Causal Cognition in Human and Nonhuman Animals: A ...

ANRV296-PS58-05 ARI 17 November 2006 1:22

Causal Cognition in Humanand Nonhuman Animals:A Comparative, CriticalReviewDerek C. Penn and Daniel J. PovinelliCognitive Evolution Group, University of Louisiana, Lafayette, Louisiana 70504;email: [email protected]

Annu. Rev. Psychol. 2007. 58:97–118

First published online as a Review inAdvance on October 9, 2006

The Annual Review of Psychology is onlineat http://psych.annualreviews.org

This article’s doi:10.1146/annurev.psych.58.110405.085555

Copyright c© 2007 by Annual Reviews.All rights reserved

0066-4308/07/0203-0097$20.00

Key Words

causal reasoning, animal cognition, causal learning, contingencylearning, causal Bayes nets, causal power, associationism

AbstractIn this article, we review some of the most provocative experimentalresults to have emerged from comparative labs in the past few years,starting with research focusing on contingency learning and finish-ing with experiments exploring nonhuman animals’ understandingof causal-logical relations. Although the theoretical explanation forthese results is often inchoate, a clear pattern nevertheless emerges.The comparative evidence does not fit comfortably into either thetraditional associationist or inferential alternatives that have domi-nated comparative debate for many decades now. Indeed, the simi-larities and differences between human and nonhuman causal cog-nition seem to be much more multifarious than these dichotomousalternatives allow.

97

Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Contents

INTRODUCTION. . . . . . . . . . . . . . . . . 98LEARNING ASSOCIATIONS. . . . . . 99

Retrospective RevaluationEffects . . . . . . . . . . . . . . . . . . . . . . . . 99

Theoretical Accounts ofRetrospective RevaluationEffects . . . . . . . . . . . . . . . . . . . . . . . . 99

The Debate Between Associationistand Inferential Accounts . . . . . . . 100

ESTIMATING CAUSAL POWER. . 101Ceiling Effects . . . . . . . . . . . . . . . . . . . 101The Power PC Model of “Causal

Power” . . . . . . . . . . . . . . . . . . . . . . . 101Do Ceiling Effects Require an

Inferential Explanation? . . . . . . . 102INTERVENING ON CAUSAL

STRUCTURES . . . . . . . . . . . . . . . . . 103Causal Bayes Nets . . . . . . . . . . . . . . . . 103Intervention Versus Observation in

a Nonhuman Subject . . . . . . . . . . 104Analysis of Blaisdell et al.’s (2006)

Results . . . . . . . . . . . . . . . . . . . . . . . . 105REASONING ABOUT CAUSAL

MECHANISMS . . . . . . . . . . . . . . . . . 106The Role of “Intuitive Theories” in

Human Causal Cognition . . . . . . 106Nonhuman Animals’

Understanding of Tools,Support, and Gravity . . . . . . . . . . 107

The Case for Diagnostic CausalReasoning in NonhumanApes . . . . . . . . . . . . . . . . . . . . . . . . . . 109

CONCLUSIONS, PROBLEMS,SUGGESTIONS . . . . . . . . . . . . . . . . 110

INTRODUCTION

Animals of all taxa have evolved cognitivemechanisms for taking advantage of causalregularities in the physical world. Many arealso quite adept at using and manufacturingsimple tools. But the way that human sub-jects cognize causal regularities is clearly agood deal more sophisticated than that of anyother animal on the planet. This much, at

least, seems indisputable. Which specific cog-nitive mechanisms human beings share withother animals, however, and which—if any—are uniquely human is an age-old questionthat is still very much unresolved (see, for ex-ample, Castro & Wasserman 2005, Chappell2006, Clayton & Dickinson 2006, Reboul2005, Vonk & Povinelli 2006).

In our opinion, substantive progress onthis fundamental question has been greatlyhindered by the dichotomous debate between“associationist” and “inferential” theories ofcausal cognition that has dominated compar-ative research for many decades and still holdssway in some parts of town (see Shanks 2006for a review). Associationists have long arguedthat all of nonhuman causal cognition andmost of human causal cognition as well canbe reduced to a kind of contingency learningbased on stimulus-bound, associative mech-anisms similar to those that govern Pavlo-vian conditioning (for reviews, see Dickinson2001, Pearce & Bouton 2001, Shanks 1995,Wasserman & Miller 1997). More generouscomparative researchers, on the other hand,claim that even nonhuman animals are capa-ble of reasoning about “causal-logical” rela-tions in a human-like fashion (Call 2004) andemploy “controlled and effortful inferentialreasoning processes” to do so (Beckers et al.2006).

In this article, we review some of themost provocative experimental results to haveemerged from comparative labs in recentyears. We start with experiments focusingon contingency learning and finish withresearch that explores nonhuman animals’understanding of unobservable causal mech-anisms and causal-logical relations. The the-oretical explanation for these provocative re-sults is often inchoate at best. Nevertheless, aclear pattern emerges. The available compar-ative evidence does not fit comfortably intoeither the traditional associationist or classi-cally inferential alternatives. Indeed, the sim-ilarities and differences between human andnonhuman causal cognition seem to be muchmore multifarious and fascinating than these

98 Penn · Povinelli

Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


dichotomous alternatives allow (for similarsuggestions, see Chappell 2006, Clayton &Dickinson 2006, Gallistel 2003, Heyes &Papineau 2005, Povinelli 2000, Shettleworth1998).

LEARNING ASSOCIATIONS

Retrospective Revaluation Effects

We begin our bottom-up review of the recentcomparative literature on causal cognitionwith a phenomenon that has challenged asso-ciationist theory from within. “Retrospectiverevaluation” is an umbrella term that associa-tionist researchers use to refer to contingencylearning scenarios in which the associativestrength between a conditioned stimulus (CS)and an unconditioned stimulus (US) changeseven though the CS in question is absent onthe relevant training episodes. The paradig-matic example of a retrospective revaluationeffect is backward blocking.

In the case of forward blocking, a cue is firstpaired with a US (e.g., A+) and then the firstcue is presented in compound with a target cueand the US (e.g., AX+). Kamin (1969) showedthat rats presented with A+ and then AX+ tri-als subsequently exhibited a weaker responseto X than did rats who were only exposed toAX+ trials—as if learning that the reward wascontingent on A “blocked” the rats from sub-sequently learning that the reward might alsobe contingent on X. The venerable Rescorla-Wagner (1972) model of associative learn-ing was developed, in large part, to accountfor cue competition effects such as forwardblocking.

In the case of backward blocking, the com-pound cue is trained first (AX+) and thenthe competing cue is presented alone (A+).As in forward blocking, the response to Xalone is “blocked” on subsequent trials. Thistime, however, the blocking effect has oc-curred even though the X stimulus was absenton the critical A+ trials. There is compellingevidence for a variety of retrospective reval-uation effects in nonhuman as well as humansubjects (e.g., Balleine et al. 2005, Blaisdell &

Within-compoundassociations:associations formedbetween cues thatoccur in closephysical and/ortemporal proximityas distinct fromassociations thatform between cuesand outcomes

Miller 2001, Denniston et al. 2003, Dickinson& Burke 1996, Miller & Matute 1996, Shanks1985, Shevill & Hall 2004, Wasserman &Berglan 1998, Wasserman & Castro 2005).

Theoretical Accounts ofRetrospective Revaluation Effects

Traditional associationist theories, such asthe Rescorla-Wagner model, cannot accountfor retrospective revaluation effects becausethey assume that only cues present on agiven trial can undergo a change in theirresponse-eliciting potential (see discussionsin Dickinson 2001, Shanks 2006, Wasserman& Castro 2005). Thus, there have beennumerous attempts to revise associationistmodels in order to account for retrospec-tive revaluation effects (e.g., Chapman 1991,Dickinson & Burke 1996, Van Hamme &Wasserman 1994). These revised associa-tionist models postulate that subjects formwithin-compound associations between CSsthat have occurred together in addition to as-sociations between a CS and a US.

One common characteristic of these re-vised models is that the associative strengthof an absent CS can be updated only if it haspreviously been associated with a CS that isactually present on the given trial. Recently,however, researchers have shown that the con-ditioned response to a US is also sensitive tothe relation between stimuli that have neveractually co-occurred but are only indirectlylinked to each other through a web of interme-diary associations (e.g., De Houwer & Beckers2002a,b; Denniston et al. 2001, 2003; Macho& Burkart 2002). For example, Dennistonet al. (2003) presented rats with AX+ trials andthen with XY+ trials. Rats who subsequentlyreceived A- extinction trials responded lessstrongly to the Y cue than did rats who re-ceived no such extinction trials even thoughthe A and Y stimuli never occurred together.As Wasserman & Castro (2005) point out,none of the revised associative models is ableto account for higher-order effects such asthese since the relevant cues never actuallyoccurred together.

www.annualreviews.org • Causal Cognition in Animals 99

Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


To inferentially minded researchers,higher-order retrospective revaluation lookslike it requires “higher-order reasoningprocesses” and “conscious propositionalknowledge” (De Houwer et al. 2005). Ac-cording to an inferential account, after the A-extinction trials, the rats in Denniston et al.’s(2003) experiment learned that A was not thetrue cause. Given this “propositional knowl-edge,” the rats deduced that X, not A, was theactual causal stimulus in the AX+ pairing andthen, by further deductive inference, that Ymust not have been the true causal stimulusin the XY+ pairing (De Houwer et al. 2005).

A higher-order reasoning account of ret-rospective revaluation certainly provides onepossible explanation for the rats’ behavior andhas undeniable appeal from a folk psycholog-ical point of view. However, it is not the onlypossible explanation. Denniston et al. (2001),for example, propose an alternative hypoth-esis that does not require propositional rep-resentations or inferential reasoning (see alsoBlaisdell et al. 1998; Denniston et al. 2003;Stout & Miller, manuscript submitted).

