http://psr.sagepub.com

Personality and Social Psychology Review

DOI: 10.1207/s15327957pspr0802_11 2004; 8; 173 Pers Soc Psychol Rev

William J. McGuire A Perspectivist Approach to Theory Construction

http://psr.sagepub.com/cgi/content/abstract/8/2/173 The online version of this article can be found at:

Published by:

http://www.sagepublications.com

On behalf of:

Society for Personality and Social Psychology, Inc.

can be found at:Personality and Social Psychology Review Additional services and information for

http://psr.sagepub.com/cgi/alerts Email Alerts:

http://psr.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

Personality and Social Psychology Review2004, Vol. 8, No. 2, 173-182

Copyright 2004 byLawrence Erlbaum Associates, Inc.

A Perspectivist Approach to Theory Construction

William J. McGuireDepartment ofPsychology

Yale University

A perspectivist approach is taken to the theory-construction process in psychologicalresearch. This approach assumes that all hypotheses and theories are true, as all arefalse, depending on the perspectivefrom which they are viewed, and that the purposeof research is to discover which are the crucial perspectives. Perspectivism assumes

also that both the a priori conceptualphase ofresearch and the a posteriori empiricalphase have both discovery and testing functions. Topics discussed include how theperspectivist approach can improve methodology training and practice (particularlyas regards theory construction); what researchers accept as theoretical explanations;the nature ofmediational theories; how theories can beformalized, expressed in mul-

tiple modalities andfor various scaling cases; and how experimental designs can beenriched by theory-guided mediational and interactional variables.

Psychologists have co-opted as their just demesne a

fascinating range of topics for study, but the yield oftheir studies often falls disappointingly short of theirpromise, evoking worry that psychologists possess a

Sadim (reverse Midas) touch such that every hypothe-sis we touch turns to dross. Part of the problem is thatwe fail to create explanatory theories that are as inter-esting as the topics they are supposed to explain. Here Ipropose improvements in our methods and substancecourses that would empower psychologists to con-

struct nonobvious, even counterintuitive, theories thatare as fascinating as the topics that earlier enticed us

into psychology.

Scope of a Perspectivist Approach toTheory Construction in Psychology

A perspectivist reconceptualization of methodol-ogy (McGuire, 1989, 1999) is proposed here betterto realize the creative potential of one's theorizing.Beginning with definitions of basic concepts such as

variables, hypotheses, and theories, I describe cre-

ative theorizing processes, especially mediationaltheories, as regards their ubiquity, logical structure,subtypes, and alternatives. I discuss theory desider-ata that are often ignored and even deplored, such as

the usefulness of formalizing theories and of ex-

pressing them in multiple modalities and for diversecases as regards scaling of their variables. Finally, Idiscuss how a perspectivist approach can guide our

theory construction to add interesting mediationaland interactional variables to one's experimental de-sign before one begins the labor-intensive empiricalinvestigation of one's hypotheses and their theoreti-cal explanation.

The perspectivist approach re-establishes as a re-

searcher's main responsibility and opportunity the cre-

ation rather than the testing of theory. I describe howone's research can be done not only better but more

joyously, theory creation being the ultimate pleasure ina life in science. As Gerard Manley Hopkins(1918/1998) in his sonnet "To R. B." explains to RobertBridges that what sustained him in his often grim cre-

ation was the roll, the rise, the carol, the creation thatwarmed his winter world. We need this, he wrote, theone rapture of an inspiration, the fine delight that fa-thers thought ... and leaves yet the mind a mother ofimmortal song.

The widow of an insight lost she lives, with aimNow known and hand at work now never wrong.

(Hopkins, 1918/1998)

Miss this glorious undertext and one misses my mes-

sage. By exploiting perspectivism, one can make re-

search a class act.

Some Basic Concepts: Variables,Hypotheses, and Theories

A major purpose of psychological research is togenerate and evaluate knowledge representations,usually formalized verbally as propositions (hypothe-ses, predictions) that state relations between variablesof interest. Psychological variables are aspects (in-

173

Reprint requests should be sent to William J. McGuire, Depart-ment of Psychology, Yale University, P.O. Box 208205; New Haven,CT 06520-8205. E-mail: william.mcguire@yale.edu

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

McGUIRE

cluding thoughts, feelings, and actions) on whichpeople differ from one another. Our usual simplemain-effect monotonic hypotheses assert a positive ornegative (direct or inverse) relation between an inde-pendent variable, IV, from which we are predictingand a dependent variable, DV, to which we are pre-dicting, DV = f (IV). Variables are related to the ex-tent that knowing where people are located on the IVallows us to improve, by more than a chance amount,the accuracy of our predictions of where these peopleare located on the DV. Other types of variables besideIV and DV to be considered later are mediating vari-ables (MV), interactional variables (jIV), control vari-ables (CV), exploratory variables (EV), and serendip-itous variables (SV).

Variables of each type, which together make up themolecules of one's knowledge, can usefully be given aconceptual definition and an operational definition.For example, if one is studying aggressiveness, onemight begin developing a conceptual definition as "atendency deliberately to hurt another person," and onemight begin an operational definition in terms of somespecific set of observable operations of a person (theparticipant) in a designated situation (e.g., how manytimes an angered preschooler kicks an inflated Bobodoll). Conceptual definitions tend to be more useful forgenerating knowledge (hypotheses, theory) and opera-tional definitions tend to be more useful for evaluatingthis knowledge.A main-effect hypothesis is a proposition that

states a predicted relation between variables, usuallyhaving as its logical structure IV -+'--> DV. A hypoth-esis uses knowledge (a) of where a sample of peoplefall on the IV and (b) of a theorized relation betweenIV and DV to predict (c) where these people fall onthe DV. A theory is a proposition or a set of proposi-tions that explain (account for) such a hypothesizedrelation. Definitions and labels other than those previ-ously mentioned for variable, hypothesis, and theoryare tenable (and some alternatives are touched onlater), but these definitions are powerful and widelyheld at least implicitly.

