Affective-Cognitive Consistency and the Effect of SalientBehavioral Information on the Self-Perception of Attitudes

Shelly ChaikenUniversity of Toronto

Ontario, Canada

Mark W. BaldwinUniversity of Waterloo

Ontario, Canada

Subjects with well-defined or poorly defined prior attitudes toward being anenvironmentalist/conservationist were identified by assessing the structural con-sistency between the affective and cognitive components of their attitudes. Aftersubjects completed one of two versions of a questionnaire designed to make salienteither past pro-ecology or past anti-ecology behaviors, their final attitudes wereassessed. The hypothesis that the self-perception account of attitude expressionholds primarily for individuals with poorly defined prior attitudes was supported:Low-consistency subjects, with presumably poorly defined attitudes, but not high-consistency subjects, with well-defined attitudes, expressed postmanipulation en-vironmentalist attitudes that were congruent with the pro- or anti-ecology be-haviors made salient by the questionnaire manipulation. The additional findingthat high-consistency (vs. low-consistency) subjects' beliefs on five ecology-re-lated issues were more highly intercorrelated supported the assumption that theconsistency construct appropriately indexes the degree to which individuals pos-sess well-defined attitudes. A comparison of theory and research on self-schematawith research on the affective-cognitive consistency variable suggested that thelatter may be a useful measure of attitude schematicity.

Self-perception theory (Bern, 1972) as-serts that people often infer their attitudes(and other internal states) from observationsof their overt behaviors and the contexts inwhich these behaviors occur. Empirical sup-port for this theory is fairly widespread. Forexample, researchers studying counteratti-tudinal advocacy (e.g., Bern, 1967, 1972;Bern & McConnell, 1970), pro attitudinaladvocacy (e.g., Fazio, Zanna, & Cooper,1977; Kiesler, Nisbett, & Zanna, 1969), andintrinsic motivation (e.g., Calder & Staw,1975; Deci, 1971; Lepper, Greene, & Nis-bett, 1973; Ross, 1976) have demonstrated

The present research was supported by a grant to thefirst author from the Humanities and Social SciencesCommittee of the Research Board of the University ofToronto. The authors thank John Bassili, Alice Eagly,Jonathan Freedman, Michael Ross, Diane Ruble, WendyWood, Mark Zanna, and two anonymous reviewers fortheir comments on an earlier draft of this manuscript.

Requests for reprints should be sent to Shelly Chai-ken, Department of Psychology, Erindale College, Uni-versity of Toronto, Mississauga, Ontario, Canada L5L1C6.

that the less people perceive their behaviorsto be under the control of contextual stimulisuch as reward or justification, the greateris their tendency to express attitudes thatcorrespond to, or are consistent with, thesebehaviors.

Most self-perception research has exam-ined the relationship between expressed at-titudes and behaviors that subjects are in-duced to perform immediately prior to thetime their attitudes are assessed. More re-cently, Salancik (1974; Salancik & Conway,1975) demonstrated that attitude inferencescan also be influenced by manipulating per-ceivers' recall of past attitudinally relevantbehaviors. For example, before assessingsubjects' religious attitudes, Salancik andConway used a linguistic device to vary thesaliency of subjects' past proreligious andantireligious behaviors. Subjects for whompast proreligious behaviors had been madesalient perceived themselves as more reli-gious and expressed more positive attitudestoward being religious than did subjects for

Journal of Personality and Social Psychology, 1981, Vol. 41, No. I, 1-12Copyright 1981 by the American Psychological Association, Inc. 0022-35I4/8I/4IOI-OOOIS00.75

SHELLY CHAIKEN AND MARK W. BALDWIN

whom antireligious behaviors had been madesalient.

Notwithstanding Salancik and Conway's(1975) demonstration that information aboutpast as well as recent behaviors affects at-titude judgments, their findings are consis-tent with other self-perception research insuggesting that persons' attitude expressionsrepresent "top of the head" (cf., Taylor &Fiske, 1978) evaluative judgments, con-structed largely on the basis of contempo-raneous and salient contextual and behav-ioral information. This research tradition,then, has tended to regard attitudes as epi-phenomena and has focused on the externalnature of the information underlying atti-tude judgments. In contrast, most other at-titudinal research (e.g., persuasion, attitude-behavior relations) has been guided by thetraditional theoretical view (e.g., Allport,1935; McGuire, 1969) that attitudes repre-sent relatively enduring learned predisposi-tions of persons. Consistent with thisconceptualization, information-processingmodels of attitude such as those proposed byFishbein and Ajzen (1975) and others (e.g.,Anderson, 1968; Wyer, 1974) have assumedthat the locus of much information under-lying attitude judgments is internal in thesense that it consists of people's privateknowledge of or memory for prior beliefsabout and past affective reactions to the at-titude object (for an elaborated discussionof self-perception vs. traditional attitude the-ory perspectives, see Greenwald, 1968, andWood, 1980).

Both conceptualizations of attitude andassumptions regarding the primary locus ofinformation underlying attitude judgmentshave received empirical support, and mostresearchers would probably concur that thetwo viewpoints can be subsumed under amore general cognitive theory explicitly ac-knowledging the influence of both internaland external cue information on attitudejudgments (cf., Eagly & Himmelfarb, 1978;Wood, 1980). Yet, despite this probable con-sensus and despite Bern's (1972) originalproviso that the self-perception process hedescribed held only to the extent that "in-ternal cues are weak, ambiguous, or unin-terpretable" (p. 2), self-perception researchhas been strikingly successful in demonstrat-

ing that subjects' attitude inferences can beso strongly influenced by contemporaneousexternal cues and that prior attitudes areoften not salient for subjects (Bern &McConnell, 1970). Such empirical successleads one to wonder when, if ever, internalcue information (i.e., prior attitudes) mightapproach the salience of contemporaneousexternal cues and thus diminish or overridethe impact of the latter information on peo-ple's attitude judgments.