According to this “extended comparatorhypothesis,” the response to a given CS resultsfrom a comparison between the representa-tion of the US directly activated by the targetCS and the representation of the US indirectlyactivated by other CSs with which the targetCS has been directly or indirectly associatedin the past. For example, in Denniston et al.’s(2003) experiment, the extended comparatorhypothesis suggests that the rats’ response toY was modulated by the second-order associa-tion between Y and A as well as the first-orderassociations between Y and X and between Xand A (see Denniston et al. 2001 for a detailedexposition).

The extended comparator hypothesis isbased firmly on traditional associative princi-ples like spatio-temporal contiguity and “se-mantically transparent associations” (Fodor2003, cited by Shanks 2006). However, con-trary to traditional associationist theories, thecomparator hypothesis posits that cues do not

compete for associative strength when theyare learned; rather, they compete for controlover the subject’s behavior when they are eval-uated. Indeed, colloquially speaking, the ex-tended comparator hypothesis proposes that asubject’s response to a target cue is diminishedby the extent to which it is able to “think” ofan alternative cause or predictor of the out-come in question (Stout & Miller, manuscriptsubmitted). In short, in order to explain theeffects of higher-order retrospective revalu-ation, the extended comparator hypothesisposits the kind of performance-focused, struc-tured information-processing capabilities thatassociationists have traditionally eschewed.

The Debate Between Associationistand Inferential Accounts

Which hypothesis best explains higher-orderretrospective revaluation effects in human andnonhuman animals is a matter of vigorous de-bate (Aitken & Dickinson 2005; Beckers et al.2005; De Houwer et al. 2005; Denniston et al.2003; Melchers et al. 2004; Wasserman &Castro 2005). Unfortunately, only the asso-ciative side of the dispute has provided a for-mal specification of its claims. Van Overwalle& Timmermans (2001), for example, haveproposed a connectionist implementation ofDickinson & Burke’s (1996) model of first-order retrospective revaluation. And Stout &Miller (manuscript submitted) have submitteda formal computational specification of the ex-tended comparator hypothesis. Higher-orderreasoning accounts of retrospective revalua-tion, on the other hand, have only been for-mulated in a “verbal manner rather than for-malized mathematically” (De Houwer et al.2005). As De Houwer et al. (2005) frankly ad-mit, “The most troubling implication of thislack of precision is that it becomes difficult torefute higher-order reasoning accounts.” Ob-viously, an important future challenge for ad-vocates of inferential accounts is to providea formal, computational specification of theirclaims.


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Regardless of which theoretical accountprevails, the existing evidence has alreadydemonstrated that nonhuman animals are ca-pable of feats of causal learning once de-nied them by traditional associationist the-ory. Even cognitively minded researchers mayneed to revise their assessment of nonhu-man causal cognition upwards. Visalberghi &Tomasello (1998), for example, once arguedthat nonhuman primates are unable to un-derstand the “web of possibilities” that con-nects causes and effects. “Associative learn-ing,” Visalberghi & Tomasello (1998) wenton to explain, “does not involve a web ofpossible connections, but only a one-to-oneconnection between antecedent and conse-quent.” Evidence of higher-order retrospec-tive revaluation in rats demonstrates that atleast some nonhuman animals are, in fact,sensitive to higher-order associations betweenabsent cues. And the extended comparator hy-pothesis demonstrates that the principles ofassociative learning can, in fact, be revisedto take this web of possible connections intoaccount.

ESTIMATING CAUSAL POWER

Ceiling Effects

If an effect, E, always occurs at its maximallevel in a given context, regardless of whethera particular cause, C, is present or not, it is im-possible to draw any inferences about whetherthe given cause has the power to produce theeffect or not (Cheng 1997, Cheng & Holyoak1995). The rational response to this state ofaffairs is for the subject to remain agnosticas to the causal power of the candidate causein question. A subject interested in evaluatingwhether or not C prevents E, however, couldinfer that C is indeed noncausal. Ceiling ef-fects are not symmetrical for generative andpreventive causes (Cheng et al. 2006). It hasnow been well established that both humanand nonhuman subjects are sensitive to ceil-ing effects when learning about contingen-

Causal power: theprobability withwhich a candidatecause, when present,actually produces orprevents an effect inquestion as distinctfrom the frequencywith which the causeand the effecthappen to co-occur

Preventive cause: acause that has thepower to prevent (orreduce theprobability of) theoccurrence of a giveneffect

Outcomemaximality: thehighest previouslevel at which a givenoutcome has beenexperienced by thesubject

Candidate cause:of all the possiblecauses for a giveneffect, the one that iscurrently beingevaluated by thesubject

Base rate: the rateat which a giveneffect occurs in theabsence of thecandidate cause

Focal set: the set ofevents that a subjectselects as relevant forcomputing the causalpower of a candidatecause in a givencontext

cies and treat generative and preventive casesdifferently (see Cheng 1997 for a detaileddiscussion).

Researchers have recently shown that bothhuman and nonhuman subjects are also sen-sitive to outcome maximality and additivityeffects (Beckers et al. 2005, 2006; De Houweret al. 2002; Lovibond et al. 2003; Vandorpeet al. 2005). For example, Lovibond et al.(2003) showed that human subjects exhibitsignificantly stronger blocking when the can-didate causes are described as having an ad-ditive effect. And Beckers et al. (2006) haveshown that forward blocking in rats is atten-uated when the intensity of the US presentedduring test trials is the same as the maximumintensity of the US experienced during priortraining trials.

The Power PC Model of “CausalPower”

The traditional model for computing the sta-tistical contingency between two cues is the�P model (Jenkins & Ward 1965, Rescorla1968),

�P = p(E|C) − p(E | ∼ C ),

where p(E |C) is the probability of observingthe effect given the presence of the candidatecause, and p(E |∼C) is the probability of ob-serving the effect in the absence of the candi-date cause.

The �P model of statistical contingencydoes not account for ceiling effects or theasymmetry between generative and preven-tive causes. Cheng (1997) showed that thenormative model for estimating the contin-gency between a cause and an effect must con-sider the base rate probability of the effectin the absence of the candidate cause. In thecase of binary causes and effects (i.e., causesand effects that are either present or absent),and assuming that the candidate cause is inde-pendent of any alternative causes in the sub-ject’s “focal set,” Cheng’s model for generative


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Outcomeadditivity: acharacteristic of cuesthat produce agreater intensityoutcome whencombined with otheradditive cues thanwhen presentedalone

causes is given by

q = �P1 − p(E | ∼ C)

,

where �P is the standard model of statisti-cal contingency described above, p(E | ∼ C) isthe probability of observing the effect in theabsence of the candidate cause, and q is an es-timate of the unobservable “causal power” ofthe candidate cause in question.

In principle, the Power PC theory pro-vides a normative model of causal inductionin the absence of prior domain-specific causalknowledge that can handle both ceiling ef-fects and the asymmetry between generativeand preventive causes. It also handles retro-spective revaluation effects by positing thatsubjects calculate causal power over the ap-propriate “focal set” of cases as specified bythe theory (Cheng & Holyoak 1995).

To be sure, whether or not human causalintuitions actually conform to the Power PCmodel’s predictions for intermediate base-rate probabilities is a matter of some dispute(Allan 2003, Buehner et al. 2003, Griffiths& Tenenbaum 2005, Lober & Shanks 2000,Perales & Shanks 2003). Worse, at least froma comparative point of view, there still havebeen no experiments testing the Power PCmodel’s central predictions about the interac-tion between �P and base-rate probabilitieson nonhuman subjects. In the absence of suchevidence, it is still too soon for advocates ofthe Power PC theory to rest on their laurels.

Do Ceiling Effects Require anInferential Explanation?

Regardless of whether or not the Power PCtheory winds up being an appropriate modelof causal induction for human or nonhumansubjects, the evidence for outcome maximalityand additivity effects suggests that both hu-man and nonhuman animals are not simplylearning about observable contingencies butare sensitive to the unobservable constraintsspecific to causal inference. Both human and

nonhuman animals appear to understand tac-itly that covariation only implies causation un-der special circumstances.

But does ruling out associationist expla-nations and acknowledging a profound sim-ilarity between human and nonhuman causalinduction mean that rats necessarily employ“controlled and effortful inferential reason-ing processes” and “conscious propositionalknowledge” (Beckers et al. 2006, De Houweret al. 2005)?

Beckers et al. (2006) do not presentany computational arguments to justify theirpropositional attributions. Instead, they argueby analogy to human psychology: i.e., in hu-man causal learning, sensitivity to outcomeadditivity and maximality seems to involveconscious propositional inferences, so nonhu-man animals who exhibit a similar sensitivitymust be employing similar mental processes(for an extended critique of this venerable ar-gument, see Povinelli et al. 2000).

There are numerous reasons to be skep-tical of this particular analogy. Almost allof the evidence cited in support of the roleof conscious inferential processes in humancausal cognition has no parallel in nonhumanstudies. For example, Vandorpe et al. (2005)showed that the “verbal self-reports” of hu-man subjects are consistent with an inferen-tial account of blocking; blocking in humansubjects is sensitive to secondary task diffi-culty (De Houwer & Beckers 2003, Vandorpeet al. 2005); and verbal information providedto human subjects after all learning trialshave concluded nevertheless influences block-ing effects (De Houwer 2002). These resultscertainly suggest a role for “controlled andeffortful inferential reasoning” and “propo-sitional knowledge” in human causal cogni-tion. But there is no comparable evidence fornonhuman subjects. Moreover, such verbalreports in human beings may sometimes bejust posthoc redescriptions of effects initiallygenerated through implicit nonpropositionalmechanisms (Povinelli et al. 2000). It is worthnoting, in this respect, that Cheng (1997) has


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


consistently argued that subjects “implicitly”use a “qualitative” version of her model.