Specific characteristics on which people vary (ag-gressiveness, helpfulness, intelligence, exposure tothe mass media, popularity, political attitudes, andso on) can fall into any of these types of variables(e.g., IV, DV, MV, J1V, CV, EV, SV) by virtue ofwhere they enter a specific argument under study.For example, aggression may be an IV in one hy-pothesis (argument), as when we are predicting howpersons' levels of aggression will affect how likedthe person will be (Agg -+-> Lik); in another argu-ment aggression may be a DV, as when we are pre-dicting how exposure to televised violence will af-fect viewer aggressiveness (TV -+-> Agg). The IV isoften but not always the cause and the DV the effectin the hypothetical relation.

Assembling the Variables on WhichOne's Research Program Will Focus

Definitions and terminology like those describedpreviously can be developed to describe rigorously andmanageably how psychologists (and other scientists)can create knowledge. First, he or she selects andmakes explicit to self and intended audience the basictopic of the inquiry. This is usually the DV (and some-times the IV) point of departure for a hypothesis (i.e.,for a predicted relation between variables) and for itsexplanatory theory. For example, if the researcher is apersonality theorist or a student of human conflict, heor she might start by choosing aggressiveness as thebasic variable to be studied. Second, if media oriented,the theorist might generate the hypothesis that expo-sure to televised violence will increase people's levelsof aggressiveness (IVt -+-> DVa). Third, he or shemight generate a mediating theory (IV -* MV1 -* DV)to explain, or account for, this hypothesized relation.

Ordinarily the researcher should cast a wider net toinclude a broader range of variables than these three inthe experimental design before proceeding to an em-pirical investigation. However, a researcher might de-cide at this point (e.g., for pedagogic purposes) tomove somewhat prematurely into data collection anddata analysis. If so, the researcher would measure arepresentative sample of N participants on the threevariables, IVt, DVa, and MV1 (amount of exposure totelevised violence, level of aggressive behavior, anddegree to which aggression is perceived by the viewerto be legitimate), each variable measured on a multi-level scale. These 3 x N data points would then be sub-jected to at least four analyses: (a) an IVt -+-> DVacorrelational analysis, rt, a, could evaluate the initial hy-pothesis, that the more violence seen on TV the moreaggressively the participants will behave; (b) two fur-ther analyses, rt, 1 and rl, a, could evaluate the two pre-mises that constituted the mediational theory, IVt -+->MV1 and MV1 -+-> DVa; (c) a partial correlational anal-ysis, rt,a.1, could then be done to estimate how much ofthe initial-hypothesis covariance may be attributed tothe theorized MV1 mediator, rt,a - rt,a.1; (d) an analysisof how much of the initial hypothesis IVt -+-> DVacovariance remains even after this legitimization theo-retical mediator, MV1, is partialled out, rt, a.l, would in-dicate how much the initial hypothesized relation ob-tains for other reasons in addition to legitimizationtheory.

A Popular Form of Explanation:Mediational Theories

When the researcher has formulated an IV -+' ->DV hypothesized relation (e.g., televised violence -f->viewer aggression), he or she is usually expected, be-

174

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

PERSPECTIVIST APPROACH TO THEORY CONSTRUCTION

fore going to a test of this prediction, to generate one ormore theories that could account for the hypothesis.For example, the researcher might theorize that this TV-> Agg relation is accounted for (in part, at least) asdue to a legitimization process. That is, the more vio-lence persons are exposed to in television programs themore legitimate they perceive it to be to behave vio-lently and therefore the more aggressively they them-selves behave. This perceived-legitimacy-as-mediatortheoretical explanation of the hypothesis can be ex-pressed as a chain of variables, IVt -+-> MVI - +-> DVa.This explanation can alternatively be formalized syllo-gistically as [(IVt-+-> MVI) & (MV1 -+-> DVa)] -> (IVt-+-> DVa), where IVt is the independent variable(amount of violent television viewed), MVI is the theo-rized mediating variable (how legitimate aggression isperceived to be), and DVa is the dependent variable(how aggressively the viewer behaves). Current ortho-doxy is that experiments should be guided by and de-signed to test (or at least investigate) not only an ex-plicit a priori hypothesis but should be preceded by aspecified a priori theoretical explanation, IVt -+-> MV1-+-> DVa, from which the hypothesized relation, IVt-+->DVa, can be derived and accounted for. This apo-theosis of theory in the psychological Establishmentgoes back at least to the mid-20th century when the in-tellectual migration of the Wienerkreis popularizedlogical empiricism in American psychology via H.Feigl at Minnesota, G. Bergmann at Iowa, and C.G.Hempel at Yale and Princeton.

Ubiquity of Mediational Theories

Mediational IV -* MV -> DV theories are not theonly type of theoretical explanation, but I focus heavilyon them here because I have found that in practice psy-chologists usually generate theories that have thismediational logical structure. This ubiquity is illus-trated when I use perspectivist worksheet exercises(McGuire, 1989, 2004b) to guide research studentssuccessively from the notion of people's scores onvariables, through hypothetical relations between peo-ple's scores on the variables, to theoretical explana-tions of the hypothesized relations. Specifically, stu-dents are asked on my worksheets to choose someinteresting variable on which people differ (e.g., ag-gression, altruism, depression, creativity, and so on),then to generate a second variable that they conjectureto be interestingly related to the first variable (e.g., sex,mass media exposure, sibling order, anxiety, or what-ever), and then to hypothesize a monotonic relation be-tween the two variables. This yields a hypothesis in theform IV -+I--> DV. Each student is led through this hy-pothesis-generating procedure reiteratively to producea half-dozen varied hypotheses. Then the student isasked to write down several theoretical explanationsthat could account for each of the half-dozen hypothe-

sized relations. Hence, the 25 or so students in each se-mester's research class together come up with severalhundred theoretical explanations for a wide variety ofself-generated hypotheses. During the past decadethese methodology students of mine have togethercome up with several thousand explanatory theories. Alogical analysis of these thousands indicates that wellover three quarters of them have the logical structure ofmediational theories.