It might be that this concern that peopleoften underutilize internal cue informationwhen queried about their attitudes is un-warranted. As others have noted (e.g., Kel-ley, quoted in Harvey, Ickes, & Kidd, 1978;Salancik & Conway, 1975; Wood, 1980),self-perception studies often employ experi-mental tasks or attitude topics that are novelor unfamiliar to subjects. Given the likeli-hood that subjects have not yet formulatedattitudes toward such stimuli, it is under-standable that their attitude inferences wouldbe strongly affected by contemporaneousexternal cues. However, at least some self-perception research has used more familiarexperimental tasks (e.g., drawing: Lepper etal., 1973) or attitude objects (e.g., "beingreligious": Salancik & Conway, 1975) to-ward which most subjects might reasonablybe expected to have prior attitudes. Yet thisresearch, too, has revealed the strong impactof contemporaneous informational cues onattitude judgments.

The present study was explicitly designedto test Bern's (1972) proviso concerning themoderating impact of internal cues, or priorattitudes, on the influence of contempora-neous external cues on attitude inferences.Guiding the present research was the as-sumption that, independent of the extremityof people's scores on some standard premea-sure of attitude, they will differ in thestrength or degree of definition of those at-titudes. In the experiment, we assessed theextent to which our subjects possessed well-defined prior attitudes. We assumed that in-ternal cue information would rival or over-ride the salience of external cues for subjectswith well-defined prior attitudes but not forsubjects with poorly defined attitudes. Thus,we hypothesized that only the attitude in-ferences of the latter subject group would

AFFECTIVE-COGNITIVE CONSISTENCY AND SELF-PERCEPTION

be strongly influenced by salient behavioralinformation.

In the study, we investigated subjects' at-titudes toward "being an environmentalist/conservationist." At the first of two experi-mental sessions, subjects with well-defined(vs. poorly defined) attitudes on this topicwere identified by assessing the structuralconsistency between the affective and cog-nitive components of their attitudes (Rosen-berg, 1960, 1968; Rosenberg & Hovland,1960). According to Rosenberg (1960,1968),people whose attitudes are characterized byhigh affective-cognitive consistency (i.e.,similar scores on affective and cognitive in-dices of attitude) are likely to have "well-articulated," "well-thought-out" attitudesreflecting a stable underlying disposition,whereas people whose attitudes are low insuch consistency are likely to have "poorlyarticulated" attitudes that do not reflect astable disposition toward the attitude object.Consistent with this hypothesis, Rosenberg(1968) showed that greater affective-cog-nitive consistency was associated with greaterattitudinal stability and resistance to per-suasion. In addition, Norman (1975), al-though not Fazio and Zanna (1978), foundthat higher affective-cognitive consistencywas predictive of higher attitude-behaviorconsistency. At our first experimental ses-sion, we attempted to adduce further evi-dence for Rosenberg's hypothesis by assess-ing subjects' beliefs on a number of ecology-related issues (e.g., nuclear power, solar en-ergy). We expected that the relationshipbetween subjects' environmentalist attitudesand their ecology-related beliefs, as well asthe relationship among these beliefs, wouldbe stronger for subjects exhibiting high (vs.low) affective-cognitive consistency.

At the second experimental session, weemployed the linguistic device developed bySalancik and Conway (1975) to make eitherpast pro-ecology or anti-ecology behaviorssalient for subjects. Immediately afterward,we reassessed subjects' environmentalist self-perceptions and attitudes. Self-perceptiontheory predicts that subjects for whom pro-ecology (vs. anti-ecology) behaviors are madesalient will express more positive attitudestoward being environmentalists/conserva-tionists and, to a greater extent, perceive

themselves as being environmentalists. Weanticipated, however, that this predictionwould be borne out only among subjects pre-viously identified as low in affective-cogni-tive consistency.

Method

Subjects

Subjects were male and female University of Torontoundergraduates who participated during regular lecturemeetings of the first author's introductory psychologycourse. At the first experimental session, all 99 studentsin attendance completed questionnaires. Ninety-fourstudents attended class the evening of the second session,and all responded to the experimental materials. Datafrom 14 of these subjects, who did not attend the firstsession, were discarded. Both sessions occurred early inthe semester and were conducted by the senior author.The experiment and its findings were subsequently dis-cussed during a lecture on attitudes.

Procedure and Measuring Instruments

First experimental session. After describing thestudy as a "survey of people's attitudes," the courseinstructor distributed a nine-page questionnaire to sub-jects. On page 1, subjects indicated their name, sex, andage, and responded to five items pertaining to the at-titude object, "being an environmentalist/conservation-ist." Subjects first indicated their favorabitity (vs. un-favorability) toward the attitude object (11-point scale)and then rated the attitude object on five 7-point bipolaradjective scales (good vs. bad, wise vs. foolish, pleasantvs. unpleasant, healthy vs. sick, beneficial vs. harmful).Subjects' responses to the five scales were summed toform one semantic differential index of attitude. Boththe favorability rating and the semantic differential in-dex assessed subjects' affect toward the attitude objectand, in addition to serving as attitude premeasures, wereused in determining affective-cognitive consistency (seebelow). The next item (self-perception premeasure) re-quested subjects to rate the extent to which they con-sidered themselves an environmentalist/conservationist(11-point scale). Finally, subjects indicated the extentto which they had thought about the attitude object inthe past and how personally important they consideredthe attitude object (9-point scales).

The next four questionnaire pages consisted of 22items designed to assess the cognitive component of sub-jects' attitudes. Following Rosenberg (1960,1968), eachitem pertained to 1 of 22 values (e.g., happiness, a worldof beauty, a sense of accomplishment, world harmony,pleasure, human welfare, good health) and consisted oftwo 9-point rating scales.' On the first scale, subjects

1 These 22 values were selected on the basis of pre-testing with a group of subjects (n = 26) who indicated,for each of a large set of values, whether the value wasrelevant to the attitude object. These subjects also in-dicated whether there were redundancies among values.

SHELLY CHAIKEN AND MARK W. BALDWIN

rated the extent to which attainment of a particularvalue (e.g., happiness) would be helped (vs. hindered)by "being an environmentalist/conservationist." On thesecond scale, subjects rated how positive (vs. negative)they considered the value. The weighted cognitive indexof subjects' attitudes was calculated by multiplying theirtwo ratings for each item and then summing the re-sulting 22 products (Rosenberg, 1960). Along with theaffective indices described earlier, this index was usedin determining affective-cognitive consistency (see be-low). Prior to the experiment, 89 pilot subjects com-pleted the 22-item instrument. Analysis of their dataindicated that the instrument possessed satisfactory re-liability (average item total r = .58, coefficient a = .88).