Beckers et al. (2005) do not present a for-mal specification of their hypothesis, so it isdifficult to know exactly what they mean whenthey claim that outcome maximality and ad-ditivity effects in rats reflect the operation of“symbolic causal reasoning processes.” Cer-tainly, Beckers et al.’s (2005) evidence suggeststhat rats are able to manipulate representa-tions that stand in for objective properties ofstimuli in the world (e.g., the maximum in-tensity level of a given US) and are able tocompute numeric operations over these values(e.g., updating the strength of a “blocked” cueas a function of the actual versus expected out-come). There are good reasons for believingthat such information-processing operationsrequire the ability to manipulate symbols andvariables (see Marcus 2001 for a lucid discus-sion). Indeed, Buehner et al. (2003) make asimilar claim with respect to the Power PCmodel. So if this is what Beckers et al. (2005)mean by “symbolic causal reasoning” pro-cesses, they seem to be on solid ground.

On the other hand, if what Beckers et al.(2005) mean by “symbolic causal reason-ing” is equivalent to the kind of higher-order propositional inferences they attributeto humans, their claim is much more ten-dentious. In other domains—such as for-aging, spatial cognition, and instrumentallearning—comparative researchers have pro-posed a variety of information-processingarchitectures that could account for the ob-served outcome maximality and additivity ef-fects in rats without positing the need forhigher-order inferential reasoning or propo-sitional representations (e.g., Clayton et al.2001, Dickinson & Balleine 2000, Gallistel2003, Shettleworth 1998). Given the species-specific evidence for conscious higher-orderinferential reasoning in human cue competi-tion effects (e.g., De Houwer et al. 2005), thesuggestion that “parallel processes” are oper-ating in humans and rodents (Beckers et al.2006) appears premature.

INTERVENING ON CAUSALSTRUCTURES

Causal Bayes Nets

An increasingly influential approach to causalinduction proposes that subjects solve causalreasoning problems in a manner consistentwith the “causal Bayes net” formalism orig-inally developed in computer science andstatistics (Pearl 2000, Spirtes et al. 2001). Themathematical details of the causal Bayes netapproach are far beyond the scope of thepresent review (but see Glymour 2003 for ashort, nontechnical introduction). The basicidea, however, can be expressed easily enoughin nonmathematical terms.

Causal Bayes nets represent causal struc-tures as directed acyclic graphs in which nodesrepresent events or states of the world, and theconnections between nodes (called “edges”)represent causal relations. Some simple ex-amples of possible causal structures includecommon-cause, common-effect, and causalchain models (see Figure 1).

The causal Bayes net formalism is predi-cated on a set of core assumptions. The mostimportant of these assumptions is the causalMarkov condition. The causal Markov condi-tion says that if one holds all the direct causesof a given variable constant, then that variablewill be statistically independent of all othervariables in the causal graph that are not its ef-fects. For example, in a simple common-causemodel in which two effects occur with a cer-tain probability given a particular value of C,the state of the two effects is independent ofeach other if the value of C is fixed.

The causal Bayes net formalism placesspecial importance on the distinction be-tween interventional and observational pre-dictions (Danks 2006, Hagmayer et al. 2007,Pearl 2000, Spirtes et al. 2001, Woodward2003). Interventions are formally modeled asan external independent cause that fixes thevalue of a given node, thereby “cutting off ”all other causal influences on that manipu-lated node. Pearl (2000) has aptly baptized


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


X

Y Z Y Z

X

Y Z

X

common-effectmodel

causal-chainmodel

common-causemodel

Figure 1Three basic causal structures represented as directed acyclic graphs. In a common-cause structure, X isthe common cause of Y and Z. In a causal-chain structure, X influences Y, which influences Z. And in acommon-effect structure, X and Y both influence a common effect, Z.

this procedure “graph surgery.” Causal Bayesnets provide a formal, computational spec-ification for how to derive interventionalpredictions from observational learning andvice-versa.

Different versions of the causal Bayesnet approach propose different learning al-gorithms for inferring the causal structureunderlying a given set of covariation data.Bottom-up models focus on providing algo-rithms for inferring causal structures basedon statistical data in the absence of any othercues (e.g., Gopnik et al. 2004, Spirtes et al.2001). The “causal model” approach empha-sizes the role of top-down domain-general as-sumptions that constrain and inform the in-duction process (Waldmann 1996, Waldmann& Hagmayer 2001, Waldmann & Holyoak1992). The “theory-based” approach focuseson the influence of domain-specific priorknowledge (Tenenbaum & Griffiths 2003,Tenenbaum et al. 2006).

The causal Bayes net formalism is arguablythe most powerful formal account of humancausal inference available today (see, for ex-ample, Danks 2005, Gopnik et al. 2004, Hag-mayer et al. 2007, Lagnado et al. 2005, Tenen-baum & Griffiths 2003). Whether or not it is apsychologically accurate description of causalcognition in any nonhuman subject is anothermatter.

Intervention Versus Observationin a Nonhuman Subject

To our knowledge, only a single published ex-perimental paper has explicitly claimed thatnonhuman animals reason about causal re-lations in a manner consistent with a causalBayes net approach. In the crucial experimentin this paper, Blaisdell et al. (2006) presentedrats with stimuli whose conditional dependen-cies purportedly corresponded to one of twoalternative causal structures. Rats presentedwith a common-cause model were given pair-ings of a light, L, followed by a tone, T, and,then separately, the same light, L, followed bya food reward, F. Rats presented with a causal-chain model were given pairings of T followedby L and then, separately, L followed by F.

During the test phase, each of the twogroups of rats was divided randomly into oneof two test conditions, and a lever that had notbeen previously present was inserted into thetest chamber. Rats in condition intervene-Treceived a presentation of T each time theypressed the lever. Rats in condition observe-T observed presentations of T independentlyof their own actions on the lever. The exper-imenters recorded the number of nose pokesthat the rats made into the magazine where Fhad been delivered during the training phase(there was no actual food in the magazine dur-ing the test phase).


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


The authors found that rats in conditionintervene-T who had witnessed the common-cause model made fewer nose pokes thandid rats in condition observe-T. In contrast,there was no significant difference betweenthe intervene-T and observe-T conditions forrats in the causal-chain group. Based on theseresults, the authors concluded:

Rats made correct inferences for instrumen-tal actions on the basis of purely obser-vational learning, and they correctly dif-ferentiated between common-cause models,causal-chains and direct causal links. Theseresults contradict the view that causal learn-ing in rats is solely driven by associativelearning mechanisms, but they are consis-tent with causal Bayes net theories. The corecompetency of reasoning with causal modelsseems to be already in place in animals, evenwhen elaborate physical knowledge may notyet be available. (Blaisdell et al. 2006)

Analysis of Blaisdell et al.’s (2006)Results

We are not surprised that nonhuman causalinduction is consistent with at least some pre-dictions of a causal Bayes net approach. Howcould it not be? The causal Bayes net formal-ism provides an exceptionally powerful linguafranca that can give posthoc explanations fornearly any nonpathological causal inference(Danks 2006). The question of whether hu-man or nonhuman learning is consistent witha causal Bayes net approach is meaningless un-less one specifies the particular model at stake.Thus, what is most intriguing about Blaisdellet al.’s results is the particular assumptions thatone must make in order to claim that the rats’behavior is consistent with a causal Bayes netapproach.

For example, during the initial trainingphase, the rats in the common-cause groupwere never presented with L followed by bothT and F as would happen if both cues wereactually effects of a common cause. Instead,the L → T pairings were perfectly negatively

correlated with the L → F pairings: i.e., ev-ery instance of L was followed by either Lor T but never both. In other words, ratspurportedly presented with a common-causemodel were, in fact, never shown covariationinformation consistent with a common-causemodel. The fact that the rats, nevertheless,acted as if they had inferred a common-causestructure should give causal Bayes net enthu-siasts reason to pause.

Adopting a simple common-cause modelbased solely on the observed data wouldviolate the causal Markov condition sinceT and F are not independent conditionalon the state of L. In order for the rats’behavior to be consistent with the causalMarkov condition, one must posit that therats had some sort of prior bias that influ-enced their causal judgments. One possibil-ity is that the rats were working under thetacit assumption that simpler causal struc-tures (e.g., a simple common-cause model)are more likely than complex causal structures(e.g., a common-cause model with inhibitoryedges between effects). Such a prior biaswould be consistent with those causal Bayesnet approaches that allow for the influence oftop-down, domain-general assumptions suchas the causal model approach advocated byWaldmann and colleagues (Waldmann 1996,Waldmann & Hagmayer 2001, Waldmann &Holyoak 1992), but it would not be consistentwith any causal Bayes net account that gener-ates causal inferences in a purely bottom-upfashion based solely on observed covariations.To be sure, while Waldmann and colleagueshave provided extensive evidence in supportof their causal model hypothesis with respectto human subjects, the evidence for extendingthis hypothesis to rats is much more tenuous.

Blaisdell et al.’s (2006) own explanation forwhy the rats inferred a common-cause struc-ture given the anomalous data is not based onWaldmann et al.’s causal model hypothesis.Instead, the authors proposed that rats whohave observed L → T pairings and then ob-serve an L → F pairing “conservatively treatthe absent but expected events [i.e., T] as


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


possibly present but missed” (Blaisdell et al.2006). Given the fact that the T stimulus wasa highly salient tone or noise of 10 seconds induration, the claim that the rats assumed theyhad somehow “missed” this cue cries out forfurther experimental corroboration.

