biased assimilation and attitude polarization: the effects ... · biased assimilation and attitude...

Journal of Personality and Social Psychology1979, Vol. 37, No. 11, 2098-2109

Biased Assimilation and Attitude Polarization: The Effects ofPrior Theories on Subsequently Considered Evidence

Charles G. Lord, Lee Ross, and Mark R. LepperStanford University

People who hold strong opinions on complex social issues are likely to examinerelevant empirical evidence in a biased manner. They are apt to accept "con-firming" evidence at face value while subjecting "discontinuing" evidence tocritical evaluation, and as a result to draw undue support for their initial posi-tions from mixed or random empirical findings. Thus, the result of exposingcontending factions in a social dispute to an identical body of relevant em-pirical evidence may be not a narrowing of disagreement but rather an in-crease in polarization. To test these assumptions and predictions, subjectssupporting and opposing capital punishment were exposed to two purportedstudies, one seemingly confirming and one seemingly disconfirming their exist-ing beliefs about the deterrent efficacy of the death penalty. As predicted, bothproponents and opponents of capital punishment rated those results andprocedures that confirmed their own beliefs to be the more convincing andprobative ones, and they reported corresponding shifts in their beliefs as thevarious results and procedures were presented. The net effect of such evalua-tions and opinion shifts was the postulated increase in attitude polarization.

The human understanding when it has once adoptedan opinion draws all things else to support and agreewith it. And though there be a greater number andweight of instances to be found on the other side,yet these it either neglects and despises, or else bysome distinction sets aside and rejects, in order thatby this great and pernicious predetermination theauthority of its former conclusion may remaininviolate. (Bacon, 1620/1960)

Often, more often than we care to admit,our attitudes on important social issues reflectonly our preconceptions, vague impressions,and untested assumptions. We respond tosocial policies concerning compensatory edu-cation, water fluoridation, or energy conser-

This research was supported in part by ResearchGrants MH-26736 from the National Institute ofMental Health and BNS-78-01211 from the Na-tional Science Foundation to Mark R. Lepper andLee Ross. The authors would like to express theirappreciation to Daryl Bern, Richard Nisbett, RobertZajonc, and Mark Zanna for their thoughtful com-ments and suggestions on earlier drafts of thisarticle, and to Lisa Burns for acting as the experi-menter.

Requests for reprints should be sent to any ofthe authors, Department of Psychology, StanfordUniversity, Stanford, California 94305.

vation in terms of the symbols or metaphorsthey evoke (Abelson, 1976; Kinder &Kiewiet, Note 1) or in conformity with viewsexpressed by opinion leaders we like or re-spect (Katz, 19S7). When "evidence" isbrought to bear it is apt to be incomplete,biased, and of marginal probative value—typically, no more than a couple of vivid,concrete, but dubiously representative in-stances or cases (cf. Abelson, 1972; Nisbett& Ross, in press). It is unsurprising, therefore,that important social issues and policiesgenerally prompt sharp disagreements, evenamong highly concerned and intelligent citi-zens, and that such disagreements often sur-vive strenous attempts at resolution throughdiscussion and persuasion.

An interesting question, and one thatprompts the present research, involves theconsequences of introducing the opposingfactions to relevant and objective data. Thisquestion seems particularly pertinent for con-temporary social scientists, who have fre-quently called for "more empirically based"social decision making (e.g., Campbell, 1969).Very likely, data providing consistent and

Copyright 1979 by the American Psychological Association, Inc. 0022-3514/79/3711-2098$00.75

2098

BIASED ASSIMILATION 2099

unequivocal support for one or another posi-tion on a given issue can influence decisionmakers and, with sufficiently energetic dis-semination, public opinion at large. But whateffects can be expected for more mixed orinconclusive evidence of the sort that isbound to arise for most complex social issues,especially where full-fledged experimentsyielding decisive and easy-to-generalize resultsare a rarity? Logically, one might expectmixed evidence to produce some moderationin the views expressed by opposing factions.At worst, one might expect such inconclusiveevidence to be ignored.

