by PET. Journal of Cognitive Neurosci-

ence, 12, 174–187.

Reuter-Lorenz, P.A. and Lustig, C. (2005). Brain aging:

reorganizing discoveries about the aging mind.

Current Opinion in Neurobiology, 15 (4) 245–251.

Reuter-Lorenz, P. A., Marshuetz, C., Hartley, A.,Jonides, J., & Smith, E. E. (2001). Neurocog-

nitive ageing of storage and executive processes.

European Journal of Cognitive Psychology, 13,

257–278.

Reuter-Lorenz, P.A. & Mikels, J. A. (2006). The Aging

Brain: The implications of enduring plasticity for

behavioral and cultural change. In Baltes, P.,

Reuter-Lorenz, PA., & Roesler, F. (Eds.), Life-

span development and the brain: The perspec-

tive of biocultural co-constructivism (pp. 255–277). New York: Cambridge University Press.

Reuter-Lorenz, P. A., & Sylvester, C. Y. (2004). The

cognitive neuroscience of aging and working

memory. In R. Cabeza, L. Nyberg, & D. Park

(Eds.), The cognitive neuroscience of aging (pp.

186–217). New York: Oxford University Press.

Rypma, B., Berger, J. S., & D’Esposito, M. (2002). The

influence of working-memory demand and sub-

ject performance on prefrontal cortical activity.


Rypma, B., & D’Esposito, M. (2001). Age-related

changes in brain–behavior relationships: Evi-

dence from event-related functional MRI stud-

ies. European Journal of Cognitive Psychology,

13, 235–256.

Rypma, B., Prabhakaran, V., Desmond, J. E., &

Gabrieli, J. D. E. (2001). Age differences in

prefrontal cortical activity in working memory.


Rypma, B., Prabhakaran, V., Desmond, J. E.,

Glover, G. H., & Gabrieli, J. D. E. (1999).

Load-dependent roles of frontal brain regions in

the maintenance of working memory. Neuro-

Image, 9, 216–226.Sakai, K., Rowe, J. B., & Passingham, R. E. (2002).

Active maintenance in prefrontal area 46 cre-

ates distractor-resistant memory. Nature Neuro-

science 5, 479–484.

Schmolesky, M. T., Wang, Y., Pu, M., & Leventhal,

A. G. (2000). Degradation of stimulus selectiv-

ity of visual cortical cells in senescent rhesus

monkeys. Nature Neuroscience, 3, 384–390.

Smith, E. E., Geva, A., Jonides, J., Miller, A., Reuter-

Lorenz, P. A., & Koeppe, R. A. (2001). Theneural basis of task-switching in working mem-

ory: Effects of performance and aging. Proceed-

ings of the National Academy of Sciences USA,

98, 2095–2100.

Smith, E. E., Jonides, J., & Koeppe, R. A. (1996).

Dissociating verbal and spatial working mem-

ory using PET. Cerebral Cortex 6, 11–20.

Smith, E. E., Jonides, J., Koeppe, R. A., Awh, E.,

Schumacher, E. H., & Minoshima, S. (1995).

Spatial vs. object working memory: PET inves-tigations. Journal of Cognitive Neuroscience 7,

337–356.

Speer, N. K., Jacoby, L. L., & Braver, T. S. (2003).

Strategy-dependent changes in memory: Ef-

fects on brain activity and behavior. Cogni-

tive, Affective, & Behavioural Neuroscience, 3,

155–167.



Wager, T. D., Jonides, J., & Reading, S. (2004).

Neuroimaging studies of shifting attention: A

meta-analysis. NeuroImage, 22, 1679–1693.

Wager, T. D., & Smith, E. E. (2003). Neuroimaging

studies of working memory: A meta-analysis.

Cognitive, Affective, & Behavioral Neuroscience,

3, 255–274.

Yantis, S. (2003). Cortical mechanisms of attention

shifts: Space, features, and objects. Abstracts of

the Psychonomic Society, 8, 47.




11

Specialized Verbal Working Memory

for Language Comprehension

DAVID CAPLAN, GLORIA WATERS, and GAYLE DEDE

The concept of working memory (WM) waspopularized in, if not introduced into, con-temporary cognitive psychology by Baddeleyand Hitch (1974), who argued that a short-duration, capacity-limited, ‘‘working memory’’system was used in a variety of cognitive tasks.This system, which both maintained represen-tations and performed computations on them,was thought to be responsible, at least in part,for limitations on human performance in awide range of cognitive tasks. The linking of themaintenance of a limited amount of infor-mation for a short period of time with com-putational operations on that informationconstituted a significant change from the then-prevailing view that the function of ‘‘short-termmemory’’ was to allow entry of information into

‘‘long-term memory.’’ In the 30 years sinceBaddeley and Hitch’s article, the concept of WM has evolved. In some researchers’ formu-lations, it has come almost full circle, to theview that WM capacity is largely a measure of attentional control that is related to the encod-ing and activation of representations in long-

term memory. We will return to some of thesemore recent formulations in the discussionsection of this chapter, but we begin with whatwe take to be the original Baddeley and Hitchconcept of WM as a capacity-limited, short-duration store in which computations are per-formed in the service of task goals.

If WM is characterized this way, languagecomprehension requires WM. Regardless of whether language is written or spoken, the inputbecomes available over time and temporallydiscontinuous parts of the input must be relatedto one another for language to be understood.The need for WM applies to the construc-tion of all levels of language—segmental andlexical phonological representations, morphol-ogy, intonational structure, syntax, and dis-

course.Our work has focused on the WM require-ments of syntactic processing. We begin thischapter by illustrating the WM requirements as-sociated with syntactic processing, and then doc-ument the existence of variability in subjects’efficiency in handling these WM demands. We

272



then explore the relationship between subjects’efficiency in handling syntactic WM demandsand individual differences in WM capacity dueto normal variation in healthy college students,

aging, and various types of neuropathology. Wethen briefly review other types of data (e.g.,neuroimaging) that we have collected that areconsistentwiththeresultsfromthesepopulations.

To anticipate, we have not found that effi-ciency in handling syntactic WM demands ispredicted by individual differences in WMcapacity using standard tests of this function,such as the reading span test. On the otherhand, we have replicated other researchers’

findings of a relationship between other aspectsof language processing, such as text memoryand comprehension, and standard tests of WM.These findings have led us to argue for a frac-tionation of the WM system used in languageprocessing into a specialized verbal WM system(svWM) that supports specific aspects of lan-guage processing that we refer to as interpretiveprocessing, and a more general verbal workingmemory system (gvWM) that supports other

aspects of language processing that we refer toas post-interpretive processing (Caplan & Wa-ters, 1999a, 1999b, 2002). We conclude with adiscussion of the implications of these results,couched in terms of answers to the four ques-tions that serve as themes for this volume.

WORKING MEMORY AND SYNTAX

It is clear that sentences differ in terms of howeasy they are to understand. Virtually all readersor listeners can understand sentences such as(1) and (2) below. However, very few readers orlisteners can understand sentences such as (3),even though such sentences are grammaticallyacceptable.

1. The mouse that the dog bit ran away.

2. The dog that the cat scratched bit themouse.3. The mouse that the dog that the cat

scratched bit ran away.

Sentences 1 and 2 each have one embeddedclause (1: that the dog bit, 2: that the cat

scratched), while Sentence 3 has two embed-ded clauses (that the dog bit, that the catscratched). The inability of most language usersto understand sentences such as (3) is thought

to result from limitations in WM capacity,making it difficult for them to structure thesentence syntactically (Gibson, 1998; Just &Carpenter, 1992; Lewis, 1996).

There is also considerable evidence thateven syntactic structures that can be under-stood by most or all language users differ in theease with which they can be understood. Thisis true whether sentence processing is mea-sured using an ‘‘off-line’’ task, in which subjects

are asked to make a judgment after hearinga sentence, or ‘‘on-line tasks,’’ in which theamount of time required to process each wordor phrase is measured on-line. A prototypicalexample of sentences that differ in terms of how easily they can be understood, despitethe fact that they contain the same lexicalitems, are sentences with subject- and object-extracted relative clauses (often referred to assubject and object relatives). Subject-extracted

sentences are syntactically simpler than object-extracted sentences. In subject-extracted sen-tences, the subject of the relative clause hasbeen extracted from the clause; in object-extracted sentences the object has been ex-tracted. These are illustrated in Sentences 4and 5:

4. The scout warmed the cabin that con-tained the firewood. (Subject-extracted)

5. The cabin that the scout warmed con-tained the firewood. (Object-extracted)

The WM demands of object-extracted sen-tences are thought to be greater than those of subject-extracted sentences because they in-volve more storage or reference to more itemsheld in memory (i.e., keeping track of par-tially processed syntactic dependencies that are

awaiting their second element for the sentenceto be grammatical) and more integration of ormore computations on items held in memory(connecting a newly input word into the struc-ture that has been built so far) (Gibson, 1998).

In off-line tasks in which subjects are askedto judge the acceptability of sentences such as

Verbal Working Memory for Language Comprehension 273



(4) and (5) that are intermixed with unaccept-able sentences (e.g., ‘‘The criminal cursed thejudge who astonished the verdict’’), even collegestudents take longer to judge the acceptability of

sentences with object-extracted clauses than forsubject-extracted clauses (e.g., Waters, Caplan,& Hildebrandt, 1987). Moreover, performanceon on-line tasks, such as self-paced reading orlistening, in which the sentences are dividedinto words or phrases and subjects press a buttonfor the successive presentation of each segment,has shown that reading and listening times arelonger at the very regions of object-extractedsentences where WM load is thought to be

higher. The assumption underlying these tasks isthat reading or listening times to words orphrases presented one at a time reflect the timeit takes to integrate lexical items into an accru-ing syntactic and semantic structure and aretherefore longer when this integration is moredifficult. Thus, longer reading or listening timesfor a particular part of the sentence are thoughtto be a reflection of increased WM or processingdemands.

Typical results from a self-paced listeningstudy of ours are shown in Figure 11.1. The on-line sentence-processing paradigm we used inthis study was the auditory moving windowsparadigm (Ferreira, Henderson, Anes, Weeks, &McFarlane, 1966), in which sentences are re-corded as a whole with normal prosody, digi-tized, and segmented into words or phrases,and subjects press a button to receive succes-sive items.1 The stimuli consisted of three dif-ferent pairs of sentences, illustrated in Table11.1, in which one member of each pair con-sisted of a syntactic structure that contained asubject-extracted relative clause and the othermember contained a more difficult object-extracted relative clause. Each pair of sentencescontained the same words and differed only inthe order of the words. The sentences wereintermixed with unacceptable sentences and

subjects were asked to pace their way througheach sentence as quickly as possible and then tomake a judgment about whether the sentencewas an acceptable sentence in English. Figure11.1 shows the segment listening times foracceptable sentences that were correctly judgedas such. As can be seen, listening times were

Figure 11.1. Word-by-word listening times for col-lege students on three pairs of syntactically simpleand complex sentences. Intro¼ introductory phrase;

NP1¼first noun phrase; Pro¼ pronoun; NP2¼second noun phrase; V ¼ verb.




virtually identical for the two members of eachpair, for all segments other than the verb.

Why do processing times increase at theverb in object-extracted sentences? Processingdifficulty occurs at points in sentences where alarger amount of information must be inte-grated and where such integration must occurover longer distances (Gibson, 1998; King &Just, 1991; MacDonald, Pearlmutter, & Sei-denberg, 1994). All procedural models of pars-ing recognize both a higher memory load and ahigher computational load at the verb in object-relativized clauses than at any point in subject-

relativized clauses. The increase in listeningtime seen at the verb is taken as being due to thedifficulty associated with integrating the verb of the object-relativized clause into the accruingrepresentation of the structure and the meaningof the sentence, since it involves reference tomore items held in memory and more compu-

tations on those items than is the case at thecorresponding points in subject-relativizedclauses.

One other potential source of the increase inlistening time at the verb in object-relativizedclauses is that the verb in these structures alsocorresponds to the final word of the clause orin some cases to the final word of the sentence.Numerous studies have documented longerreading times at the end of clauses (Balogh,Zurif, Prather, Swinney, & Finkel, 1998). Theincrease in reading time at the end of a clauseis referred to as a wrap-up effect. Therefore, an

additional comparison that can be made to en-sure that the increase in listening time at theverb in object-relativized sentences is due to theWM load imposed at this point is to comparelistening time at this point (V1) with the wordthat is the last word of the clause in the subject-relativized sentence (i.e., the second noun

TABLE 11.1. Examples of the Three Pairs of Sentences Used in the Auditory MovingWindows Task

Sentence Type

Subject Relative Cleft Subject (CS)

PhraseIntro NP1 Pro V1 NP2 Continuation

It was/ the book/ that/ interested/ the teenager/ because it was a romance.

Object Relative Cleft Object (CO)

PhraseIntro NP1 Pro NP2 V1 Continuation

It was/ the teenager/ that/ the book/ interested/ because it was a romance

Subject Relative Subject Subject (SS)

PhraseNP1 Pro V1 NP2 V2 NP3

The law/ that/ favored/ the millionaire/ frustrated/ the workers.

Object Relative Subject Object (SO)

PhraseNP1 Pro NP2 V1 V2 NP3

The law/ that/ the millionaire/ favored/ frustrated/ the workers.

Subject Relative Object Subject (OS)

PhraseNP1 V1 NP2 Pro V2 NP3

The millionaire/ favored/ the law/ that/ frustrated/ the workers.

Object Relative Subject Object (SO)

PhraseNP1 Pro NP2 V1 V2 NP3The law/ that/ the millionaire/ favored/ frustrated/ the workers.

Intro¼ introductory phrases; NP1¼first noun phrase; Pro¼ pronoun; V1¼first verb; V2¼ second verb; NP2¼ second nounphrase; continuation¼ final phrase.




phrase, or NP2, which is the clause final word).If the effect is truly one due to the WM load atthe verb, then listening times at the verb inobject-relativized clauses should be longer than

both those at the verb and the second nounphrase of the corresponding subject-relativizedsentence. Thus, the increase in listening timeseen at the verb in both comparisons is likelydue to the greater WM demands at the em-bedded verb of an object-relativized sentencethan at the corresponding points in a subject-relativized sentence.

How much do individuals vary in theirabilities to meet these WM needs? Table 11.2

shows the mean listening time (correctedfor the length of the segment) at two differ-ent points in subject- and object-relativizedsentences—one in which there should be nodifference in processing load or integrationcosts (the first noun phrase, NP1) and one inwhich processing and integration should bemore difficult in the object-relativized sentence(the first verb, V1). Table 11.2 shows that, evenin college students, there is considerable indi-

vidual variability in listening times for wordsand phrases in sentences containing subject-and object-extracted relative clauses. This var-iability (as shown by the standard error andcoefficient of variability) is much greater inobject-extracted (cleft object, CO, and subject

object, SO) than in subject-extracted (cleftsubject, CS; object subject, OS; and subjectsubject, SS) sentences and at the embeddedverb of object-extracted relative clauses (V1)

than at the verb of subject-extracted relativeclauses. Table 11.3 shows that the greatestvariability in listening time differential betweenmatched segments in object- and subject-ex-tracted relative clauses is in the difference be-tween listening times for the embedded verbof object-extracted relative clauses (V1) and thecorresponding embedded verb (V1) and rela-tive-clause final noun (NP2) in subject-ex-tracted relative clauses. Variability across these

segments is much greater than across controlsegments (NP1) that should not differ in inte-gration costs or processing load. These datashow that there is considerable individualvariability in the efficiency with which subjectsdeal with the local WM demands (i.e., WMdemands at the point of difficulty) of the object-extracted relative-clause construction. We shallrefer to this as variability in WM utilization.

Table 11.4 shows the correlation between

the difference scores for different sentencepairs and comparisons that we take to be goodmeasures of syntactic WM utilization in thisparadigm for the data from the group of collegestudents shown in Figure 11.1. The fact thatthere are high, significant correlations between

TABLE 11.2. Variability in Segment Listening Times in College Students

Segment Mean StandardError Coefficient of Variability Minimum Maximum

CS NP1 279.8 17.5 .43 14.2 590.9CO NP1 241.9 20.3 .58 88.5 608.7CS V1 331.8 16.9 .35 76.6 580.3CO V1* 604.5 47.6 .54 119.6 1714.3SS NP1 280.8 19.1 .47 52.0 537.1SO NP1 299.3 19.2 .44 22.4 613.3SS V1 323.1 14.8 .32 130.6 530.8SOV1* 484.9 41.1 .59 145.4 1775.4OS NP1 308.8 17.1 .38 16.1 577.1

SO NP1 299.3 19.2 .44 22.4 613.3OS V1 326.9 16.6 .35 108.3 607.1SOV1* 484.9 41.1 .59 145.4 1775.4

CS¼ cleft subject; CO¼ cleft object; SS¼ subject subject; OS¼ object subject; SO¼ subject object; V1¼firstverb; NP1¼first noun phrase; NP2¼ second noun phrase.

*indicates segment at which processing load is high.

Data from Waters and Caplan (2005).




all of the difference scores supports the hy-pothesis that individual differences in syntacticWM utilization that are measured in this par-adigm are stable and reliable. The question weshall explore in the next sections of this chapter

is whether this variability is related to WMcapacity as measured by standard tests of thisfunction.

Before reviewing our work on this topic, wenote two important methodological and inter-pretive issues regarding the relationship be-tween gvWM and syntactic processing. First,the effort to relate individual differences ingvWM capacity to individual differences in theefficiency of syntactic processing requires that

there be an equally reliable way to measuregvWM in individual subjects. One of the mostwidely used WM tasks is the Daneman andCarpenter (1980) reading or listening span taskin which participants read aloud or listen to

increasingly longer sequences of sentence andthen recall the final words of all of the sen-tences in each sequence. Working memoryspan is taken to be the longest list length atwhich recall of the sentence final words iscorrect on the majority of trials. Despite wide-spread use of the reading span task, severalbasic psychometric properties of this task re-main incompletely characterized or inade-quate. For example, we found that the task’s

TABLE 11.3. Difference in Listening Time between Matched Segments of Sentenceswith Subject- and Object-Relative Clauses

Segment DifferenceStandard

Error Minimum Maximum

CO NP1–CSNP1 37.9 7.5 149.7 84.6CO V1–CSV1* 272.7 40.3 253.2 1218.7COV1–CSNP2* 171.6 43.9 322.8 1368.4SO NP1–SSNP1 18.5 10.4 164.3 248.3SOV1–SSV1* 161.8 35.3 41.7 1373.5SOV1–SSNP2* 116.8 32.6 138.7 1362.6SO NP1–OSNP1 9.5 13.1 381.5 304.2SOV1–OSV1* 158.0 36.2 127.3 1460.0SOV1–OSNP2* 78.3 31.7 179.9 1214.9

CS¼ cleft subject; CO¼ cleft object; SS¼ subject subject; OS¼object subject; SO¼ subject object; V1¼first verb;

NP1¼ first noun phrase; NP2¼ second noun phrase.*indicates contrast at which processing load difference between sentence types is high.


TABLE 11.4. Correlation between Measures of Syntactic Working Memoryuse across Three Different Sets of Sentences

COV1–CSV1

COV1–CSNP2

SOV1–SSV1

SOV1–SSNP2

SOV1–OSV1

SOV1–OSNP2

Difference Score

COV1–CSV1 — .96 .69 .67 .67 .65COV1–CSNP2 — — .71 .69 .72 .69

SOV1–SSV1 — — — .94 .97 .94SOV1–SSNP2 — — — — .92 .92SOV1–OSV1 — — — — — .93SOV1–OSNP2 — — — — — —

Note: All correlations are significant. CS¼ cleft subject; CO¼ cleft object; SS¼ subject subject; OS¼ objectsubject; SO¼ subject object; V1¼first verb; NP1¼first noun phrase; NP2¼ second noun phrase.





test–retest reliability and the stability of cate-gorization of subjects into WM groups overshort time periods through use of the readingspan task were extremely poor (Waters & Ca-

plan, 1996b, 2003). The approach we havetaken to resolve this problem in our researchhas changed over time. In some of our earlierstudies we used our own variant of the task,which required participants to make a plausi-bility judgment about each sentence beforerecalling the final words of all of the sentencesin a set (Waters & Caplan, 1996b). This ver-sion has somewhat better test–retest reliabilityand stability of subject classification than the

task by Daneman and Carpenter. In our morerecent studies, we have used a composite WMscore that is based on four different WMmeasures (alphabet span, Craik, 1986; subtract2 span, Salthouse, 1988; two versions of read-ing span, Waters & Caplan, 1996b) that con-firmatory factor analysis suggests reflect acommon factor. We found these measures tohave the best test–retest reliability and stability of subject classification over a 6-week period (Wa-

ters & Caplan, 2003). Another important issue concerns the type

of results that would provide evidence of anycognitive operation relying on gvWM. Moststudies have used off-line tasks, such as makingjudgments at the end of a sentence. However,off-line tasks do not measure syntactic com-plexity effects as they occur and thus give onlyan indirect view of syntactic processing abili-ties. Furthermore, end-of-sentence measuresare likely to include effects of syntactic com-plexity associated with reviewing sentences tosatisfy task requirements such as making plau-sibility or grammaticality judgments or match-ing sentences to pictures (i.e., post-interpretiveprocesses). For these reasons, most researchersbelieve that on-line tasks are necessary to char-acterize first-pass syntactic processing (seeMacDonald et al., 1994, for a review).