The next three questionnaire pages consisted of 5opinion statements on ecology-related issues and 14 fillerstatements on other topics. Subjects indicated theiragreement with these opinion statements on 15-pointscales. The ecology-related statements were: "More nu-clear power stations should be built," "Nonreturnablesoft drink bottles should be outlawed," "Aerosol spraycans are damaging to the environment and should bebanned," "The government should make the develop-ment of solar energy technology its highest energy prior-ity," and "nonbiodegradable consumer goods such ascolored tissue and toilet paper should be outlawed."

Finally, the last questionnaire page consisted of the33-item Crowne-Marlowe social desirability scale(Crowne & Marlowe, 1964). This scale was includedin order to explore, and hopefully to rule out, the pos-sibility that social desirability concerns underlie persons'tendencies to exhibit affective-cognitive consistency.

After subjects had completed and handed in theirquestionnaires (a task requiring approximately 30 min-utes), the instructor began the evening's lecture. Whilenoting that the questionnaires would be discussed laterin the term, the instructor made no mention of a secondexperimental session.

Second experimental session. At the beginning ofclass 2 weeks later, subjects were asked to complete twoone-page questionnaires. On the first questionnaire, sub-jects indicated, for each of 26 ecology-related behavioralstatements, whether or not the particular behavior wasself-descriptive: Subjects were instructed to place acheck mark next to behaviors they considered "true" forthem and to leave blank behaviors they considered "nottrue." Two versions of this questionnaire, representingthe two levels of the pro-ecology (vs. anti-ecology) be-havior saliency manipulation (see below), were ran-domly distributed to subjects. On the second question-naire, subjects indicated their name and responded tothe experiment's major dependent measures. Except forminor differences in scale length, these postmanipula-tion measures were identical to the self-perception andtwo attitude premeasures: Subjects indicated the extentto which they considered themselves environmentalists/conservationists, their favorability (vs. unfavorability)toward being an environmentalist/conservationist (both15-point scales), and rated "being an environmentalist/conservationist" on the five 7-point adjective scales de-scribed earlier.

Independent VariablesAffective-cognitive consistency. As suggested by

Rosenberg (1968) and Norman (1975), affective-cog-

nitive consistency was determined in the following way:Subjects (N - 99) were rank ordered on the basis of anaffective index (average of each subject's Z scores onthe two attitude premeasures) and also on the basis ofthe overall favorability implied by the weighted cogni-tive index. The absolute value of the discrepancy be-tween each subject's standing in the two rankings wasdefined as his or her level of affective-cognitive consis-tency. Lower (higher) discrepancy scores reflectedhigher (lower) consistency. Subjects (n = 50) whosediscrepancy scores were below the median (18.4, range =1-98) were considered high in consistency, whereas sub-jects (n = 49) whose scores fell above the median weredesignated low in consistency.

Saliency manipulation. The first questionnaire dis-tributed at the second session consisted of 13 pro-ecologyand 13 anti-ecology behavioral statements of the generalform "I do (or refuse to do) X," where X is pro- or anti-ecology behavior. Using the linguistic device developedby Salancik and Conway (1975), we prepared two ver-sions of this questionnaire, one designed to make pro-ecology behaviors salient and the other designed to makeanti-ecology behaviors salient. The linguistic device isbased on the tested assumption (Salancik & Conway,1975) that the probability of endorsing a statement "Ido X on occasion" is higher than the probability ofendorsing a statement "I do X frequently." Implemen-tation of this device involved systematically varying thewording of the 26 behavioral statements such that, inthe pro-ecology version, pro-ecology behaviors werepaired with "on occasion" (or "occasionally") whereasanti-ecology behaviors were paired with "frequently"(e.g., "I occasionally pick up other persons' garbage andtake it to the trash can," "I occasionally carpool ratherthan drive separately," "I frequently litter," "I fre-quently leave on lights in rooms I'm not using"). Op-posite pairings were used to create the anti-ecology ver-sion of the questionnaire (e.g., "I frequently pick upother persons' garbage . . . ," "I frequently carpool. . . ," "I occasionally litter," "I occasionally leave onlights . . .").

Results

The design included two levels each of thesaliency manipulation, affective-cognitiveconsistency, and subject sex. Since prelimi-nary analyses yielded no significant effectsinvolving sex, all reported analyses ignoredthis variable.

Check on Experimental Conditions

Saliency manipulation. To assess thesuccess of the saliency manipulation, ananalysis of variance was performed on a"salient behavior" index. This index wasderived from subjects' responses to the 26-item questionnaire designed to manipulatetheir pro-ecology (vs. anti-ecology) behav-ioral endorsements and represented thenumber of pro-ecology statements minus the

AFFECTIVE-COGNITIVE CONSISTENCY AND SELF-PERCEPTION

number of anti-ecology statements that sub-jects actually endorsed. As expected, themain effect for the saliency manipulationproved highly significant on this index, F( 1,76) = 29.40, p < .0001: Subjects who com-pleted the pro-ecology version of the ques-tionnaire endorsed significantly more pro-ecology (vs. anti-ecology) behaviors as self-descriptive than did subjects who completedthe anti-ecology version (ATs = 5.69 vs. .58).Separate analyses on each component of theindex revealed that both were affected bythe manipulation: pro-ecology behaviors,F(l, 76) = 8.32, p < .005, and anti-ecologybehaviors, F(l, 76) = 36.11, p< .0001. Al-though a marginally significant Consis-tency X Saliency Manipulation interactionon the salient behavior index, F(l,76) = 3.11, p = .09, suggested that the ma-nipulation had a slightly greater impact onthe pro- and anti-ecology behavior endorse-ments of low-consistency (vs. high-consis-tency) subjects, the main effect for the sa-liency manipulation proved significant withinboth subject groups (ps < .025 and .001 forhigh- and low-consistency subjects, respec-tively). Further, when the influence of sub-jects' initial attitudes and self-perceptions ontheir behavioral endorsements was partialedout in an analysis of covariance on the salientbehavior index, the saliency manipulationmain effect remained highly significant, F(1, 73) = 26.21, /x.OOOl, whereas theConsistency X Saliency Manipulation inter-action disappeared, F(l, 73) = 1.42,p = .24.It should also be noted that the two-way in-teraction proved nonsignificant in the anal-ysis of variance on the pro- and anti-ecologycomponents of the salient behavior index(ps = .17 and .12, respectively). Finally, theanalyses of the salient behavior index andits two components revealed no overall dif-ferences between high- and low-consistencysubject samples in terms of their means(Fs < 1.0) or variances (Fs < 1.33).