The present study examines a rather differ-ent thesis—one born in an analysis of thelayperson's general shortcomings as an intui-tive scientist (cf. Nisbett & Ross, in press;Ross, 1977) and his more specific short-comings in adjusting unwarranted beliefs inthe light of empirical challenges (cf. Ross,Lepper, & Hubbard, 1975). Our thesis is thatbelief polarization will increase, rather thandecrease or remain unchanged, when mixedor inconclusive findings are assimilated byproponents of opposite viewpoints. This"polarization hypothesis" can be derived fromthe simple assumption that data relevantto a belief are not processed impartially. In-stead, judgments about the validity, reliability,relevance, and sometimes even the meaningof proffered evidence are biased by the ap-parent consistency of that evidence with theperceiver's theories and expectations. Thusindividuals will dismiss and discount empiricalevidence that contradicts their initial viewsbut will derive support from evidence, of nogreater probativeness, that seems consistentwith their views. Through such biased assim-ilation even a random set of outcomes orevents can appear to lend support for anentrenched position, and both sides in a givendebate can have their positions bolstered bythe same set of data.

As the introductory quotation suggests, thenotions of biased assimilation and resultingbelief perseverance have a long history.Beyond philosophical speculations and awealth of anecdotal evidence, considerableresearch attests to the capacity of preconcep-tions and initial theories to bias the consider-

ation of subsequent evidence, including workon classic Einstellung effects (Luchins, 1942,1957), social influence processes (Asch,1946), impression formation (e.g., Jones &Goethals, 1971), recognition of degradedstimuli (Bruner & Potter, 1964), resistanceto change of social attitudes and stereotypes(Abelson, 1959; Allport, 1954), self-fulfillingprophecies (Merton, 1948; Rosenhan, 1973;Snyder, Tanke, & Berscheid, 1977), and thepersistence of "illusory correlations" (Chap-man & Chapman, 1967, 1969). In a partic-ularly relevant recent demonstration, Ma-honey (1977) has shown that trained socialscientists are not immune to theory-basedevaluations. In this study, professional re-viewers' judgments about experimental pro-cedures and resultant publication recom-mendations varied dramatically with thedegree to which the findings of a study underreview agreed or disagreed with the reviewers'own theoretical predilections.

Thus, there is considerable evidence thatpeople tend to interpret subsequent evidenceso as to maintain their initial beliefs. Thebiased assimilation processes underlying thiseffect may include a propensity to rememberthe strengths of confirming evidence but theweaknesses of disconfirming evidence, to judgeconfirming evidence as relevant and reliablebut disconfirming evidence as irrelevant andunreliable, and to accept confirming evidenceat face value while scrutinizing disconfirmingevidence hypercritically. With confirming evi-dence, we suspect that both lay and profes-sional scientists rapidly reduce the complexityof the information and remember only a fewwell-chosen supportive impressions. With dis-confirming evidence, they continue to reflectupon any information that suggests less dam-aging "alternative interpretations." Indeed,they may even come to regard the ambiguitiesand conceptual flaws in the data opposingtheir hypotheses as somehow suggestive ofthe fundamental correctness of those hypoth-eses. Thus, completely inconsistent or evenrandom data—when "processed" in a suitablybiased fashion—can maintain or even rein-force one's preconceptions.

The present study was designed to examineboth the biased assimilation processes that

2100 C. LORD, L. ROSS, AND M. LEPPER

may occur when subjects with strong initialattitudes are confronted with empirical dataconcerning a controversial social issue and theconsequent polarization of attitudes hypoth-esized to result when subjects with differinginitial attitudes are exposed to a common setof "mixed" experimental results. The socialcontroversy chosen for our investigation wasthe issue of capital punishment and its effec-tiveness as a deterrent to murder. This choicewas made primarily because the issue is thesubject of strongly held views that frequentlydo become the target of public education andmedia persuasion attempts, and has been thefocus of considerable social science researchin the last twenty years. Indeed, as our basichypothesis suggests, contending factions inthis debate often cite and derive encourage-ment from the same body of inconclusivecorrelational research (Furman v. Georgia,1972; Sarat & Vidmar, 1976; Sellin, 1967).