In addition, gvWM may affect performanceon both off-line and on-line tasks for a numberof reasons (Waters & Caplan, 1996b). There-fore, evidence that gvWM has a specific effecton a linguistic (e.g., syntactic) operation re-quires that both off-line and on-line measuresbe more affected by linguistically more de-

manding (syntactically more complex) struc-tures (i.e., that group differences increase whenthe demands made by structure-building andinterpretive operations increase). If on-line mea-

sures are taken at multiple points in a sentence,as in self-paced reading or listening, it is there-fore necessary to compare points of high andlow demand across sentences. Support for a roleof gvWM in syntactic processing would comefrom the finding of a greater effect on perfor-mance in subjects with low gvWM at the com-plex parts of complex sentences (such as at theverb in object-relativized sentences in the exam-ples outlined above) than at other positions (such

as at the first noun phrase in the examples out-lined above).

SYNTACTICALLY BASED SENTENCECOMPREHENSION AND WORKINGMEMORY IN YOUNG SUBJECTS

In this chapter, we focus on research into theWM demands of syntactic processing that has

capitalized on the differences in the WM de-mands associated with processing sentenceswith subject-extracted and object-extractedrelative clauses. Studies with other syntacticstructures have yielded similar results and con-clusions (see Waters & Caplan, 1996b, for areview). However, because of space limitationswe will focus on work with sentences with rel-ative clauses.

The first experiment in the literature to in-vestigate the ability of college students whodiffer in WM capacity to process sentences withsubject- and object-extracted clauses was carriedout by King and Just (1991, Experiment 1).They reported self-paced word-by-word read-ing times for high- and low-span subjects forthese two sentence types in a Daneman andCarpenter–type sentence span task. In thisstudy, King and Just (1991) present a graph in-

dicating that the biggest reading time differ-ences between high- and low-span subjects arein the syntactically critical area of the object-relative sentences. However, no statistical ana-lyseswerereportedtosupportthecontentionthatthe difference in reading times between high-and low-span subjects is specifically localized




to the region of object-relative sentences wherethere is the greatest processing load (Waters &Caplan, 1996b). In addition, the data presentedby King and Just (1991) did not isolate perfor-

mance on sentences in which the subjects didnot have to retain the sentence-ending words(no analyses were reported by King and Just onthe sentences in the zero-load condition alone).

Our first experiment examined the relation-ship between working memory capacity as mea-sured using our variant of the reading span taskand on-line syntactic processing, as measuredusing the auditory moving windows task, in agroup of 100 college students (Waters & Ca-

plan, 2004). In this study, subjects were testedon two of the three pairs of sentences shown inTable 11.1 (cleft-subject [CS] vs. cleft-object[CO] and object-subject [OS] vs. subject-ob-ject [SO]). The stimuli in this study differedfrom those in Table 11.1 in that different lex-ical items were used in the matched pairs of sentences and the effects of lexical frequencywere regressed out. In addition, the CS/COsentences did not contain an extra phrase at the

end (continuation in Table 11.1). Acceptablesentences were intermixed with unacceptablesentences and subjects made a judgment aboutacceptability at the end of each sentence. Theuse of these two pairs of sentences allowedfor two separate tests of the hypothesis thatthere is a relationship between working mem-ory capacity and on-line sentence processingefficiency using different sets of stimulus ma-terials.

Subjects were classified as high, medium, orlow WM span on the basis of their perfor-mance on the Waters and Caplan (1996a)reading span task. Figure 11.2 shows the meanlistening time for each phrase of the two typesof subject- and object-extracted sentences forthe three span groups. Comparison of listeningtimes at each phrase showed that there werethe predicted increases in these times at the

more capacity-demanding phrases of the morecomplex sentences. For the cleft subject–cleftobject sentences this increase occurred at theverb. For object subject–subject object sen-tences the increase occurred at both verbs andwas carried through the end of the sentence. Ascan be seen in Figure 11.2, all three groups of

subjects showed the expected pattern. Fur-thermore, listening times were also longer on

the verb of the more complex object-extractedsentences than on the final noun phrase (NP2)of the subject-extracted sentences, indicatingthat the increase at the verb is not simply dueto its clause-final position. Examination of thereaction time and error data on the plausibilityjudgment at the end of the sentence showed

Figure 11.2. Word-by-word listening times for col-lege students who differ in WM capacity on syntac-tically simple and complex sentences. Intro¼introductory phrase; NP1¼ first noun phrase; NP2¼second noun phrase; V1¼first verb; V2¼ secondverb; NP3¼ third noun phrase.




that, as expected, all subjects took longer tomake judgments about sentences with object-extracted forms than sentences with subject-extracted forms. However, these effects were

not larger in low-span subjects. Together, theseresults suggest that subjects who differ in WMcapacity as measured by the reading span testdo not differ in their ability to structure sen-tences syntactically.

Because of our concern about the reliabilityof WM span measures outlined above, we car-ried out a second study of 48 college students inwhich subjects were tested on four measures of working memory capacity. These measures

consisted of two versions of the reading span taskused in the experiment outlined above (one inwhich the stimuli consisted of syntacticallysimple sentences and one in which they con-sisted of syntactically complex sentences); analphabet span task (Craik, 1986) in which sub-jects were presented with a list of words in ran-dom order and required to repeat them back inalphabetical order; and the subtract 2 span task(Salthouse, 1988), in which subjects were pre-

sented with digits and required to repeat themback after subtracting 2 from each. These taskswere chosen because they represent a range of processing operations that overlap computa-tionally to different degrees with the on-linecomprehension task and because our previousstudy showed that a composite WM score basedon these four tasks resulted in the best test–retestreliability (r ¼ .85). In this experiment, all sub-jects were tested up to a common span size onthe working memory measures, and the numberof items recalled in correct serial order was usedas the measure of working memory capacity incorrelational analyses.

The stimulus materials were also modi-fied in this experiment. To eliminate end-of-sentence effects on critical phrases (the verbin CO sentences), an additional (continuation)phrase was added at the end of CS and CO

sentences. Sample stimuli from this study areshown in Table 11.1. A third type of sentenceused in a number of other studies (e.g., King &Just, 1991), subject subject (SS), was added inthis study. The subject subject–subject objectcomparison is thought to be a purer test of the difference between object and subject

relativization than the object subject–subjectobject comparison because, in both subject-object and subject-subject sentences, the sec-ond verb (the main verb) must be connected to

the first noun over the intervening relativeclause, which is not the case in object-subjectsentences. For example, in the examples inTable 11.1 in both the subject-subject andsubject-object sentences the second verb fru-strated must be connected to the first nounphrase the law, whereas in the object-subjectsentence there is no intervening material be-tween the law and frustrated. To controlfor lexical effects, the subject-subject, object-

subject and subject-object sentences were de-veloped in triplets using the same words.

In addition, in this study all subjects werealso tested on a more global test of languagecomprehension, the Nelson Denny ReadingTest. This test is one of the most commoncriterion tasks used in studies of the relation-ship between working memory and languagecomprehension, and several previous studieshave indicated that it is related to working

memory capacity. The use of the Nelson DennyReading Test serves to determine whether therelationship that is usually found betweenworking memory and general measures of com-prehension is found in the subjects studiedhere.

The on-line data for this study are thoseseen in Figure 11.1. As outlined above, thesesubjects showed the typical pattern of on-line effects seen in studies of sentences withsubject- and object-extracted relative clauses.Examination of the WM data for the subjectsin this study showed that they differed widelyin terms of their working memory capacities asmeasured by the four WM tasks. Using thesame criteria used to classify subjects into dis-crete memory-span groups as in Experiment 1,12 were classified as low span, 20 as mediumspan, and 16 as high span on the basis of their

scores on the reading span task with cleft sub-ject sentences.The relationship between on-line perfor-

mance and WM capacity was addressed in cor-relational analyses in this study. PearsonProduct Moment Correlation Coefficients werecomputed between working memory, measured




Figure 11.4. Once again, listening times were

longer overall for the elderly subjects than forthe young subjects. As in the previous study,listening times were consistently longer at thecapacity-demanding regions of the harder thanthe matched simpler sentence types for allsubjects. However, once again, this effect wasnot larger in the older subjects or in subjects

who had reduced gvWM. Table 11.5 shows thecorrelation between the composite WM scorebased on four WM tasks and the on-line mea-sures, as well as measures taken at the end of sentence (i.e., reaction time and errors in

Figure 11.3. Word-by-word listening times for fivegroups of subjects who differ in age on syntacticallysimple and complex sentences. Intro¼ introductoryphrase; NP1¼first noun phrase; NP2¼ secondnoun phrase; V1¼first verb; V2¼ second verb;NP3¼ third noun phrase.

Figure 11.4. Word-by-word listening times for 48young (YC) and 48 elderly (EC) subjects. Intro ¼introductory phrase; NP1¼first noun phrase; Pro¼Pronoun; NP2¼ second noun phrase; V1¼first verb;

V2¼ second verb.




making acceptability judgments to more com-plex object-relativized sentences compared tosimple subject-relativized sentences). In thiscase, there was a significant correlation be-tween WM and processing times at the verb of cleft-object sentences. However, the fact thatthe correlation was not significant when theverb in cleft-object sentences was compared tothe final noun phrase in cleft-subject sentencessuggests that the effect is an end-of-clause ef-fect, rather than a syntactic effect. In addition,for the cleft subject–cleft object comparisonand for the object subject–subject object com-parison, subjects who made more errors on the

acceptability judgment task for the more com-plex object-relativized sentences tended to havelower WM scores, as shown by the significantcorrelation between WM and these end-of-sentence measures. Finally, as in our studyof young subjects, there were significant cor-

relations between the WM measures and per-formance on the Nelson-Denny ReadingComprehension Test.

Taken together, the results of our studies and

those of others show that, although older indi-viduals perform more poorly on standard tests of working memory, this decrement in workingmemory capacity is not related to a decrement inthe on-line construction of syntactic form. Stan-dard measures of working memory are related,however, to on-line measures that are likely dueto clause-final ‘‘wrap-up’’ processes, to off-linemeasures taken at the end of sentences, and tomore global measures of language comprehen-

sion in both young and elderly individuals.These findings suggest that gvWM is involved inoperations that occur after the initial processingof a sentence has taken place.

STRUCTURAL EQUATIONMODELING

We have further explored the relationship be-

tween WM capacity and interpretive and post-interpretive processing through the use of structural equation modeling. We modeledboth the relationship between WM measuresand on-line language processing (LP in Fig.11.5) and between WM measures and globallanguage processing (ND in Fig. 11.5) in agroup of young and elderly subjects. Workingmemory measures included alphabet span(Alph), backward digit span (Back), subtract 2span (Subt), running item span (Run), and asentence span measure formed by combiningperformance on the final-word recall compo-nent of two sentence span tasks (WM comp).On-line language processing was measured bythe increase in processing time at the capacity-demanding portion of the complex comparedto the simple sentence types (V CO-CS and V1SO-OS in Fig. 11.5). Global language process-

ing was measured with the Nelson DennyReading Comprehension Test. In both cases wewere able to find a good fit for the model. Inthe model of on-line sentence processing,all parameter estimates were significant otherthan those between memory and language

TABLE 11.5. Correlation betweenComposite Working Memory Measureand Language Measures

Type of Measure

Working

Memory

On-line Measures

COV1–CSV1 .23*COV1–CSNP2 .11SOV1–SSV1 .02SOV1–SSNP2 .00SOV1–OSV1 .01SOV1–OSNP2 .11

End-of-Sentence Measures

CO–CS response time .05

SO–OS response time .11SO–SS response time .07CO–CS accuracy .23*SO–OS accuracy .26*SO–SS accuracy .06

Discourse Measure

Nelson Denny comprehension .27*Nelson Denny rate .25*

*p<.05 CS¼ cleft subject; CO¼ cleft object; SS¼ subjectsubject; OS¼ object subject; SO ¼ subject object; V1¼

first verb; NP1¼first noun phrase; NP2¼ second nounphrase.





processing. In contrast, in the model withNelson Denny Reading Comprehension, allparameter estimates were significant, includingthose between memory and comprehension. Insubsequent models, we have replicated this re-sult using measures of on-line languageprocessing that are not confounded by end-of-

clause wrap-up effects (i.e., COV–CS NP1;SOV–OS N1) (DeDe, Caplan, Kemtes, &Waters, 2004). These data are consistent withthe view dividing language processing into twofunctional areas that are and are not related toWM as measured by standard tests.

SENTENCE COMPREHENSION ANDWORKING MEMORY IN PATIENTSWITH ALZHEIMER DEMENTIA

Patients with Alzheimer dementia (AD) providean excellent population in which to examinethe effects of a reduction in gvWM on sentence

comprehension. Patients with AD have impair-ments in the central executive component of gvWM, as shown by performance on both spanand other tasks (Morris, 1984). The central ex-ecutive is thought to play a critical role in thestorage and processing of items in gvWM and

Figure 11.5. Structural equation models of the relationship between working memory spanmeasures and measures of on-line and global language processing. Alph¼ alphabet span; Back¼backward digit span; Subt¼ subtract 2 span; Run¼ running item span; Sent Span¼sentence span; Mem¼working memory span measures; LP¼ language processing; ND¼NelsonDenny Reading Comprehension Test.




has been argued to play a very important role instandard WM tests (Just & Carpenter, 1992).The gvWM limitations seen in AD patients areoften severe and are stable. Thus, if gvWM as

measured by standard WM tests plays an im-portant role in structuring sentences syntacti-cally, then patients with AD should performvery poorly on sentence comprehension tasksthat require them to structure sentences syn-tactically.

We tested 26 AD patients in the mild-to-moderate stages of the disease and 26 age- andeducation-matched controls on the six WMtasks outlined above in the structural equation

modeling work. All of the AD patients were ableto perform all of the tasks, although the tasksranged in terms of difficulty. Several patientswere able to perform the sentence-processingand word recall components of the reading spantask alone but were unable perform the two taskstogether. General verbal WM spans of the pa-tients were significantly lower than thoseof controls on all of the tasks. Internal consis-tency was high for all of the tasks. Test–retest

reliability for a composite measure based onperformance on six of the span measures for thepatients was high (r ¼ .95 for Time I and TimeII). These data show that the working memoriesof AD patients are considerably more reducedthan those of matched normal subjects. Theproblem of reliably classifying subjects with re-spect to gvWM does not arise with AD patients,particularly when a composite measure basedon several different gvWM tests is used.

Given the severity of the WM decrementsseen in AD, if AD patients with gvWM im-pairments do not show impairments in syntacticprocessing, this would provide strong evidencethat this functional ability does not rely on thegvWM system. Conversely, evidence from ADcould show that gvWM is involved in syntacticprocessing. This argument requires two steps.First, AD patients must be shown to have dis-

proportionate impairments on syntacticallymore complex sentences (see above). Second,these impairments must be related to the gvWMlimitations of the patients, either through cor-relational analyses or other experimentalmanipulations (such as contrasting visual pre-

sentation of entire sentences with auditorypresentation; see Caplan & Waters, 1990, fordiscussion).

Studies of the sentence comprehension

abilities of AD patients have been inconsis-tent. Some authors have asserted that sentencecomprehension is impaired (e.g., Kontiola,Laaksonen, Sulkava, & Erkinjuntt, 1985) andothers that it is preserved (e.g., Schwartz, Marin,& Saffran, 1979). Careful examination of theresults of these studies suggests that AD patientsmay have performed poorly in many studiesbecause of deficiencies in their ability to accesssemantic knowledge, to enact responses, and to

accomplish other post-interpretive require-ments of many of these tasks (see Rochon, Wa-ters, & Caplan, 1994, for discussion). Somestudies have reported that comprehension isdifferentially impaired in AD on syntacticallymore complex sentence structures, providingevidence for a syntactic processing deficit (e.g.,Grober & Bang, 1995), while other have notfound this to be the case (e.g., Rochon et al.,1994; Waters, Caplan, & Rochon, 1995).

However, the relationship between poorerperformance on syntacticallymore complexsen-tences and reductions in gvWM in AD is notclear. For example, Grober and Bang (1985)reported that AD patients performed morepoorly on syntactically complex sentences in asentence picture–matching test, but they ar-gued that the impairment was not related to animpairment in gvWM, since the syntactic def-icit persisted even when a written sentence anda picture were in view simultaneously. Gross-man, Mickanin, Onishi, and Hughes (1995)found that AD patients had an impairment intasks assessing quantifier–noun agreement butargued that the deficit could not be explainedby their short-term memory deficits. Small,Kemper, and Lyons (2000) reported that ADpatients were worse at repeating syntacticallycomplex sentences and they attributed the im-

pairment to reduced gvWM in the patients, butthe correlations between gvWM and perfor-mance were not greater for the syntacticallycomplex than for the syntactically simple sen-tences. Our own work using off-line tasks hasfound that AD patients are not disproportionately




affected by increased syntactic complexity insentence–picture matching or acceptability judg-ment compared to elderly controls. We showedthat AD patients perform more poorly overall on

several off-line measures, such as sentence–picture matching and sentence acceptabilityjudgment, when the stimuli consist of syntacti-cally simple sentences (subject relatives) andcomplex sentences (object relatives) (Rochon etal., 1994; Waters et al., 1995). However, wehave consistently found that they do not dodisproportionately more poorly on the morecomplex sentence structures.

As elsewhere, more persuasive evidence re-

garding AD patients’ capacities would comefrom on-line measures. Although it initiallyseemed possible that AD patients might not beable to perform on-line tasks, several researchgroups have shown such measures can be ob-tained in AD. These studies show surprisinglygood on-line performance in some syntacticcomprehension tasks, and little effect of gvWMon performance in this population. For in-stance, MacDonald and colleagues used an on-

line cross-modal naming task, and found that Alzheimer patients showed the same increasein reading times for violations of subject–verbagreement and verb transitivity as that of normalcontrols (Kempler, Almor, Tyler, Andersen, &MacDonald, 1998). Moreover, the difference inreading times for the grammatical andungrammatical continuations did not correlatewith a composite measure of gvWM in thesestudies. In other work, these investigatorshave used a task in which subjects had to readaloud a sentence containing a heteronym (e.g.,dove, bow, wind) whose correct pronunciationdepended on prior syntactic or semantic in-formation (Stevens et al., 1998). The perfor-mance of the AD patients and elderly controlswas similar and showed that both groups reliedon the semantic and syntactic context to resolvethe ambiguity. Almor, MacDonald, Kempler,

Andersen, and Tyler (2001) reported that ADpatients and controls showed the same sensi-tivity to violations in subject–verb numberagreement in a short-sentence condition andsimilar degradation of this sensitivity in a long-sentence condition. Performance in neither

condition was related to gvWM. These re-sults suggest that many aspects of on-line syn-tactic processing are normal in Alzheimerpatients.

Two additional studies by MacDonald andcolleaguessuggest,however,thatwhilethemem-ory impairment seen in AD does not interferewith on-line grammatical processing withinsentences, it may affect on-line processing acrosssentences. In one study (Almor, Kempler,MacDonald, Andersen, & Tyler, 1999), ADpatients and controls heard an introductory two-sentence passage and the beginning of a thirdsentence, followed by a written pronoun that was

either contextually appropriate or inappropriate.Both AD patients and controls had longerreading times for inappropriate than appropriatepronouns, but the magnitude of the effect wassmaller in AD patients and was correlated with ameasure of gvWM. One feature of this study thatbears noting is that the control subjects answeredquestions about the appropriateness of the pro-noun after every passage whereas the patientsonly answered questions after every fourth pas-

sage, which may have led to the greater sen-sitivity of the controls to the pronouns’appropriateness. A second experiment (Almoret al., 2001) showed that AD patients were lesssensitive than controls to pronoun–antecedentnumber agreement violations across sentences,and that their performance was correlated withmeasures of gvWM.

Altogether, two studies show that AD patientshave difficulties with pronominal coreference inshort discourses, and the literature otherwiseindicates that AD patients in the early to middlestages of the disease show normal on-line syn-tactic processing. However, to date, very few on-line studies have been done.

We tested 20 AD patients and 20 controlson six tests of gvWM and on the auditory mov-ing windows task outlined above in which thestimuli consisted of pairs of sentences seen in

Table 11.1 (Waters & Caplan, 2001; see alsoWaters & Caplan, 1997). Patients had lowergvWM scores than controls, with the averagereading span of the patients being only 1.2compared to 3.1 for the controls. The patientsperformed more poorly than the controls on




the end-of-sentence acceptability judgments.However, patients were not more affected thancontrols by the syntactic complexity of a sen-tence in these judgments. Figure 11.6 shows

that AD patients and elderly controls showedsimilar effects of syntactic structure in the lis-tening time data. Table 11.6 shows the corre-lation between the on-line and end-of-sentencemeasures and the composite WM measures.These results are very similar to those in thestudy of young vs. elderly subjects outlinedabove. There was a significant correlation be-tween listening time at the verb of cleft-subjectsentences and WM. However, as in the study of

normals, this effect seemed to be an end-of-clause effect rather than a syntactic effect. Inaddition, there was also some evidence for arelationship between WM and accuracy inmaking judgments at the end of the sentencein this group of elderly subjects. These resultsindicate that early-stage AD patients are notimpaired in their ability to assign syntacticstructure and to use it to determine aspects of sentence meaning, despite their reduced work-

ing memories.