Indeed, Blaisdell et al. (2006) provide verylittle evidence that rats inferred a common-cause model from the data using a causal Bayesnet approach at all. Since the pairings weredeterministic, the only information a sub-ject could use to distinguish a common-causestructure from a causal-chain structure wasthe temporal ordering of the cues: i.e., T andF both appeared 10 seconds after L. Whilesome causal Bayes net theorists have empha-sized the importance of temporal informationfor causal learning (see Lagnado et al. 2005for an example), the use of temporal informa-tion to distinguish between alternative causalstructures is certainly not unique to the causalBayes net formalism.

This said, the most critical finding inBlaisdell et al.’s (2006) experiment clearly un-dermines a traditional associationist accountof the rats’ behavior (see also Clayton &Dickinson 2006). Rats in the intervene-T con-dition of the common-cause group were lessinterested in F than rats in the observe-Tcondition. Blaisdell et al.’s provocative resultschallenge any theory of causal cognition thatcannot explain how subjects derive novel in-terventional predictions from purely observa-tional learning. But ruling out a traditional as-sociationist explanation of the rats’ behaviordoes not necessarily mean that rats tacitly cog-nize their own interventions in a human-likefashion or use the causal Markov condition todo so.

There is an abundance of evidence demon-strating that human subjects are able touse their own interventions in a deliberatelyepistemic fashion (Danks 2006; Hagmayeret al. 2007; Lagnado & Sloman 2002, 2004;Povinelli & Dunphy-Lelii 2001; Steyverset al. 2003; Waldmann & Hagmayer 2005;Woodward 2003). Steyvers et al. (2003), forexample, showed that human subjects do

not intervene randomly when the numberof interventions they are allowed to makeis constrained; instead, they choose theirinterventions in order to provide the most di-agnostic test of their initial hypotheses. In-deed, human subjects seem to plan their in-terventions like quasi-experiments in orderto eliminate confounds and distinguish be-tween possible causal structures when obser-vational data alone are ambiguous (Lagnadoet al. 2005).

Blaisdell et al.’s (2006) results are consis-tent with Pearl’s concept of graph surgery ifone interprets this term as meaning nothingmore than implicitly treating two associatedevents as independent once the subject hasintervened on the consequent event (MichaelWaldmann, personal communication). Cru-cially, however, Blaisdell et al.’s (2006) re-sults do not provide any evidence that ratstacitly cognize their own interventions in anepistemic fashion, are sensitive to the causalMarkov condition, or plan their own interven-tions in a quasi-experimental fashion to elu-cidate ambiguous causal relations as humansubjects do. If what the rats did in the resultsreported by Blaisdell et al. (2006) counts asgraph surgery, it is graph surgery by an acci-dental surgeon.

REASONING ABOUT CAUSALMECHANISMS

The Role of “Intuitive Theories” inHuman Causal Cognition

All of the theories reviewed to this pointhave focused on causal induction and the partplayed by domain-general causal assump-tions. Researchers who focus on the problemsof causal induction—whether from an as-sociationist, causal power, or causal Bayesnet perspective—have largely tabled anydiscussion of domain-specific prior knowl-edge. Many have explicitly stipulated that thenettlesome problem is outside the scope oftheir models (e.g., Cheng 1997, Dickinson2001). Even those theorists who argue for the


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


importance of top-down knowledge havelargely focused on tightly canalized, domain-general assumptions rather than learneddomain-specific representations (e.g.,Lagnado et al. 2005, Waldmann 1996).

Nevertheless, nearly everyone admits thatprior domain-specific knowledge is an inte-gral aspect of human causal cognition outsideof the laboratory. Human subjects almost al-ways use their prior domain-specific knowl-edge to evaluate novel causal relations ratherthan bootstrap their way up from observedcovariation information and domain-generalassumptions alone (Tenenbaum & Griffiths2003, Tenenbaum et al. 2006). Indeed, humansubjects often seek out and prefer informationabout underlying mechanisms rather than relysolely on information about covariation (Ahnet al. 1995).

Many cognitively minded researchersclaim that human children possess ab-stract, coherent, rule-governed representa-tions about the unobservable causal mech-anisms at work in specific domains such asphysics, biology, and psychology. Some the-orists argue that this “core knowledge” ishighly canalized due to heritable mecha-nisms (Carey 1985, Keil 1989, Spelke 1994),whereas others posit that it is largely learnedon the fly (Gopnik & Meltzoff 1997). In ei-ther case, it is widely agreed that a child’scausal knowledge is “theory-like” in the sensethat it provides principled, allocentric, coher-ent, abstract explanations for the unobserv-able causal mechanisms that govern a givendomain.

We use the term “intuitive theory” to referto a subject’s coherent domain-specific knowl-edge about unobservable causal mechanisms(Gopnik & Schulz 2004, Tenenbaum et al.2006). By “unobservable,” we mean that thesecausal mechanisms are based on the struc-tural or functional relation between objectsrather than on perceptually based exemplars(cf. Vonk & Povinelli 2006). One of the salientcharacteristics of unobservable causal mecha-nisms, such as gravity and support, is that theycan be generalized freely to disparate concrete

examples that share little to no perceptuallybased featural similarity

We have no doubt that intuitive theoriesabout unobservable causal mechanisms play aformative role in human causal cognition. Inthe remainder of this article, we review thecomparative evidence to ascertain to what ex-tent this is true of nonhuman animals as well(see also Povinelli 2000, Vonk & Povinelli2006).

Nonhuman Animals’ Understandingof Tools, Support, and Gravity

To date, the strongest positive claims con-cerning nonhuman animals’ intuitive theo-ries about the physical world have come froma series of seminal experiments carried outby Hauser and colleagues on nonhuman pri-mates’ understanding of tools (Hauser 1997;Hauser et al. 1999, 2002a,b; Santos et al. 2003,2006).

Hauser (1997) showed that adult tamarinmonkeys reliably preferred cane-like toolswhose shape is the same as a previously func-tional tool to cane-like tools with a novelshape but familiar color and texture. Basedon these results, Hauser (1997) claimed thattamarin monkeys can distinguish causally rel-evant from causally irrelevant properties ofa tool and thus possess a “functional con-cept of artifacts.” Similar results have beendocumented for infant tamarin monkeys withminimal prior exposure to manipulable ob-jects (Hauser et al. 2002a) as well as rhesusmacaques, vervet monkeys, and lemurs (seeHauser & Santos 2006 for a review). Anal-ogous results have been shown in the do-main of food (Santos et al. 2001, 2002), wherethe sets of relevant and irrelevant features arereversed.

These results add to the growing body ofevidence that nonhuman animals indeed dopossess evolved domain-specific predisposi-tions that bias how they perceive and ma-nipulate objects in the world in the absenceof observed covariation information or directinstrumental learning (see Shettleworth 1998


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


for a review). On the other hand, a heritablediscriminative bias is not the same thing asan intuitive theory. Nothing in Hauser et al.’sresults suggests that monkeys possess any in-sight into why one set of features is morerelevant to tools than to food. Nor is thereany evidence that their discriminatory biasesare abstract, allocentric, or theory-like in thesense attributed to human children. All ofHauser’s results to date are consistent with amore modest hypothesis; i.e., nonhuman pri-mates are predisposed to perceive certain clus-ters of features as more salient than otherswhen selecting among potential tools withoutunderstanding anything about the underlyingcausal mechanisms involved.

There is not simply an absence of evidencethat nonhuman primates possess an intuitivetheory about tools, there is also consistentevidence of an absence. Povinelli and col-leagues have performed an extensive series oftests on chimpanzees’ understanding of phys-ical causal mechanisms (see Povinelli 2000).The general conclusion of these experimentsis that when tool-use tasks are carefully con-strued to tease apart observable and unobserv-able relations, chimpanzees consistently focussolely on the observable relations and fail tocognize the unobservable causal mechanismsat stake (see also Vonk & Povinelli 2006).

In one of these experiments, for exam-ple, Povinelli (2000, Chapter 10) replicatedPiaget’s (1952) cloth-pulling experiment, inwhich subjects are asked to pull a piece ofcloth toward them in order to obtain a re-ward that is lying on the cloth but out ofreach. Hauser et al. (1999) had previously re-ported that tamarin monkeys successfully dis-tinguished between rewards lying on or off thecloth and understood the support relation “atan abstract level, tolerating all featural trans-formations.” Povinelli and colleagues, how-ever, systematically varied the featural cuesavailable to the chimpanzees and found thatthey were only sensitive to certain perceptualrelations—such as the degree of surface con-tact between the cloth and the reward—andwere insensitive to the actual structural rela-

tion casually relevant to obtaining the reward.In particular, the chimpanzees appeared to beoblivious to whether or not the cloth was ac-tually supporting the reward as opposed tosimply being in contact with it.

A series of seminal experiments byVisalberghi et al. provides further evidencefor the absence of any intuitive theory amongnonhuman primates about purely abstractcausal mechanisms such as gravity and sup-port (Limongelli et al. 1995, Visalberghiet al. 1995, Visalberghi & Limongelli 1994,Visalberghi & Trinca 1989). For example,Visalberghi & Limongelli (1994) tested ca-puchin monkeys’ ability to retrieve a piece offood placed inside a transparent tube using astraight stick. In the middle of the tube, therewas a highly visible hole with a small trans-parent cup attached. If the subject pushedthe food over the hole, the food fell into thecup and was inaccessible. After about 90 tri-als, only one out of the four capuchin mon-keys learned to push the food away from thehole; and even this one learned the correctbehavior through trial and error. Worse, onceVisalberghi et al. rotated the tube so that thetrap-hole was now facing up and causally irrel-evant, the only successful capuchin still per-sisted in treating the hole as if it needed tobe avoided—making it obvious that even thissubject did not understand the causal relationbetween the trap hole and the retrieval of thereward. By way of comparison, it should benoted that children as young as three yearsof age successfully solve the trap-tube taskafter only a few trials (Visalberghi &Tomasello 1998).