In the present experiment, we presentedboth proponents and opponents of capitalpunishment first with the results and thenwith procedural details, critiques, and rebut-tals for two studies dealing with the deterrentefficacy of the death penalty—one study con-firming their initial beliefs and one study dis-confirming their initial beliefs. We anticipatedbiased assimilation at every stage of this pro-cedure. First, we expected subjects to ratethe quality and probative value of studiesconfirming their beliefs on deterrent efficacymore highly than studies challenging theirbeliefs. Second, we anticipated correspondingeffects on subjects' attitudes and beliefs suchthat studies confirming subjects' views wouldexert a greater impact than studies discon-firming those views. Finally, as a function ofthese assimilative biases, we hypothesized thatthe net result of exposure to the conflictingresults of these two studies would be anincreased polarization of subjects' beliefs ondeterrent efficacy and attitudes towardscapital punishment.

Method

Subjects

A total of 1S1 undergraduates completed an in-classquestionnaire that included three items on capitalpunishment. Two to four weeks later, 48 of these

students were recruited to participate in a relatedexperiment as partial fulfillment of a course require-ment. Twenty-four were "proponents" who favoredcapital punishment, believed it to have a deterrenteffect, and thought most of the relevant researchsupported their own beliefs. Twenty-four were "op-ponents" who opposed capital punishment, doubtedits deterrent effect, and thought that the relevantresearch supported their views.

Procedure

Upon entering the experiment, mixed groups ofproponents and opponents were seated at a largetable. The experimenter, blind to subjects' attitudes,told them that they would each be asked to read2 of 20 randomly selected studies on the deterrentefficacy of the death penalty and asked them to usetheir own "evaluative powers" in thinking aboutwhat the author(s) of the study did, what the criticshad to say, and whether the research provided sup-port for one side or the other of this issue.

The experimenter next showed subjects a set of10 index cards, each containing a brief statement ofthe results of a single study. Each subject was askedto choose one card and read it silently. In reality,all 10 cards in any one session were identical, pro-viding either prodeterrent information, for example:

Kroner and Phillips (1977) compared murder ratesfor the year before and the year after adoption ofcapital punishment in 14 states. In 11 of the 14states, murder rates were lower after adoption ofthe death penalty. This research supports thedeterrent effect of the death penalty.

or antideterrent information, for example:

Palmer and Crandall (1977) compared murderrates in 10 pairs of neighboring states with differ-ent capital punishment laws. In 8 of the 10 pairs,murder rates were higher in the state with capitalpunishment. This research opposes the deterrenteffect of the death penalty.

To control for order effects, half of the proponentsand half of the opponents saw a "prodeterrence"result first, and half saw an "antideterrence" resultfirst. The studies cited, although invented specificallyfor the present study, were characteristic of researchfound in the current literature cited in judicialdecisions.

After reading one of these "result cards," subjectsanswered two sets of questions, on 16-point scales,about changes in their attitudes toward capitalpunishment (from —8 = more opposed, to 8 = morein favor) and their beliefs about the deterrentefficacy of the death penalty (from —8 = less beliefthat capital punishment has a deterrent effect, to 8 =more belief in the deterrent effect). One set of ques-tions examined change occasioned by the single pieceof information they had just finished reading; asecond set of questions assessed the cumulative change


produced by all of the materials read since the startof the experiment.1

Next the experimenter distributed detailed researchdescriptions bearing code letters corresponding tothose on the result cards. The descriptions gavedetails of the researchers' procedure, reiterated theresults, mentioned several prominent criticisms ofthe study "in the literature," listed the authors'rebuttals to some of the criticisms and depicted thedata both in table form and graphically. After read-ing this more detailed description and critique of thefirst study, subjects were asked to judge how wellor poorly the study had been conducted (from —8 =very poorly done, to 8 = very well done), and howconvincing the study seemed as evidence on thedeterrent efficacy of capital punishment (from —8 =completely unconvincing, to 8 = completely convinc-ing)." Following this evaluation, subjects were askedto write why they thought the study they had justread did or did not support the argument that capitalpunishment is a deterrent to murder, and then toanswer a second set of attitude and belief changequestions on the effects of the description alone andthe effects of all experimental materials (i.e., theresults and subsequent description and critique) upto that point in time.

Following completion of these questions, the entireprocedure was repeated, with a second fictitious studyreporting results opposite to those of the first. Again,subjects initially received only a brief description ofthe results of this second study but were then pro-vided with a detailed presentation of the procedure,results, and critiques. As before, subjects were askedto evaluate both the impact of each single piece ofevidence and the impact of all experimental materialsup to that point in the experiment on their attitudestoward capital punishment and their beliefs concern-ing its deterrent efficacy.