STUDIES OF APHASIC PATIENTSWITH REDUCED RESOURCES FORSYNTACTIC PROCESSING

The studies outlined to this point have all reliedupon data from normal individuals or from pa-tients who have impairments in gvWM. How-ever, there are other patients who seem to haveimpairments in the svWM system that is used forsyntactic processing. Patients with aphasia pro-vide such cases. Research into the nature of thesentence comprehension impairments seen inpatients who are aphasic subsequent to a lefthemisphere stroke has shown that many suchpatients have disturbances affecting their abilityto use syntactic form to determine the meaning

of a sentence (see Berndt, Mitchum, & Haen-diges, 1996, for review). Several aspects of theperformance of these patients suggest that onereason for this impairment is a reduction in theprocessing resources that a patient can apply tothis task. One piece of evidence that favors this

view is that groups of patients have been shownto have difficulty understanding sentences withmore complex syntactic structures (Caplan, Ba-ker, & Dehaut, 1985). Second, factor analyses

have shown that a single factor on which allsentence types load accounts for about two-thirdsof the variance in many syntactic comprehen-sion tasks in aphasic groups (Caplan et al.,1985). Third, cluster analysis shows that patientstend to be grouped according to their overalllevel of performance in these tasks, with perfor-mance in more impaired clusters showinggreater effects of syntactic complexity (Caplanet al., 1985). Finally, some patients have been

able to interpret sentences when either of twosyntactic features was present, but not when bothwere found in a sentence (Hildebrandt, Caplan,& Evans, 1987). These patterns of performanceare consistent with the hypothesis that the prob-lem in syntactic processing in sentence com-prehension that is seen in many aphasic patientsresults in part from reductions in their ability toallocate processing resources to the syntacticcomprehension task.

Study of these patients has allowed us toapproach the question of the relationship be-tween working memory capacity and syntacticprocessing in sentence comprehension from an-other angle, by examining the effect of a con-current memory load on syntactic comprehen-sion in patients in whom there is evidence for areduction in the resources available for syntacticprocessing in sentence comprehension. Therationale for such studies is that, if syntacticprocessing relies on svWM, a concurrent gvWMload would not further reduce syntactic pro-cessing performances in individuals whosesyntactic processing is already compromised byreduced svWM; if syntactic processing relies ongvWM, a concurrent gvWM load would furtherreduce syntactic processing performances. Thesestudies are relevant here because they usevariability in WM, in this case svWM, as a source

of data about the structure of the central execu-tive. We examined the effect of a concurrentdigit load on the sentence comprehension per-formance of aphasic patients (Caplan & Waters,1996). Over 100 aphasic patients were screenedto ensure that we only tested patients who




showed effects of syntactic complexity on a sen-tence picture–matching task, whose perfor-mance was below ceiling and above chance onthat task, and whose abilities to repeat singlewords permitted them to be tested on a digit spantask. We selected 10 patients who met these cri-teria, and tested them on a sentence picture–matching task in no-interference and concurrentload conditions (repeating a string of digits thatwas equivalent to their span or to their span –1).The data in Figure 11.7 show the typical patternseen in aphasic patients (when tested without anyconcurrent load) of decreasing levels of perfor-mance as the syntactic complexity of the sen-tence increases. Although aphasic patientsshowed large effects of syntactic complexitywhen tested without a concurrent load, theseeffects were not exacerbated by the addition of a

memory load. Their performance on the con-current memory task was poorer with larger digitloads, but the effect of syntactic complexity wasnot exacerbated. These results provide strikingevidence for the separation of the resources usedin syntactic processing in sentence comprehen-sion and those required for span tasks.

STUDIES OF THE NEURAL BASIS OFSYNTACTIC COMPREHENSION

A last approach to investigating the possibilitythat svWM, not gvWM, underlies syntacticprocessing is to examine neural responses tosvWM and gvWM load. We have used the firstof these approaches, and tested subjects whodiffer in either gvWM or svWM to see if theirneurovascular responses to svWM load differ.

Our approach to the study of the functionalneuroanatomy of syntactic processing in sen-tence comprehension has been to comparePET activity associated with processing syntac-tically more complex objected-extracted sen-tences to that with simpler subject-extractedsentences as in examples (4) and (5) outlinedabove. Experimental controls and counterbal-

ances were used to ensure that the two con-ditions differed only along the syntactic di-mension. Behavioral data (reaction times andaccuracy) were collected on a plausibilityjudgment task in all PET runs. Establishedmethods were used for PET acquisition, imagereconstruction, normalization, and statistical

TABLE 11.6. Correlation between Composite Working MemoryMeasure and Language Measures for Elderly Controls, Patientswith Alzheimer Dementia, and Combined Group of Subjects (All)

Elderly

Controls

Alzheimer

Group All

On-line Measures

COV1–CSV1 .28 .21 .44*COV1–CSNP2 .17 .11 .10SOV1–SSV1 .01 .06 .07SOV1–SSNP2 .02 .16 .13SOV1–OSV1 .03 .01 .14SOV1–0SNP2 .12 .19 .15

End-of-Sentence Measures

CO–CS response time .48* .24 .11

SO–OS response time .01 .21 .17SO–SS response time .05 .25 .21CO–CS accuracy .28 .24 .12SO–OS accuracy .49* .06 .15SO–SS accuracy .35 .09 .09

CS¼ cleft subject; CO¼ cleft object; SS¼ subject subject; OS¼ object subject; SO¼ subjectobject; V1¼first verb; NP1¼ first noun phrase; NP2¼ second noun phrase.





analysis. Our work with young subjects showeda replicable pattern of increased regional ce-rebral blood flow (rCBF) in Broca’s area whensubjects made plausibility judgments aboutthe more complex object-extracted sentencescompared to that with the simpler subject-extracted sentences (Caplan, Alpert, & Waters,1998; Caplan, Alpert, Waters, & Olivieri, 2000;Stromswold, Caplan, Alpert, & Rauch, 1996).

We followed up these findings with a study of young subjects who were matched for language-processing efficiency but who differed in gvWM(Waters, Caplan, Alpert, & Stanczak, 2003). Westudied nine pairs of subjects who differed intheir verbal gvWM as measured by a compos-ite score on the tests outlined above that have

been found to be the most reliable measures of gvWM. The groups were closely matched forage and sex and for level of performance on ascreening test of syntactic processing in whichreaction times and errors in making plausibilityjudgments about sentences were measured.

Analysis of the high- and low-gvWM subjects’

behavioral responses showed no differences inaccuracy or reaction times on the plausibilityjudgments made during scanning. High- andlow-gvWM subjects both activated the same

areas in the left and right inferior frontal cortex,as well as midline structures. We then reviewedthe gvWM and syntactic processing screen per-formances of the subjects and regrouped themto form two new groups, each containing eightsubjects, who were matched for gvWM and whodiffered in proficiency of syntactic processingon a screening test. As in previous studies, high-proficiency subjects activated the inferior fron-tal cortex, and low-proficiency subjects acti-

vated posterior structures. The resultsdemonstrate that individual differences in thegvWM system measured by standard gvWMtests are not associated with differences inneural hemodynamic responses to syntacticprocessing, while differences in proficiency of processing are. Taken together, the results of this study provide converging evidence for thenotion that individual differences in gvWMcapacity are not related to syntactic processing

ability.

SUMMARY OF THE EMPIRICALDATA

To summarize the empirical data, we havefound the following:

Standard measures of WM capacity arenot related to the efficiency with whichsubjects assign syntactic structures on-line.

Patients with reduced ability to use WMin syntactic comprehension are not furtheraffected by a concurrent verbal memoryload.

Regional CBF effects in sentence com-prehension are modulated by the profi-ciency of syntactic processing but not by

WM capacity.

We have interpreted the failure to find sig-nificant interactions between age (or gvWM),syntactic complexity, and region in our on-linestudies as evidence for a specific WM re-source used in the interpretation of syntactically

Figure 11.7. Performance of aphasic patients on asentence picture–matching task with no concurrentinterference task, while remembering a sequence of digits equal to their span minus 1, and while re-membering a sequence of digits at span. Sentenceson ordinate are listed in order of increasing syntacticcomplexity. A ¼ active; ACTh¼ active conjoinedtheme; TP¼ truncated passive; P¼passive; CO¼

cleft object; Con¼ conjoined; OS¼ object subject;

SO¼ subject object.




complex sentences. However, there are sev-eral methodological and theoretical issues thatmust be considered in relation to accepting thenull hypothesis.

Salthouse and Coon (1994) have pointedout that postulation of a separate process on thebasis of the lack of significant interactions be-tween age and an experimental manipulationrequires evidence that the measure presumedto reflect the added process (V in object-extracted sentences) have at least a moderateamount of variance that is independent of thevariance in the original measure (V in subject-extracted sentences) assumed to reflect the

operation of all other processes involved in thetask. Our data seem to meet this criterion. Asseen in Table 11.2, comparison of matchedsegments in object- and subject-extracted sen-tences shows that there is much morevariability in listening times for the verbs thanfor other segments.

A related issue is the extent to which thefailure to find a relationship between gvWMand syntactic processing reflects a lack of sta-

tistical power, rather than a true absence of such a relationship. One issue concernswhether sufficient numbers of subjects weretested to provide adequate power. The self-paced studies measuring listening times forsubject and object relatives were initially car-ried out on 100 college students and 130 indi-viduals ranging from 18 to 80þ years of age.They were subsequently replicated with a newset of stimulus materials in a study of 48 collegestudents and 48 elderly individuals and a studyof 20 AD patients and 20 matched controls. Inthese studies, listening times at the mostcapacity-demanding portion of the syntacticallymore complex sentence are typically twice aslong as those seen on the same word when itappears in a syntactically simpler sentence.Thus, these are very large effects. In addition,all of these studies have tested subjects on two

and in some cases three pairs of sentences thatprovide a test of the hypothesis. The results of these studies with several hundred subjectsand dozens of comparisons have been over-whelmingly consistent with the view that theincrease in processing time at the verb in more

complex sentences is not associated withgvWM.

Another issue is whether gvWM measures arereliable enough to classify subjects and whether

the division of subjects into arbitrary gvWMgroups based on span measures may mask finerdistinctions in gvWM among subjects who aremembers of the same group (Miyake, Emerson,& Friedman, 1999). In all of our on-line studies,we have measured gvWM using a variety of gvWM measures (Waters & Caplan, 2001), andin our more recent studies we have classifiedsubjects on the basis of a composite score thatreflects performance on several different gvWM

measures and that has good test–retest reliabil-ity (Waters & Caplan, 2003, 2004, 2005). In ad-dition, in several studies we have tested allsubjects on all span sizes from 2 to 8 on thegvWM measures and have used the total num-ber of items recalled in correct serial position,rather than span, in correlational analyses withthe listening time measures (Waters & Caplan,2004, 2005). This approach is more likely todetect relationships among measures, if they ex-

ist, than factorial analyses with group as a factor.The results of all of these analyses have beenconsistent and have not suggested a relationshipbetween gvWM and on-line syntactic proces-sing.

A related question is whether the measuresof gvWM that we and other researchers haveused are measuring the wrong WM construct.While tests that have been developed to mea-sure WM have generally invoked the Baddeleyand Hitch concept of WM that emphasizes thestorage and computation functions of WM,many of them seem to us to be better measuresof the Engle-type WM that emphasize thestorage and divided-attention components of WM. For instance, complex span tasks (e.g.,reading span) all require ongoing division of attention across two often completely unrelatedtasks, and many do not include a measure

of the computations in the task in the mea-sure of WM. Other working memory tasks (e.g.,alphabet span, subtract 2 span) require thememory and processing components of the taskto be related to each other, as they are in nat-urally occurring tasks that require working




memory, and are more naturally related to theBaddeley and Hitch notion of WM. Measure-ments of WM that use the latter type of taskdo not show relationships with syntactic

processing.

DISCUSSION

Our discussion of these findings is organized interms of the four questions posed to all con-tributors to this volume. An overview of ouranswers to the four questions is presented inBox 11.1.

Theory of Working Memory

The theoretical framework we operate withinis that of Baddeley and Hitch (1974), whichpostulates the existence of a short-duration,capacity-limited, memory system that is capableof both maintaining representations and per-forming computations on them. Our focuswithin this framework is on the central execu-

tive component of WM. Our work suggests thatthere is a specialization within the central ex-ecutive for the assignment of the syntacticstructure of a sentence and the use of that struc-ture in conjunction with the meanings of thewords in the sentence to determine aspects of the propositional meaning of the sentence. Sev-eral points need to be made about this putativespecialization.

First, if it exists, it is likely to apply only tothe initial assignment of syntactic structure.Revisions of that structure, as occur in gardenpath effects, are not necessarily independentof working memory capacity as usually mea-sured.

Second, if this specialization exists, we sug-gest that it is unlikely that its application is re-stricted to this one aspect of comprehension.Rather, we hypothesize that a specialized WM

system supports a wider set of processes that ac-complish the initial, automatic, on-line, oblig-atory, unconscious processes that assign thestructure and the literal, preferred, discourse-congruent meaning of utterances. We havecalled these interpretive processes. We suggest

that processes that take as input the products of interpretive processes— specifically, that take asinput the propositional content of sentences andthe relations between propositions in the pre-

ferred interpretation of a discourse—and operateon these representations in the service of func-tions such as reasoning, entry into semanticmemory, etc., do not use this specialized WMsystem.

One question that arises about this hypoth-esis is what the set of operations is that fallswithin the scope of svWM. There are threeaspects to our answer to this question. The firstpoint is that the answer requires commitment

to a theory of linguistic structure and its pro-cessing. For instance, the studies in Fodor andFerreira (1998) present a variety of models of features of the structure of sentences and dis-course that do and do not trigger controlledrevisions. Each of these models yields a slightlydifferent characterization of interpretive pro-cesses. Second, we think that progress can bemade by examining phenomena for whichthere is good evidence that they involve auto-

matic first-pass processing. Third, we note thatthe major contrast we propose is between as-signment of initial meaning and the use of thatmeaning to accomplish tasks. It is thereforepossible to test our model by varying task de-mands, as well as by contrasting initial auto-matic and controlled sentence processing.Overall, though there are unquestionably grayareas where it is not clear whether a processthat is relevant to establishing meaning is au-tomatic or controlled, the theory seems to us tobe sufficiently well articulated to be tested inmany uncontroversial ways.

A second question relates to the notion of automaticity. There are two issues here. First,automaticity of a process is usually said to pre-clude its having resource demands, but thiscannot be an appropriate way to conceptualizeautomaticity if processes such as on-line com-

prehension can be shown to require suchresources but are to be considered to be au-tomatized. We believe that the notion of auto-maticity needs to be revised to exclude the ideathat automatic processes are resource-free.Second, one could see the data as indicating




that syntactic processing is so automatized thatit requires very little in the way of WM re-sources, rather than as pointing to the existenceof a separate WM system dedicated to syntactic

comprehension (and, by hypothesis, related as-pects of comprehension).Though the ‘‘automatized’’ view is possible,

we are inclined toward the ‘‘separate system’’view because of two features of the results. First,the sentence types we have used to measureWM use are at the limit of human compre-hension abilities. Most listeners cannot under-stand object-extracted relative clauses thatcontain one more embedding than those used

in our materials, such as Sentence 6:

6. The boy that the baby that the girlhugged kissed slipped.

Second, some subjects, such as AD patients,who showed normal effects of syntactic WMdemand in these structures have extremely re-duced WM capacity.

Taken together, these two facts indicate that

individuals with extremely reduced WM havenormal capacity to handle the greatest WMdemands associated with syntactic processingthat can ordinarily be handled. Though thiscould occur if syntactic processing were so au-tomatized that the largest universally tractableWM demands it imposed were within the ca-pacity of a WM system that is reduced as muchas is ever seen, it seems to us to shift the burdenof argument onto those who favor the ‘‘auto-matized’’ view.

Sources of Variabilityin Working Memory

Our results indicate that variability in utiliza-tion of WM in on-line first-pass syntactic pro-cessing is not related to overall WM capacity asmeasured by standard tests, or to age. We

comment briefly on both these findings.One consideration regarding the findingthat ‘‘standard’’ WM capacity does not predictWM utilization in syntactic processing is that‘‘standard’’ WM measures are measures of WMcapacity and the measures of on-line process-ing are measures of utilization. A relationship

between the two would not be expected if re-source demands do not affect processing whenthey fall within the range of resource avail-ability. If there is only one resource available to

meet the WM demands of syntactic pro-cessing, this conclusion would entail that thesedemands are minimal, which raises the issuesconsidered immediately above in our discus-sion of automaticity. If there is a specializedWM system for syntactic processing, the WMdemands made on this system by the sentenceswe have used are within subjects’ capacity,since the reaction time measures we use arebased on sentences that subjects responded to

correctly and error rates were relatively low. Itstill may be the case that individual differencesin utilization of this putative WM resource arerelated to individual differences in this spe-cialized resource’s capacity. To see if this is thecase, we need a measure of the capacity of thisspecialized resource, something like a mea-sure of span rather than a measure of its utili-zation.

Using the example of span (or complex

span), such a measure would indicate the levelof syntactic WM load at which sentence com-prehension fails for each individual. Thiswould require that subjects be tested for com-prehension of sentences that differ by smallincrements in their peak WM load, and thepoint at which they fail to reliably understandthe sentences be determined. Ideally, their on-line performances would also be measured toensure that comprehension failure is associatedwith abnormal on-line processing at points of peak load. The psychometric properties of themeasure would have to be established. Even-tually, if the data have the hoped-for properties,a measure based on comprehension data couldbe used as a measure of syntactic WM capacity.This is a daunting research program, althoughits scope is certainly not greater than that of research programs into constructs such as ‘‘fluid

intelligence.’’ Until a sufficient part of it isaccomplished to have a preliminary measure of this sort, the question of whether individualdifferences in the utilization of a putative WMresource for syntactic processing are related toindividual differences in this resource’s capac-ity remains unanswered.




In addition to asking whether individualdifferences in utilization of a putative special-ized WM resource are related to individualdifferences in the resource’s capacity, we can

also ask whether they are related to the utili-zation, rather than the capacity, of a ‘‘standard’’WM resource. To answer this question, weneed a measure of WM utilization, not WMcapacity. Perhaps a measure can be found insome chronometric value, such as the dura-tion of the initial pause before list productionin recall or the duration of inter-item pauses inrecall trials. The operational problem we seehere is that a ‘‘standard WM resource’’ involves

many types of operations (attention, inhibition,search, etc.) and it may be necessary to mea-sure different ones differently.

Turning to the age variable, the fact that agedoes not predict utilization of this putativespecialized WM system indicates that syntacticprocessing groups with measures of crystallizedintelligence, like vocabulary size, not fluid in-telligence, like performance on the Raven’smatrices, with respect to its relationship to age.

Operationally, syntactic processing seems to liebetween doing the Raven’s test and retainingvocabulary. It resembles the Raven’s test and isdistinguished from vocabulary retrieval insofaras it involves computations that relate differentitems. It differs from the Raven’s insofar as thecomputations it requires for the most part in-volve a relatively small set of highly overprac-ticed operations.2 The Raven’s test requires newcomputations, at least in its mid- and late items.Syntactic processing seems to lie between pro-totypical tasks that are weighted toward crystal-lized intelligence and those weighted towardfluid intelligence. We return to this point inrelationship to Question 3.

What, then, does determine individualvariability in use (and maybe capacity) of svWM?We see two main possibilities. One is that expo-sure and practice leads to the development of

reliance on an independent resource and thatthese effects of experience and practice asymp-tote relatively early in the life cycle. However, inaddition to exposure and practice, we think thatan innateness factor is needed to account for thehighly idiosyncratic nature of the representationsconstructed in parsing and that there are very

likely to be genetic factors that partially deter-mine variability in svWM use. The relative rolesof these two factors are unknown. (Disease alsoaffects svWM use, as seen in aphasic cases).

Extending our perspective, considerations re-garding both experience and innate endowmentare consistent with the speculation that svWM,if it exists, supports interpretive processing andnot just syntactic comprehension. Consideringexposure and practice, extracting the literal,preferred, discourse-congruent meaning of anutterance involves constructing a limited num-ber of types of representations using a limitednumber of operations, and constructing these

representations using these operations consti-tutes one of the most overpracticed sets of relatedcognitive functions that humans undertake. Afterit has been recovered from an utterance, lit-eral,preferred,discourse-congruentpropositionalmeaning engages different cognitive operationsas a function of the purpose of listening—entryinto semantic memory; use in immediate plan-ning of action, etc.—and these operations are farless practiced than the extraction of this repre-

sentational set from the signal, simply becausethe number of times they are engaged is a propersubset of the number of times interpretive oper-ations are engaged. However, practice patternsalone cannot explain the fact that interpretiveoperations co-occur the way they do. Why, forinstance, is phonological form automaticallyconverted into lexical and lexical semantic rep-resentations but not automatically entered intolong-term memory? Again, innately specified,genetically and/or epigenetically determinedprocesses seem to be involved to delineate thebasic architecture of processes that use linguisticrepresentations.

Other Sources of Variability inWorking Memory and Connectionto Other Work Reportedin this Volume

Our work suggests that the central executive isfractionated, and suggests that variability in sub-parts of the central executive may not be sharedwith that in other parts. It also provides clues asto lines along which the central executive maybe explored and fractionated.