Affective-cognitive consistency. As ex-pected, the correlation between the affectiveand cognitive indices used to determine af-fective-cognitive consistency was significantfor high-consistency subjects (r = .83, p <.001) but nonsignificant for low-consistencysubjects (r = -.20). It is important to notethat the two consistency groups did not differsignificantly in terms of their means or vari-

ances on the affective and cognitive indicesof their attitudes (all Fs < 1.0). Thus, priorto the saliency manipulation, high- and low-consistency subjects evidenced no differencein the extremity or variability of their atti-tudes toward being an environmentalist/conservationist.

Analyses on the remaining measures as-sessed at the first experimental session alsorevealed little difference between the twoconsistency samples. High- and low-consis-tency subjects did not differ in the extent towhich they reported thinking about theirenvironmental attitudes in the past or in theextent to which they considered the attitudeobject personally important (Fs < 1.0).Whereas low-consistency (vs. high-consis-tency) subjects indicated greater agreementwith the statement "More nuclear powerstations should be built" (p < .01), the opin-ions of the two subject groups did not differon the remaining four ecology-related issues(Fs = 1.23 or smaller). Finally, the fact thatthe two consistency groups did not differ intheir Crowne-Marlowe scores (F < 1.0) sug-gests that social desirability concerns do notunderlie persons' tendencies to manifest highor low affective-cognitive consistency.

Self-Perceptions and Attitudes

Analysis of variance revealed no signifi-cant differences among experimental con-ditions on either the self-perception premea-sure or the two attitude premeasures(favorability ratings, semantic differentialindex). Thus, the three corresponding post-manipulation measures were treated byanalyses of variance.2

Cell means for the postmanipulation mea-sures appear in Table 1. The saliency ma-nipulation main effect was significant on allthree post measures: favorability ratings,F(l, 76) = 8.48, p < .005; semantic differ-ential index, F( 1, 76) = 10.05, p < .005; andself-perceptions, F( 1,76) = 4.08, p < .05.Overall, subjects for whom pro-ecology (vs.anti-ecology) behaviors were made salientin the context of the second experimental

3 Each postmanipulation measure was also submittedto an analysis of covariance, using its correspondingpremeasure as the covariate. The results of these anal-yses were virtually identical to those reported in the text.

6 SHELLY CHAIKEN AND MARK W. BALDWIN

Table 1Mean Postmanipulation Self-Perception and Attitude Scores as a Function of Affective-CognitiveConsistency and Salience of Pro- Versus Anti-Ecology Behaviors

High consistency Low consistency

Dependent variable

Self-perceptionsSemantic differential

indexFavorability ratings

P-E

9.00

29.1311.73

A-E

9.68

28.7211.00

P-E

11.33

31.2512.17

A-E

8.25

26.3810.62

Note. P-E = pro-ecology behaviors salient; A-E = anti-ecology behaviors salient. Higher numbers indicate a greatertendency to describe oneself as an environmentalist/conservationist and more positive attitudes toward being anenvironmentalist/conservationist. Cell its ranged from 15 to 25.

session reported more positive attitudes to-ward "being an environmentalist/conserva-tionist" and, to a greater extent, perceivedthemselves as being environmentalists. Thesefindings replicate those reported by Salancikand Conway (1975), who studied self-per-ceptions and attitudes about religiosity, andthus lend further support to the self-percep-tion account of attitude expression.

More important, however, the Consis-tency X Saliency manipulation interactionproved significant on subjects' semantic dif-ferential scores, F(l, 76) = 7.01, p<.0\,and their self-perception scores, F(l, 76) =9.64, p < .005. The patterning of the inter-action on these variables, as well as a similarbut nonsignificant patterning on subjects'favorability ratings, F(l, 76) =1.06, sup-ports our hypothesis that the saliency ma-nipulation would more strongly influence theself-reported attitudes and self-perceptionsof low-consistency (vs. high-consistency)subjects (see Table 1). Indeed, planned com-parisons revealed that for low-consistencysubjects, the constrast between pro- andanti-ecology conditions was significant on allthree postmanipulation measures (Fs ̂ 7.91,ps < .01 or smaller). For high-consistencysubjects, however, the pro- versus anti-ecol-ogy contrast proved nonsignificant on allthree postmeasures (Fs <, 1.71, ps > .15 orlarger).

Correlational analyses provided addi-tional evidence regarding the differential in-formational determinants of high- and low-consistency subjects' postmanipulation re-sponses. As expected, correlations computedbetween subjects' salient behavior scores(i.e., number of pro- minus anti-ecology be-

havioral endorsements on the manipulationquestionnaire) and their postmanipulationself-perceptions and attitudes were generallyhigher for low-consistency (vs. high-consis-tency) subjects: self-perceptions, rs = .59versus .38, p < .10; favorability ratings, rs =.57 versus .34, p < .10; and semantic differ-ential scores, rs = .25 versus .25, ns. In con-trast, correlations between subjects' pre-manipulation and postmanipulation self-perceptions/attitudes tended to be greaterfor high-consistency (vs. low-consistency)subjects: self-perceptions, rs = .61 versus.50, p = .23; favorability ratings, rs = .58versus .15, p < .025; and semantic differ-ential scores, rs = .33 versus .18, p = .23.3

To explore more directly the relative influ-ence of initial attitudes/self-perceptions ver-sus contemporaneous behavioral informa-

3 In evaluating these findings, it is important to notethat the high- and low-consistency samples differed littlein terms of their within-group standard deviations onthe seven variables included in the correlational anal-yses. The F tests yielded only two significant group dif-ferences (all other comparisons were nonsignificant,Fs[39, 39] = 1.40 or smaller.ps > .10 or larger): High-consistency (vs. low-consistency) subjects manifestedless variability in their premanipulation semantic dif-ferential scores (SDs = 3.49 vs. 5.19, p < .01) butgreater variability in their postmanipulation favorabilityratings (SD = 2.05 vs. 1.48, p < .05). To the extent thata restricted range on a variable reduces the magnitudeof the correlation that may be observed between thatvariable and others, we may have overestimated (in fa-vor of our hypothesis) the difference in correlations be-tween consistency groups in one case (r between pre-manipulation and postmanipulation favorability ratings)and underestimated (in opposition to our hypothesis)group differences in two cases (r between premanipu-lation and postmanipulation semantic differential scoresand r between salient behavior index and postmanipu-lation favorability ratings).