The perspectivist innovation that all hypotheses,even pairs of contraries, are true (at least from someperspective) leads to further worksheet exercises thatguide the student to assert the contrary proposition tohis or her initial hypothesis and to generate two ormore theoretical explanations of each of these contrarypropositions. Again, well over three quarters of the the-oretical explanations given for these contrary hypothe-ses also fit a mediational-theory logical structure.

For example, if the worksheet asks a student toname a psychological variable of interest, he or shemight suggest "aggression" or "antisocial behavior,"and when asked to name a related variable, he or shemight respond with "exposure to televised violence."Asked further to hypothesize a monotonic relation be-tween the two variables, he or she tends to say, "Themore televised violence the person is exposed to (IVt)the more (-+->) aggressively he or she behaves (DVa),"hypothesizing a monotonic positive relation. Then,when asked to give a theoretical explanation that mightaccount for this lVt -+-> DVa hypothesized relation, thestudent in more than three quarters of the cases gives amediational theory as explanation, for example, "Be-cause seeing all that violence makes the student think itis legitimate to aggress" (MVI), or "Because seeing allthat violence gets a person excited" (MVe), or someother explanation that similarly can be spelled out toapproximate the logical form of a mediational theory.

Logical Structure of MediationalTheories (and, More Broadly, What Is

It "To Explain"?)

Three typical, logically equivalent expressions for amediational theory are

1. IVt -+-> DVa .-MV1: The more televised vio-lence people watch (IVt), the more (-f->) aggressivelythey behave (DVa) because (-. ) viewing all that violencemakes one feel that it is legitimate to aggress (MVI).

2. IVt -+-> MVI -+-> DVa: = The more televised vi-olence people are exposed to, the more legitimate theyperceive aggression to be, and so the more aggressivelythey behave.

3. [(IVt -+-> MVI) & (MV, -+-> DVa)] * (IVt-+->DVa): = The more televised violence people watch,the more legitimate they feel it is to aggress; also, the

175

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

McGUIRE

more legitimate they feel it is to aggress, the more ag-gressively they behave; therefore, the more televisedviolence people are exposed to, the more aggressivelythey behave.

These three explanations are logically equivalent.Number 3 is the full, formally correct syllogisticmediational theory, which few research students (ex-cept some trained in formal logic) spell out com-pletely. Most students verbalize their mediational ex-planation in an abridged enthymemic form such asNumber 1 or 2, omitting one of the premises andleaving it as understood.

That mediational theories (or at least close approxi-mations of them) are so popular among psychology re-searchers as a way of explaining the world as they find itsurprises me, because mediational theories have onlymodest explanatory power. When one explains a hy-pothesis (IVt -+-> DVa)by amediational theory (1Vt -+- >MV, -+-> DVa), little additional predictive clarificationor insight is provided except that the MV1 serves as astepping stone, allowing one's thinking to get from IVtto DVa in two easy baby steps instead ofone bigger step.More descriptive work is needed on the logical structureof mediational and other types of explanatory theoriesand on the nature of explanation in general. Explainingexplanation is a timely challenge (Doise, 1997;McGuire, 2004a; Ruben, 1990, 1993).

A Variety of Mediational Theories

Mediational theories can be elaborated in severalways beyond the simple one-MV type mentioned pre-viously, IV -X MV, -> DV. A common polysyllogisticlinear elaboration lengthens the chain, IV -* MV, -MV2 -4 MV3 -e DV. The additional MVs may beadded between IV and MV1, or between MVI and DV,or both. We can add depth to a mediational/syllogisticexplanation, alternatively, by deriving syllogisticallyeach of the two premises of our first-order mediationaltheory: [(IVt -+- > MVo) & (MVo -+-> MVI)] -* (1Vt-+-> MV1). A more dramatic perspectivist elaborationof mediational theories is to generate multiple-pathmediational theories that allow additional causal pathsfrom IVt to DVa. McGuire (1997) described 49 creativeheuristics to aid the researcher to generate numerousmediating theories to explain a given hypothesis.

Here the hypothesis IVt -+-> DVa (the more televi-sion violence to which one is exposed the more aggres-sively one behaves) can be theorized to be due to eitheror both MV1 (e.g., the viewer's cognition that it is legit-imate to behave aggressively) or by MVe (e.g., theviewer's affect, such as feeling excited). There are of-ten direct as well as mediated paths between IV andDV (e.g., attitudes may affect behavior both directly

Alternatives to Mediational Theories:Systems and Set-Inclusionist Theories

As the number of mediators increases and interrela-tions emerge among the mediators themselves, the the-orist may advance from the unidirectional linearmediational theories so far considered to more com-plex systems theories (e.g., with bidirectional rela-tions, alternative pathways, and so on, among vari-ables), which are more complex both tactically andstrategically, for example, by calling for collectingtime-series data and analyzing by causal modeling.

Had the research students whom I asked to carry outmy perspectivist worksheet exercises lived in the clas-sical or medieval eras rather than in the modern era (i.e.in the Aristotelian rather than the Galilean era, to useKurt Lewin's terms), most of the students might havegenerated set-inclusionist theories rather than themediational theories generated by my 20th and 21stcenturies students. That is, asked to explain the deathof Socrates, they might theorize, "Because he is hu-man" or, more fully and explicitly, "All humans aremortal; and Socrates belongs to the class of humans;therefore Socrates is mortal." Such set-inclusionist the-ories are even more limited than mediational theories,but they continue to be popular in philosophical andother nonempirical disciplines that still send their stu-dents out hunting for black swans.