Models of WM—whether they be the origi-nal Baddeley and Hitch model that postulatedthe existence of the central executive but did notcharacterize it or more recent models that link

performance on WM tasks to processes that aresupported by the central executive (e.g., atten-tional control)—seem to us to generally un-derstate the ‘‘processing’’ that the central execu-tive accomplishes. Many models emphasize therole of attention, inhibition and other aspects of control in central executive function withoutmuch emphasis on the processing that occursonce attention has been deployed or inhibitionachieved. Some models specify processing, but

characterize it at a relatively general level. Forinstance, Oberauer et al. (Chapter 3) charac-terize processing as ‘‘binding’’ but say littleabout what binding consists of; Cowan refers to‘‘search’’ as an operation that takes place duringpauses between retrieval of words in span tasks,but search is surely only one of many processingoperations that the central executive accom-plishes. In part this emphasis may stem from theattempt to characterize performance on WM

tasks or, in the case of complex span tasks, on therecall measures in such tasks, which do not in-volve complex computational processes. How-ever, even in work that varies processing load inthe non-recall task in complex span tasks, whatsubjects have to do to meet the processing loadis rarely as relevant to the issues explored as thatthe processing load is varied. A considerableamount of work in this volume also attempts topredict general intellectual capacity (fluid in-telligence) using factor analyses of task perfor-mance, rather than mechanistic descriptions of the operations that subjects use to perform tasks,as a measure of this construct. We think theseapproaches, while important, miss one poten-tially equally important aspect of central exec-utive function—how tasks are accomplished.This is the middle ground that lies between fluidand crystallized intelligence referred to earlier,

that is occupied by models of tasks such assyntactic comprehension.There are some papers that consider tasks in

relation to WM theory that provide much moreelaborate descriptions of the processes that go onin accomplishing those tasks (Kieras Meyer,

Mueller, and Seymour, 1999). These papers andour orientation fit well into the general frame-work developed by Ericsson and his colleagues(Ericsson & Delaney, 1999), that emphasizes

the importance of activated knowledge and op-erational routines in long term memory in de-termining how tasks are accomplished. Unlikethe expert performances that Ericsson andcolleagues have studied, which involve highlytrained specialized domains such as chess play-ing, our interest has been in a domain in whichexpertise is widespread in neurologically intactindividuals who have had normal exposure tolanguage and normal opportunity for its use.

Despite this difference in how expertise is ac-quired and some rather fascinating differencesin the ability to expand the skill,3 skilled per-formances in these different domains all involvehighly efficient utilization of complex, domain-specific, operations and knowledge. We thinkthat many aspects of ordinary cognitive func-tioning in which normal individuals showhighly skilled performance (e.g., visual catego-rization, reaching and throwing, and so on) in-

volve similar domain-specific knowledge andoperations and, in some cases, application of such knowledge and operations to representa-tions that have been activated at temporally dis-continuouspoints;i.e.,aWMsysteminthesensewe started with. We believe our work points toa need to include detailed models of the pro-cessing requirements of tasks—especially every-day ecologically utilitarian tasks—into modelsof the central executive.

Additionally, along these lines, our resultssuggest that the central executive may be frac-tionated along lines that are defined by thenature of these processes, not just the nature of the representations to which the processesapply. That is, within the domain of verbal func-tions, processes related to syntactic compre-hension appear to dissociate with respect toWM utilization from processes that underlie

performance on other verbally mediated tasks.While one can characterize the processes thatare needed in each domain in broad terms suchas‘‘binding,’’theactualoperationsthattakeplaceand the representations that are constructed ineach domain have important individual features.




This is clearly seen in the area of syntax, wherethe particular structures that are constructed areunlike those seen in any other cognitive domain.These operations may define components of the

central executive.While our work is closely aligned with that of Ericsson, Kieras, Meyer and others, it is not easyto relate it to other work presented in this vol-ume; specifically, to research into the impor-tance of the domain-general operations such asattentional control, inhibition, search and oth-ers in WM. For instance, Engle and colleagues(Engle, Kane, & Tuholski, 1999) emphasizethe role of attentional control in performing

WM tasks and the importance of this abilityin explaining the predictive value of WM forfluid intelligence. It is not easy to see how at-tentional control (or inhibition, or search) isrelated to handling the WM demands imposedby syntactic processing. While sentence com-prehension requires that sentences be attendedto, it is unclear that any variation in attentionalcontrol occurs as a function of local syntacticWM load. Rather, once attended to, initial

sentence processing proceeds as best it can,limited by factors that we do not understand. Inthis sense, if the notions ‘‘automatic’’ and‘‘uncontrolled’’ apply to any cognitive functionbeyond perceptual identification, many as-pects of the comprehension of a sentence—theones we are focusing on—are automatic anduncontrolled once a sentence is attended. Onecan, of course, reformulate the notion of ‘‘attention’’ or ‘‘attentional control’’ to includewhatever underlies coping with local variationin WM demand in sentences. This would beconsistent with the tradition of conceiving of ‘‘at-tention’’ as a ‘‘resource,’’ as in Kahneman (1973).If one takes this approach, the difference be-tween the Baddeley and Hitch and Englemodels evaporates. In that case, our modelwould have to be reformulated as claiming thatthere is a specialized attentional resource sub-

component.Other researchers (not represented in thisvolume) go further, and deny the existence of workingmemory(orWMlimitations)altogether.MacDonald and Christiansen (2002) argue thatprocedure-based symbol-manipulating models

impose limits on the combination of computa-tional activity and memory storage by artificialmeans; in their view, a limited WM capacity is aDeus ex Machina. They claim that connec-

tionist models provide a more realistic compu-tational mechanism for cognitive processes andthat, in such models, limits on computationalcapacity are determined by the experience of the system. We strongly disagree with this view.Connectionist models are limited by their in-trinsic, a priori, architectures (cf., the addedpower associated with hidden units), which are,at present, just as arbitrary as WM capacity limitson symbolic manipulations. There are three is-

sues that need to be explored. What are the limitson human information processing? Are theselimits domain-specific (and, if so, what arethe domains)? Why do these limits arise? Deny-ing the existence of limits is not an option.

There are also empirically based challengesto the need for WM, if not to the concept itself.Salthouse, for instance, has argued that age-related variation in processing speed accountsfor age-related variation in WM (see Salthouse,

1996, for a review). An important question iswhether processing speed reduction, or othersimilar mechanisms such as sensory thresholdelevation, are alternatives to working memoryaccounts of reduced processing efficiency. Onemechanism that Salthouse (1996) has sug-gested whereby processing speed might affectperformance levels is if the temporal course of the activation of representations is slowed be-yond a point that allows incoming material tobe integrated into a developing set of repre-sentations (sensory threshold elevation couldoperate through a similar mechanism). How-ever, this will only happen if incoming and/orconstructed materials cannot be adequatelystored in a memory system, which leads directlyback to WM limitations. Our research could berecast as demonstrating domain-specific decre-ments in processing speed; it is less easy to

see how our results could be consistent withelevated sensory thresholds being responsiblefor decreasing processing efficiency, given thatsensory thresholds apply ubiquitously to stimuliand we have documented domain-specific dec-rements in processing.




Contribution to General WorkingMemeory Theory

Our answer to this question can be brief

because it is contained in our previous dis-cussion. We think that our work makes a con-tribution in that it argues for the existence of atleast one specialization within the central ex-ecutive. It suggests that operational domains inthe central executive may be defined compu-tationally as well as according to types of representation (verbal, visual). Our work alsosuggests that characterization of the actual op-erations required to perform a task is an im-

portant part of characterizing the nature of

the central executive. Attention, inhibition,search, etc. all must apply to representationsthat are subject to computations; an under-standing of the central executive requires un-

derstanding these representations and computa-tions. Indeed, to go out on a bit of a limb, itseems worthwhile to ask whether the some-times not inconsiderable variance in compositemeasures of fluid intelligence and factors like‘‘g’’ that remains unaccounted for after vari-ance due to attention, inhibition, search, etc. isremoved might not reflect individual differ-ences in the ability to deal with task-specificcomputations on particular types of represen-

tations.


1. THE OVERARCHING THEORY OF

WORKING MEMORY

We operate within a Baddeley and Hitch–type

framework in which there is a short-duration,

capacity-limited, verbal working memory sys-

tem capable of maintaining representations and

performing computations on them. Our focus is

on the central executive component of WM. We

hypothesize that the central executive is divided

into at least two components—one that is in-

volved in the initial assignment of meaning, and

one that uses that meaning to accomplish tasks.


MEMORY VARIATION

Variability in use of WM in the domain of

syntactic comprehension that we are interested

in is not related to overall WM capacity as

measured by standard tests, or to age. There

seem to be two main possibilities for what

determines individual variability in use (andmaybe capacity) of svWM. One is that expo-

sure and practice lead to the development of

reliance on an independent resource that as-

ymptotes relatively early in the life cycle. In

addition, an innateness factor likely accounts

for the highly idiosyncratic nature of the rep-

resentations constructed in parsing and a ge-

netic factor likely helps determine variabilityin svWM utilization. The relative roles of these

factors are unknown.


MEMORY VARIATION

In our view, the central executive is fraction-

ated. This fractionation may occur along lines

defined by the nature of the representations and/

or by the nature of the processes involved in atask. Variability in subparts of the central exec-

utive may not be shared with that in other parts.



Our work argues for the existence of at least

one specialization within the central executive.

It suggests that operational domains in the cen-

tral executive may be defined computationallyas well as according to types of representation

(verbal, visual). Characterization of the actual

operations required to perform a task may be an

important part of characterizing the nature of

the central executive.




Notes

1. Ferreira and colleagues have shown that this

paradigm is sensitive to the same sorts of linguistic

variables, and yields similar results, as the widelyused self-paced reading paradigm (Ferreira et al.,

1996).

2. Note that this is an area in which the choice of

parsing model makes an enormous difference.

The characterization of parsing as using a small

number of overpracticed operations is predicated

upon parsing being a matter of assigning higher-

order categories and links between these cate-

gories to lexical syntactic categories and features

that are activated as part of lexical access. If parsing consists of word-to-word associations, as

posited by some connectionist modelers (e.g.,

Christiansen and Chater, 1999), most sentences

would require novel operations, much more akin

to those required by the Raven’s test.

3. Skills seem to asymptote at higher levels in spe-

cialized domains. Even extensive practice (in

anyone) does not seem to improve the ability to

deal with multiply embedded object extractions

anywhere near as much as practice (in some peo-ple) improves the ability to look ahead in chess.

References

Almor, A., Kempler, D., MacDonald, M. C.,

Andersen, E. S., & Tyler, L. K. (1999). Why

do Alzheimer’s patients have difficulty with

pronouns? Working memory, semantics, and

reference in comprehension and production of

Alzheimer’s disease. Brain and Language, 67,

202–228.

Almor, A., MacDonald, M. C., Kempler, D.,

Andersen, E. S., & Tyler, L. K. (2001). Com-

prehension of long distance number agreement

in probable Alzheimer’s disease. Language and

Cognitive Processes, 16, 35–63.



of learning and motivation: Advances in researchand theory (Vol. 8, pp. 47–89). New York: Ac-

ademic Press.

Balogh, J., Zurif, E. B., Prather, P., Swinney, D., &

Finkel, L. (1998). Gap filling and end of sentence

effects in real-time language processing: Implica-

tions for modeling sentence comprehension in

aphasia. Brain and Language, 61, 169–182.

Baum, S. (1991). Sensitivity to syntactic violations

across the age span: Evidence from a word-

monitoring task. Journal of Clinical Linguisticsand Phonetics, 5, 317–328.

Berndt, R., Mitchum, C. C., & Haendiges, A. N.

(1996). Comprehension of reversible sentences

in ‘‘agrammatism’’: A meta-analysis. Cognition,

58, 289–308.

Caplan, D., Alpert, N., & Waters, G. S. (1998). Effects

of syntactic structure and propositional number

on patterns of regional cerebral blood flow.


Caplan, D., Alpert, N., Waters, G., & Olivieri, A.(2000). Activation of Broca’s area by syntactic

processing under conditions of concurrent artic-

ulation. Human Brain Mapping, 9, 65–71.

Caplan, D., Baker, C., & Dehaut, F. (1985).

Syntactic determinants of sentence comprehen-

sion in aphasia. Cognition, 21, 117–175.

Caplan, D., & Waters, G. (1990). Short-term

memory and language comprehension: A crit-

ical review of the neuropsychological literature.

In T. Shallice and G. Vallar (Eds.), Theneuropsychology of short-term memor y (pp.

337–389). Cambridge, UK: Cambridge Uni-

versity Press.

Caplan, D., & Waters, G. S. (1996). Syntactic

processing in sentence comprehension under

dual-task conditions in aphasic patients. Lan-

guage and Cognitive Processes, 11, 525–551.

Caplan, D., & Waters, G. S. (1999a). Verbal

working memory and sentence comprehension.

Behavioral and Brain Sciences, 22, 77–94.

Caplan, D., & Waters, G. S. (1999b). Verbal working

memory and sentence comprehension [dis-

cussion]. Behavioral and Brain Sciences, 22,

95–126.

Caplan, D., & Waters, G. S. (2002). Working

memory and connectionist models of parsing:

A response to MacDonald and Christiansen.


Carpenter, P. A., Miyake, A., & Just, M. A. (1994).Working memory constraints in comprehen-

sion: Evidence from individual difference, apha-

sia and aging. In M. A. Gernsbacher (Ed.),

Handbook of psycholinguistics (pp. 1075–1122).

New York: Academic Press.




Christiansen, M. H. and Chater, N. (1999). Toward a

connectionist model of recursion in human

linguistic performance. Cognitive Science, 23:

157–205.

Craik, F. I. M. (1986). A functional account of age differences in memory. In F. Klix & H.

Hagendorf (Eds.), Human memory and cogni-

tive capabilities (pp. 409–421). New York:

Elsevier.

Daneman, M., & Carpenter, P. (1980). Individual

differences in working memory and reading.


ior, 19, 450–466.

DeDe, G., Caplan, D., Kemtes, K., & Waters, G. S.

(2004). The relationship between age, verbalworking memory and language comprehension.


Engle, R. W., Kane, M. J., & Tukholski, S. W. (1999).

Individual differences in working memory ca-

pacity and what they tell us about controlled

attention, general fluid intelligence, and func-

tions of the pre-frontal cortex. In A. Miyake &

P. Shah (Eds.), Models of working memory:

Mechanisms of active maintenance and executive

control (pp. 102–134). New York: CambridgeUniversity Press.

Ericsson, A., & Delaney, P. F. (1999). Long-term




Models of working memory (pp. 257–297). New


Fier, C. D., & Gerstman, L. J. (1980). Sen-

tence comprehension abilities throughout the

adult life span. Journal of Gerontology, 35, 722–

728.

Ferreira, F., Henderson, J. M., Anes, M. D., Weeks, P.

A., & McFarlane, D. K. (1996). Effects of lexical

frequency and syntactic complexity in spoken

language comprehension: Evidence from the

auditory moving window technique. Journal of

Experimental Psychology: Learning, Memory,


Fodor, J. D., & Ferreira, F. (1998). Reanalysisin sentence processing. Boston: Kluwer.

Gibson, E. (1998). Linguistic complexity: Locality

of syntactic dependencies. Cognition, 68, 1–76.

Grober, E., & Bang, S. (1995). Sentence compre-

hension in Alzheimer’s disease. Developmental


Grossman, M., Mickanin, J., Onishi, K., & Hughes,

E. (1995). An aspect of sentence processing in

Alzheimer’s disease: Quantifier-noun agree-

ment. Neurology, 45, 85–91.

Hildebrandt, N., Caplan, D., & Evans, K. (1987).The man left without a trace: A case study of

aphasic processing of empty categories. Cogni-

tive Neuropsychology, 4, 257–302.

Just, M. A., & Carpenter, P. A. (1992). A capacity

theory of comprehension: Individual differ-

ences in working memory. Psychological Re-

view, 99, 122–149.

Kahneman, D. (1973). Attention and effort. Engle-

wood Cliffs, NJ: Prentice Hall.

Kemper, S. (1986). Imitation of complex syntacticconstructions by elderly adults. Applied Psycho-

linguistics, 7, 277–287.

Kemper, S., & Kemtes, K. (1999). Limitations on

syntactic processing. In S. Kemper & R. Kliegl

(Eds.), Constraints on language: Aging, gram-

mar, and memory (pp. 79–105). Boston: Kluwer.

Kempler, D., Almor, A., Tyler, L. K., Andersen, E. S.,

& MacDonald, M. C. (1998). Sentence compre-

hension deficits in Alzheimer’s disease: A com-

parison of off-line vs. on-line sentence processing.Brain and Language, 64, 297–316.

Kemtes, K. (1999). Decomposing adults’ sentence

comprehension: The role of age, working memory,

inhibitory functioning, and perceptual speed.

Unpublished doctoral dissertation, University

of Kansas, Lawrence, KS.

Kemtes, K., & Kemper, S. (1997). Younger and

older adults’ on-line processing of syntactically

ambiguous sentence. Psychology and Aging, 12,

362–371.

Kieras, D. E., Meyer, D. E., Mueller, S., & Seymour,

T. (1999). Insights into working memory from

the perspective of the EPIC architecture for

modeling skilled perceptual-motor and cogni-

tive human performance. In A. Miyake & P.

Shah (Eds.), Models of working memory: Mech-

anisms of active maintenance and executive

control (pp. 183–223). New York: Cambridge

University Press,King, J. W., & Just, M. A. (1991). Individual

difference in syntactic processing: The role of

working memory. Journal of Memory and

Language, 30, 580–602.

Kontiola, P., Laaksonen, R., Sulkava, R., & Erkinjuntt,

T. (1985). Patterns of language impairment is




different in Alzheimer disease and multi-infarct

dementia. Brain and Langauge, 38, 364–383.

Lewis, R.I. (1996). Interference in short-term mem-

ory: The magical number two (or three) in

sentence processing. Journal of PsycholinguisticResearch, 25, 93–115.

MacDonald, M. E., & Christiansen, M. H. (2002).

Reassessing working memory: A reply to Just

and Carpenter and Waters and Caplan. Psy-

chological Review, 109,

MacDonald, M. C., Pearlmutter, N. J., & Seiden-

berg, M. S. (1994). Lexical nature of syntactic

ambiguity resolution. Psychological Review,

101, 676–703.

Miyake, A., Emerson, M. J., & Friedman, N. P.(1999) Good interactions are hard to find.

Behavioral and Brain Sciences, 22, 108–109.

Morris, R.G. (1984). Working memory in Alzheimer-

type dementia. Neuropsychology, 8, 544–554.

Rochon, E., Waters, G. S., & Caplan, D. (1994).

Sentence comprehension in patients with Alzhei-

mer’s disease. Brain and Language, 46, 329–349.

Salthouse, T. A. (1988). The role of processing re-

sources in cognitive aging. In M. L. Howe & C. J.

Brainerd (Eds.), Cognitive development in adult-hood (pp. 185–239). New York: Springer-Verlag.




Salthouse, T. A., & Coon, V. E. (1994). Interpre-

tation of differential deficits: The case of aging

and mental arithmetic. Journal of Experimental


20, 1172–1182.

Schwartz, M. F., Marin, O. S., & Saffran, E. M.

(1979). Dissociations of language function in

dementia. A case study. Brain and Language, 7,

277–306.

Small, J., Kemper, S., & Lyons, K. (2000). Sentence

repetition and processing resources in Alzhei-

mer’s disease. Brain and Language, 75, 232–258.

Stevens Dagerman, K., Almor, A., MacDonald,

M.E., Kempler, D., & Andersen, E. (1998).

Preserved use of semantic and syntactic contextin Alzheimer’ disease. Brain and Language, 65,

87–132.

Stine-Morrow, E., Loveless, M. K, & Soederberg,

L. M. (1996) Resource allocation in on-line

reading by younger and older adults. Psychology

and Aging. 11, 475–486.

Stine-Morrow, E. A. L., Ryan, S., & Leonard, J. S.

(2000). Age differences in on-line syntactic pro-

cessing. Experimental Aging Research, 26, 315–

322.

Stromswold, K., Caplan, D., Alpert, N., & Rauch, S.(1996). Localization of syntactic comprehen-

sion by positron emission tomography. Brain


Waldstein, R., & Baum, S. (1992 ). The influence of

syntactic and semantic context on word moni-

toring latencies in normal aging. Journal of

Speech Language Pathology and Audiology, 6,

217–222.

Waters, G. S., & Caplan, D. (1996a). The measure-

ment of verbal working memory capacity and itsrelation to reading comprehension. Quarterly

Journal of Experimental Psychology, 49A, 51–75.

Waters, G. S., & Caplan, D. (1996b). The capacity

theory of sentence comprehension: Critique of

Just and Carpenter (1992). Psychological Re-

view, 103, 761–772.

Waters, G. S., & Caplan, D. (1997). Working

memory and on-line sentence comprehension

in patients with Alzheimer’s disease. Journal of

Psycholinguistic Research, 26, 377–400.Waters, G. S., & Caplan, D. (2001). Age, working

memory and on-line syntactic processing in

sentence comprehension. Psychology and Ag-

ing, 16, 128–144.