To control for possible differences in the inherentplausibility of the two studies, two sets of materialswere employed that interchanged the ostensibleresults of the two invented experiments. The overalldesign was thus completely counterbalanced withrespect to subjects' initial attitudes, order of confirm-ing vs. discontinuing evidence, and the association ofthe "before-after" vs. "adjacent states" designs withpositive or negative results.3 At the end of the pro-cedure, subjects were carefully debriefed concerningthe fictitious nature of the studies and were askednot to reveal this deception to others. In additionto insuring that subjects understood the fictionalnature of the experimental materials, this debriefingincluded discussion of the processes underlying theassimilation of evidence to previous theories andreassurance that a skeptical reaction to poorlydesigned research is often a praiseworthy cognitiveresponse.

Results

Evaluations oj the Two Studies

Our first hypothesis was that subjects hold-ing different initial positions would differ-

entially evaluate the quality and "convincing-ness" of the same empirical studies and find-ings. The relevant evaluations, presented inTable 1, revealed strong support for thehypothesized bias in favor of the study thatconfirmed subjects' initial attitudes.

A two-way analysis of variance (InitialAttitude X Order of Presentation) on thedifferences between ratings of convincingnessof the prodeterrence and antideterrence stud-ies yielded only a main effect of initial atti-tude, F(l , 44) = 32.07, p < .001. Proponents

1 Since most of our subjects had reported initialpositions at, or very close to, the ends of theattitude and belief scales used for selection pur-poses, our initial plan to assess attitude polariza-tion—in terms of difference scores assessing changesin subjects' attitudes and beliefs on these samescales from these initial measures to the completionof the experiment—proved impossible. As a substi-tute, we employed three sorts of measures to assessattitude change. First, we asked subjects, after eachnew piece of information, to indicate any changesin their attitudes and beliefs occasioned by thatsingle piece of information. Second, we asked sub-jects, at these same points, to report on "cumula-tive" changes in their attitudes and beliefs sincethe start of the experiment. Third, subjects wereasked to keep "running records" of their attitudesand beliefs on enlarged versions of the scales ini-tially used for selection purposes. Although all ofthese measures individually raise some problems,the congruence of data across these different mea-surement devices gives us some confidence con-cerning the results reported. Indeed, because theresults obtained on the "running record" measureso completely parallel the findings obtained on thecumulative change question depicted in Figures 1and 2, in terms of both the array of means and theobtained significance levels, the data from thismeasure will not be reported separately.

2 Subjects were also asked, at this point, whetherthey thought the researchers had favored or op-posed the death penalty and whether they thoughtan unbiased consideration should lead one to treatthe study as evidence for or against capital punish-ment. Analyses on the first question showed onlythat subjects believed the researchers' attitudes tocoincide with their stated results. Analyses on thesecond question proved wholly redundant withthose presented for the "convincingness" and "welldone" questions.

3 Preliminary analyses were conducted to see ifthe particular association of positive versus nega-tive results with either the before-after or ad-jacent-states designs would affect the results ob-tained. There were no significant effects or inter-actions involving this variation in stimulus materials;hence, the data were collapsed across this factor.


Table 1Evaluations of Prodeterrence and AntideterrenceStudies by Proponents and Opponents ofCapital Punishment

Study Proponents Opponents

Mean ratings of how well the twostudies had been conducted

ProdeterrenceAntideterrence

Difference

1.5-1.6

3.1

-2.1- .3-1.8

Mean ratings of how convincing the two studies wereas evidence on the deterrent efficacy of

capital punishment

ProdeterrenceAntideterrence

Difference

1.4-1.8

3.2

-2.1.1

-2.2

Note. Positive numbers indicate a positive evalua-tion of the study's convincingness or procedure.Negative numbers indicate a negative evaluation ofthe study's convincingness or procedure.

regarded the prodeterrence study as signifi-cantly more convincing than the antideter-rence study, i(23) = S.I8, p < .001,4 regard-less of whether it was the "before-after"design that suggested the efficacy of capitalpunishment and the "adjacent states" designthat refuted it, or vice versa. Opponents, bycontrast, regarded the prodeterrence study assignificantly less convincing than the anti-deterrence study, <(23) = -3.02, p < .01,again irrespective of which research designwas purported to have produced which typeof results. The same was true of the differencebetween ratings of how well done the twostudies had been, F(l, 44) = 33.52, p <.001.6 As above, proponents found the pro-deterrence study to have been better con-ducted than the antideterrence study, £ (23)= S.37, p < .001, whereas opponents foundthe prodeterrence study to have been less wellconducted, / (23) = —2.80, p < .05. As onemight expect, the correlation between the"convincingness" and "well done" questionswas substantial, r — .67, p < .001.