On Advantages of Formalizing One'sTheories

Current research students usually offer mediationaltheories to explain a hypothesis but express this theoryin an informal manner, often leaving one theoreticalpremise implicit so the argument is given as anenthymeme rather than a complete syllogism. Myworksheets encourage students to go beyond suchabridged statements of mediational theories by statingtheir explanatory argument in full and explicit syllogis-tic form. This may seem pedantic, because it is com-monly assumed that the essence of scientific or mathe-matical creativity lies in one's initial cryptic andinformal "eureka!" discovery of the gist of the insightand that the subsequent working out of the formal proofisjust an uncreative tidying up for publication ofthe ini-tial vague insight. However, I find that in empirical theo-rizing the transformation of the initial gist of themediational theory into a full formal syllogistic argu-ment facilitates the theorist's exploiting the creative po-tential ofthe explanation. This can be illustratedby putt-ing the MV1 explanation into valid syllogistic form:

Minor premise. The more televised violence thatpeople view (1Vt), the more legitimate they per-ceive it to be to behave aggressively (MV1); that

and via a behavioral-intention mediated link).

176

is, 1vt -+->MVI.

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

PERSPECTIVIST APPROACH TO THEORY CONSTRUCTION

Major premise. The more legitimate theviewer perceives aggression to be (MV1), themore aggressively he or she behaves (DVa); thatis, MVi -+-> DVa.

.-. Conclusion (and initial hypothesis). There-fore, the more televised violence that peopleview (IVt), the more aggressively they behave(DVa); that is, IVt -+-> DVa.

In this syllogistic formalization, the conclusion is the re-searcher's initial hypothesis, whereas the minor and ma-jor premises are the researcher's mediational explana-tory theory to account for this initial hypothesis. Layingout the theory in this explicit syllogistic form facilitatesthe researcher's performing elegantly at least nine cre-atively demanding processes:

1. It facilitates doing a validity check on the vari-ables by checking if the MV is stated essentially thesame way in both minor and major premises, if the IVis stated the same way in both the minor premise andthe conclusion, and if the DV is stated the same way inboth the major premise and the conclusion.

2. It guides doing a validity check on the relationsby checking if the two premises have the same sign(both asserting direct or both inverse relations), thenthe sign of the conclusion's relation should be a plus;and if the two monotonic premises have the oppositesigns (one plus, the other minus), then the conclusionshould have a negative relation.

3. It detects tautology by checking if the MV is freefrom conceptual overlap with the IV and with the DV.Overlap is a sign of tautology by revealing that the pur-ported theory repeats the hypothesis rather than explains it.

4. It allows an a priori diagnostic plausibility check.How a priori compelling (obvious) is the major versus theminor premise? Differential plausibility identifies possi-ble weak links (premises) in the theory or alternativelyidentifies which is the desirably nonobvious premise.

5. It allows elaboration, clarification, and apprecia-tion of the variables in hypothesis and theory by allow-ing linguistic exploration of IV, MV, and DV by play-ing word games (McGuire, 1989) that suggest, forexample, where one might usefully partition a variableinto subvariables that relate differently to other vari-ables in one's experimental design.

6. It permits a check of the independence of alter-native theories purporting to give different explana-tions of the initial IV -+-> DV hypothesis by a pinpoint-ing analysis of whether MV, . MV2.

7. Particularly enriching, each premise suggests(especially with the help of McGuire's, 1997, 49 cre-ative heuristics) multiple situational and dispositionalinteractional hypotheses, interesting in their own right,and clarifying the theory. The formalization suggests atleast four types of interaction variables (each type in-cluding numerous variables) as follow:

a. Multiple dispositional (across-personal)interactional variables predicted to intensifythe minor premise.

[iIVd x (IVt e> MV01) [iIVd x (IVt -> DVa)]b. Multiple dispositional interactional variables

predicted to intensify the major premise.[iIVD X (MV1 DVa)] [iIVD X (JVt1 DVa)]

c. Multiple situational (across-conditions)interactional variables predicted to intensifythe minor premise.

[iIVs x (IVt MVI)] - [iIVs x (IVt DVa)]d. Multiple situational interactional variables pre-

dicted to intensify the major premise.[iIVs x (MV1 DVa)] [iIVs x (IVt DVa)]

8. It analyzes each mediational theory into two ormore premises, which allows a posteriori diagnosticempirical tests of the weaknesses and strengths of thecomponent propositions of the overall theory. HenceNumber 8, this a posteriori plausibility check supple-ments the a priori Number 4 plausibility conceptualcheck.

9. This formalization allows grounding one's the-ory in depth by deriving each premise as a conclusionexplained by its own mediational syllogistic theory.That is, it suggests how each premise of one's initialtheory can itself be explained as the conclusion of anantecedent level of theory.

For example, [(IVt -+-> MVO) & (MVO -+-> MV1)](IVt -+-> MVI)

The research student can be given worksheet exer-cises (McGuire, 2004b) that guide him or her throughsuch utilizations of theory-formalization until he or shecan gain facility in creatively exploiting them, possiblywith the help of creative heuristics such as the 49 de-scribed in McGuire (1997).

The elegant symbolism and terminology we haveworked out and the procedures we have expressed ex-plicitly and formally here are neither pedanticformalisms nor distractions but rather facilitate the re-searcher's creativity. Several other procedures recom-mended in this article similarly illustrate how formaliz-ing a variety of complex procedures also allows theresearcher to grapple vigorously and to set prioritieswithin daringly complicated experimental designs (e.g.,see Figures 1 and 2). Elegant formalization does not sti-fle or replace creativity but rather releases it for use ingrappling with higher, not-yet-formalizable tasks.