Waters, G. S., & Caplan, D. (2002). Working

memory and on-line syntactic processing in

Alzheimer’s disease: Studies with auditory

moving windows presentation. Journal of Ger-

ontology B: Psychological Sciences and Social

Sciences, 57, P298–311.

Waters, G. S., & Caplan, D. (2003). The reliability

and stability of verbal working memory mea-

sures. Behavior Research Methods, Instruments

& Computers, 35, 550–564.

Waters, G. S., & Caplan, D. (2004). Verbal working

memory and on-line syntactic processing: Evi-

dence from self-paced listening. Quarterly

Journal of Experimental Psychology, 57, 129–

164.Waters, G. S., & Caplan, D. (2005). The relation-

ship between age, processing speed, working

memory capacity, and language comprehen-

sion. Memory, 13, 403–413.

Waters, G. S., Caplan, D., Alpert, N., & Stanczak, L.

(2003) Individual differences in rCBF correlates




of syntactic processing in sentence comprehen-

sion: Effects of working memory and speed of

processing. NeuroImage, 19, 101–112.

Waters, G. S., Caplan, D., & Hildebrandt, N.

(1987). Working memory and written sen-tence comprehension. In M. Coltheart (Ed.),

Attention and Performance XII: The psychol-

ogy of reading (pp. 531–555). Hillsdale, NJ:

Erlbaum.

Waters, G. S., Caplan, D., & Rochon, E. (1995).

Processing capacity and sentence comprehension

in patients with Alzheimer’s disease. Cognitive


Zurif , E., Swinney, D., Prather, P., Wingfield, A., &Brownell, H. (1995). The allocation of memory

resources during sentence comprehension: Ev-

idence from the elderly. Journal of Psycholin-

guistic Research, 24, 165–182.




Author Index

Note: Page numbers followed by f and t refer to figures and tables, respectively.

Abeles, N., 229, 238, 247 Abrams, R.A., 137, 158, 197, 199, 200, 201, 201f, 202,

203, 206, 207, 208, 217, 219, 222 Ackerman, P.L., 49, 55, 57, 61, 67, 68, 70, 218, 221 Adams, A., 115, 131 Adams, D.R., 206, 223 Adams, J.L., 168, 177, 191 Adams, L.M., 239, 245 Aguiar, A., 173, 185, 189 Aguirre, G.K., 88, 104 Ahmed, A., 173, 189 Alba, J.W., 236, 243 Alexander, G.E., 167, 190 Alloway, T.P., 115, 131 Allport, D.A., 37, 45, 54, 59, 70, 237, 243 Almor, A., 287, 299, 300, 301 Alpert, N., 291, 299, 301 Altmann, E.M., 37, 45

Andersen, E.S., 287, 299, 300, 301 Andersen, R.A., 219, 223 Anderson, J.R., 50, 70, 180, 190, 227, 228,

233, 234, 235, 243 Anderson, K.C., 180, 193 Anderson, M.C., 32, 45 Anderson, N.D., 228, 244 Anes, M.D., 274, 299n1, 300

Arbuckle, T.Y., 237, 243 Armilio, M.L., 230, 249 Arnsten, A.F.T., 92, 103, 239, 243 Aston-Jones, G., 239, 243 Astur, R.S., 200, 224 Atkinson, R.C., 7, 15, 21, 45 Atkinson, T.M., 60, 74 Awh, E., 29, 46, 200, 219, 221, 223, 253, 254, 256, 257,

268, 270, 271

Babcock, R.L., 54, 55f, 58, 74, 138, 141, 143, 145, 160,198, 221, 260, 268

Bacon, W.F., 36, 45Baddeley, A.D., xiii, xiv, xvi, 6, 7, 8, 11, 15, 21, 22, 38, 39,

40, 45, 50, 51, 54, 58, 59, 66, 70, 71, 72, 79, 103, 109,110, 111, 113, 127, 129, 130, 131, 135, 136, 137, 138,139, 140, 141, 142, 144, 145, 147f, 148, 149, 150, 151,154, 156, 157, 158, 159, 188, 190, 194, 195, 196, 197,

199, 200, 202, 205, 207, 208, 216, 218, 219, 221, 222,227, 228, 237, 242, 243, 250, 253, 268, 272, 293, 299

Badre, D., 89, 104Baillargeon, R., 173, 185, 190Baker, C., 288, 299Baker, S., 111, 132Baldwin, J.M., 5, 13, 15Ballard, D., 88, 104

303



Balogh, J., 275, 299Balota, D.A., 230, 232, 242, 244Bang, S., 286, 300Bangert, A., 262, 269Banich, M.T., 263, 269

Bara, B.G., 64, 72Barad, V., 263, 269Barch, D.M., 77, 78, 80, 82, 92, 93, 94,

101, 103, 105, 239, 243, 258, 260, 268Barnes, L.L., 29, 46Barrouillet, P., xiv, xvi, 24, 45, 124, 125, 128, 131,

153, 156, 158, 159Barto, A.G., 168, 177, 191Bates, E.A., 163, 191Batista, A.P., 219, 223Baum, S., 281, 299, 301Baumeister, R.F., 242, 242n1, 247

Bava, S., 259, 269Bayliss, D.M., xiv, xvi, 59, 71, 113, 129, 131, 134,

140, 141, 144, 145, 147f, 148, 149, 151, 156, 157Bayliss, G.C., 36, 42, 43, 48Beauducel, A., 52, 53, 66, 69, 72, 74Becker, J.T., 251, 256, 259, 270Beier, M.E., 49, 55, 57, 61, 67, 68, 70, 218, 221Bell, M.A., 163, 190Bellugi, U., 137, 161Belopolsky, A.V., 263, 269Benishay, D., 242n1, 248Bennett, F., 5, 15Bennett, P.J., 261, 270Bennetto, L., 138, 157Benson, D., 163, 193Benton, A.L., 185, 190Berger, J.S., 78, 105, 235, 248, 254, 259, 265, 270, 271Berish, D.E., 60, 74Berkefeld, T., 261, 269Bernadin, S., xiv, xvi, 24, 45, 124, 125, 128, 131, 153, 156Berndt, R., 288, 299Berninger, V., 149, 152, 160Berridge, C.W., 239, 243, 245

Berry, E.M., 238, 247Bibi, U., 240, 246Bindra, D., 9, 15Binet, A., 5, 15Bishop, D.V.M., 135, 150, 157Bjorklund, D.F., 195, 221, 229, 230, 245Blankenship, A.B., 6, 15Bleckley, M.K., 33, 36, 42, 43, 45, 46, 60, 62, 63, 72,

200, 222, 228, 229, 236, 246Blennerhassett, A., 114, 115, 132Bodenhausen, G.V., 236, 243Boivin, D.B., 230, 244

Bolten, T.L., 5, 15Boone, K.B., 51, 75Booth, J.R., 170, 193Botvinick, M.M., 77, 78, 82, 103, 260, 268Bouchard, T.J., 230, 246Bower, G.H., 227, 228, 233, 234, 235, 243Bowles, R.P., 234, 243

Boyle, M.O., 49, 55, 57, 61, 67, 68, 70, 218, 221Brainerd, E.J., 195, 221Brandimonte, M.A., 111, 133Braver, T.S., xiv, xvi, 60, 72, 76, 77, 78, 79, 80, 82, 86, 87f,

88, 89, 92, 93, 94, 95, 96, 97, 97f, 99, 101, 103, 104,

105, 138, 157, 163, 167, 170, 178f–179f, 179, 180,181f–183f, 184, 185, 190, 192, 193, 239, 243, 258,259, 260, 265, 268, 269, 271

Brizzolara, D., 137, 161Broadbent, D.E., 7, 15Bronik, M.D., 129, 132, 137, 145, 158, 197, 207, 208,

217, 222Brooks, L.R., 197, 221Brown, G.D.A., 57, 71Brown, J., 7, 15Brown, J.W., 82, 88, 103, 105Brown, R.G., 163, 169, 190

Brownell, H., 282, 302Brush, L.N., 235, 243Buchanan, M., 111, 131, 136, 154, 156Buckner, R.L., 96, 105, 238, 246Budde, M., 261, 268Bull, R., 148, 157Bundesen, C., 37, 47Bunting, M.F., xiv, xvi, 24, 27f, 28, 32, 42, 43, 45, 49, 55,

60, 61, 68, 70n2, 71, 134, 135, 136, 140, 145, 148, 157,185, 190, 235, 242n1, 243, 244

Burgess, G.C., 76, 89, 92, 96, 97, 97f, 99, 101, 103, 104Burgess, N., 51, 57, 71Burgess, P., 59, 74, 237, 248Burt, C., 5, 15Busby, R.S., 51, 73Butler, K.M., 33, 45, 228, 236, 243Butler, R.W., 185, 190Buxton, R.B., 257, 268Byrne, R.M.J., 63, 72

Cabeza, R., 93, 103, 228, 238, 244, 261, 268Cacciopo, J.T., xiv, xvii, 3, 8, 16Callender, G., 187, 190

Cameli, L., 229, 244Camos, V., xiv, xvi, 24, 45, 124, 125, 128, 131, 153,156, 159

Cannon, T.D., 259, 269Canter, D., 52, 71Cantor, J., 30, 46, 123, 140, 157, 235, 244Caplan, D., xiv, xvi, 139, 161, 218, 221, 272, 273, 274,

278, 279, 281, 282, 285, 286, 287, 288, 291, 292, 299,300, 301, 302

Carlesimo, G.A., 137, 161Carlson, M.C., 228, 232, 244Carpenter, P.A., xiii, xiv, xvi, 7, 8, 13, 16, 22, 23, 29, 46,

54, 64, 71, 110, 112, 115, 119, 132, 134, 136, 139, 157,198, 217, 221, 227, 231, 234, 244, 246, 273, 277, 278,281, 286, 299, 300

Carroll, J.B., 69, 71Carson, S.H., 238, 244Carstensen, L.L., 238, 244Carter, C.S., 41, 47, 77, 78, 82, 92, 93, 103, 260, 268

304 Author Index



Carullo, J.J., 30, 46, 123, 235, 244Case, R., xiii, xvi, 7, 13, 15, 24, 45, 112, 113, 113f, 114,

115, 118, 119, 131, 131n1, 135, 138, 139, 145, 157Casey, B.J., 170, 184, 190Cattell, J.M., 5, 15

Cepeda, N.J., 60, 71, 263, 269Cerella, J., 194, 196, 206, 221, 223Cermakian, N., 230, 244Chabris, C.F., 60, 72, 89, 92, 97, 104, 231, 232, 247,

265, 269Chafee, M.V., 253, 268Changeux, J.P., 173, 190Chapman, J.P., 206, 221Chapman, L.J., 206, 221Chater, N., 299, 299n2Chein, J.M., 251, 256, 259, 270Chen, S., 239, 243

Chetwynd, A., 60, 74, 219, 223Chiappe, P., 229, 234, 244Christal, R.E., 22, 47, 49, 55, 61, 69, 73, 134, 138, 140,

141, 144, 156, 159Christiansen, M.H., 30, 47, 297, 299, 299n2, 301Christoff, K., 51, 71Chuah, Y.M.L., 137, 146, 150, 154, 157Clifford, E., 231, 232, 247Cohen, J., 239, 243Cohen, J.D., xiv, xvi, 41, 47, 51, 75, 77, 78, 79, 80, 88, 92,

93, 103, 105, 138, 157, 163, 167, 170, 171, 178f–179f,179, 180, 181f–183f, 183, 184, 185, 190, 192, 193, 240,247, 258, 260, 268

Cohen, M.S., 259, 269Cohen, N.J., 263, 269Colcombe, S., 238, 248Coles, M.G.H., 260, 269Collins, D.L., 217, 223Collins, P.F., 95, 104Colom, R., 49, 71Connelly, S.L., 228, 232, 244Connor, C., 165, 188, 193Conrad, R., 7, 15, 136, 157

Constantinidis, C., 165, 188, 193, 255, 268Contreras-Vidal, J.L., 168, 177, 190Conway, A.R.A., xiv, xvi, 3, 21, 22, 24, 25, 26, 26f, 27f, 28,

30, 31, 32, 33, 34, 42, 43, 45, 46, 47, 49, 55, 57, 60, 61,62, 63, 68, 70n2, 71, 72, 134, 135, 136, 137, 139, 140,144, 145, 148, 153, 156, 157, 158, 195, 200, 214, 221,222, 228, 229, 236, 238, 242n1, 244, 246, 251, 269

Coon, V.E., 292, 301Cooney, J.W., 264, 269Copeland, D.E., 126, 132Corbetta, M., 218, 219, 221, 251, 256, 268Corley, R.P., 60, 72

Cornoldi, C., 60, 61, 71, 137, 157, 217, 221Courtney, S.M., 253, 268Cowan, N., xiii, xiv, xvi, 6, 15, 22, 24, 27f, 28, 32, 38, 39,

41, 42, 45, 46, 48, 49, 50, 51, 52, 55, 60, 61, 68, 70n1,70n2, 71, 83, 103, 121, 123, 124, 126, 127, 132, 133,134, 135, 136, 140, 145, 148, 154, 157, 242n1, 244, 251,253, 268

Cowan, W.B., 61, 73Craik, F.I.M., 24, 48, 228, 230, 238, 244, 245, 249, 260,

268, 269, 278, 280, 300Cronbach, L.J., xiv, xvi, 3, 8, 15, 22, 46Crowder, R.G., 6, 16, 21, 40, 46

Crutchfield, J.M., 36, 45Csikszentmihalyi, M., 231, 244Curiel, J.M., 248Curran, T.E., 206, 221Curtis, C., 89, 104, 235, 248

Dalla Vecchia, R., 137, 157Daneman, M., xiii, xiv, xvi, 7, 8, 13, 16, 22, 23, 29, 46, 54,

57, 71, 110, 112, 114, 115, 119, 132, 134, 136, 139, 157,198, 217, 221, 233, 234, 244, 277, 278, 300

Daselaar, S.M., 261, 268Datta, A.K., 236, 247

Davelaar, E.J., 39, 46, 123, 132David-Jurgens, M., 261, 269Davidson, B.J., 83, 105Davidson, N.S., 251, 270Davidson, R.J., xiv, xvii, 3, 8, 16, 95, 103Dayan, P., 79, 84, 105, 168, 177, 179, 193De Beni, R., 60, 61, 71De Jong, R., 33, 37, 46De Menezes Santos, M., 51, 75DeDe, G., 218, 221, 272, 285, 300Deese, J., 228, 244DeFries, J.C., 60, 72Dehaene, S., 173, 190, 251, 270Dehaut, F., 288, 299Delaney, P., 14, 16, 126, 127, 129, 132, 296, 300Delgado, M.R., 251, 256, 259, 270DeLong, M.R., 167, 190Dempster, F.N., 151, 157, 195, 196, 221, 230, 234, 244Denburg, N.L., 229, 238, 247DePisapia, N., 82, 104Depue, R.A., 95, 104Dermen, D., 118, 119, 132Desgranges, B., 229, 245

Desimone, R., 163, 165, 185, 188, 192, 254, 255, 270Desmond, J.E., 259, 265, 271DeSoto, C., 51, 71D’Esposito, M., 78, 88, 91, 104, 105, 187, 192, 200, 223,

228, 235, 248, 250, 254, 255, 259, 261, 264, 265, 269,270, 271

Dewhurst, S.A., 111, 133Di Pellegrino, G., 255, 269Diamond, A., 101, 104, 173, 176, 180, 184, 186, 187, 190DiGirolamo, G.J., 263, 269Ding, Y.S., 239, 249Dinse, H.R., 261, 269

Dixon, P., 139, 158Do, N., 134, 140, 159Dobbs, A.R., 260, 269Dodd, M.D., 240, 246Dolcos, F., 228, 244Dominey, P.F., 180, 190Donaldson, D.I., 88, 96, 103, 105

Author Index 305



Donchin, E., 260, 269Dorfman, J., 51, 71Drain, M., 29, 46Drevets, W.C., 97, 104Driver, J., 35, 47

Druin, D., 187, 190Druzgal, T.J., 250, 254, 269, 270Duchek, J.M., 230, 232, 242, 244Dueker, G.L., 230, 246DuFault, C., 194Duff, S.C., 136, 158Dulcos, F., 261, 268Dunbar, K., 183, 185, 190Duncan, J., 60, 66, 71, 78, 89, 104, 183, 190, 265, 269Durso, F.T., 36, 45Durstewitz, D., 168, 190Durston, S., 170, 184, 190

D’Ydewalle, G., 56, 64, 74Dykes, M., 242n1, 244Dywan, J., 232, 244

Earles, J.L., 229, 233, 247Ebbinghaus, H., 4–5, 16Edgar, D.D., 239, 244Egan, M.F., 101, 105Egeth, H.E., 36, 45Einstein, G.O., 83, 104Ekstrand, B.R., 43, 48Ekstrom, R.B., 118, 119, 132Elliott, E.M., 121, 123, 124, 126, 127, 132, 140, 145, 157Elman, J.L., 163, 191Emerson, M.J., xiv, xvii, 60, 73, 112, 132, 139, 144, 158,

229, 230, 235, 238, 247, 292, 301Emery, L., 194, 195, 199, 202, 203, 205, 209, 210, 211,

217, 221, 223Emslie, H., 66, 71, 78, 89, 104Engle, R.W., xiv, xvi, 8, 16, 17, 21, 22, 23, 23f, 24, 25, 26,

26f, 27, 28, 28f, 30, 31, 32, 33, 34, 35, 36, 38, 40, 41, 42,43, 44–45, 45, 45n2, 46, 47, 48, 49, 50, 57, 58, 60, 62,63, 64, 71, 72, 74, 75, 88, 89, 90, 105, 123, 134, 136,

137, 139, 140, 141, 144, 148, 150, 152, 153, 156, 157,158, 159, 160, 195, 196, 200, 214, 217, 218, 219, 221,222, 223, 228, 229, 234, 235, 236, 237, 238, 244, 246,251, 253, 269, 297, 300, n1

Erickson, C.A., 163, 185, 188, 192Ericsson, K.A., xiv, xvi, 14, 16, 126, 127, 129, 130, 132,

296, 300Eriksen, C.W., 9, 16, 36, 46Erkinjuntt, T., 286, 300Eustache, F., 229, 245Evans, A.C., 217, 223Evans, K., 288, 300

Eysenck, H.J., 242n1, 245

Farah, M.J., 170, 184, 187, 191, 192Farde, L., 239, 246Farrand, P., 137, 150, 154, 159Farrell, S., 58, 72Faust, M.E., 229, 230, 234, 245

Fellous, J.M., 168, 191Fera, F., 101, 105Fernandes, M.A., 228, 245Ferraro, F.R., 206, 223Ferreira, F., 274, 293, 299n1, 300

Fier, C.D., 281, 300Fiez, J.A., 9, 16Fiez, J.T., 251, 256, 259, 270Filion, D.L., 242n1, 247Finkel, L., 275, 299Fiss, W.H., 23, 44n1, 47Fletcher, P.C., 86, 104Flores-Mendoza, C., 49, 71Fodor, J., 293, 300Fodor, J.A., 129, 132Foong, N., 230, 247Foote, S.L., 239, 243, 245, 248

Forman, S.D., 258, 268Fossella, J.A., 170, 184, 190Fowler, J.S., 239, 249Fox, N.A., 163, 190Frackowiak, R.S.J., 86, 88, 104, 105, 256, 270Frank, L.R., 257, 268Frank, M.J., 98, 105, 163, 167, 168, 177, 179, 191, 192Freedman, D.J., 180, 191, 192Freer, C., 60, 71, 78, 89, 104, 265, 269French, J.W., 118, 119, 132Friedman, N.P., xiv, xvii, 22, 30, 46, 47, 58, 59, 60, 65, 72,

73, 145, 159, 184, 185, 191, 214, 222, 229, 230, 235,238, 240, 245, 247, 292, 301

Frieske, D., 229, 233, 247Friston, K.J., 86, 104Frith, C.D., 86, 104, 256, 270Frith, U., 138, 158Fry, A.F., xiv, xvi, 129, 132, 135, 137, 138, 145, 147, 158,

159, 194, 196, 197, 206, 207, 208, 209, 212, 213, 214,217, 218, 221, 222

Frye, D., 171, 175, 193Furey, M.L., 176, 191Fuster, J., 78, 88, 104, 163, 191

Gabrieli, J.D.E., 259, 265, 271Gaines, C., 229, 233, 247Galanter, E., 6, 16Galton, F., 5, 15, 16Garavan, H., 219, 221Garner, W.R., 9, 16Garrard, P., 164, 192Gathercole, S.E., 22, 45, 115, 131, 137, 146, 149, 156,

158, 160, 199, 207, 221Gauthier, I., 176, 193Gavens, N., 153, 158

Gazzaley, A., 264, 269Gehring, W.J., 260, 269Geisler, M.W., 239, 245Gelade, G., 35, 48Gentner, D., 64, 72Georgieff, N., 180, 191Gerard, L., 228, 245

306 Author Index



Gernsbacher, M.A., 229, 230, 234, 245Gerstman, L.J., 281, 300Geva, A., 263, 271Ghaffar, O., 87, 105Gibson, E., 273, 275, 300