These differing opinions of the quality ofthe two studies were also reflected in subjects'written comments. At the risk of opening our-selves to a charge of "biased assimilation,"we present a set of subjects' comments—

selected for dramatic effect but not unrepre-sentative in content—in Table 2. As thesecomments make clear, the same study canelicit entirely opposite evaluations frompeople who hold different initial beliefs abouta complex social issue. This evidence of biasin subjects' evaluations of the quality andconvincingness of the two studies is consistentwith the biased assimilation hypothesis andsets the stage for testing our further predic-tions concerning attitude and belief polariza-tion.

Overall Attitude Polarization

Given such biased evaluations, our primaryhypothesis was that exposure to the "mixed" "data set comprised by the two studies wouldresult in a further polarization of subjects'attitudes and beliefs rather than the con-vergence that an impartial consideration ofthese inconclusive data might warrant. Totest this hypothesis requires a consideration

4 All p values reported in this article are based ontwo-tailed tests of significance.

5 In order to examine possible main effects ofeither study direction or initial attitude on sub-jects' ratings of how convincing and how well donethe studies were—findings that would not be por-trayed in the difference score analysis reported—athree-way analysis of variance (Initial Attitude XOrder of Presentation X Direction of Study) wasalso performed. There were no main effects of studydirection on either measure. A main effect of initialattitude—indicating that opponents evaluated thetotal set of evidence more negatively than did pro-ponents—proved significant for the "well done"question, F( 1,44) =4.69, p < .05, but not for the"convincing" question, F(l,44) = 1.53, ns.

6 The term mixed, we should emphasize, refers tothe fact that one study yielded evidence confirmingthe deterrent efficacy of the death penalty, whereasthe other study yielded evidence disconfirming suchefficacy (with appropriate counterbalancing of pur-ported procedures and purported results). Subjects,regardless of initial position, clearly recognized thisdiscrepancy between results, as will be apparent inour analyses of their responses to the simple state-ments of the study's main findings. We do notmean to imply that the subjects "phenomenologi-cally" judged the two studies to be of equal pro-bative value; indeed, as indicated in the precedingdiscussion, identical procedures were clearly judgedto differ in their probativeness depending on thecongruity between the study's outcomes and thesubject's initial beliefs.


Table 2Selected Comments on Prodeterrence and Antideterrence Studies by Proponentsand Opponents of Capital Punishment

Comments on

Subject Prodeterrence study Antideterrence study

Set 1 materials

S8 Proponent "It does support capital punishmentin that it presents facts showingthat there is a deterrent effectand seems to have gathered dataproperly."

S24 Proponent "The experiment was well thoughtout, the data collected was valid,and they were able to come upwith responses to all criticisms."

535 Opponent "The study was taken only 1 yearbefore and 1 year after capitalpunishment was reinstated. To bea more effective study they shouldhave taken data from at least 10years before and as many years aspossible after."

536 Opponent "I don't feel such a straightforwardconclusion can be made from thedata collected."

"The evidence given is relativelymeaningless without data abouthow the overall crime rate wentup in those years."

"There were too many flaws in thepicking of the states and too manyvariables involved in the experi-ment as a whole to change myopinion."

"The states were chosen at random,so the results show the averageeffect capital punishment hasacross the nation. The fact that 8out of 10 states show a rise inmurders stands as good evidence."

"There aren't as many uncontrolledvariables in this experiment as inthe other one, so I'm still willingto believe the conclusion made."

Set 2 materials514 Proponent "It shows a good direct comparison

between contrasting death penaltyeffectiveness. Using neighboringstates helps to make the experi-ment more accurate by usingsimilar locations."

515 Proponent "It seems that the researchersstudied a carefully selected groupof states and that they were care-ful in interpreting their results."