On Advantages of Expressing One'sTheories in Multiple Modalities

So far I have discussed constructing and expressingtheories verbally, in natural language (except that I oc-casionally repeat or abbreviate the natural language ex-

177

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

Levels of scaling ofthe Independent Variable (IV,) and Dependent Variable (DV,)

Modality of a IV, Continuous (Multilevel) b. IV, (lDi)chotomous c. IV, (Di)chotomousExpression DV. Continuous (Multilevel) DV, Continuous (Multilevel) DV. (Di)chotomous

1. Verbal The more tv violence to which People with above-average ex- People with above-average exposure(Natural people are exposed, the more posure to tv violence behave to tv violence are more likely toSpeech) aggressive their behavior more aggressively than those aggress than people with below-

_with below-average exposure averase exposure

IV, --'--> DV,; or Ab < A, (IVbDV, + IVDVb) < (IVbDVb + IV.DV,)2. Abstract DV. = f' IV, where A = mean aggression

Symbolic a = above-average tv exposureb = below-average tv exposure

IV,:tv Below AboveDV, D xposure Average Average

3. Pictorial DV, (Aggres- Ag-(Graphical) (aggression : sion)

Yes ny ny0 1 23 ... below above

IV, (TV violence exposure) average average No nbN non nNIV, (tv violence)

S IVt DV IV, (tv violence) (T. AN + Tb AY) < (Tb AN + T. AY)1 t, a, below above 1 1 1 1

4. Tabular 2 t1 a2 avY2rge t averagetv 1 1 1 13 t3: a3 A, A, 1 1 1 1

A, A, I IA4 1 1

n to a. Lb < A. nN nby nbN ny

Coeff. of Correlation (r) Difference between Means Coeff. of Contingency (C)5. Descriptive o < r,

Statistics .A

6. InferentialStatistics t, ANOVA Chi-Square

Figure 1. Six alternative modalities for expressing the relation in the initial, main-effect hypothesis (IVt-+-> DMO), shown for threecommon cases as regards scaling the IV and DV variables.

Type Mediational Hypotheses, Theories ("because") Interactional Hypotheses ("especially when")HypothesisModality of a IV, MV, and DV all b. (Dichotomous IV; c. Continuous (Multilevel) d. Di)chotomous IV, ,IVExpression Continuous (Multilevel) Continuous (Multilevel) IV, DV; (Di)chotomous JIV Continuous (Multilevel)

MV. DV DV1. Verbal "The more tv violence people "People with above-average "People's aggression goes up "People who view more(Natural are exposed to, the more tv violence exposure perceive with exposure to tv violence than average tv violenceSpeech) legitimate they perceive aggression as more legitimate especially when good guys behave more aggressively

aggression as being, and so than do people with below- commit the violence on tv." than those who view lessthe more aggressively they average exposure and so than average tv violence,behave." behave more aggressively." especially if the good guys

perpetrate the violence."2. Abstract IVt -+-> MV, -'->DV, [(Lb < L,) & (MV, -+->DV)] LIVb (IV, -`->DV,)] < LIVb (DVb < DV,)] <Symbolic or IV, -+->MV, -> (T, < T) LIV, (IV, -a->DV) JIV, (DVb <DV]

MV, -+->DV,IV. -"-DV.

V X DV WMV = DV'

Ii D( V IV3. Pictorial 0~ s.0s IV"'T MV (L) Wvw DV(Graphical) IV, MV (legit) (W.-)

- DV CV N IV(tvviolence) W, l ll_

(A IV (atv9V c A a

O 1 2 ,, ~~~~~~~Tb POP-"p (bed) <(good)IV(tv) IV (tv)

S IX, M, V IV (tv violence) abed [44 1IVb badgg. gQrooguI T, L, A, Below Ag AboveAvg. S IV V IV DV IV, tvi Att, b-tI

4. Tabular 2 T2 L2 A2 M- DV EY V 2 t2 a2 I t, a, a3 a4 a2 a,IL IA 3L 3A 5 t, a, 3 t3 al a4 a, a2 a32L 2A 4L 4A 4 t4a4

5L 5A :N TN LN AN, badrM goodr,, A.AAbeN A,,

f& EA.

5.Descrip- Partial Coeff. of [adjL(Ab < A,)] < bad r,. < good r, Diff. Between Diff. intive Correlation vs. [unadjL(A, < A,)] MeansStatistical first order corr. (Diff. between adjusted [bad guys (Ab < A,)] <

r, .,, < r,. vs. unadjusted means) [good guys (Ab < A,)]

6. Inferential tr - (CANOVA with MV, as t, - ANOVA (m x n intraction)Statistical the covariate

Figure 2. Six alternative modalities for expressing mediational theories and interactional hypotheses, each shown for a pair ofcommonscaling cases.

178

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

PERSPECTIVIST APPROACH TO THEORY CONSTRUCTION

pression in abstract symbols). And why not? As T. S.Eliot remarked, we have to use words when we talk.However, although one usually expresses one'sthoughts in natural language, it is enabling to recognizethat there are additional, sometimes more powerfulmodalities for expressing and constructing one's theo-ries. For specifiable purposes, it may be advantageousto use other-than-verbal modalities, so researchersshould be able to switch among modalities. When Iteach methods, I try to enhance the students' power andcomfort in constructing and expressing their thinkingin each of six different modalities: (a) verbal (or naturallanguage), (b) abstract symbolic, (c) pictorial (orgraphic), (d) tabular, (e) descriptive statistical, and (f)inferential statistical. These six constitute the rowheadings of Figures 1 and 2.

Skill in expressing one's theory in multiple modali-ties is advocated here, not to be pedantic or to engage inshow-off tours de force, but because multimodal ex-pressions enrich one's grasp of the theory and increaseone's likelihood of noticing its implications, of recog-nizing its similarities and contrasts with other formula-tions, of detecting its gaps and weaknesses. Also, re-searchers vary among themselves in regard to themodality in which they have the greatest personal cog-nitive strength. For example, researchers differ in pref-erence for geometric versus algebraic styles of thoughtexpression, with the geometric stylists thinking moreeffectively when they express their theory in the picto-rial (e.g., graphical or flow chart) modality, whereas al-gebraic stylists think more powerfully in the abstractsymbolic modality. The neophyte (or even the old prowhen dealing with a new topic) may think better in thenatural speech modality, but with more familiar topicsthe theorist may think better in the pictorial modality.For working out one's experimental design, the ab-stract symbolism modality may be more helpful, butfor working out the statistical analysis, the tabular mo-dality may be more facilitating. A given researchermay find the verbal expression more facilitating whenengaged in a priori extending or qualifying his or hertheory but may find the tabular expression more usefulwhen deciding on an appropriate experimental design.