Gick, M.L., 260, 269Giffard, B., 229, 245Gilbert, S.J., 59, 72Glahn, D.C., 259, 269Glass, J., 29, 46Glick, M.L., 181, 186, 191Glover, G.H., 259, 271Gmeindl, L., 257, 268Gobbini, M.I., 176, 191Godde, B., 261, 269Gold, D.P., 237, 243Goldberg, J., xiii, xvi, 7, 15, 24, 45, 113, 113f, 114, 115,

118, 119, 131, 131n1, 135, 138, 139, 145, 157Goldberg, T.E., 101, 105Goldman, P.S., 163, 191Goldman-Rakic, P.S., 169, 180, 184, 190, 191, 193, 250,

253, 259, 268, 269, 270Goldstein, D., 230, 233, 235, 243n3, 245, 246, 249Goldstein, J.H., 235, 248Gonzalez de Sather, J.C.M., 60, 71Gore, J.C., 259, 270Goschke, T., 84, 104Goss, B., 260, 269Govoni, R., 228, 244Grady, C.L., 238, 245, 261, 270Grant, P., 153, 158Grasby, P.M., 86, 104Gray, J.A., 94, 104Gray, J.R., 60, 72, 76, 89, 92, 95, 96, 97, 97f, 101, 103,

104, 265, 269Greenberg, R., 235, 245Grigorenko, E.L., 10, 17Grober, E., 286, 300Grossman, M., 286, 300Gunn, D.M., xiv, xvi, 59, 71, 113, 129, 131, 140, 141, 144,

145, 147f, 148, 151, 156, 157Gunning-Dixon, F.M., 263, 269Gunturkun, O., 168, 191Gur, R.C., 239, 249Gutchess, A.H., 262, 269

Haarmann, H.J., 39, 46, 51, 75, 123, 132Haendiges, A.N., 288, 299Hager, L.D., 184, 193Hahn, C., 243n3, 245Hake, H.W., 9, 16Hale, S., xiv, xvi, 100, 105, 129, 132, 135, 137, 138, 140,

145, 147, 158, 159, 194, 195, 196, 197, 199, 200, 202,203, 205, 206, 207, 208, 209, 210, 211, 212, 213, 214,216, 217, 218, 221, 222, 223

Halford, G.S., 50, 51, 56, 72, 153, 158Hall, L.K., 136, 159Halldin, C., 239, 246Halliday, M.S., 110, 111, 132

Hambrick, D.Z., 8, 16, 21, 22, 24, 25, 27, 28f, 38, 43,45n2, 47, 49, 57, 58, 72, 140, 141, 159

Hamilton, G., 140, 157Hamilton, Z., 121, 122, 123, 124, 126, 127, 128, 132, 133,

140, 157

Hamm, V.P., 229, 231, 233, 245Handel, S., 51, 71Happe, F., 138, 158Hariri, A.R., 101, 105Harman, H.H., 118, 119, 132Harnishfeger, K.K., 195, 221, 229, 230, 245Hartley, A., xiv, xvii, 238, 248, 261, 262f, 264, 271Hartman, M., 234, 245Hasbroucq, T., 62, 72Hasher, L., xiv, xvii, 24, 32, 43, 46, 47, 53, 60, 72, 73, 156,

159, 195, 208, 222, 223, 224, 227, 228, 229, 230, 231,232, 233, 234, 235, 236, 237, 238, 240, 242, 243, 243n3,

244, 245, 246, 247, 248, 249, 264Haxby, J.V., 176, 191, 253, 268Hayes, J.R., 64, 73Healy, A., 10, 16Hebb, D.O., 6, 16Hecht, S.A., 117, 133Hedden, T., 251, 262, 269, 270Hegarty, M., xiv, xvii, 10, 16, 22, 47, 58, 59, 65,

73, 135, 145, 158, 159, 214, 222Heitz, R.P., 44–45, 48, n1Henderson, J.M., 33, 45, 228, 236, 243, 274,

299n1, 300Hendry, L., 136, 150, 160Henry, L., 114, 115, 132, 138, 160Henson, R.N.A., 51, 57, 72Hermer, L., 175, 191Hermer-Vasquez, L., 175, 191Hernandez, L., 230, 239, 240, 248Heron, C., 184, 193Hewes, A.K., 135, 137, 154, 158, 159Hewitt, J.K., 60, 72Higgins, D.M., 238, 244Higuchi, S., 239, 245

Hildebrandt, N., 274, 288, 300, 302Hill, J.M., 185, 190Hismajatullina, A., 145, 157Hitch, G.J., xiii, xiv, xvi, xvii, 7, 8, 11, 15, 21, 22, 24,

46, 51, 57, 71, 109, 110, 111, 116, 117, 118, 119, 119f,120, 121, 122, 123, 124, 125, 126, 127, 128, 129,131, 132, 133, 135, 136, 137, 138, 139, 140, 144, 145,148, 153, 154, 156, 157, 158, 159, 160, 194, 237,243, 250, 268, 272, 293, 299

Hitzeman, R., 239, 249Hochreiter, S., 167, 191Hodges, J.R., 164, 192

Hofman, M.A., 239, 245Hofstadter, M.C., 173, 191Holyoak, K.J., 51, 64, 71, 72, 74, 75, 181, 186, 191Hommel, B., 62, 72Horn, D., 51, 75Horne, J.A., 230, 246Horton, N., 127, 133

Author Index 307



Houk, J.C., 168, 177, 191Houston-Price, C.M.T., 112, 113, 133, 136, 160Howerter, A., xiv, xvii, 60, 73, 229, 230, 235, 238, 247Hoyer, C.M., 88, 103Hoyer, W.J., 196, 222, 232, 248

Hsieh, S., 54, 59, 70Hugdahl, K., xiv, xvii, 3, 8, 16Hughes, E., 286, 300Hull, A.J., 7, 15Hulme, C., 57, 71, 111, 132, 137, 149, 158, 159Hummel, J.E., 64, 72Humphrey, D.G., 60, 73, 219, 222Humpstone, H.J., 5, 16Hunt, E., 112, 133Hunt, R.R., 40, 48Hur, Y., 230, 246Huttenlocher, P.R., 180, 184, 191

Hutton, U., xiv, xvii, 24, 46, 116, 117, 118, 119, 119f, 120,122, 123, 124, 125, 126, 128, 132, 133, 135, 140, 144,145, 153, 154, 156, 158, 160

Huttunen, M., 259, 269

Ishai, A., 176, 191

Jacobs, J., 5, 11, 13, 16Jacobsen, C.F., 253, 255, 269Jacoby, L.L., 81, 86, 87f, 101, 104, 105, 259, 271Jager, A.O., 52, 53, 66, 69, 72James, W., 16Jarrold, C., xiii, xiv, xvi, 3, 59, 71, 113, 129, 131, 134, 135,

137, 138, 140, 141, 144, 145, 147f, 148, 149, 151, 154,156, 157, 158, 159, 160

Jenkins, L., 100, 105, 135, 140, 159, 195, 197, 206, 208,209, 217, 222, 223

Jennings, J.M., 260, 268Jensen, A.R., 26, 42, 46Jessie, K.A., 212, 222Jimenez, R., 231, 232, 247Joanisse, M.F., 170, 191Joel, D., 168, 177, 191

Joerding, J.A., 197, 206, 208, 217, 222Johnson, D.M., 15n2, 16Johnson, M., 51, 73Johnson, M.H., 163, 191Johnson, M.J., 170, 176, 192Johnson, M.K., 43, 46, 228, 248Johnson, R., 60, 71, 78, 89, 104, 265, 269Johnson-Laird, P.N., 56, 63, 64, 72, 74Johnston, R.S., 148, 157Jonas, D., 228, 232, 246Jones, D., 137, 150, 154, 159Jonides, J., xiv, xvi, xvii, 29, 46, 78, 88, 89, 91, 103, 104,

105, 137, 138, 154, 157, 160, 163, 190, 193, 200, 218,219, 221, 223, 230, 238, 239, 240, 247, 248, 250, 253,254, 256, 257, 259, 260, 261, 262f, 263, 264, 265, 268,269, 270, 271

Josephs, O., 88, 105Just, M.A., xiv, xvi, 64, 71, 227, 231, 246, 273, 275, 278,

279, 280, 281, 286, 299, 300

Kahneman, D., 238, 246, 297, 300Kail, R., 26, 42, 46, 136, 159, 194, 196, 213, 222, 232, 246Kan, I.P., 256, 270Kane, M.J., xiv, xvi, 3, 8, 16, 21, 22, 23f, 24, 25, 27, 28f, 31,

32, 33, 34, 35, 38, 40, 41, 42, 43, 45, 45n2, 46, 47, 49,

50, 57, 58, 60, 62, 63, 71, 72, 73, 88, 89, 90, 105, 139,140, 141, 158, 159, 196, 200, 217, 218, 219, 222, 228,229, 234, 235, 236, 237, 246, 247, 297, 300

Kanwisher, N., 176, 191Karat, J., 65, 72Karlsson, P., 239, 246Karmiloff-Smith, A., 163, 191Kastner, S., 256, 257, 270Katsnelson, A.S., 175, 191Katz, S., 256, 268Kaye, J.A., 92, 93, 103Keele, S.W., 59, 73

Kele, M., 168, 191Keller, C.V., 121, 132, 154, 157, 268Keller, T.A., 121, 132, 154, 157, 268Kelley, C.M., 81, 104Kelly, J.E., 96, 105Kemper, S., 281, 282, 286, 300, 301Kempler, D., 287, 299, 300, 301Kempter, T.L., 261, 270Kemtes, K., 218, 221, 282, 285, 300Keppel, G., 7, 16, 235, 246Kerkhof, G.A., 230, 246Kerrouche, N., 229, 245Kesner, R.P., 169, 193Ketelaar, E., 95, 105Keys, B.A., 92, 93, 103Khana, M.M., 36, 45Kieras, D.E., 296, 300Kim, J., 259, 269Kim, S., 228, 230, 231, 246Kimberg, D.Y., 184, 187, 191Kincade, J.M., 218, 219, 221, 251, 256, 268King, J., 275, 278, 279, 280, 300Kintsch, W., xiv, xvi, 10, 16, 130, 132

Kirkham, N., 176, 190Klatsky, R.L., 110, 132Klein, K., 23, 44n1, 47Kliegl, R., xiv, xvii, 154, 156, 159Klotz, W., 61, 73Knowlton, B.J., 51, 75Kocinski, D., 136, 150, 160Koeppe, R.A., xiv, xvi, xvii, 137, 154, 160, 238, 248,

253, 254, 256, 259, 261, 262f, 263, 264, 265, 268, 269,270, 271

Kok, A., 239, 248Kolodny, J., 66, 71

Komori, M., 126, 133Kontiola, P., 286, 300Kornblum, S., 62, 72Kosslyn, S.M., xiv, xvii, 3, 8, 16Kotary, L., 196, 222Kotovsky, K., 64, 73Kramer, A.F., 60, 71, 73, 219, 222, 263, 269

308 Author Index



Kray, J., 60, 73Kroger, J.K., 51, 71Kupfer, D.J., 239, 247Kurland, M.D., xiii, xvi, 7, 15, 24, 45, 113, 113f, 114, 115,

118, 119, 131, 131n1, 135, 138, 139, 145, 157

Kusbit, G.W., 236, 248Kyle, F., 140, 159Kyllonen, P.C., 22, 47, 49, 55, 61, 68, 69, 73, 134, 138,

140, 141, 144, 156, 159

La Pointe, L.B., 136, 150, 159Laaksonen, R., 286, 300Lacey, J.F., 121, 123, 124, 126, 127, 132, 140, 157Lacey, S.C., 230, 239, 240, 248Lakoff, G., 51, 73Lambon-Ralph, M.A., 164, 191Lancel, M., 230, 246

Langacker, R.W., 51, 73Lange, E., 58, 74, 218, 219, 223Larish, J.F., 60, 73, 219, 222Larsen, R.J., 95, 96, 97, 97f, 101, 104, 105Lauber, E.J., 29, 46Laughlin, J.E., 24, 26, 26f, 28, 42, 46, 49, 57, 72, 134, 136,

137, 139, 140, 144, 148, 156, 158, 214, 221, 238, 244,251, 269

Lautenschlager, G., 229, 233, 247, 251, 270Laver, G.D., 229, 246Lawrence, A., 111, 132Lawrence, B.M., 194, 195, 197, 200, 201, 201f, 202, 216,

217, 219, 222Lease, J., 88, 104Leather, C.V., 114, 115, 132Lebiere, C., 50, 70, 180, 190Leeke, T., 135, 158LeFevre, J.-A, 139, 158Leigh, E., 113, 131, 147f, 149, 156, 157Leighton, E.A., 238, 247Lenartowicz, A., 228, 233, 248Lencz, T., 242n1, 248Leonard, G., 217, 223

Leonard, J.S., 282, 301Lepage, M., 87, 105Lepine, R., 156, 159Lesperance, D., 232, 247Leung, H.C., 259, 270Leventhal, A.G., 261, 271Levitt, J.B., 169, 179, 192Lewandowsky, S., 58, 72Lewis, D.A., 169, 179, 184, 192Lewis, G.H., 236, 247Lewis, R.I., 273, 300Lewis, V.J., 111, 131, 145, 156, 199, 221

Li, K.Z.H., 235, 249Li, S., 228, 232, 239, 246, 261, 264, 270Lieberman, K., 111, 131Lieberman, M.D., 96, 105Lien, M.-C, 61, 73Lima, S.D., 195, 196, 206, 216, 222, 223Lindenberger, U., 60, 73, 261, 264, 270

Lisman, J.E., 168, 191Littler, J.E., 111, 132Liu, L.L., 262, 269Liu, Y., 239, 245Locke, H.S., 88, 105

Lockenhoff, C.E., 238, 244Logan, G.D., 37, 47, 60, 61, 73, 219, 222Logan, J.M., 238, 239, 246, 249Logie, R.H., 136, 137, 145, 151, 158, 159, 195, 196, 197,

222, 223Logue, V., 6, 17London, M., 51, 71Lonnqvist, J., 259, 269Loughry, B., 163, 167, 168, 177, 191Lovallo, W.R., xiv, xvii, 3, 8, 16Love, T., 257, 268Loveless, M.K., 282, 301

Luck, S.J., 265, 270Lund, J.S., 169, 179, 192Lustig, C., xiv, xvii, 24, 43, 47, 60, 73, 156, 159, 227, 228,

229, 230, 232, 235, 238, 245, 246, 248, 261, 271Lykken, D.T., 230, 246Lyons, K., 286, 301

MacDonald, A.W., III, 41, 47MacDonald, M.C., 30, 47, 275, 278, 287, 297, 299,

300, 301MacDonald, M.E., 287, 301Mack, A., 231, 232, 246MacLeod, C.M., 240, 246Maeda, A., 239, 245Maisog, J.M., 253, 268Malmo, R., 255, 270Manly, T., 236, 247Manninen, M., 259, 269Marchetti, C., 151, 159Marin, O.S., 286, 301Marsden, C.D., 163, 169, 190Marshuetz, C., xiv, xvi, xvii, 238, 247, 248, 259, 261, 262f,

263, 264, 269, 270, 271

Mattay, V.S., 101, 105Mattox, S., 145, 157Mattson, D., 95, 105May, C.P., xiv, xvii, 24, 43, 47, 53, 60, 72, 73, 156, 159,

227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237,238, 240, 242, 245, 246, 247, 248, 249

Mayberry, M.T., 134, 137, 140, 146, 150, 153, 154, 157,158, 159

Mayr, U., 59, 73McAuley, E., 116, 132, 138, 158, 263, 269McAvoy, M.P., 96, 105McClelland, J.L., 163, 164, 170, 176, 183, 185, 190,

192, 193McConnell, J., 111, 133McDaniel, M.A., 83, 104McDermott, K.B., 228, 248McDowd, J.M., 242n1, 247McElree, B.D., 81, 104McFarlane, D.K., 274, 299n1, 300

Author Index 309



McGeoch, J.A., 6, 16McGhie, A., 242n1, 244McIntosh, A.R., 261, 270McLaughlin, J., 165, 188, 193McLean, J.F., 148, 159

McNamara, D.S., 30, 47Meck, W.H., 230, 246Meiran, N., 60, 73Melton, A.W., 6, 16, 40, 47Merikle, P.M., xiv, xvi, 22, 46, 57, 71, 134, 157,

233, 244Meyer, D.E., 260, 269, 296, 300Mickanin, J., 286, 300Miezin, F.M., 96, 105Mikels, J.A., 238, 247, 263, 271Milham, M.P., 263, 269Miller, A., 261, 262f, 263, 271

Miller, E.K., 163, 165, 180, 184, 185, 188, 191, 192, 193,240, 247, 254, 255, 270

Miller, G.A., 6, 7, 8, 11, 15n3, 16, 41, 47Miller, M.B., 206, 221Milner, B., 6, 16, 137, 159Minear, M., 261, 262, 264, 269, 270Minkoff, S., xiv, xvi, 24, 27f, 28, 42, 45, 49, 55, 61, 68,

70n2, 71, 134, 135, 136, 140, 145, 148, 157Minoshima, S., 253, 254, 256, 270, 271Mintum, M.A., 253, 256, 270Mishkin, F.S., 51, 75Mitchum, C.C., 288, 299Miyake, A., xiii, xiv, xv, xvii, 3, 4, 8, 10, 13, 16, 17, 22, 24,

27, 30, 46, 47, 48, 58, 59, 60, 65, 68, 72, 73, 74, 112,118, 123, 124, 128, 132, 133, 134, 135, 139, 140, 141,144, 145, 158, 159, 160, 164, 184, 185, 191, 192, 214,217, 222, 223, 229, 230, 235, 238, 240, 245, 247, 281,292, 299, 301

Mizumori, S.J.Y., 169, 193Moberg, P.J., 239, 249Monheit, M.A., 170, 191Monk, T.H., 239, 247Monsell, S., 35, 47, 54, 59, 74, 219, 223

Montague, P.R., 79, 84, 105, 168, 177, 179, 193Montgomery, J.W., 137, 159Moreno, M.V., 121, 123, 124, 126, 127, 132, 140, 157Morey, C., 145, 157Morris, J.C., 238, 246Morris, N., 137, 150, 154, 159Morris, R.G., 260, 269, 285, 301Morton, J.B., 162, 171, 176, 180, 184, 192Moscovitch, M., 50, 73, 196, 208, 223, 228, 238,

245, 247Most, S.B., 231, 232, 247Motohashi, Y., 239, 245

Mozer, M., 164, 192Mueller, S., 296, 300Muir, C., 111, 132Multhaup, K.S., 234, 247Munakata, Y., xiv, xvii, 162, 163, 164, 170, 171, 173, 174,

174f, 176, 180, 184, 185, 192, 193Muraven, M., 242, 242n1, 247

Murphy, K.J., 230, 249Murphy, T.D., 36, 46Murphy, W.E., 232, 244Muter, V., 149, 159Myerson, J., 100, 105, 135, 137, 140, 158, 159, 194,

195, 196, 197, 199, 200, 201, 201f, 202, 203,205, 206, 207, 208, 209, 210, 211, 216, 217, 219, 221,222, 223

Nairne, J.S., 6, 16, 39, 40–41, 45n3, 47, 199, 218, 223Naveh-Benjamin, M., 228, 244, 256, 270Navon, D., 231, 247Nee, D.E., 91, 104Neely, J.H., 32, 45Nelson, J.K., 239, 247, 260, 270Neumann, O., 61, 73Newell, A., 63, 73, 83, 105, 180, 192

Nichols, T.E., 230, 239, 240, 248Nieder, A., 180, 192Nigg, J.T., 229, 230, 247Nishizaki, Y., 126, 133Niv, Y., 168, 177, 191Noelle, D., 170, 178f–179f, 179, 180, 181f–183f, 193Noll, D.C., xiv, xvi, 78, 88, 103, 138, 157, 163, 190,

258, 268Norman, D.A., 7, 17, 23, 47, 59, 73, 78, 79, 105Norris, D., 51, 58, 72, 74Nugent, L.D., 121, 132, 154, 157, 268Nyberg, L., 87, 105, 228, 244, 261, 268Nyberg, S.E., 235, 249Nystrom, L.E., xiv, xvi, 78, 88, 103, 138, 157, 163, 190,

258, 268

Oakhill, J., 140, 159Oberauer, K., xiv, xvii, 49, 50, 51, 52, 53, 54, 55, 55f, 56f,

57, 58, 61, 62, 65, 66, 67, 67t, 68, 69, 73, 74, 75, 136,137, 140, 143, 144, 154, 156, 159, 160, 206, 218,219, 223

O’Hare, A.W., 153, 158O’Kane, G., 232, 249

Olivieri, A., 291, 299Onishi, K., 286, 300Oonk, H.M., 197, 200, 201f, 217, 219, 222O’Reilly, R.C., 51, 73, 77, 78, 98, 105, 162, 163, 164, 167,

168, 170, 177, 178f–179f, 179, 180, 181f–183f, 184, 185,190, 191, 192, 193

Osaka, M., 126, 133Osaka, N., 126, 133Osherson, D., 228, 248Oshinsky, M.L., 239, 243Osman, A.M., 62, 72Ostberg, O., 230, 246

Ozonoff, S., 138, 160

Page, M.P.A., 51, 58, 72, 74Palladino, P., 60, 61, 71Papagno, C., 22, 45, 137, 161Pardo, J., 95, 105Parisi, D., 163, 191