S25 Opponent "The data presented are a randomlydrawn set of 10. This fact seemsto be the study's biggest problem.Also many other factors are notaccounted for which are veryimportant to the nature of theresults."

S38 Opponent "There might be very differentcircumstances between the sets oftwo states, even though they weresharing a border."

"I don't think they have completeenough collection of data. Also,as suggested, the murder ratesshould be expressed as percentages,not as straight figures."

"The research didn't cover a longenough period of time to provethat capital punishment is not adeterrent to murder."

"The murder rates climbed in all buttwo of the states after new lawswere passed and no strong evidenceto contradict the researchers hasbeen presented."

"These tests were comparing thesame state to itself, so I feel itcould be a fairly good measure."

of subjects' final attitudes, after exposure toboth studies and related critiques and rebut-tals, relative to the start of the experiment.

The relevant data provide strong supportfor the polarization hypothesis. Asked fortheir final attitudes relative to the experi-


Table 3Mean Attitude and Belief Changes for aSingle Piece of Information

Issue and study

Initial attitudes

Proponents Opponents

Results only

Capital punishmentProdeterrenceAntideterrenceCombined

Deterrent efficacyProdeterrenceAntideterrenceCombined

1.3-0.7

0.6

1.9-0.9

1.0

0.4-0.9-0.5

0.7-1.6-0.9

Details, data, critiques, rebuttals

Capital punishmentProdeterrence 0.8 -0.9Antideterrence 0.7 -0.8Combined 1.5 —1.7

Deterrent efficacyProdeterrence 0.7 -1.0Antideterrence 0.7 -0.8Combined 1.4 -1.8

Note. Positive numbers indicate a more positiveattitude or belief about capital punishment and itsdeterrent effect. Negative numbers indicate a morenegative attitude or belief about capital punishmentand its deterrent effect.

ment's start, proponents reported that theywere more in favor of capital punishment,t ( 2 3 ) = S.07, p < .001, whereas opponentsreported that they were less in favor ofcapital punishment, t(23) = -3.34, p < .01.In a two-way analysis of variance (InitialAttitude X Order of Presentation), the effectof initial attitude was highly significant,F (1, 44) = 30.06, p < .001, and neither theorder effect nor the interaction approachedsignificance. Similar results characterizedsubjects' beliefs about deterrent efficacy.Proponents reported greater belief in thedeterrent effect of capital punishment, i(23)= 4.26, p < .001, whereas opponents reportedless belief in this deterrent effect, <(23) =-3.79, p<.OQl. Final attitudes towardcapital punishment and beliefs concerningdeterrent efficacy were highly correlated,r= .88, p < .001.

Such results provide strong support for themain experimental hypothesis that inconclu-

sive or mixed data will lead to increasedpolarization rather than to uncertainty andmoderation. Moreover, the degree of polar-ization shown by individual subjects was pre-dicted by differences in subjects' willingnessto be less critical of procedures yielding sup-portive evidence than of procedures yieldingnonsupportive evidence. Significant correla-tions were found between overall attitudechange regarding capital punishment and dif-ferences in ratings of both how convincing,r = .56, p < .001, and how well done, r -.56, p < .001, the studies were. Overallchanges in beliefs in deterrent efficacy pro-duced comparable correlations of .53 and .57,both ps < .001.

Components 0} Attitude Polarization

In view of this strong evidence of overallattitude polarization, it is worth examiningthe course of attitude polarization as subjects'opinions were successively assessed afterexposure to the first study, the details andcritiques of the first study, the results of thesecond study, and the details and critiques ofthe second study. At each stage, it will berecalled, subjects were asked about theimpact of the single piece of information theyhad just considered and the cumulativeimpact of all information presented to thatpoint. Let us first examine the reportedeffects of single segments of evidence andthen the effects of accumulated evidence overtime.