The cost/utility of the various modalities for ex-pressing a theory depends on the researcher's personaldispositions and on the situational circumstances of theresearch problem. One should generally choose themodalities that fit one's strengths, but there will be oc-casions for venturing on roads less traveled. Hence weshould help our students to develop some facility intranslating theories from one to another modality. Tothis end, our methods courses (and to some extent oursubstantive courses also) should (a) convey the conceptof diverse modalities of theory expression, (b) describea variety of contrasting modalities such as the six de-scribed previously, and (c) discuss comparative advan-tages of various modalities.

On Advantages of Expressing One'sTheories for Diverse Scaling Cases

Another dimension of complexity (and of opportu-nity) in theory expression has to do with how the vari-ables in our hypotheses and theories are scaled. We ex-pect the relation between our variables (e.g., IVt -+->DVa) to remain constant in principle across alternativescalings of the related variables, but the expressions ofthese relations often look dramatically different de-pending on how the variables are scaled, thus present-ing alternative creative challenges and provocations.The student can be sensitized to scaling effects and op-portunities by contrasting the expression of a relationwhen its variables are given maximally differentscales, for example, by contrasting dichotomous ver-sus continuous (or at least multilevel) scaling of thevariables involved in the relation. Such contrasts arehighlighted by comparing adjacent column entries inFigure 1 and 2. A pedagogically powerful device forempowering the research student to make creative useof both the modality and the scaling of the variables inone's hypothesis and theories is to present matricessuch as Figures 1 and Figure 2 whose row headings arethe six modalities and whose column headings are di-chotomous versus multilevel scaling of the variables.(The columns here differ also in the logical structure ofthe propositions being expressed.) In a methodscourse, the cells in these matrices would be left emptyto be filled in by the student as an exercise, but here thecells are filled in for expository purposes.

Figure 1 expresses the researcher's initial main-ef-fect hypothesis (e.g., IVt -+-> DVa; the greater expo-sure to TV violence, the more aggressive the viewer'sbehavior will be) in each of the six row modalities. Thethree columns (a, b, and c) in Figure 1 show thismain-effect hypothesis as expressed for each of itsthree most common scaling cases. Column A ex-presses this IVt -+-> DVa relation in all six modalitieswhen both IVt and DVa are scaled continuously (or atleast on multilevels); Column B expresses this relationwhen IVt is scaled dichotomously and DVa is multi-level scaled; and Column C expresses this relationwhen both IVt and DVa are scaled dichotomously.

Figure 2 presents two similar matrices with ourusual 6 row modalities x 2 column scalings for the twoother most popular logical types of hypotheses in-volved in theory construction, namely, mediationaltheories and interactional propositions. The twoleftmost columns in Figure 2 express the student'smediational theory (IVt -+ MVI -e DVa) in all six mo-dalities for the two most common scaling cases,namely the case in which all three variables, IVW, MV1,DVa, are multilevel scaled (in the leftmost Column A),and for the case in which both MV1 and DVa are multi-level and IVt is dichotomous (in the second from leftColumn B). The two rightmost Columns C and D of

179

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

McGUIRE

the four columns in Figure 2 yield a matrix that pro-vides the student with practice in the 6-row modalitiesx 2-column scalings for the third type of logical propo-sitions that enter theorizing, namely, interaction hy-potheses, iIV x (IVt -+I--> DVa). Column C expressesthe interaction relations for the scaling case in whichIV and DV are scaled on multilevels and J1V is scaleddichotomously. Column D expresses the interactionalrelation for the scaling case in which DV is multipliedand IV and J1V are dichotomous.

Together Figures 1 and 2, by laying out systemati-cally the most common logical structure, modality, andscaling options in theory construction and expression,familiarize the research student with most frequentstyles of theorizing with which he or she will be wres-tling. When the cells in matrices like those in Figures 1and 2 are left blank, the exercise of filling them in pro-vides the student with tailored practice in translatinghis or her own theories into these different formats.With practice, the unique creative potentials of eachcell in the matrices becomes comfortably accessible tothe student. Space limitation here restricts us to a fewexamples of how working with Figures 1 and 2 matri-ces can advance one's theory-construction powers, butfor more details see McGuire (1989, 1999, 2004b).

Theoretical Elaborations of One'sExperimental Design Prior to

Beginning the Empirical Investigation

The most sophisticated and popular psychology ofscience prior to perspectivism, the Wienerkreis logicalempiricism movement, prescribed that the researchshould be deductive in using conceptual analysis to de-rive a hypothesis (and a theory from which this hypoth-esis could be derived) before setting out inductively onthe labor-intensive empirical investigation.Perspectivism is even more demanding regarding thescope of the a priori conceptualization. It prescribesthat before collecting data on the relations the re-searcher should consider explicitly not only the initialhypothesis but also its contrary, and not only one ex-planation but several explanations of each hypothesis,both the initial hypothesis and its contrary. Thus the re-searcher should elaborate the experimental design ofhis or her first empirical study by selecting additionalvariables sufficient to investigate both the initial hy-pothesis and its contrary and at least two distinct theo-ries to explain each of these hypotheses. Typically thismore elaborate experimental design would involveadding interactional hypotheses suggested uniquely byeach theoretical explanation (asserting the hypothesiswould obtain especially if ...) and adding theoreticalmediational explanations (asserting that the hypothesiswould obtain because ...).

Using A Priori Theories to GenerateInteractional Variables, "Especially If..."