310 Author Index



Park, D.C., 229, 233, 238, 247, 251, 261, 262, 264,269, 270

Park, R., 261, 264, 270Park, Y.-S, 213, 222Pascual-Leone, J., 7, 13, 17

Passingham, R.E., 88, 105, 255, 271Patalano, A.L., 29, 46Patterson, K., 164, 192Paulesu, E., 256, 270Paus, T., 217, 223Paxton, J.L., 93, 94, 101, 105Payne, T.W., 8, 16, 27, 28f, 38, 43, 45n2, 47, 57, 58, 72,

140, 141, 159Pazzaglia, P., 60, 61, 71Peacock, K., 122, 124, 128, 133Peaker, S.M., 137, 160Pearlmutter, N.J., 275, 278, 301

Pearson, D.G., 39, 47Pellegrino, J.W., 112, 133Pennington, B., xiv, xvii, 10, 16, 138, 157, 160Penpeci, C., 261, 270Persad, C.C., 229, 238, 247Perstein, W.M., 78, 88, 103Peters, A., 217, 223Petersen, S.E., 251, 270Peterson, J.B., 238, 244Peterson, L.R., 7, 17Peterson, M.J., 7, 17Petit, L., 253, 268Petrides, M., 163, 192, 259, 270Pezzini, G., 137, 161Phillips, C., 135, 137, 158, 159Phillips, L.H., 60, 75Phillips, N.A., 229, 232, 244, 247Phillips, S., 50, 51, 56, 72Piaget, J., 173, 192Pickering, S.J., 137, 160Pietrini, P., 176, 191Pike, B., 217, 223Pineda, J.A., 239, 245, 248

Piolino, P., 229, 245Plaut, C.C., 170, 192Plude, D.J., 232, 248Plunkett, K., 163, 191Poggio, T., 180, 191Polich, J., 239, 245, 248Polk, T.A., 238, 247, 261, 264, 270Poole, B., 35, 47Posner, M.I., 83, 105, 153, 160, 251, 270Postle, B.R., 78, 91, 104, 105, 200, 223, 235, 243, 248, 250,

254, 255, 259, 269, 270Postman, L., 235, 246

Poutanen, V.P., 259, 269Prabhakaran, V., 51, 71, 259, 265, 271Prather, P., 275, 282, 299, 302Pratt, R.T.C., 6, 17Preece, T., 57, 71Prevor, M., 187, 190Pribram, K.H., 6, 16

Prince, S.E., 228, 244, 261, 268Proctor, R.W., 61, 73Pu, M., 261, 271

Quinn, J.G., 111, 133

Rabbitt, P.M.A., 60, 74, 219, 223Racine, C.A., 93, 94, 101, 103, 105Radvansky, G.A., 126, 132, 208, 224, 228, 245, 248, 249Raffone, A., 51, 74Ragozzino, K.E., 169, 192Ragozzino, M.F., 169, 192Rahhal, T., 228, 232, 238, 246, 247, 248Raichle, M.E., 95, 96, 97, 104Raine, A., 242n1, 248Rajah, M.N., 261, 270Rajkowski, J., 239, 243

Ranganath, C., 228, 248Ransdell, S., 117, 133Rao, S.G., 169, 193Rapus, T., 171, 175, 193Rathunde, K., 231, 244Rauch, S., 291, 301Ravizza, S.M., 251, 256, 259, 270Raz, N., 238, 239, 248, 262, 263, 269, 270Reading, S., 257, 271Rebollo, I., 49, 71Reder, L., 236, 248Reitman, J., 7, 17Rettinger, D.A., xiv, xvii, 22, 47, 58, 59, 65, 73, 112, 132,

139, 144, 145, 158, 159, 214, 222Reuter-Lorenz, P.A., xiv, xvi, xvii, 29, 46, 238, 239, 247,

248, 250, 257, 259, 260, 261, 262f, 263, 264, 268, 269,270, 271

Reynolds, J.R., 88, 103, 105Reznick, J.S., 173, 191Rhee, S.H., 137, 158, 195, 197, 199, 200, 203, 206, 207,

208, 217, 222, 223Rice, H.J., 228, 244Richardson, J.T., 195, 223

Riesenhuber, M., 180, 191Rissman, J., 264, 269Robbins, T.W., 92, 103Roberts, M.J., 60, 75Roberts, R.J., xiv, xvii, 184, 193Robertson, I.H., 236, 247Robin, N., 51, 74Rochon, E., 286, 287, 301, 302Rock, I., 231, 232, 246Roediger, H.L., III, 228, 248Rogers, R.D., 35, 47, 54, 59, 74, 219, 223Rogers, S.J., 138, 157

Rogers, T.T., 164, 192Rokeach, M., 238, 248Roodenrys, S., 137, 158Rorsman, I., 185, 190Rose, R., 3, 8, 16Rosen, V.M., 31, 32, 34, 42, 47, 60, 74Rosenthal, R., 96, 105

Author Index 311



Rossman, E., 153, 160Rosvold, H.E., 163, 191Rougier, N.P., 170, 178f–179f, 179, 180, 181f–183f, 193Rowe, G., 228, 233, 235, 248Rowe, J.B., 88, 105, 255, 271

Roy, J.A., 148, 157Rubin, D.C., 239, 248Ruffman, T., 173, 190Rule, B.G., 260, 269Rumelhart, D.E., 163, 193Ruppin, E., 168, 177, 191Rush, B.K., 93, 103Russell, J., 138, 159, 160Ryan, E.B., 115, 116, 133, 138, 160Ryan, S., 282, 301Rypma, B., 259, 261, 265, 271

Saffran, E.M., 286, 301Sahni, S.D., 174, 174f, 185, 193Saito, S., xiv, xvii, 118, 123, 124, 128, 133Sakai, K., 255, 271Salthouse, T.A., 10, 16, 24, 26, 42, 46, 48, 54, 55f, 58, 60,

61, 74, 100, 105, 138, 141, 143, 145, 160, 194, 196, 198,218, 221, 222, 223, 224, 229, 232, 233, 234, 243, 248,260, 268, 278, 280, 292, 297, 301

Sander, N., 49Sanders, A.L., 238, 246Satpute, A.B., 93, 103Saults, J.S., 121, 123, 124, 126, 127, 132, 140, 145, 157Savage, A., 261, 264, 270Scarpa, A., 242n1, 248Schacher, S., 239, 248Schaefer, A., 96, 97, 97f, 101, 104Schaeken, W., 56, 64, 74Scheier, I.H., 9, 15Schmidhuber, J., 167, 191Schmolesky, M.T., 261, 271Scholey, K.A., 137, 154, 160Scholl, B.J., 231, 232, 247Schouten, J.L., 176, 191

Schrock, J.C., 44–45, 48, 60, 75, n1Schulkind, M.D., 61, 73Schultz, W., 79, 84, 105, 168, 177, 179, 190, 193Schulze, R., 49, 54, 58, 66, 67, 67t, 68, 69, 74, 75, 136,

137, 159, 160, 218, 223Schumacher, E.H., 29, 46, 254, 256, 268, 271Schwartz, M.F., 286, 301Schweickert, R., 205, 223Scott, J.L., 30, 47Seamans, J., 168, 191Seidenberg, M., 163, 170, 191, 193, 275, 278, 301Sejnowski, T.J., 168, 191

Sekuler, A.B., 261, 270Sekuler, R., 261, 270Servan-Schreiber, D., 171, 184, 185, 190Seymour, T., 296, 300Shah, P., xiii, xiv, xv, xvii, 4, 8, 10, 16, 17, 22, 24, 27, 47,

48, 58, 59, 65, 68, 73, 74, 134, 135, 139, 140, 141, 145,158, 159, 160, 164, 192, 214, 217, 222, 223

Shallice, T., 6, 17, 23, 47, 59, 72, 73, 74, 78, 79, 105,237, 248

Shaw, R.J., 54, 55f, 58, 74Sheard, E.D., 240, 246Shell, P., 64, 71

Shiffrin, R.M., 7, 15, 21, 45Shilling, V.M., 60, 74, 219, 223Shin, R.K., 88, 104Shulman, G.L., 218, 219, 221, 251, 256, 268Siegel, L.S., 229, 234, 244Siegel, S., 115, 116, 133, 138, 160Siegler, R., 170, 176, 177, 192, 193Sikstrom, S., 239, 246, 261, 264, 270Simon, H.A., 63, 64, 73Simon, T., 5, 15Simons, D.J., 231, 232, 247Simpson, J., 86, 104

Sliwinski, M., 140, 159, 206, 223Small, J., 286, 301Smith, A.D., 229, 233, 247, 251, 270Smith, E.E., xiv, xvi, xvii, 29, 46, 78, 89, 91, 104, 105, 137,

138, 154, 157, 160, 163, 190, 193, 200, 218, 219, 223,230, 239, 240, 247, 248, 253, 254, 256, 257, 259, 260,261, 262f, 263, 265, 268, 269, 270, 271

Smith, G., 239, 249Smith, L.B., 176, 193Smith, M.R., 261, 264, 270Smith, P.K., 251, 270Smith, R.E., 83, 105Smyth, M.M., 137, 154, 160Snowling, M., 149, 159Snyder, A.Z., 238, 246Snyder, C.R.R., 83, 105Snyder, L.H., 219, 223Soederberg, L.M., 282, 301Soto, R., 51, 73Speer, N.K., 86, 87f, 101, 105, 259, 271Spelke, E., 175, 191Spiegel, D, 3, 8, 16Spiegel, W.R., xiv, xvii

Spooner, A., 149, 156Stanczak, L., 291, 301Standertskjold-Nordenstam, C.G., 259, 269Stedron, J., 174, 174f, 185, 193Steinmetz, M., 165, 188, 193, 255, 268Stenger, V.A., 41, 47Sternberg, R.J., 10, 17, 63, 74Sternberg, S., 7, 17, 86, 105, 258, 271Stevens Dagerman, K., 287, 301Stine-Morrow, E., 282, 301Stoltzfus, E.R., 195, 223, 233, 247Strayer, D.L., 60, 73, 219, 222

Strick, P.L., 167, 190Stromswold, K., 291, 301Stuart, G., 137, 150, 154, 159Stuss, D., 163, 193, 230, 249Styles, E.A., 54, 59, 70Sulkava, R., 286, 300Suri, R.E., 168, 177, 193

312 Author Index



Suß, H.-M, 49, 52, 53, 54, 55, 55f, 56f, 57, 58, 61, 66, 67,67t, 68, 69, 72, 74, 75, 136, 137, 140, 143, 144, 159,160, 206, 218, 223

Swales, M., 60, 71, 265, 269Swanson, H.L., 149, 152, 160

Swick, D., 239, 248Swinney, D., 275, 282, 299, 302Sylvester, C.Y., 230, 239, 240, 247, 248, 260, 270, 271

Tallis, F., 92, 105Talmy, L., 51, 75Tardif, T., 134, 157Tarr, M.J., 176, 193Taylor, S.F., 238, 247Tehan, G., 136, 150, 160Tentori, K., 228, 248Tessitore, A., 101, 105

Thelen, E., 176, 193Therman, S., 259, 269Therriault, D., xiv, xvi, 24, 27f, 28, 42, 45, 49, 55,

61, 68, 70n2, 71, 134, 135, 136, 140, 145,148, 157

Thessing, V.C., 230, 232, 242, 244Thompson-Schill, S.L., 89, 104, 256, 270Thomson, N., 111, 131, 132, 136, 154, 156Thorndike, E.L., 6, 17Tonev, S., 232, 245Toni, I., 88, 105Toth, J.P., 40, 48Towse, J.N., xiii, xiv, xvi, xvii, 3, 24, 46, 109, 112, 113, 116,

117, 118, 119, 119f, 120, 121, 122, 123, 124, 125, 126,127, 128, 129, 132, 133, 135, 136, 139, 140, 144, 145,153, 154, 156, 158, 160, 194

Treisman, A.M., 35, 48Tressoldi, P.E., 137, 157Tucker, L.R., 66, 75Tuholski, S.W., xiv, xvi, 8, 16, 22, 23f, 24, 26, 26f,

27, 28, 28f, 35, 36, 38, 42, 43, 45n2, 46, 47,48, 49, 50, 57, 58, 71, 72, 134, 136, 137, 139,140, 141, 144, 148, 156, 158, 159, 214, 217,

221, 238, 244, 251, 269, 297Tulving, E., 87, 105Tuma, R., 185, 190Tun, P.A., 232, 249Turcotte, J., 228, 235, 248Turley-Ames, K.J., 30, 44n1, 48Turner, M.L., 8, 17, 23, 24, 48, 152, 160Twilley, L.C., 139, 158Tyler, L.K., 287, 299, 300

Underwood, B.J., xiv, xvii, 7, 8, 9, 16, 17, 21, 43, 48, 228,235, 245, 249

Underwood, W.D., 196, 223Ungerleider, L.G., 253, 256, 257, 268, 270Unsworth, N., 38, 44–45n1, 48, 60, 64, 75Usher, M., 39, 46, 51, 75, 123, 132

Valderrama, S., 228, 233, 248 Vallar, G., 111, 131, 137, 145, 156, 161, 199, 221

Van der Stigchel, S., 88, 105 Van Erp, T.G.M., 259, 269 Vecchi, T., 217, 221 Vecera, S.P., 170, 191 Verhaeghen, P., 206, 218, 223, 224

Vicari, S., 137, 161 Visscher, K.M., 96, 105 Vogel, E.K., 265, 270 Volkow, N.D., 239, 249

Wager, T.D., xiv, xvii, 60, 73, 229, 230, 235, 238, 239,240, 247, 248, 254, 257, 271

Waldstein, R., 281, 301Walker, P., 111, 133Wallace, M.A., 170, 191Wallis, J.D., 180, 193Waltz, J.A., 51, 75

Wang, G.J., 239, 249Wang, P.P., 137, 161Wang, X.-J., 168, 191, 193Wang, Y., 261, 271Ward, G., 60, 75Warrington, E.K., 6, 15, 17Washburn, D.A., 200, 224Waters, G.S., xiv, xvi, 139, 161, 218, 221, 272, 273, 274,

278, 279, 281, 282, 285, 286, 287, 288, 291, 292, 299,300, 301, 302

Watkins, M.J., 228, 235, 249Watkins, O.C., 228, 235, 249Watson, P.C., 236, 247Waugh, N.C., 7, 17Webb, A., 263, 269Weeks, P.A., 274, 299n1, 300Wegner, D.M., 229, 230, 249Weinert, D., 239, 249Weiss, C.S., 137, 158, 197, 199, 200, 203, 206, 207, 208,

217, 222Weissman, D.H., 228, 244, 263, 269Welsh, R.C., 262, 269Wenzlaff, R.M., 229, 230, 249

West, R., 208, 224, 230, 238, 249Whalen, S., 231, 244White, D.A., 195, 199, 205, 209, 217, 223Whiteley, H.E., 111, 133Whitfield, M.M., 30, 44n1, 48Whytock, S., 127, 133Wickelgren, W.A., 7, 17Widaman, K.F., 100, 105Wilhelm, O., 8, 16, 27, 28f, 38, 43, 45n2, 47, 49, 53, 54,

55, 55f, 56f, 57, 58, 61, 62, 63, 66, 67, 67t, 68, 69, 72,74, 75, 136, 137, 140, 141, 144, 159, 160, 218, 223

Williams, G.V., 169, 193

Williams, P., 78, 89, 104Willis, C., 115, 131Wilson, D.E., 240, 246Wilson, W.H., 50, 51, 56, 72Wingfield, A., 232, 249, 282, 302Winocur, G., 196, 208, 223, 230, 235, 236, 238, 240,

247, 249

Author Index 313



Wirz-Justice, A., 239, 249Wise, S.P., 255, 269Wittmann, W.W., 49, 53, 54, 55, 55f, 56f, 57, 58, 61, 66,

67, 67f, 67t, 68, 69, 74, 75, 137, 140, 144, 160, 218, 223Witzki, A.H., xiv, xvii, 60, 73, 229, 230, 235, 238, 247

Wolfe, J.M., 36, 48Wolters, G., 51, 74Wong, E.C., 257, 268Wood, N.L., 32, 48, 121, 132, 154, 157, 268Wood, P.K., 121, 132, 154, 157, 268Woodin, M.E., 111, 132Woodworth, R.S., 11, 17Wszalek, T.M., 263, 269Wundt, W., 5–6, 17, 39, 48Wylie, G., 37, 45

Yantis, S., 256, 257, 265, 271

Yarkoni, T., 96, 97, 97f, 101, 104 Yee, P.L., 112, 133 Yerys, B.E., 173, 192 Yoon, C., 230, 233, 235, 236, 237, 242, 249

Yoshioka, T., 169, 179, 192 Young, S.E., 60, 72 Yuasa, T., 239, 245

Zacks, J.L., 232, 249

Zacks, R.T., 32, 33, 45, 46, 53, 60, 72, 195, 208, 222,223, 224, 227, 228, 229, 230, 231, 232, 233,234, 235, 236, 238, 240, 243, 244, 245, 246,247, 249, 264

Zald, D., 95, 105Zarahn, E., 88, 104Zavortink, B., 235, 246Zechmeister, E.G., 235, 249Zelazo, P.D., 171, 175, 193, 243n3, 245Zhao, Z., 51, 71Zheng, Y., 100, 105Zhu, Y., 239, 243

Zijdenbos, A., 217, 223Zucco, G.M., 137, 159, 195, 196, 197, 222Zurif, E., 275, 282, 299, 302Zwahr, M., 229, 233, 247

314 Author Index



Subject Index

Note: Page numbers followed by f and t refer to figures and tables, respectively.

A-not-B task, 173 Abstraction, prefrontal cortex and, 180–81, 183 Access function of inhibition, 231–33, 243n4 Activated representations

in executive attention theory, 40–41in factor model of working memory, 50, 50f

Active maintenanceattentional control and, 256–58computational trade-offs, 164–66, 164f–166f definition, 163developmental variations, 170–75

Affect, control processes and, 94–98, 97f Aging

attentional control, 260–64, 262f cognitive control strategies, 92–94, 94f, 98individual differences in working memory, 214–16, 216f inference generation, 229inhibitory processes, 232, 233–35, 236, 238, 263–64

memory deficits, role of content, 238neural alterations, 261, 262–63processing speed, 195syntactic processing, 281–84, 283f, 284tverbal/visuospatial memory span, 208–12, 209f–211f, 217

Alphabet span task, 280 Alzheimer’s dementia, syntactic processing in, 285–88,

289f, 290t

Analogical reasoning, 64 Anterior cingulate cortex, 78, 96–98, 97f Antisaccade task, 33, 200, 201f, 236 Aphasia, sentence comprehension in, 288, 290, 291f Apperception, 6, 39 Apprehension, span of, 5–6, 15n1, 39 Arousal patterns

inhibitory processes, 232, 233–34, 236, 243n3lifespan differences, 230

Attentionaging and, 260–64, 262f deficits, working memory span and, 115focus of

capacity limitations, 41–42in factor model of working memory, 50, 50f,

51–52historical perspective, 5–6individual differences, 31–34

interference variation, 31–32, 43measuring, 38–39in service of selection and activation, 256–58shifts. See also Task-set switching

visuospatial memory span and, 200–202, 201f in short-term memory span tasks, 38–39, 251–52,

258–60task demands and, 258–60

315



Attention (continued)working memory capacity and, 24–34working memory representations and, 252–56in working memory span tasks, 23–24, 38, 44–45n1

Attentional inhibition. See Inhibition

Atypical cognitive development, 115, 138, 149–50,151–52

Auditory moving windows paradigm, 274, 274f, 275t,299n1

Autism, 138 Automaticity

proactive control and, 84syntactic processing and, 293–94

AX-CPT delayed-response paradigm, 79–80, 81,93–94, 94f

Backward span, 205, 209, 210f

Basal ganglia. See also Prefrontal–basal ganglia workingmemory (PBWM) model

dynamic gating in. See Dynamic gatingin reversal learning, 169in working memory, 98, 163

Behavioral approach system (BAS), 95–98, 97f Behavioral inhibition system (BIS), 95–98, 97f Behavioral research, as complement to computational

modeling, 9–10Berlin Intelligence Structure model, 53Bilateral brain activity, aging and, 264Binding, relational integration and, 50, 51, 56–59Brain imaging. See Neuroimaging studiesBrinley plot, 206Broca’s area, 291Brodmann’s areas, 254f Brunswik lens model, 66–67, 67f Buffer(s)

episodic, 40, 110, 126–27, 129storage, 7, 39–40, 253

Capacity limitationsattentional focus, 41–42

short-term memory, 4–6, 15n3working memoryhistorical perspective, 4–8, 15n3measurement, 38–39. See also Working memory

span tasksnature, 11

Card-sorting task, 171–73, 171f, 188n1Central executive, 11, 23f, 110, 137–38, 197,

250–51, 295–96. See also Executive controlprocesses

Children. See Development of working memoryCingulate cortex, 78, 96–98, 97f

Cognitive aging. See AgingCognitive control processes. See Executive control

processesCombination span task, 112–13Compensation hypothesis, 263–64Complex span tasks. See Working memory span tasksComputational cognitive neuroscience, 162–63

Computational modelsas complement to behavioral research, 9–10dual mechanisms of control theory and, 98–99of serial recall, 58

Computational trade-offs

in active maintenance, 164–66, 164f–166f between proactive and reactive cognitive control,