Povinelli (2000, Chapter 4) replicatedVisalberghi’s trap-tube setup with sevenchimpanzees. Only a single chimp performedabove chance on the normal trap tube condi-tion. When tested on the inverted trap con-dition, this chimp—like the single successfulcapuchin in Visalberghi’s originalexperiment—failed to take the positionof the trap into account. More recently still,Santos et al. (2006) has replicated many ofPovinelli’s (2000) experiments with tamarin


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


and vervet monkeys and found convergentresults. Based on these experiments, Santoset al. (2006) now conclude that nonhumanprimates’ comprehension of tools “is morelimited than previously stated.”

The failure to understand unobservablecausal mechanisms such as support and gravityin an abstract fashion is not limited to nonhu-man primates. Seed et al. (2006) recently pre-sented eight rooks, a species of corvid, witha clever modification to the traditional trap-tube task. Seven out of eight rooks learned theinitial version of the modified trap-tube taskquite rapidly. Nevertheless, when presentedwith a series of transfer tasks in which the vi-sual cues that predicted success in the initialtask were absent or confounded, only one ofthe seven subjects passed. In a separate follow-up experiment (Tebbich et al. 2006), none ofthe rooks passed the transfer task.

Seed et al.’s (2006) results add to the grow-ing evidence that corvids are quite adept at us-ing stick-like tools, perhaps even more adeptthan nonhuman primates (see, for example,Chappell & Kacelnik 2002, 2004; Weir et al.2002). But as Seed et al. (2006) point out, theseresults also suggest that rooks, like nonhumanprimates, do not have a species-universal un-derstanding of “unobservable causal proper-ties” like gravity and support. Instead, they ap-pear to solve tool-use problems based on theobservable features of the task and evolved,task-specific expectations about what featuresare likely to be most salient (see also Chappell2006 on the importance of interindividualdifferences).

The Case for Diagnostic CausalReasoning in Nonhuman Apes

Until recently, there has been a consensus thatnonhuman animals do not seek out diagnosticcausal explanations (Povinelli 2000, Povinelli& Dunphy-Lelii 2001, Premack & Premack1994, Visalberghi & Tomasello 1998). Break-ing with this comparative consensus, Call(2004, 2005) has recently argued that non-human apes are, in fact, quite good at seek-

ing out diagnostic causal explanations basedon “causal-logical relations” and “quite bad atassociating arbitrary stimuli and responses.”These tendentious claims are based on a setof experiments testing apes’ inferences aboutthe location of food (Call 2004).

In these experiments, Call (2004) pre-sented 4 bonobos, 12 chimpanzees, 6orangutans, and 8 gorillas with two opaquecups, one of which was baited with a foodreward. In the first experiment, the experi-menter either showed the contents of bothcups to the subjects or shook both of thecups. Unsurprisingly, subjects strongly pre-ferred the cup in which they had seen the re-ward. Of the 27 original subjects, 9 also pre-ferred the cup in which the shaking motionwas associated with a noise.

In the second experiment, the 9 success-ful subjects from experiment 1 were retestedin a condition in which only one of the twocups was shaken. In some trials, the baited cupwas shaken; in other trials, the empty cup wasshaken. Out of the 9 subjects, 3 were abovechance on the crucial condition in which onlythe empty cup was shaken and the ape had toinfer “by exclusion” that the food was in theother cup.

In subsequent experiments, Call (2004)tested a number of arbitrary noises (such astapping on the cup or playing the recordedsound of a shaking noise) against the ac-tual noise produced by shaking the cup. Theapes chose the baited cup more frequently on“causal” conditions than on “arbitrary” ones.Based on this evidence, Call (2004) arguedthat the apes had understood the “causal-logical relation between the cup movement,the food, and the auditory cue” and under-stood that “the food causes the noise.”

There are numerous problems with Call’sinterpretation. We do not know enough aboutthe learning history of the subjects involved inthese experiments to rule out the alternativehypothesis that they were simply respondingon the basis of previously learned contingen-cies. It seems quite plausible, for example, thatthese captive apes had previously learned that


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


a shaking noise (N) combined with a shak-ing motion (M) is jointly indicative of a re-ward (NM+), whereas a shaking noise with-out a shaking motion (N-) or a shaking motionwithout a shaking noise (M-) is not. Positivepatterning of this kind is a well-documentedphenomenon in the animal conditioning lit-erature (see Wasserman & Miller 1997 for adiscussion). None of Call’s numerous manip-ulations ruled out this obvious alternative hy-pothesis. As we discussed above, there is ampleevidence that nonhuman animals often reasonabout causal relations in a fashion that is in-explicable in associationist terms. Ironically,though, a quite traditional associative expla-nation suffices in this particular case.

Indeed, Call’s (2004) claim that the apesunderstood the contingencies in a “causal-logical” fashion appears to be refuted by Call’sown results. In experiment 3, Call presentedthe apes with an empty “shaken silent cup”and an empty “rotated silent cup” (i.e., turnedupside down and then right side up again). Inexperiments 1 and 2, the same shaking motionproduced an audible rattling noise when thecup contained food. Nevertheless, the sub-jects strongly preferred the silent shaken cupto the silent rotated cup. If the apes had infact understood the causal-logical relation-ship involved, they would have inferred thatneither cup contained food and would havechosen randomly between the two cups or,if anything, would have preferred the rotatedcup. Call provides no “causal-logical” expla-nation for why the apes would strongly pre-fer a shaken silent cup to a rotated silent cup.Once again, an explanation based on simpleassociative conditioning seems to fit the bill.

CONCLUSIONS, PROBLEMS,SUGGESTIONS

We hope that the comparative evidence wehave reviewed over the course of this ar-ticle has demonstrated why the venerabledichotomy between associationist and infer-ential explanations of nonhuman causal cog-nition is both specious and unproductive. We

agree with Chappell (2006): The real situationseems to be much more complicated, multi-farious, and fascinating.

Many aspects of both human and nonhu-man causal learning are parsimoniously ex-plained in terms of some form of associativeconditioning. On the other hand, both hu-man and nonhuman animals are also sensi-tive to constraints specific to causal relationssensu strictu—such as ceiling effects and theasymmetry between generative and preven-tive causes. This implies that causal induc-tion is not simply reducible to contingencylearning in either human or nonhuman sub-jects. Even nonhuman animals employ cog-nitive mechanisms that distinguish betweencausality and covariation.

With respect to nonhuman animals’ un-derstanding of their own instrumental ac-tions, the evidence again suggests that nonhu-man causal cognition lies somewhere outsidethe associationist and inferential alternatives.On the one hand, nonhuman animals’ capac-ity for flexible goal-directed actions suggeststhat they explicitly represent the causal rela-tion between their own action and its conse-quences as well as the value of the expectedoutcome (see Dickinson & Balleine 2000).Moreover, Blaisdell et al.’s (2006) provoca-tive results suggest that rats tacitly differen-tiate between the consequences of interven-tions on different kinds of causal structures.On the other hand, there is still no convincingevidence that nonhuman animals of any taxaseek out diagnostic explanations of anomalouscausal relations or deliberately use their owninterventions in order to elucidate ambiguouscausal dependencies. For the moment, suchdiagnostic, inferential reasoning abilities ap-pear to be uniquely human.

Most importantly, there appears to be afundamental discontinuity between humanand nonhuman animals when it comes tocognizing the unobservable causal mecha-nisms underlying a given task or state ofaffairs. While many species (including hu-mans) have a tacit understanding that someevents have the unobservable “power” to


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


cause other events (Cheng 1997), nonhumananimals’ causal beliefs appear to be largelycontent-free; that is, their causal beliefs donot incorporate an abstract representationof the underlying generative mechanisms in-volved (Dickinson & Balleine 2000). Reason-ing about the unobservable causal-logical re-lation between one particular causal belief andanother appears to be a uniquely human trait.

This is not to say that all nonhuman an-imals are uniform in their cognitive abili-ties or that human subjects always reasonin abstract, inferential terms. Evolution hasclearly sculpted cognitive architectures toserve specific functions in specific species(Shettleworth 1998). The remarkable tool-using abilities of corvids stand as a stark re-minder that the humanist notion of a scalanaturae in causal cognition is wishful thinking.Our hypothesis is simply that in the panoply ofanimal cognition, the ability to reason aboutunobservable, domain-specific causal mecha-nisms in a causal-logical fashion is a specifi-cally human specialization (see also Povinelli2000, 2004; Vonk & Povinelli 2006).

Why is it that only human animals areable to acquire and use representations aboutunobservable causal mechanisms? Unfortu-nately, answering this comparative explanan-dum is hampered by the fact that our un-derstanding of how intuitive theories work inhuman subjects is inchoate at best.

Causal Bayes net models are often toutedas the way to reconcile bottom-up causalinduction and top-down causal knowledge(Danks 2005, Gopnik et al. 2004, Gopnik &Schulz 2004). Upon closer inspection, how-ever, it is clear that causal Bayes nets are not upto the task. As Tenenbaum et al. (2006) pointout, any formal specification of intuitive causaltheories of human causal cognition must beable to account for the hierarchical coherenceamong causal relations at various levels of ab-straction. Lien & Cheng (2000), for example,showed that human subjects are more likely tojudge a candidate cause to be genuinely causalif it is “hierarchically consistent” with theirprior knowledge about superordinate causal

relations. Unfortunately, the edges and nodesof the causal Bayes net formalism are not “suf-ficiently expressive” to evaluate what makes agiven concrete causal relation more or less co-herent with superordinate mechanisms at dif-ferent levels of abstraction (Tenenbaum et al.2006).