AFFECTIVE-COGNITIVE CONSISTENCY AND SELF-PERCEPTION

tion on subjects' final attitudes/self-perceptions, we performed three multipleregression analyses for each consistencygroup. Each analysis regressed one post-manipulation measure (i.e., self-perception,favorability, or semantic differential scores)on (a) its corresponding premanipulationmeasure and (b) the salient behavior index.The results of these analyses are shown inTable 2. Examination of the F (and asso-ciated p) values, partial rs, and beta weightsfor the two predictors (where comparison of/32 for the two predictors reflects their rela-tive importance in explaining variation in thecriterion variable; cf. McNemar, 1969, p.195) reveals that the postmanipulation self-perceptions and attitudes of high-consis-tency subjects were better predicted by theirinitial self-perceptions and attitudes than bythe environmental behaviors they were in-duced to endorse. In contrast, low-consis-tency subjects' postmanipulation responseswere better predicted by the environmentalbehaviors they endorsed than by their initialself-perceptions and attitudes.

Affective-Cognitive Consistency andEcology-Related Beliefs

Correlational analyses explored the rela-tionship between subjects' (premanipula-

tion) attitudes toward being an environmen-talist/conservationist and their beliefs onfive ecology-related issues as well as the in-terrelatedness of these beliefs. These anal-yses included data from all 99 subjects whoattended the first experimental session.

Attitude-belief relationship. Subjects'agreement with the five ecology-relatedopinion statements were summed to formone index, with higher scores implyinggreater favorability toward the environment.As expected, the correlations between thisopinion index and the two attitude indices(i.e., affective and weighted cognitive index)were larger for high-consistency (vs. low-consistency) subjects: affective index, rs =.55 versus 12, p < .01; and cognitive index,rs = .43 versus .18, p < .08 (both one-tailedzs). Further, when the two attitude measureswere correlated with subjects' opinions oneach of the five issues, the correlation coef-ficient was larger for high-consistency sub-jects in 9 out of 10 possible group compar-isons. The average correlations betweensubjects' ecology-related beliefs and the af-fective (cognitive) index were .398 (.313)and .068 (.114) for high- and low-consis-tency subjects, respectively.

Relationship among beliefs. As we an-ticipated, the ecology-related beliefs of high-

Table 2F Values and ft Weights for Predictor Variables and Partial Correlations Between Predictor andCriterion Variables in Multiple Regression Equations

Criterion variable andpredictor variables

High-consistency subjects(« = 40)

F value weight Partial r

Low-consistency subjects(n = 40)

F value 0 weight Partial r

Postmanipulationself-perceptions

Initial self-perceptionsSalient behavior index

Postmanipulationfavorability ratings

Initial favorabilityratings

Salient behavior indexPostmanipulation semantic

differential scoresInitial semantic

differential scoresSalient behavior index

17.78****3.30

15.08****2.38

4.08*2.00

.55

.24

.52

.21

.31

.22

.57****

.28**

.54****

.24

.32**

.23

5.01**11.56***

.6217.86****

.391.54

.31

.47

.10

.56

.10

.21

.34**49****

.13

.57****

.10

.20

Note. For tabled F values, dfs = 1, 37; for tabled partial rs (one-tailed tests), dfa* / > « .06. **p< .05. ***p< .01. ****p< .001.

37.

8 SHELLY CHAIKEN AND MARK W. BALDWIN

Table 3Correlations Among Ecology-Related Beliefs for Combined Samples of High- and Low-Consistency Subjects

1 2 3 4

1 . Nuclear power2. Solar energy3. Nonreturnable

bottles4. Aerosol sprays5. Biodegradable

products

3.71 (3.58).19*

.32***

.21*

.34***

-.053.02 (3.28)

.40***

.13

.31***

-.16.26**

3.45 (3.11).39***

.58***

.06.19*

.55***3.20 (3.05)

.57***

.19*

.15

.42***

.45***

3.40 (3.81)

Note. Intercorrelations for high-consistency subjects (n = 94) appear below the diagonal and those for low-con-sistency subjects (n = 94) appear above the diagonal. Within-group standard deviations (shown without [with]parentheses for high-consistency [low-consistency] subjects) for the belief measures appear along the diagonal. Nosignificant group variance differences were obtained (Fs = 1.26 or smaller). Higher scores on the belief measuresindicated less agreement that more nuclear plants should be built and greater agreement that solar energy technologyshould be supported, nonreturnable bottles outlawed, aerosol spray cans banned, and nonbiodegradable productsoutlawed (see Method section for exact wording of belief statements).* p< .05. **p<.01. ***p<.005.

consistency (vs. low-consistency) subjectstended to be more highly interrelated. Theaverage intercorrelation among beliefs was.347 for high-consistency subjects and .238for low-consistency subjects, and in 7 of 10possible group comparisons, the pair-wisecorrelation coefficient was higher for high-consistency subjects. Because pilot subjects(n = 89) had also completed the opinionquestionnaire and the procedure used to clas-sify subjects into high and low consistencygroups was identical for the experimentaland pilot samples (i.e., the same median con-sistency score obtained in both samples), thetwo samples were combined. For this largersample, the average relationship among be-liefs was .354 and .216 for high- and low-consistency subjects, respectively. Further,as Table 3 indicates, 8 of the 10 group com-parisons yielded a larger pair-wise correla-tion for high-consistency subjects (p < .10,one-tailed, by Mann-Whitney U test).4

Among the filler items appearing on theopinion questionnaire completed by experi-mental subjects were five statements dealingwith the United States' cultural dominationover Canada (e.g., "The number of U.S.programs shown on Canadian televisionshould be reduced"). If, as Rosenberg (1960)claimed, the consistency construct is domainspecific rather than a reflection of a generalindividual difference, high- and low-consis-tency subjects (whose consistency was as-

sessed with respect to their environmentalattitudes) should show no difference in thedegree to which their beliefs about culturaldomination are interrelated. In line withthis reasoning, the average intercorrelationamong these beliefs proved quite similar forhigh- and low-consistency subjects (meanrs = .343 and .391, respectively). Further,of 10 possible group comparisons, 4 yieldeda larger pair-wise correlation for high-con-sistency subjects, whereas 5 yielded a largercorrelation for low-consistency subjects (onecomparison yielded a tie).