Interactional theoretical elaborations stay withinone's initial mediational explanation but exploit itsanalysis into minor and major premises by using eachpremise as a springboard for generating additionalinteractional hypotheses (perhaps by using some ofMcGuire's, 1997, 49 creative heuristics). For example,using legitimization theory, MV1, to account for the hy-pothesized aggression-enhancing effect of exposure totelevised violence, IVt -+-> MV1 -+-> DVa, one formal-izes the theory into two syllogistic premises, the minor,IVt -+-> MVi (the more violence people are exposed toon television, the more legitimate they perceive the useof aggression) and the major, MVI -+-> DVa premise(the more legitimate people perceive the use of aggres-sion, the more aggressive their behavior tends to be).Each of these two premises suggests interaction vari-ables that multiply the premise relation and thereforethe conclusion relation, which is the initially predictedhypothesis, IVt -+-> DVa. This helps the researcher touse each premise to recognize several interestinginteractional variables for adding to the empirical de-sign. For example, an interactional variable iIVg is sug-gested by the minor premise of the legitimization the-ory, which can be expressed symbolically as iIVg x (IVt-+-> DVa). That is, the more televised violence towhich viewers are exposed, the more legitimate theviewers perceive the use of aggression to be, especiallyif the violence is depicted as perpetrated by the goodguys. Note that logically if the interacting variable,iIVg (perceived goodness of the depicted violence per-petrator), multiplies the relation in a premise, it alsomultiplies (although with probabilistically diminishedforce) the relation specified in the conclusion (that is,the relation predicted in one's initial hypothesis). Eachpremise of the explanation suggests numerous situa-tional and dispositional interaction effects, interestingin their own right and clarifying the meaning of the ex-planatory theory. Several particularly deserving ofthese interaction variables from each theory should beadded to the experimental design.

Using A Priori Theories to GenerateMediational Variables, "Because ..."

A more dramatic theoretical elaboration is the useof multiple (mediational) theories to account for theinitial IVt -+-> DVa hypothesis. One of the striking pos-tulates in my perspectivist epistemology is that all hy-potheses are true. That is, all relations among vari-ables, even mutually contradictory relations, obtain (atleast in some contexts, from some perspectives) andeach of the hypothetical relations obtains for multipletheoretical reasons, as does the contrary relation(McGuire, 1989, 1999). In terms of William Blake's

180 © 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution.

by on November 16, 2007 http://psr.sagepub.comDownloaded from

PERSPECTIVIST APPROACH TO THEORY CONSTRUCTION

\Causal Amount of Violence Viewed Affects ->Co Direc- Amount ofAggressive Behaviorvaria- ionation Mediational Advocating InteDirection Theory (MV) Theorist Var

Positive

(Direct)

Co-

varia-

tion

Negative

(Inverse)5. Catharsis(hydrolic,vicarious)

QUADRANT BAristotleS. FreudS. Feshbach

alternative outletfantasy life

Amount of Aggressive Behavior Affects ->Amount ofViolence Viewed

eractionalriable Implication (CI)

MediationalTheory (MV)

Advocating InteractionalTheorists Variable Implication (jIV)

QUADRANT C7. Ostracism R. Huesmann number of sibs

8. Predilection A. Fenigstein control over program

QUADRANT D9. Conventionality R. Jessor entertainment value

Co-6. Time pre-

variation emptionJ. Robinson active in sports

Figure 3. Four hypothesized directions of relations between amount of exposure to televised violence (TVt) and amount of aggresivebehavior by the viewer (DV.), each as accounted for by any of several mediational theories, each theory advocated by multiple theorists,and each theory implying several multiplying interactional variables.

Proverbs of Hell, "Everything possible to be believedis an image of truth." Hence, whenever the researcherundertakes to predict and account for (explain, theorizeabout) a hypothesized relation between variables (e.g.,IVt -+-> DVa, that the persons' levels of exposure totelevised violence is positively related to the viewer'slevels of aggressive behavior), the researcher shouldcast a wider net in predicting methodically other im-plicit directions of relations between the two variables.At the least, the researcher should consider and theo-retically account for not only the direction of the initialhypothesis (IVt -+-> DVa) but also the direction of allfour of the 2 x 2 directions shown in Figure 3, includ-ing (IVt -+-> DVa), (IVt- --> DVa), (DVa -`-> IVt), and(DVa ---> IVO). Moreover, the researcher should gener-

ate multiple theories to account for each of these fourpossible directions of the relations. The four (2 x 2) di-rections of the relation include two valences (direct, +,versus inverse, -) x two directions of causality (televi-sion viewing causing aggressiveness versus aggres-siveness causing television viewing). The four quad-rants in Figure 3 illustrate these four directions.

This discussion of theory enrichment is conserva-

tive in that it assumes that any IVt to DVa relation isconfined to that in the upper-left quadrant of Figure 3.That is, we assumed a positive (not inverse) relationbetween IVt and DVa; and we assumed a direction ofcausality such that TV viewing caused viewer aggres-sion (not aggression caused TV viewing). This narrow

focus is not surprising considering that the upper-leftQuadrant A relation is conventional wisdom. Still, as a

card-carrying perspectivist, I would urge more broad-ness, a full-court press, by considering the IV-DVrelation in each of the four directional quadrants andconsidering multiple theories to explain the hypotheti-cal relation postulated in each of the four quadrantsshown in Figure 3 that constitute a 2 x 2 matrix, 2covariation directions x 2 causal directions:

Quadrant A. Exposure to TV violence tends to in-crease aggressiveness (Hypothesis TV -+->Agg).Quadrant B. Exposure to TV violence tends to de-crease aggressiveness (Hypothesis TV --->Agg).Quadrant C. Aggressiveness tends to increase ex-

posure to TV violence (Hypothesis Agg -+->TV).Quadrant D. Aggressiveness tends to decreaseexposure to TV violence (Hypothesis Agg --->TV).

A perspectivist, when hypothesizing any one of thesefour relations, should investigate the other three also.