In the absence of any well-established for-mal account of intuitive theories in humancausal cognition, it is well nigh impossible togive a formal explanation for the discontinu-ity between human and nonhuman abilities inthis area. Nevertheless, in the interests of pro-voking future research and debate, we close byproposing a preliminary hypothesis.

Our hypothesis is that abstract causalreasoning—i.e., causal cognition that involvesreasoning about the relation between causalpredicates at various levels of generality—isintimately bound up with the dynamics ofanalogical reasoning. It is well known thathuman subjects often learn about novel andunobservable causal relations by analogy toknown and/or observable ones: The structureof the atom, for example, is often describedby analogy to the solar system; electricity isconceived of as analogous to a flowing liq-uid; gravity is like a physical force. As Lien &Cheng (2000) suggest, the process of acquir-ing and predicating abstract causal relationsseems akin to analogical inference; i.e., novelcausal relations are often learned “by analogy”to known superordinate relations, and su-perordinate causal schemas are often learnedby systematically abstracting out the func-tional elements common to superficially dis-parate causal regularities. If this hypothesis isright, computational models of analogical in-ference may provide the missing link betweenbottom-up and top-down processes in humancausal cognition (French 2002, Gentner et al.2001, Holyoak & Thagard 1997).

Our hypothesis provides a computationalexplanation for why human and nonhumanabilities differ so dramatically when it comesto reasoning about unobservable and ab-stract causal mechanisms. With the excep-tion of a single unreplicated experiment on


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


one language-trained chimp (i.e., Gillan et al.1981), there is no evidence that any nonhu-man animal is capable of analogical reasoning.Ex hypothesi, the reason why only human sub-

jects can reason about unobservable domain-specific causal mechanisms is because only hu-mans have the representational architecturenecessary to reason by analogy.

SUMMARY POINTS

1. The evidence suggests that nonhuman causal cognition is significantly more sophis-ticated than can be accounted for by traditional associationist theories. In particular,both human and nonhuman animals do not simply learn about observable contin-gencies; they appear to be sensitive to the unobservable constraints specific to causalinference.

2. On the other hand, there is a lack of compelling evidence that nonhuman animalspursue diagnostic explanations of anomalous causal relations or deliberately use theirown interventions in order to elucidate ambiguous causal dependencies.

3. Nonhuman animals do not appear capable of the kinds of causal-logical inferencesemployed by human subjects when reasoning about abstract causal relations.

4. Nonhuman casual cognition is not well served by either the traditional associationistor classically inferential alternatives that have dominated comparative debate for manydecades.

FUTURE ISSUES

1. Can the new generation of associationist models—such as the extended comparatorhypothesis—account for the richness of nonhuman causal cognition without givingup the representational-level parsimony that has been the hallmark of traditionalassociationist theory?

2. Can advocates of a higher-order inferential account of causal cognition provide aformal, computational specification of their claims?

3. Are nonhuman animals sensitive to the effects of intermediate base-rate probabilitiesas predicted by the Power PC theory?

4. To what extent can nonhuman animals use their own interventions in a deliberatelydiagnostic manner? For example, can they use their interventions to elucidate am-biguous causal structures?

5. Can nonhuman animals infer the presence of hidden and/or unobservable causes? Arecausal Bayes nets a useful formalism for describing the cognitive processes responsiblefor nonhuman causal cognition?

6. What representational-level limitations account for the inability of nonhuman animalsto reason about unobservable causal relations?

7. What is the computational role of analogical inference in human causal cognition?


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


ACKNOWLEDGMENTS

We are grateful to the many researchers who provided us with preprints of their papers toreview. We would particularly like to thank Aaron Blaisdell, Patricia Cheng, Keith Holyoak,and Michael Waldmann for many invaluable discussions as well as their critical comments andsuggestions on early drafts. This work was supported in part by a Centennial Fellowship fromthe James S. McDonnell Foundation to DJP.

LITERATURE CITED

Ahn W, Kalish CW, Medin DL, Gelman SA. 1995. The role of covariation versus mechanisminformation in causal attribution. Cognition 54:299–352

Aitken MR, Dickinson A. 2005. Simulations of a modified SOP model applied to retrospectiverevaluation of human causal learning. Learn. Behav. 33:147–59

Allan LG. 2003. Assessing Power PC. Learn. Behav. 32:192–204Balleine BW, Espinet A, Gonzalez F. 2005. Perceptual learning enhances retrospective reval-

uation of conditioned flavor preferences in rats. J. Exp. Psychol.: Anim. Behav. Process.31:341–50

Beckers T, De Houwer J, Pineno O, Miller RR. 2005. Outcome additivity and outcome max-imality influence cue competition in human causal learning. J. Exp. Psychol.: Learn. Mem.Cogn. 31:238–49

This importantpaper shows thatrats are sensitive tooutcome additivityand maximalityeffects.

Beckers T, Miller RR, De Houwer J, Urushihara K. 2006. Reasoning rats: forward block-ing in Pavlovian animal conditioning is sensitive to constraints of causal inference.J. Exp. Psychol.: Gen. 135:92–102

Blaisdell AP, Bristol AS, Gunther LM, Miller RR. 1998. Overshadowing and latent inhibitioncounteract each other: support for the comparator hypothesis. J. Exp. Psychol.: Anim. Behav.Process. 24:335–51

Blaisdell AP, Miller RR. 2001. Conditioned inhibition produced by extinction-mediated recov-ery from the relative stimulus validity effect: a test of acquisition and performance modelsof empirical retrospective revaluation. J. Exp. Psychol.: Anim. Behav. Process. 27:48–58

A provocative andseminal experimentsuggesting that thecausal Bayes netformalism may beappropriate fornonhuman as wellas human subjects.

Blaisdell AP, Sawa K, Leising KJ, Waldmann MR. 2006. Causal reasoning in rats. Science311:1020–22

Buehner MJ, Cheng PW, Clifford D. 2003. From covariation to causation: a test of the as-sumption of causal power. J. Exp. Psychol.: Learn. Mem. Cogn. 29:1119–40

The authorpresents evidencethat nonhumanapes are able toreasondiagnostically in a“causal-logical”fashion.

Call J. 2004. Inferences about the location of food in the great apes (Pan paniscus, Pantroglodytes, Gorilla gorilla, and Pongo pygmaeus). J. Comp. Psychol. 118:232–41

Call J. 2006. Descartes’ two errors: reason and reflection in the great apes. In Rational Animals?,ed. S Hurley, M Nudds, pp. 219–34. Oxford, UK: Oxford Univ. Press

Carey S. 1985. Conceptual Change in Childhood. Cambridge, MA: MIT PressCastro L, Wasserman EA. 2005. Associative learning in animals and humans. Presented

at Online Conf. Causality, organized by Inst. Cogn. Sci., Lyon and Univ. Geneva.http://www.interdisciplines.org/causality/

Chapman GB. 1991. Trial order affects cue interaction in contingency judgment. J. Exp. Psy-chol.: Learn. Mem. Cogn. 17:837–54

Chappell J. 2006. Avian cognition: understanding tool use. Curr. Biol. 16:244–45Chappell J, Kacelnik A. 2002. Tool selectivity in a nonprimate, the New Caledonian crow

(Corvus Moneduloides). Anim. Cogn. 5:71–78Chappell J, Kacelnik A. 2004. Selection of tool diameter by New Caledonian crows. Anim.

Cogn. 7:121–27


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Cheng PW. 1997. From covariation to causation: a causal power theory. Psychol. Rev. 104:367–405

Cheng PW, Holyoak KJ. 1995. Complex adaptive systems as intuitive statisticians: causality,contingency and prediction. In Comparative Approaches to Cognition, ed. JA Meyer, HLRoitblat, pp. 271–302. Cambridge, MA: MIT Press

Cheng PW, Novick LR, Liljeholm M, Ford C. 2006. Explaining four psychological asymme-tries in causal reasoning: implications of causal assumptions for coherence. In Topics inContemporary Philosophy (Vol. 4): Explanation and Causation, ed. M O’Rourke. Cambridge,MA: MIT Press

Clayton NS, Dickinson A. 2006. Rational rats. Science 9:472–74Clayton NS, Griffiths DP, Emery NJ, Dickinson A. 2001. Elements of episodic-like memory

in animals. Philos. Trans. R. Soc. Lond. B Biol. Sci. 356:1483–91Danks D. 2006. Causal learning from observations and manipulations. In Thinking with Data,

ed. M Lovett, P Shah. Mahwah, NJ: Erlbaum. In pressDanks D. 2005. The supposed competition between theories of human causal inference. Philos.

Psychol. 2:259–72De Houwer J. 2002. Forward blocking depends on retrospective inferences about the presence

of the blocked cue during the elemental phase. Mem. Cogn. 30:24–33De Houwer J, Beckers T. 2002a. Higher-order retrospective revaluation in human causal

learning. Q. J. Exp. Psychol. B 55:137–51De Houwer J, Beckers T. 2002b. Second-order backward blocking and unovershadowing in

human causal learning. Exp. Psychol. 49:27–33De Houwer J, Beckers T. 2003. Secondary task difficulty modulates forward blocking in human

contingency learning. Q. J. Exp. Psychol. B 56:345–57De Houwer J, Beckers T, Glautier S. 2002. Outcome and cue properties modulate blocking.