Discussion

Our hypothesis that the self-perceptionaccount of attitude expression holds primar-ily for individuals who do not possess well-defined prior attitudes toward the target at-titude object was supported. At the secondexperimental session, the attitude judgmentsand self-perceptions of subjects with poorly

4 A separate analysis of the beliefs expressed by high-consistency (vs. low-consistency) pilot subjects revealedthat the average belief intercorrelation was .358 (vs..215) and that 8 of the 10 possible group comparisonsyielded a larger pair-wise correlation for high-consis-tency subjects. It should also be noted that the pair-wisecorrelations accounting for the 2 reversals to the overallpattern (larger rs for high-consistency subjects) in thepilot sample and the pair-wise correlations accountingfor the 3 reversals in the experimental sample showedno overlap.

AFFECTIVE-COGNITIVE CONSISTENCY AND SELF-PERCEPTION

defined prior attitudes were strongly affectedby the contemporaneous behavioral cuesmade available to them.: Low-consistencysubjects for whom past pro-ecology (vs. anti-ecology) behaviors were made salient ex-pressed significantly more favorable atti-tudes toward being environmentalists andsignificantly heightened perceptions of beingenvironmentalists. Further, the regressionanalyses revealed that the postmanipulationresponses of low-consistency subjects werebetter predicted by the pro- or anti-ecologynature of the behaviors they were inducedto endorse than by their prior self-percep-tions and environmentalist attitudes.

In contrast to these findings, subjects whowere high in affective-cognitive consistencywere not significantly affected by the pro-ecology (vs. anti-ecology) behavior manip-ulation. Although the manipulation was suc-cessful in inducing these subjects, like low-consistency subjects, to endorse pro-ecology(vs. anti-ecology) behaviors as self-descrip-tive, their postmanipulation attitude judg-ments and self-perceptions reflected no ap-preciable reliance on these contemporaneousbehavioral cues. Presumably, high-consis-tency subjects possessed strong internal cuesregarding their feelings and self-perceptionsabout being environmentalists and thus didnot need to "infer" their attitudes from cur-rently available behavioral information.Consistent with this interpretation, theregression findings indicated that the post-manipulation responses of high-consistencysubjects were better predicted by their priorattitudes and self-perceptions than by thecontemporaneous pro- and anti-ecology be-havioral cues available to them.

Although these findings strongly supportour major hypothesis, it is important thattheir implications for self-perception theorynot be overstated. We think it would be in-appropriate to conclude that the self-percep-tion account of attitude expression holdsonly for individuals with poorly defined priorattitudes or, more specifically, that personswith well-defined attitudes never express at-titudes that reflect, at least in part, the im-pact of salient and contemporaneous ex-ternal cues. (Indeed, had our saliencymanipulation been more potent, the finalattitudes expressed by high-consistency sub-

jects might have begun to reflect—albeit toa lesser extent than the final attitudes of low-consistency subjects—the impact of cur-rently available behavioral cues.) Althoughour results indicate that both internal cueinformation (i.e., prior attitudes) and exter-nal cue information can influence attitudinaljudgments, we suspect that the latter kindof information often predominates. As Tay-lor and Fiske (1978) recently suggested, in-dividuals may frequently respond with littlethought to the most salient stimuli in theirenvironment and thus may typically express"top of the head" opinions (and other judg-ments) that reflect little or no informationbeyond that available to them in the im-mediate situation.

Affective-Cognitive Consistency

The fact that high- and low-consistencysubjects responded in the predicted mannerto our saliency manipulation, as well as thefact that their postmanipulation attitudesand self-perceptions were best predicted byinternal and external cue information, re-spectively, is consistent with Rosenberg's(1968) hypothesis that the affective-cogni-tive consistency construct appropriately in-dexes the extent to which people possesswell-defined attitudes with respect to a par-ticular object. Our correlational analysesprovided further evidence regarding the va-lidity of the consistency construct. The re-lationships between subjects' environmen-talist attitudes and their ecology-relatedbeliefs, as well as the relationships amongthese beliefs, were stronger for high-consis-tency (vs. low-consistency) subjects. Thesefindings indicate that persons who score highon the affective-cognitive consistency di-mension can be distinguished from thosewho score low on this dimension on the basisof the presence (vs. absence) of an organizedset of supporting cognitions in relation to theattitude object. In this regard, it is importantto note that our two consistency groups man-ifested little difference in the degree to whichtheir beliefs on a nonecology topic (culturaldomination of Canada by the United States)were interrelated. This result is compatiblewith other research that has obtained non-significant correlations between subjects'

10 SHELLY CHAIKEN AND MARK W. BALDWIN

consistency scores in different attitudinaldomains (Norman, 1975; Chaiken & Gar-isto, Note 1) and provides additional evi-dence for the assumption that the affective-cognitive consistency construct is truly do-main specific rather than simply reflectiveof a more general individual difference.

Although the above findings support theutility and validity of the consistency con-struct as an index of attitude definition, itshould be noted that we found no differencesbetween high- and low-consistency subjectson several ancillary measures where differ-ences had been anticipated. Thus, we foundno greater tendency for high-consistency (vs.low-consistency) subjects to consider the at-titude object personally important or to re-port thinking about their attitudes in thepast. Correlations between subjects' consis-tency scores and these measures were alsolow and nonsignificant (rs = . 12 and .09 forthought and importance ratings, respec-tively).5 In addition, Fazio and Zanna (1978)found no relationship between affective-cog-nitive consistency and several other potentialindicants of attitude strength: self-reportedcertainty of attitude and number of past di-rect experiences with the attitude object.