Moreover, a perspectivist would exploit the theoret-ical riches by recognizing that in each of the four quad-rants the relation hypothesized in that quadrant can beexplained (accounted for) by multiple theories (forQuadrant A these include the four listed and others notlisted in Figure 3). So far I have taken my examplesonly from Quadrant A relations; and in Quadrant A Ihave considered only one mediating theory, legitimiza-tion theory. Figure 3 shows that legitimization theory isonly one of four theories listed to account for the up-per-left Quadrant A direction of predicted hypothesisrelation, TV -+->Agg.

181

QUADRANT AI.Legitimization L. Berkowitz good-guy perpetrator,(desensitization; M. Thomas sanitized consequencesdisinhibition)

2. Modeling A. Bandura violence rewarded,(social learning; L. Eron victim similaravailability) J. Rotter

3. Arousal L. Ross female less than male,(excitement) P. Tannenbaum chronic aggression

D. Zillman

4. Mood J-P. Leyens presence of guns,imagery

I

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from

McGUIRE

1. Legitimization (also known as desensitization ordisinhibition) theories have been advocated by L.Berkowitz and M. Thomas, among others (MVI theory= perceived legitimacy of aggression). One of the inter-action-variable multipliers of the hypothesized IVt -+->DVa relation implied by legitimacy theories is that theminor premise relation holds especially when the vio-lence is depicted as committed by the good guys, iIVg.Another interacting variable that multiplies the IVt -+->DVa relation is when the violence is shown in sani-tized, low-bloodshed style, IVs.

2. Modeling (also called social learning or avail-ability) theories shown in Quadrant A of Table 3 havebeen advocated by A. Bandura, L. Eron, J. Rotter, andothers (MVa theory = availability of aggressive re-sponse in the viewer's behavioral repertory). They im-ply as hypothesis-multiplying interaction variablesthat the depicted victim look similar to the viewer's ad-versary, IVs, and that the TV violence is shown as be-ing rewarded, iIVr.

3. Arousal (or excitement) theories have been ad-vocated by P. Tannenbaum, L. Ross, D. Zillman,among others (MVa theory = activity level in theviewer). They imply as interaction variables theviewer's chronic or acute activity level, jIVa, or evok-ing preangered or agitated excitement in the viewer.

4. Mood theories have been advocated by J.-P.Leyens, among others (MVm theory = hostile mood inthe viewer). They imply as interaction variables a vividimagination or the actual presence of a weapon, JIV,

Even confining the discussion to the legitimizationfamily of theories in Quadrant A illustrates how syllo-gistic mediational-theory formalization ofjust this onelegitimization family of its explanatory theories allowsderivation of many interaction hypotheses and numer-ous other predictions, both theory-driven and serendip-itous, about relations involving our initial variables,exposure to television violence, IVt, and viewer ag-gressiveness, DVa. A perspectivist can and should alsogenerate multiple relations between the variables andpostulate multiple (mediational) theories to accountfor each of the hypothesized relations. Hence, the richyield of the one legitimization theory will be multi-plied by the number of additional theories perceived tofall in Quadrant A.

This richness of theories is further apparent when itis recognized that the four theoretical accounts just dis-cussed are confined to explaining the relation in the up-per-left Quadrant A in Figure 3. When the yields of ad-ditional predictions from theories in the other three,Quadrants B, C, and D, are also taken into account,there is a further quadrupling of predictions, many of

which can be tested by fairly simple elaborations of theexperimental designs, often requiring only that the par-ticipants be measured on easily scored, intrinsically in-teresting interaction variables (e.g., sex of participant,type of stimulus or of effect).

It may appear that perspectivism confronts research-ers with an embarras de richesses, threatening tosmotherthem under a confusing plethora ofpredictions,but this danger is containable when the researcher isforewarned and forearmed. The theory-constructing,theory-exploiting procedures described in this articleare superficially complex but can be reduced to reiter-ated recursive routines by methodical organization. Theperspectivist can construct pedagogic worksheets(McGuire, 2004b) that provide guidance and practice indealing with these complexities, enabling the studentre-searcher quickly to develop proficiency in using theseriches for creative theory construction and empirical re-finement: "aim now known and hand at work now neverwrong" (Hopkins, 1998). Training the student re-searcher in these elegantly complex perspectivist proce-dures for constructing theory can give him or her cour-age and capacity to exploit the complexities. Furtherdevelopments ofthis perspectivist approach are likely toyield many additional ambiguities, complexities, andenrichments.

References

Doise, W. (1997). Organizing social-psychological explanations. InC. McGarty & S. A. Haslam (Eds.), The message ofsocial psy-chology (pp. 63-76). Oxford, England: Blackwell.

Hopkins, G. (1998). To R. B. In W. Davis (Ed.), Gerard ManleyHopkins: Poetry and prose (p. 100). London: Everyman Dent.(Original work published 1918)

McGuire, W. J. (1989). A perspectivist approach to the strategicplanning of programmatic scientific research. In B. Gholson,W. R. Shadish, Jr., R. A. Neimeyer, & A. C. Houts (Eds.), Thepsychology of science: Contributions to metascience(pp.214-245). New York: Cambridge University Press.

McGuire, W. J. (1997). Creative hypothesis generating in psychol-ogy: Some useful heuristics. Annual Review ofPsychology, 48,1-30.

McGuire, W. J. (1999). Constructing social psychology: Creativeand criticalprocesses. New York: Cambridge University Press.

McGuire, W. J. (2004a). Twenty question sets to be answered by re-tiring epistemologists. In R. Rosenthal (Ed.), Papers in honor ofRalph Rosnow. Manuscript in preparation.

McGuire, W. J. (2004b). Perspectivist worksheets for generating aprogram of research. In J. Jost, M. Banaji, & D. Prentice (Eds.),Perspectivism in social psychology: The yin and yang ofscien-tific progress (pp. 319-322). Washington, DC: American Psy-chological Association.

Ruben, D.-H. (1990). Explaining explanation. London: Routledge.Ruben, D.-H. (1993). (Ed.). Explanation. Oxford, England: Oxford

University Press.

182

© 2004 Society for Personality and Social Psychology, Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by on November 16, 2007 http://psr.sagepub.comDownloaded from