82–85Concentric model of working memory, 50–52, 50f Confirmatory factor analysis, 146–47, 215, 216f Conflict. See also Interference effects

familiarity, 259–60Content effects, working memory, 121, 217–18, 238Context representations, prefrontal cortex and, 79Control dilemma, 84Control processes. See Executive control processesConverging operations in research, 9–10

Counting span, 7, 24, 113–14, 113f, 115–16, 117,118, 131n1

Deductive reasoning, 63–64Delayed-response paradigm, AX-CPT, 79–80, 81,

93–94, 94f Delayed-response task, 250Deletion function of inhibition, 233–35, 243n5Dementia, syntactic processing in, 285–88, 289f, 290tDevelopment of working memory, 206–8, 207f, 208f

in atypically developing individuals, 115, 138, 149–50,151–52

controlled-attention account, 140domain specificity, 118, 129dual mechanisms of control theory, 100–101as dual-task performance, 112–13executive functions and, 136, 144, 144t, 151, 154, 156executive processing space, 112higher-level cognition and, 148–50, 151–52history of research, 7inhibitory control and, 125–26long-term knowledge representations, 126mathematical skills, 115, 148–49

processing and storage constraints, 142–44, 145–48,147f, 150–51processing and storage requirements, 126–27, 136,

139–41, 142–44processing speed, 118, 128–29, 152–53, 196, 212–14,

213f, 214f qualitative changes, 111quantitative changes, 111–12reading skills, 115, 148–49, 150, 151–52recall timing, 121–22, 127residual variance, 144, 144t, 154, 156resource-sharing account, 112, 116, 124, 139, 125

strength of representations, 170–75task-switching account, 116–20, 119f, 122, 123–25,

139–40, 207, 207f theoretical accounts, 115–20, 117f, 119f, 153–54,

156value of research approach, 135, 175–77verbal/visuospatial information, 206–8, 207f, 208f

316 Subject Index



working memory span task performance, 113–20,113f, 141–48

Developmental disordersreading skills, 149–50, 151–52short-term memory deficits, 137

Dichotic-listening task, 32–33Digit span, age effects, 209, 210f Direct-access region, in factor model of working memory,

50, 51–52, 70n1Distraction effects, age differences, 232–33Domain specificity/generality, 11

computational and developmental approach, 185–87developmental cascade model, 218, 219dual mechanisms of control theory, 100executive attention theory, 26–29, 28f, 45n2short-term memory, 28–29working memory span in children, 118, 129

Dopaminein dynamic gating, 239in inhibition, 239interactions with cortical and subcortical systems, 78,

79, 80, 82, 186–87in learning, 79in phenylketonuria, 186–87

Down syndrome, 137Dual-task performance, children’s working memory as,

112–13Dynamic gating, 167–68, 167f

learning mechanisms, 169–70, 177, 179role of dopamine and norepinephrine, 239

Episodic buffer, 40, 110, 126–27, 129Event-related potentials (ERPs), in inhibition, 239Executive control processes, 11. See also Central executive

affect/personality and, 94–98, 97f aging and, 260–64, 262f attentional. See Attentioncognitive individual differences, 88–92, 91f differentiability, 265examples, 250–51

inhibitory-based. See Inhibitioninterference expectancy and, 89–92, 91f neural dysfunction and, 92–94, 94f in short-term memory span tasks, 38–39, 251–52,

258–60within-individual variation, 85–88, 87f

Executive functionschildren’s working memory and, 136, 144, 144t, 151,

154, 156working memory capacity and, in factor model of

working memory, 59–61, 69Executive processing space, 112

Extraversion, control processes and, 95–98, 97f Eye-movement tasks, 200, 201f

Face processing, 176Facet model

of intelligence, 52of working memory, 53–54

Factor analysis, confirmatory, 146–47, 215, 216f Familiarity conflict, 259–60Fluid intelligence

control processes and, 89–92, 91f predictors, developmental cascade model, 212–16, 213f,

214f, 216f working memory and, 6–7, 25, 26f, 63–69, 67f,

67t, 70n2Forward span, 209, 210f Frontal lobe. See Prefrontal cortexFrontal–parietal executive attention system, 256–57,

264–65Functional magnetic resonance imaging

attention studies, 41inhibitory processes, 239–40

Gating. See Dynamic gating

Genetics, working memory and, 101Goals, experimenter-set, 242–43n2Graded-representations approach, 170‘‘Guided search’’ theory, 36

Higher-level cognition, predictorsin atypically developing individuals, 149–50, 151–52in typically developing individuals, 148–49

Hysteresis, in perceptual processing streams, 255

Individual differences in working memorybenefits of research, xiv, 8–9, 22cognitive control processes and, 88–92, 91f current limitations of research, 4executive attention and, 31–34lifespan changes, 212–16, 213f, 214f, 216f.

See also Aging; Development of working memoryreasoning ability and, 49–75

Inductive reasoning, 63, 64–65Inference generation task, 229Inhibition

access function, 231–33, 243n4age-comparative approach, 229

aging and, 232, 233–35, 236, 238, 263–64biological bases for variation, 238–40children’s working memory span and, 125–26deletion function, 233–35, 243n5dual mechanisms of control theory, 99–100executive attention theory and, 42–43factor model of working memory and, 53, 60–61intra-individual-differences approach, 230restraint function, 235–37time-of-day effects, 232, 233–34, 236, 243n3

Intelligencecrystallized, syntactic processing and, 295

facet model, 52fluid

control processes and, 89–92, 91f predictors, developmental cascade model, 212–16,

213f, 214f, 216f working memory and, 6–7, 8, 25, 26f, 63–69, 67f,

67t, 70n2

Subject Index 317



Interference effectsage differences, 195–96, 263–64causes, 218–19dual-task paradigms, 111verbal and visuospatial tasks, 199, 200, 202, 203,

205, 208working memory capacity and, 23, 31–32, 43, 59,

234–35Interference expectancy, executive control processes and,

89–92, 91f Interference theories, historical perspective, 6Interpretive processing, 273, 293, 295Item recognition task, 258

Language comprehensioninterpretive processing, 273, 293, 295post-interpretive processing, 273, 281

working memory for, 272–302Language impairment, short-term memory deficits, 137Latent variables, working memory and, 25Learning disabilities

reading skills, 149–50, 151–52short-term memory deficits, 137

Letter-number span tasks, 202–5, 204f, 209–11,210f, 211f

Lifespan changes in working memory, 194–224. See also Aging; Development of working memory

children, 206–8, 207f, 208f individual differences approach, 212–16, 213f,

214f, 216f older adults, 208–12, 209f–211f young adults, 197–206, 198f, 201f, 204f

Limb movements, visuospatial memory span and, 200,201f

Listening span, 114Listening studies, subject-extracted versus object-extracted

sentences, 274–78, 274t, 275t–277tLong-term knowledge representations, in children, 126Long-term memory, case studies, 9

M-space, 7Magnetic resonance imaging, functionalattention studies, 41inhibitory processes, 239–40

Manipulation span tasks, 202–6, 204f Mathematical skills, working memory and, 115,

148–49Matrices, Raven’s, 64, 90, 148, 212, 295Medial temporal lobe, connectivity with prefrontal

cortex, 78Memory deficits

aging-related, role of content, 238

short-term, in developmental disorders, 137Memory span task

complex. See Working memory span taskshistorical overview, 5–6

Mental models theory, 63–64Mental rotation task, 145Moses illusion, 236

Multiple-component model of working memory, 39–40,110–11, 136–38, 252–53

Multivariate reliability theory, 66

n-back task, 8, 89, 95–96, 258, 280

Nelson Denny Reading Test, 280Neural dysfunction

cognitive control processes and, 92–94, 94f working memory deficits and, 255

Neural mechanisms, mapping constructs onto, 184–85Neural networks. See also Computational models

objections to concept, 297rehearsal, 256–57storage sites, 254–56

Neuroimaging studiesattentional/executive control processes in older adults,

93, 94f, 260–64, 262f

inhibitory processes, 238–39neuropsychology as complement to, 9parallels between storage and perception, 254–56syntactic processing, 290–91

Neuropsychologyas complement to neuroimaging studies, 9short- versus long-term memory case studies, 9

Neuroticism, control processes and, 95–98, 97f Nomothetic theories, 8, 10, 21–22, 39, 196Norepinephrine

in dynamic gating, 239in inhibition, 239

Object-extracted sentences, 273–78, 274t, 275t–277tObsessive-compulsive disorder, cognitive control

processes, 92Occipital cortex, in spatial storage, 253Older adults. See AgingOperation span, 23–24, 117–20, 119f

P300 event-related potential, in inhibition, 239Parietal cortex

in rehearsal, 256–57

in spatial storage, 253in verbal storage, 254Parkinson’s disease, 169Perceived value and learned value (PVLV), 177Perception sites, parallels between storage sites and,

254–56Perceptual processing streams, hysteresis in, 255Perseveration, 175–76Personality, 94–98, 97f, 242n1Phenylketonuria, 186Phonological loop, 58, 110, 137, 197, 256Plans, working memory and, 6–7

Post-interpretive processing, 273, 281Posterior cortex, 78, 187–88Predictor-criterion relationships, symmetry in,

66–67, 67f Prefrontal cortex

age-related atrophy, 262–63connectivity patterns, 78

318 Subject Index



development of rule-like representations in, 178f–179f,179–84, 181f–183f

dorsolateral, in spatial storage, 253in dual mechanisms of control theory, 78–80inferior, in visual storage, 254

in inhibition, 238interactions between fluid intelligence and interference

expectancy, 89–92, 91f memory strategy effects on, 87–88, 87f protracted development, 180in working memory, 41, 163

age-related changes, 93–94, 94f, 260–61, 262f Prefrontal–basal ganglia working memory (PBWM)

model, 167–68, 167f, 177, 178f–179f, 179–84,181f–183f

Primary memory, 5Priming, semantic, 229

Proactive controlcognitive individual differences, 88–92, 91f negative consequences, 83–84prefrontal cortex and, 78–80, 81–82versus reactive control, 80–82, 81t

Proactive interferencereactive control susceptibility, 84working memory and, 23, 31–32, 59, 234–35

Problem-solving tasks, 64–65Processing constraints, developmental, 142–44, 145–48,

147f, 150–51Processing efficiency, 29–30, 139Processing requirements, working memory span tasks,

126–27, 136, 139–41, 142–44Processing speed

aging and, 195development and, 100, 118, 128–29, 152–53, 196,

212–14, 213f, 214f in developmental cascade model, 100, 194, 196, 211,

212–16, 213f, 214f, 216f working memory and, 25–26, 27f, 42, 61–63, 218

Prosaccade task, 200, 201f Psychometric research, use of experimental variation, 9

Random-generation task, 65Raven’s matrices, 64, 90, 148, 212, 295Reactive control

cognitive individual differences, 88–92, 91f negative consequences, 84–85versus proactive control, 80–82, 81t

Reading skillsin atypically developing individuals, 149–50, 151–52working memory and, 8, 115, 148–49, 151–52

Reading span, 7–8, 23, 114–15, 117–20, 119f, 202, 280Reasoning, working memory and, 6–7, 49–50, 63–69, 67f,

70n2. See also Fluid intelligenceRecall timing, developmental effects, 121–22, 126, 127Redintegration, 205Reflexive saccade task, 200, 201f Rehearsal

attentional mechanisms, 257–58parietal cortex in, 256–57

Relational integrationbinding and, 56–59reasoning and, 63–66tasks, 55f, 57

Reorienting, use of geometric versus featural information,

175Residual variance, working memory span performance in

children, 144, 144t, 154, 156Resource-sharing account, working memory development

and, 112, 116, 124, 125, 139Response competition, working memory capacity and,

32–34Response timing analysis, 121–22, 127Restraint function of inhibition, 235–37Reversal learning, basal ganglia and, 169Rotation span, 24Rotation task, mental, 145

Schemata, in concentric model of working memory, 51Scholastic Aptitude Test (SAT) scores, working memory

span and, 112Semantic priming, 229Senile dementia, syntactic processing in, 285–88,

289f, 290tSensory areas, co-opting of, for working memory, 187–88,

257–58Sentence comprehension, syntactically based, 278–90Serial recall, 57, 58Series-completion tasks, 64Short-term memory

capacity limitations, 4–6, 15n3case studies, 9developmental disorders and, 137in executive attention theory, 23f, 39fluid intelligence and, 25, 26f

Short-term memory span tasksattentional/executive control in, 38–39, 251–52,

258–60working memory span tasks versus, 24

Slave systems, 111

Slips of thought, 236Social issues, 242n1Span of apprehension, 5–6, 15n1, 39Span of prehension, 5Spatial. See also Visuospatial entriesSpatial memory, activation patterns in older adults, 261Spatial rehearsal, 256–57Spatial representations

in concentric model of working memory, 50–51, 50f storage sites, 253

Spatial span tasks, 24, 57, 58, 197–99, 198f Sternberg task, 86, 89–90

Stop-signal task, 236Storage buffers, 7, 39–40, 253Storage constraints, developmental, 142–44, 145–48,

147f, 150Storage-only tasks

attentional/executive control in, 38–39, 251–52, 258–60working memory span tasks versus, 24

Subject Index 319



Storage requirements, working memory span tasks,126–27, 136, 142–44

Storage sites, 253–56, 254f Strategies, working memory and, 30, 87–88, 87f, 92–94,

94f, 98

Stroop task, 33–34, 236–37Structural credit assignment problem, 179Structural equation modeling, 215–16, 216f, 284–85, 285f Subcortical structures, in inhibition, 238–39Subject-extracted sentences, 273–78, 274t, 275t–277tSymmetry, in predictor-criterion relationships, 66–67, 67f

Task-set switchingaging effects, 263developmental effects, 116–20, 119f, 122, 123–25,

139–40, 207, 207f executive attention theory and, 35–36, 37

factor model of working memory and, 54, 59–60Task skill, working memory and, 29–30Task-switching account, limits, 123–25Temporal cortex

connectivity with prefrontal cortex, 78in verbal storage, 254in visual storage, 254

Temporal credit assignment problem, 177, 179Time-based resource-sharing account, 124Time-of-day effects, inhibitory processes, 232, 233–34,

236, 243n3Timing of recall, developmental effects, 121–22, 126, 127Towel-pulling task, 173–75, 174f Tower of Hanoi puzzle, 64–65Transformations, mental, 63, 65

Unified psychology, 10Updating of information, 163, 166–68, 167f, 178f–179f,

179–84, 181f–183f

Verbal information storage, 254 Verbal memory, activation patterns in older adults, 261 Verbal memory span

children, 206–8, 207f, 208f individual differences, 212–16, 213f, 214f, 216f manipulation tasks, 202–6, 204f older adults, 208–12, 209f–211f tasks, 197–99, 198f

Verbal rehearsal, 256 Verbal working memory

general (gvWM), 273, 277–78, 281, 282, 284, 286–88,291, 292

aging and, 281 Alzheimer’s dementia and, 285–86

specialized (svWM), 273, 288, 290, 293–94, 295

Visual information storage, 254 Visual search, 35–37 Visuospatial memory span

attention shifts and, 200–202, 201f children, 206–8, 207f, 208f individual differences, 212–16, 213f, 214f, 216f manipulation tasks, 203, 204f, 205

older adults, 208–12, 209f–211f, 217as predictor of fluid intelligence, 215tasks, 197–99, 198f

Visuospatial processing, 200–216 Visuospatial sketchpad, 110, 137, 197

Wechsler Memory Scale, age effects, 209, 210f White-matter volume, age differences, 217Williams syndrome, 137Wisconsin Card-Sorting Task, 263Within-individual variation

cognitive control processes, 85–88, 87f inhibitory processes, 230

Woodcock Johnson Concept Formation subtest, 215Word-monitoring paradigm, 281–82Working memory

active maintenance. See Active maintenance

aging and. See Agingattentional focus and, 41, 50f, 51–52. See also Attentionbinding and, 50, 51, 56–59computational models. See Computational modelsconcentric model, 50–52, 50f conflict and, 259–60content effects, 121, 217–18, 238definitions, 3–4, 6–7, 10–11, 15n2, 50, 77–78, 110–11,

136, 216–17, 293development and. See Development of working memorydomain generality. See Domain specificity/generalitydopamine and. See Dopamineexecutive functions and, 53, 59–61, 136, 144. See also

Executive control processesfactor structure, 53–56, 55f, 56f, 57tfunctionalist and process-oriented approaches, 40–41,

45n3general, 111general verbal (gvWM), 273, 277–78, 281, 282, 284,

286–88, 291, 292aging and, 281

Alzheimer’s dementia and, 285–86genetics and, 101

historical overview, 4–8, 15n2individual differences. See individual differences inworking memory

inhibition and. See Inhibitionintelligence and. See Intelligencelatent variables and, 25lifespan changes, 194–224. See also Aging;

Development of working memorymultiple-component model, 39–40, 110–11, 136–38,

252–53plans and, 6–7prefrontal cortex in. See Prefrontal cortex

processing efficiency and, 29–30, 139processing speed and, 25–26, 27f, 42, 61–63, 218.

See also Processing speedreading skills and, 8, 115, 148–49, 149, 151–52reasoning and, 6–7, 49–50, 63–69, 67f, 70n2representational characteristics, 252–56specialized verbal (svWM), 273, 288, 290, 293–94, 295

320 Subject Index



specific, 111strategies and, 30, 87–88, 87f, 92–94, 94f, 98syntactic, 273–78, 274t, 275t–277ttask skill and, 29–30unitary versus multifaceted theories, 11

updating of information, 163, 166–68, 167f, 178f–179f,179–84, 181f–183f

Working memory period, 122Working memory span tasks

in atypically developing individuals, 149–50, 151–52in autism, 138counting span, 7, 24, 113–14, 113f, 115–16, 117, 118,

131n1creation, 7–8, 22, 110deletion and, 234–35development of performance on, 113–20, 113f,

141–48

dual-task characteristics, 112–13executive attention and, 23–24, 38, 44–45n1executive functions and, 60letter-number span, 202–5, 204f, 209–11, 210f, 211f operation span, 23–24, 117–20, 119f

processing and storage requirements, 126–27, 136,139–41, 142–44

reading span, 7–8, 23, 114–15, 117–20, 119f, 202, 280residual variance, 144, 144t, 154, 156versus serial recall alone, 57–58spatial span, 24, 197–99, 198f verbal and visuospatial, 197–99, 198f

Wrap-up effect, 275

Young adultssyntactic processing in, 278–81, 279f working memory in, 197–206, 198f, 201f, 204f

Subject Index 321



DorsolateralPFC

Low-Load

Expectancy

0.8

0.6

0.4

0.2

–0.2

–0.4

0

0 5

A v e r a g e r e s p o n s e c h a

n g e

A v e r a g e r e s p o n s e c h a

n g e

10

High-LoadExpectancy

Low-LoadExpectancy

15 20

Time (seconds)

25


30 35

AnteriorPFC

High-Load

Expectancy

0.8

0.6

0.4

0.2

–0.2

–0.4

0

0 5 10 15 20

Time (seconds)

25


30 35

Figure 4.2. Memory strategy effects on prefrontal cortex (PFC) activity (Speer et al., 2003). Leftpanel: Left dorsolateral PFC region showing anticipatory, delay-related activation in low-load ex-pectancy condition. Right panel: Left anterior PFC region showing increased probe-related activityin high-load expectancy condition. X-axis refers to the time course of activity. Y-axis refers to averagepercentage fMRI signal change (from baseline).

Figure 4.3. Interactions between general fluid intelligence (gF) and interference expectancy in lat-eral prefrontal cortex (PFC) activity (Burgess & Braver, 2004). Top panel: High-gF group shows thathigh interference expectancy leads to increased delay-related activity in left ventrolateral PFC,whereas the low gF group shows no expectancy effect. Bottom panel: In high-expectancy condition,low-gF group shows increased probe-related activity in a nearby left ventrolateral PFC region for re-cent negatives, but high-gF group does not.



RostralACC

10

.5 .5

S t a t e A c i t i v i t y ( % s

i g n a l c h a n g e )

I t e m

A c i t i v i t y ( %

s i g n a l c h a n g e )

0.0

–.5

–1.0

–1.520

BIS BAS

30 20 30 40 50

r = –.28r = .41

60–.3

–.2

–.1

.1

.2

.3

.4

–.0

CaudalACC

Figure 4.5. Contributions of personality to anterior cingulate cortex (ACC) activity (Gray et al.,2005). Left panel: Positive association between behavioral inhibition system (BIS) (punishment-sensitivity) and tonic activity in rostral ACC. Right panel: Negative association between behavioral

approach system (BAS) (reward-sensitivity) and trial-specific activity in caudal ACC. Y-axis is per-centage fMRI signal change (from baseline).



Figure 10.1. Lateral view of the left hemisphere illustrating the location of Brodmann’s areas (nu-merical codes) known to participate in aspects of working memory. Color coding is used to identifythe neuroanatomical regions in which these Brodmann’s areas are situated. SMA = supplementarymotor cortex; PFC = prefrontal cortex; DLPFC = dorsolateral PFC; VLPFC = ventrolateral PFC.

[Andrew_R._A._Conway,_Christopher_Jarrold,_Michael(BookFi.org).pdf

Documents

Transcript of [Andrew_R._A._Conway,_Christopher_Jarrold,_Michael(BookFi.org).pdf