The lack of a relationship between quan-tity of direct experiences with the attitudeobject and structural consistency does notnecessarily undermine the validity of thisconstruct. Certainly, people seem capable offorming strong, well-defined attitudes to-ward objects with which they have had nopersonal direct experiences (e.g., capitalpunishment, abortion). The possibility thatstructural consistency is unrelated to indi-viduals' self-reports of the personal impor-tance of the attitude object, the amount oftime they have spent thinking about theirattitudes in the past, and their attitudinalcertainty is somewhat more disconcerting,since it seems reasonable that people wouldbe more likely to form well-defined attitudestoward objects of greater personal impor-tance and that people possessing well-de-fined attitudes would have thought moreabout and would express greater confidencein their attitudes. Clearly, more researchexamining the relationship between affec-tive-cognitive consistency and theseand other possible indicants of attitudinal

strength or definition, as well as researchexploring possible antecedents of structuralconsistency, is necessary before the validityof this construct can be fully accepted. Giventhe appareftt lack of relationship betweenaffective-cognitive consistency and variousself-report indicants of attitudinal strength,such research might also entertain the in-teresting possibility that people whose atti-tudes are characterized by high (vs. low)structural consistency and whose behavior,in our own and previous research (Norman,1975; Rosenberg, 1968), indicates that theypossess strong, well-defined attitudes are, forwhatever reason, not especially aware of thisfact.

Before we conclude, two additional pointsabout the affective-cognitive consistencyconstruct deserve mention. First, it is im-portant to note that this construct assumesthe validity of a multicomponent view ofattitudes (cf., Bagozzi, 1978; Katz & Stot-land, 1959; Kothandapani, 1971; Ostrom,1969; Rosenberg, 1968). Although popularat one time, this conceptualization has beenlargely superseded in recent years by a uni-dimensional view that regards the cognitiveand affective (and conative) components ofattitude as alternative measures of the sameunderlying affective dimension rather thanas distinguishable constructs (cf. Fishbein& Ajzen, 1974, 1975). The present findings,as well as previous research on affective-cognitive consistency (Norman, 1975; Ro-senberg, 1968), support the idea that affectand cognition represent differentiable com-ponents of attitude. Further, using confir-matory factor analysis, Bagozzi and Burnk-rant (1979) recently substantiated thevalidity of a two-component (affective/cog-nitive), but not a single-component (affec-tive), model of attitude. The view that at-titude is a multidimensional construct clearly

5 Norman (1975), who studied attitudes toward vol-unteering for psychological research, also found no re-lationship between subjects' consistency scores and theextent to which they considered the attitude object im-portant or reported thinking about the attitude objectin the past. Also in line with our findings, Norman foundno differences between his high- and low-consistencysubjects in the extremity of their attitudes. Thus, thereis little reason to believe that individuals whose attitudesare characterized by high (vs. low) consistency will nec-essarily possess highly polarized attitudes.

AFFECTIVE-COGNITIVE CONSISTENCY AND SELF-PERCEPTION 11

deserves renewed empirical attention by at-titude researchers. To illustrate only onepossible direction that such research mighttake, we are currently pursuing Greenwald's(1968) suggestion that the affective and cog-nitive components of attitude may have sep-arate antecedents (Chaiken & Garisto,Note 1).

Finally, it is interesting to compare theaffective-cognitive consistency construct andexisting research on this variable with recenttheorizing and research in personality psy-chology regarding self-schemata (Kuiper& Rogers, 1979; Markus, 1977; Rogers,Kuiper, & Kircker, 1977). According to selftheorists (e.g., Markus, 1977), individualswho are schematic (vs. aschematic) along aparticular trait dimension (e.g., indepen-dence-dependence) are people who possesswell-articulated cognitive generalizationsabout the self in that trait domain. Becausean attitude can be regarded as a judgmentconcerning the self's orientation toward someobject, an individual whose attitude towardsome object is characterized by high (vs.low) affective-cognitive consistency mightplausibly be labeled attitude schematic (vs.aschematic).

Existing research indicates that self-sche-mata facilitate the processing of informationabout the self and that schematics (vs. as-chematics) can supply more behavioral ev-idence supporting their self-perceptions, aremore resistant to counterschematic infor-mation, and exhibit greater confidence inpredicting their behaviors on schema-relateddimensions (Kuiper & Rogers, 1979; Mar-kus, 1977; Rogers et al., 1977). Schematicsare also postulated to manifest greater cross-situational consistency in their trait-relatedbehaviors (Markus, 1977). The idea thataffective-cognitive consistency may be aschemalike concept is given credence byNorman's (1975) demonstration that peoplewho exhibit high (vs. low) structural consis-tency manifest greater attitude-behaviorcorrespondence. Also compatible with thisnotion are Rosenberg's (1968) demonstra-tion that higher affective-cognitive consis-tency confers greater resistance to persua-sion and the present study's results, sinceboth experiments could be interpreted as in-dicating that attitude-schematic individuals

are more resistant to counterschematic in-formation.

We believe that it would be premature toassert that the assessment of affective-cog-nitive consistency provides an acceptablemethod for identifying persons whose statedattitudes do or do not reflect a well-articu-lated schema in a particular attitudinal do-main. Nevertheless, we are intrigued by thetheoretical and empirical parallels that seemto exist between the affective-cognitive con-sistency literature and the literature on self-schemata, and feel that further research ex-ploring the utility of the consistency con-struct as a measure of attitude schematicityis warranted. As suggested by the literatureon self-schemata, such research mightinvestigate whether greater structural con-sistency facilitates the processing of atti-tude-relevant information, whether high-and low-consistency persons differ in theirpredictions regarding the likelihood of per-forming attitude-consistent behaviors, andwhether they differ in their ability to providebehavioral or cognitive evidence supportingtheir stated attitudes.

Reference Note

1. Chaiken, S., & Garisto, B. Experiential antecedentsof affective-cognitive consistency. Research in prog-ress, University of Toronto, 1981.

References

Allport, G. W. Attitudes. In C. Murchison (Ed.), Hand-book of social psychology. Worcester, Mass.: ClarkUniversity Press, 1935.

Anderson, N. H. A simple model for information inte-gration. In R. P. Abelson et al., (Eds.), Theories ofcognitive consistency: A sourcebook. Chicago: RandMcNally, 1968.

Bagozzi, R. P. The construct validity of the affective,behavioral and cognitive components of attitude byanalysis of covariance structures. Multivariate Be-havioral Research, 1978, 13. 9-31.

Bagozzi, R. P., & Burnkrant, R. E. Attitude organi-zation and the attitude-behavior relationship. Journalof Personality and Social Psychology, 1979,37. 913-929.