Q. J. Exp. Psychol. A 55:965–85De Houwer J, Beckers T, Vandorpe S. 2005. Evidence for the role of higher order reasoning

processes in cue competition and other learning phenomena. Learn. Behav. 33:239–49Denniston JC, Savastano HI, Blaisdell AP, Miller RR. 2003. Cue competition as a retrieval

deficit. Learn. Motiv. 34:1–31Denniston JC, Savastano HI, Miller RR. 2001. The extended comparator hypothesis: learn-

ing by contiguity, responding by relative strength. In Handbook of Contemporary LearningTheories, ed. RR Mowrer, SB Klein, pp. 65–117. Mahwah, NJ: Erlbaum

Dickinson A. 2001. Casual learning: an associative analysis. Q. J. Exp. Psychol. 54B:3–25Dickinson A, Balleine B. 2000. Causal cognition and goal-directed action. In The Evolution of

Cognition, ed. CM Heyes, L Huber, pp. 185–204. Cambridge, MA: MIT PressDickinson A, Burke J. 1996. Within-compound associations mediate the retrospective revalu-

ation of causality judgements. Q. J. Exp. Psychol. B 49:60–80Fodor JA. 2003. Hume Variations. Oxford, UK: Oxford Univ. PressFrench RM. 2002. The computational modeling of analogy-making. Trends Cogn. Sci. 6:200–5Gallistel CR. 2003. Conditioning from an information processing perspective. Behav. Process.

61:1–13Gentner D, Holyoak KJ, Kokinov BN, eds. 2001. The Analogical Mind: Perspectives from Cognitive

Sciences. Cambridge, MA: MIT PressGillan DJ, Premack D, Woodruff G. 1981. Reasoning in the chimpanzee: I. Analogical rea-

soning. J. Exp. Psychol.: Anim. Behav. Process. 7:1–17Glymour C. 2003. Learning, prediction and causal Bayes nets. Trends Cogn. Sci. 7:43–48Gopnik A, Glymour C, Sobel D, Schulz L, Kushnir T, Danks D. 2004. A theory of causal

learning in children: causal maps and Bayes nets. Psychol. Rev. 111:1–31


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Gopnik A, Meltzoff AN. 1997. Words, Thoughts, and Theories. Cambridge, MA: MIT PressGopnik A, Schulz L. 2004. Mechanisms of theory formation in young children. Trends Cogn.

Sci. 8:371–77Griffiths A, Tenenbaum JB. 2005. Structure and strength in causal induction. Cogn. Psychol.

51:334–84Hagmayer Y, Sloman SA, Lagnado DA, Waldmann MR. 2007. Causal reasoning through

intervention. In Causal Learning: Psychology, Philosophy and Computation, ed. A Gopnik, LSchulz. Oxford, UK: Oxford Univ. Press. In press

Hauser MD. 1997. Artifactual kinds and functional design features: what a primate understandswithout language. Cognition 64:285–308

Hauser MD, Kralik J, Botto-Mahan C. 1999. Problem solving and functional design features:experiments on cotton-top tamarins, Saguinus oedipus oedipus. Anim. Behav. 57:565–82

Hauser MD, Pearson H, Seelig D. 2002a. Ontogeny of tool use in cottontop tamarins, Saguinusoedipus: innate recognition of functionally relevant features. Anim. Behav. 64:299–311

Hauser MD, Santos LR. 2006. The evolutionary ancestry of our knowledge of tools: frompercepts to concepts. In Creations of the Mind: Essays on Artefacts and Their Representation,ed. E Margolis, S Lawrence. Oxford, UK: Oxford Univ. Press. In press

Hauser MD, Santos LR, Spaepen GM, Pearson HE. 2002b. Problem solving, inhibition anddomain-specific experience: experiments on cotton-top tamarins, Saguinus oedipus. Anim.Behav. 64:387–96

Heyes CM, Papineau D. 2005. Rational or associative? Imitation in Japanese quail. In RationalAnimals?, ed. M Nudds, S Hurley, pp. 187–196. Oxford, UK: Oxford Univ. Press

Holyoak KJ, Thagard P. 1997. The analogical mind. Am. Psychol. 52:35–44Jenkins HM, Ward WC. 1965. Judgment of contingency between responses and outcomes.

Psychol. Monog. 79(Suppl. 1):1–17Kamin LJ. 1969. Selection association and conditioning. In Fundamental Issues in Associative

Learning, ed. NJ Mackintosh, pp. 42–64. Halifax, Nova Scotia: Dalhousie Univ. PressKeil F. 1989. Concepts, Kinds and Cognitive Development. Cambridge, MA: MIT PressLagnado DA, Sloman SA. 2002. Learning causal structure. In Proceedings of the Twenty-Fourth

Annual Conference of the Cognitive Science Society, ed. B Bel, I Marlien, pp. 560–65. Mahwah,NJ: Erlbaum

Lagnado DA, Sloman SA. 2004. The advantage of timely intervention. J. Exp. Psychol.: LearnMem. Cogn. 30:856–76

Lagnado DA, Waldmann MR, Hagmayer Y, Sloman SA. 2007. Beyond covariation: cues tocausal structure. In Causal Learning: Psychology, Philosophy and Computation, ed. A Gopnik,L Schultz. New York: Oxford Univ. Press

Lien Y, Cheng PW. 2000. Distinguishing genuine from spurious causes: a coherence hypoth-esis. Cogn. Psychol. 40:87–137

Limongelli L, Boysen ST, Visalberghi E. 1995. Comprehension of cause-effect relations in atool-using task by chimpanzees (Pan troglodytes). J. Comp. Psychol. 109:18–96

Lober K, Shanks DR. 2000. Is causal induction based on causal power? Critique of Cheng1997. Psychol. Rev. 107:195–212

Lovibond PF, Been SL, Mitchell CJ, Bouton ME, Frohardt R. 2003. Forward and backwardblocking of causal judgment is enhanced by additivity of effect magnitude. Mem. Cogn.31:133–42

Macho S, Burkart J. 2002. Recursive retrospective revaluation of causal judgments. J. Exp.Psychol.: Learn. Mem. Cogn. 28:1171–86

Marcus GF. 2001. The Algebraic Mind: Integrating Connectionism and Cognitive Science. Cam-bridge, MA: MIT Press


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Melchers KG, Lachnit H, Shanks DR. 2004. Within compound associations in retrospectiverevaluation and in direct learning: a challenge for comparator theory. Q. J. Exp. Psychol.57B:25–53

Miller RR, Matute H. 1996. Biological significance in forward and backward blocking: reso-lution of a discrepancy between animal conditioning and human causal judgment. J. Exp.Psychol.: Anim. Behav. Process. 18:251–64

Pearce JM, Bouton ME. 2001. Theories of associative learning in animals. Annu. Rev. Psychol.52:111–39

Pearl J. 2000. Causality. London: Cambridge Univ. PressPerales JC, Shanks DR. 2003. Normative and descriptive accounts of the influence of power

and contingency on causal judgment. Q. J. Exp. Psychol. 56A:977–1007Piaget J. 1952. The Origins of Intelligence in Children. New York: NortonPovinelli DJ. 2000. Folk Physics for Apes. New York: Oxford Univ. PressPovinelli DJ. 2004. Behind the apes’ appearance: escaping anthropomorphism in the study of

other minds. Daedalus: J. Am. Acad. Arts Sci. Winter, pp. 29–41Povinelli DJ, Bering JM, Giambrone S. 2000. Toward a science of other minds: escaping the

argument by analogy. Cogn. Sci. 24:509–41Povinelli DJ, Dunphy-Lelii S. 2001. Do chimpanzees seek explanations? Preliminary compar-

ative investigations. Can. J. Exp. Psychol. 55:185–93Premack D, Premack AJ. 1994. Levels of causal understanding in chimpanzees and children.

Cognition 50:347–62Reboul A. 2005. Similarities and differences between human and nonhuman causal cognition. Pre-

sented at Online Conf. Causality, organized by Inst. Cogn. Sci., Lyon and Univ. Geneva.http://www.interdisciplines.org/causality/

Rescorla RA. 1968. Probability of shock in the presence and absence of CS in fear conditioning.J. Comp. Physiol. Psychol. 66:1–5

Rescorla RA, Wagner AR. 1972. A theory of Pavlovian conditioning: variations in the effective-ness of reinforcement and nonreinforcement. In Classical Conditioning II: Current Researchand Theory, ed. AH Black, WF Prokasy, pp. 64–99. New York: Appleton-Century-Crofts

Santos LR, Hauser MD, Spelke ES. 2001. Recognition and categorization of biologicallysignificant objects by rhesus monkeys (Macaca mulatta): the domain of food. Cognition82:127–55

Santos LR, Hauser MD, Spelke ES. 2002. Domain-specific knowledge in human children andnonhuman primates: artifacts and food. In The Cognitive Animal, ed. M Bekoff, C Allen,GM Burghardt, pp. 269–81. Cambridge, MA: MIT Press

Santos LR, Miller CT, Hauser MD. 2003. Representing tools: how two nonhuman primatespecies distinguish between the functionally relevant and irrelevant features of a tool.Anim. Cogn. 6:269–81

Researcherspresentnontool-usingmonkeys with tasksdesigned to mirrorthose that hadpreviouslychallengedchimpanzees.

Santos LR, Pearson H, Spaepen G, Tsao F, Hauser M. 2006. Probing the limits of toolcompetence: experiments with two nontool-using species (Cercopithecus aethiopsand Saguinus oedipus). Anim. Cogn. 9:94–109The authors

present anexceedingly clevertwist on thetraditionaltrap-tube task toseven rooks, anontool-usingspecies of corvid.