Bern, D. J. Self-perception: An alternative interpretationof cognitive dissonance phenomena. PsychologicalReview, 1967, 74, 183-200.

Bern, D. J. Self-perception theory. In L. Berkowitz(Ed.), Advances in experimental social psychology(Vol. 6). New York: Academic Press, 1972.

Bern, D. J., & McConnell, H. K. Testing the self-per-ception explanation of dissonance phenomena: On the

12 SHELLY CHAIKEN AND MARK W. BALDWIN

salience of premanipulation attitudes. Journal of Per-sonality and Social Psychology, 1970, 14, 23-31.

Calder, B. J., & Staw, B. M. Self-perception of intrinsicand extrinsic motivation. Journal of Personality andSocial Psychology, 1975, 31, 599-605.

Crowne, D., & Marlowe, D. The approval motive. NewYork: Wiley, 1964.

Deci, E. L. Effects of externally mediated rewards onintrinsic motivation. Journal of Personality and So-cial Psychology, 1971, 18, 105-115.

Eagly, A. H., & Himmelfarb, S. Attitudes and opinions.Annual Review of Psychology, 1978, 29, 517-554.

Fazio, R. H., & Zanna, M. P. Attitudinal qualities re-lating to the strength of the attitude-behavior rela-tionship. Journal of Experimental Social Psychol-ogy, 1978, 14, 398-408.

Fazio, R. H., Zanna, M. P., & Cooper, J. Dissonanceversus self-perception: An integrative review of eachtheory's proper domain of application. Journal ofExperimental Social Psychology, 1977,13, 464-479.

Fishbein, M., & Ajzen, I. Attitudes toward objects aspredictors of single and multiple behavioral criteria.Psychological Review, 1974, 81, 59-74.

Fishbein, M., & Ajzen, I. Belief, attitude, intention, andbehavior: An introduction to theory and research.Reading, Mass.: Addison-Wesley, 1975.

Greenwald, A. G. On defining attitude and attitude the-ory. In A. G. Greenwald, T. C. Brock, & T. M. Os-trom (Eds.), Psychological foundations of attitudes.New York: Academic Press, 1968.

Harvey, J. H., Ickes, W., & Kidd, R. F. A conversationwith Edward E. Jones and Harold H. Kelley, InJ. H. Harvey, W. Ickes, & R. F. Kidd (Eds.), Newdirections in attribution research (Vol. 2). Hillsdale,N.J.: Erlbaum, 1978.

Katz, D., & Stotland, E. A preliminary statement to atheory of attitude structure and change. In S. Koch(Ed.), Psychology: A study of a science (Vol. 3). NewYork: McGraw-Hill, 1959.

Kiesler, C. A., Nisbett, R. E., & Zanna, M. P. Oninferring one's beliefs from one's behavior. Journalof Personality and Social Psychology, 1969, / / , 321-327.

Kothandapani, V. Validation of feeling, belief, and in-tention to act as three components of attitude andtheir contribution to prediction of contraceptive be-havior. Journal of Personality and Social Psychol-ogy, 1971, 19, 321-333.

Kuiper, N. A., & Rogers, T. B. Encoding of personalinformation: Self-other differences. Journal of Per-sonality and Social Psychology, 1979, 37, 499-514.

Lepper, M. R., Greene, D., & Nisbett, R. E. Under-mining children's intrinsic interest with extrinsic re-ward: A test of the overjustitication hypothesis. Jour-nal of Personality and Social Psychology, 1973, 28,129-137.

Markus, H. Self-schemata and processing information

about the self. Journal of Personality and SocialPsychology, 1977, 35, 63-78.

McGuire, W. J. The nature of attitudes and attitudechange. In G. Lindsey & E. Aronson (Eds.), Thehandbook of social psychology (Vol. 3, 2nd ed.).Reading, Mass.: Addison-Wesley, 1969.

McNemar, Q. Psychological statistics (4th ed.). NewYork: Wiley, 1969.

Norman, R. Affective-cognitive consistency, attitudes,conformity, and behavior. Journal of Personality andSocial Psychology, 1975, 32, 83-91.

Ostrom, T. M. The relationship between the affective,behavioral, and cognitive components of attitude.Journal of Experimental Social Psychology, 1969,5, 12-30.

Rogers, T. B., Kuiper, N. A., & Kircker, W. S. Self-reference and the encoding of personal information.Journal of Personality and Social Psychology, 1977,35, 677-688.

Rosenberg, M. J. A structural theory of attitude dy-namics. Public Opinion Quarterly, 1960, 24, 319-341.

Rosenberg, M. J. Hedonism, inauthenticity, and othergoals toward expansion of a consistency theory. InR. P. Abelson, et al. (Eds.), Theories of cognitiveconsistency: A sourcebook. Chicago: Rand McNally,1968.

Rosenberg, M. J., & Hovland, C. I. Cognitive, affective,and behavioral components of attitudes. In M. J.Rosenberg, et al. (Eds.), Attitude organization andchange. New Haven, Conn.: Yale University Press,1960.

Ross, M. The self-perception of intrinsic motivation. InJ. H. Harvey, W. J. Ickes, & R. F. Kidd (Eds.), Newdirections in attribution research (Vol. 1). Hillsdale,N.J.: Erlbaum, 1976.

Salancik, G. R. Inference of one's attitude from behaviorrecalled under linguistically manipulated cognitivesets. Journal of Experimental Social Psychology,1974, 10, 415-427.

Salancik, G. R., & Conway, M. Attitude inferencesfrom salient and relevant cognitive content about be-havior. Journal of Personality and Social Psychol-ogy, 1975, 32, 829-840.

Taylor, S. E., & Fiske, S. T. Salience, attention, andattribution: Top of the head phenomena. In L. Ber-kowitz (Ed.), Advances in experimental social psy-chology (Vol. 11). New York: Academic Press, 1978.

Wood, W. The retrieval of attitudinally-relevant in-formation from memory: Effects on susceptibility topersuasion and on intrinsic motivation. Unpublisheddoctoral dissertation, University of Massachusetts,1980.

Wyer, R. S., Jr. Cognitive organization and change: Aninformation processing approach. Potomac, Md.:Erlbaum, 1974

Received June 4, 1980Revision received December 16, 1980 •