Seed AM, Tebbich S, Emery NJ, Clayton NS. 2006. Investigating physical cognition inrooks (Corvus frugilegus). Curr. Biol. 16:697–701

Shanks DR. 1985. Forward and backward blocking in human contingency judgment. Q. J. Exp.Psychol. 37B:1–21

Shanks DR. 1995. The Psychology of Associative Learning. London: Cambridge Univ. PressShanks DR. 2006. Associationism and cognition: human contingency learning at 25. Q. J. Exp.

Psychol. 59:In press


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Shettleworth SJ. 1998. Cognition, Evolution and Behavior. New York: Oxford Univ. PressShevill I, Hall G. 2004. Retrospective revaluation effects in the conditioned suppression pro-

cedure. Q. J. Exp. Psychol. 57B:331–47Spelke E. 1994. Initial knowledge: six suggestions. Cognition 50:433–47Spirtes P, Glymour C, Scheines R, eds. 2001. Causation, Prediction, and Search. Cambridge,

MA: MIT PressSteyvers M, Tenenbaum JB, Wagenmakers EJ, Blum B. 2003. Inferring causal networks from

observations and interventions. Cogn. Sci. 27:453–89Stout SC, Miller RR. 2006. Sometimes competing retrieval (SOCR): a formalization of the

comparator hypothesis. Psychol. Rev. Manuscr. submittedTebbich S, Seed AM, Emery NJ, Clayton NS. 2006. Non-tool-using rooks (Corvus frigilegus)

solve the trap-tube task. Anim. Cogn. In pressTenenbaum JB, Griffiths TL. 2003. Theory-based causal induction. In Advances in Neural

Information Processing Systems, ed. S Becker, S Thrun, K Obermayer, pp. 35–42. Cambridge,MA: MIT Press

Tenenbaum JB, Griffiths TL, Niyogi S. 2006. Intuitive theories as grammars for causal infer-ence. In Causal Learning: Psychology, Philosophy, and Computation, ed. A Gopnik, L Schulz.Oxford, UK: Oxford Univ. Press. In press

Van Hamme LL, Wasserman EA. 1994. Cue competition in causality judgments: the role ofnonpresentation of compound stimulus elements. Learn. Motiv. 25:127–51

Van Overwalle F, Timmermans B. 2001. Learning about an absent cause: discounting andaugmentation of positively and independently related causes. In Connectionist Models ofLearning, Development and Evolution: Proceedings of the Sixth Neural Computation and Psy-chology Workshop, ed. RM French, JP Sougne, pp. 219–28. London: Springer-Verlag

Vandorpe S, De Houwer J, Beckers T. 2005. Further evidence for the role of inferentialreasoning in forward blocking. Mem. Cogn. 33:1047–56

Visalberghi E, Fragaszy DM, Savage-Rumbaugh ES. 1995. Performance in a tool-using taskby common chimpanzees (Pan troglodytes), bonobos (Pan paniscus), an orangutan (Pongopygmaeus), and capuchin monkeys (Cebus apella). J. Comp. Psychol. 109:52–60

Visalberghi E, Limongelli L. 1994. Lack of comprehension of cause-effect relations in tool-using capuchin monkeys (Cebus apella). J. Comp. Psychol. 108:15–22

Visalberghi E, Tomasello M. 1998. Primate causal understanding in the physical and psycho-logical domains. Behav. Process. 42:189–203

Visalberghi E, Trinca L. 1989. Tool use in capuchin monkeys: distinguishing between perfor-mance and understanding. Primates 30:511–21

Vonk J, Povinelli DJ. 2006. Similarity and difference in the conceptual systems of primates:the unobservability hypothesis. In Comparative Cognition, ed. T Zentall, EA Wasserman,pp. 363–87. Oxford, UK: Oxford Univ. Press

Waldmann MR. 1996. Knowledge-based causal induction. In The Psychology of Learning andMotivation, ed. DR Shanks, KJ Holyoak, DL Medin, pp. 47–88. San Diego, CA: Academic

Waldmann MR, Hagmayer Y. 2001. Estimating causal strength: the role of structural knowl-edge and processing effort. Cognition 82:27–58

Waldmann MR, Hagmayer Y. 2005. Seeing versus doing: two modes of accessing causal knowl-edge. J. Exp. Psychol.: Learn. Mem. Cogn. 31:216–27

Waldmann MR, Holyoak KJ. 1992. Predictive and diagnostic learning within causal models:asymmetries in cue competition. J. Exp. Psychol.: Gen. 121:222–36

Wasserman EA, Berglan LR. 1998. Backward blocking and recovery from overshadowing inhuman causal judgement: the role of within-compound associations. Q. J. Exp. Psychol. B51:121–38


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Wasserman EA, Castro L. 2005. Surprise and change: variations in the strength of present andabsent cues in causal learning. Learn. Behav. 33:131–46

Wasserman EA, Miller RR. 1997. What’s elementary about associative learning? Annu. Rev.Psychol. 48:573–607

Weir AAS, Chappell J, Kacelnik A. 2002. Shaping of hooks in New Caledonian crows. Science297:981

Woodward J. 2003. Making Things Happen: A Theory of Causal Explanation. Oxford, UK: OxfordUniv. Press


Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.

Contents ARI 8 November 2006 21:2

Annual Review ofPsychology

Volume 58, 2007

Contents

Prefatory

Research on Attention Networks as a Model for the Integration ofPsychological ScienceMichael I. Posner and Mary K. Rothbart � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 1

Cognitive Neuroscience

The Representation of Object Concepts in the BrainAlex Martin � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 25

Depth, Space, and Motion

Perception of Human MotionRandolph Blake and Maggie Shiffrar � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 47

Form Perception (Scene Perception) or Object Recognition

Visual Object Recognition: Do We Know More Now Than We Did 20Years Ago?Jessie J. Peissig and Michael J. Tarr � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 75

Animal Cognition

Causal Cognition in Human and Nonhuman Animals: A Comparative,Critical ReviewDerek C. Penn and Daniel J. Povinelli � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 97

Emotional, Social, and Personality Development

The Development of CopingEllen A. Skinner and Melanie J. Zimmer-Gembeck � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 119

vii

Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Biological and Genetic Processes in Development

The Neurobiology of Stress and DevelopmentMegan Gunnar and Karina Quevedo � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 145

Development in Societal Context

An Interactionist Perspective on the Socioeconomic Context ofHuman DevelopmentRand D. Conger and M. Brent Donnellan � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 175

Culture and Mental Health

Race, Race-Based Discrimination, and Health Outcomes AmongAfrican AmericansVickie M. Mays, Susan D. Cochran, and Namdi W. Barnes � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 201

Personality Disorders

Assessment and Diagnosis of Personality Disorder: Perennial Issuesand an Emerging ReconceptualizationLee Anna Clark � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 227

Social Psychology of Attention, Control, and Automaticity

Social Cognitive Neuroscience: A Review of Core ProcessesMatthew D. Lieberman � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 259

Inference, Person Perception, Attribution

Partitioning the Domain of Social Inference: Dual Mode and SystemsModels and Their AlternativesArie W. Kruglanski and Edward Orehek � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 291

Self and Identity

Motivational and Emotional Aspects of the SelfMark R. Leary � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 317

Social Development, Social Personality, Social Motivation,Social Emotion

Moral Emotions and Moral BehaviorJune Price Tangney, Jeff Stuewig, and Debra J. Mashek � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 345

viii Contents

Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


The Experience of EmotionLisa Feldman Barrett, Batja Mesquita, Kevin N. Ochsner,

and James J. Gross � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 373

Attraction and Close Relationships

The Close Relationships of Lesbian and Gay MenLetitia Anne Peplau and Adam W. Fingerhut � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 405

Small Groups

OstracismKipling D. Williams � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 425

Personality Processes

The Elaboration of Personal Construct PsychologyBeverly M. Walker and David A. Winter � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 453

Cross-Country or Regional Comparisons

Cross-Cultural Organizational BehaviorMichele J. Gelfand, Miriam Erez, and Zeynep Aycan � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 479

Organizational Groups and Teams

Work Group DiversityDaan van Knippenberg and Michaéla C. Schippers � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 515

Career Development and Counseling

Work and Vocational Psychology: Theory, Research,and ApplicationsNadya A. Fouad � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 543

Adjustment to Chronic Diseases and Terminal Illness

Health Psychology: Psychological Adjustmentto Chronic DiseaseAnnette L. Stanton, Tracey A. Revenson, and Howard Tennen � � � � � � � � � � � � � � � � � � � � � � � � � � � 565

Contents ix

Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.


Research Methodology

Mediation AnalysisDavid P. MacKinnon, Amanda J. Fairchild, and Matthew S. Fritz � � � � � � � � � � � � � � � � � � � � � 593

Analysis of Nonlinear Patterns of Change with Random CoefficientModelsRobert Cudeck and Jeffrey R. Harring � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 615

Indexes

Cumulative Index of Contributing Authors, Volumes 48–58 � � � � � � � � � � � � � � � � � � � � � � � � � � � 639

Cumulative Index of Chapter Titles, Volumes 48–58 � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 644

Errata

An online log of corrections to Annual Review of Psychology chapters (if any, 1997 to thepresent) may be found at http://psych.annualreviews.org/errata.shtml

x Contents

Ann

u. R

ev. P

sych

ol. 2

007.

58:9

7-11

8. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by I

ND

IAN

A U

NIV

ER

SIT

Y -

Blo

omin

gton

on

01/1

4/10

. For

per

sona

l use

onl

y.

Causal Cognition in Human and Nonhuman Animals: A ...

Documents

Transcript of Causal Cognition in Human and Nonhuman Animals: A ...