Effect oj Exposure to the Results ofEach Study

Considering the result cards as single piecesof evidence, both proponents and opponentsreported shifting their attitudes in the direc-tion of the stated results for both the pro-deterrence, <(47) = 4.67, p < .001, and anti-deterrence, < ( 4 7 ) = -5.15, p < .001, studies.As shown in the top half of Table 3, however,subjects' responses to the two studies alfovaried with initial attitude. Proponents tendedto be influenced more by the prodeterrencestudy and opponents more by the anti-deterrence study. Thus a two-way analysisof variance (Initial Attitude X Order of Pre-


sentation) on combined change from the tworesult cards considered individually yieldedonly a main effect of initial attitude for bothattitudes toward the death penalty, F(\, 44)= 6.3S, p < .02,7 and beliefs about its deter-rent effect, F(l, 44) = 10.37, p < .01. Inter-estingly, the analysis of beliefs regardingdeterrent efficacy also showed an unantici-pated interaction effect, F ( l , 44) = 7.48, p< .01, with proponents showing a differentialresponse to results alone regardless of orderof presentation but opponents showing a dif-ferential response to results alone only whenthe confirming study was presented first.

Effect oj Exposure to Procedures and Data,Critiques and Rebuttals

When provided with a more detaileddescription of the procedures and data, to-gether with relevant critiques and authors'rebuttals, subjects seemed to ignore the statedresults of the study. As shown in the bottomhalf of Table 3, both proponents and oppo-nents interpreted the additional information,relative to the results alone, as strongly sup-porting their own initial attitudes. Detaileddescriptions of either the prodeterrence orthe antideterrence study, with accompanyingcritiques, caused proponents to favor capitalpunishment more and believe in its deterrentefficacy more, but caused opponents to opposecapital punishment more and believe in itsdeterrent efficacy less. A two-way analysis ofvariance (Initial Attitude x Order of Pre-sentation) on attitude change for the twodescriptions combined yielded only a signifi-cant main effect of initial attitude for boththe capital punishment issue, F ( l , 44) =28.10, p < .001, and the deterrent efficacyquestion, F ( l , 44) = 26.93, p < .001.

Changes in Attitudes Across Time

Subjects' reported changes in attitudes andbeliefs, relative to the start of the experiment,following exposure to each of the four sep-arate pieces of information are depicted inFigure 1 for attitudes concerning capitalpunishment and in Figure 2 for beliefs con-cerning deterrent efficacy. These data, por-trayed separately for subjects who received

first either the prodeterrence study or theantideterrence study, provide a more detailedview of the attitude polarization process.They allow, as well, an examination of thehypothesized "rebound effect," that the pro-vision of any plausible reason for discountingdata that contradict one's preconceptions willeliminate the effects that mere knowledge ofthose data may have produced.

The existence of such a "rebound effect" isobvious from examination of these figures.Whether they encountered the discontinuingresult first or second, both proponents andopponents seemed to be swayed momentarilyby this evidence, only to revert to theirformer attitudes and beliefs (and in 23% ofthe individual cases, to even more extremepositions) after inspecting the proceduraldetails and data, and the critiques and rebut-tals found in the literature. Across all sub-jects, this rebound in opinions proved signifi-cant for both the capital punishment, £ ( 4 7 )= 4.43, p < .001, and deterrent efficacy,/ ( 4 7 ) = 4.58, p < .001, issues. By contrast,no compensating rebound effects resulted fromreading the descriptions and critiques of stud-ies supporting subjects' initial attitudes, foreither capital punishment, < ( 4 7 ) = .60, ns,or deterrent efficacy, £(47) — .23, ns.

Discussion

The results of the present experiment pro-vide strong and consistent support for theattitude polarization hypothesis and for thebiased assimilation mechanisms postulated tounderlie such polarization. The net effect ofexposing proponents and opponents of capitalpunishment to identical evidence—studiesostensibly offering equivalent levels of sup-port and disconfirmation—was to increasefurther the gap between their views. Themechanisms responsible for this polarizationof subjects' attitudes and beliefs were clearly

7 In order to rule out the possibility that direc-tion of study interacted with initial attitude, athree-way analysis of variance (Initial Attitude XOrder of Presentation X Direction of Study) Wasalso performed on these data. The relevant interac-tion term did not approach significance, F(l,44) =1.62, ns.


MORE IN FAVOR OFCAPITAL PUNISHMENT

NO CHANGE

MORE OPPOSED TOCAPITAL PUNISHMENT

MORE IN FAVOR OFCAPITAL PUNISHMENT

NO CHANGE

MORE OPPOSED TOCAPITAL PUNISHMENT

1 1

H PROPONENTS

Q OPPONENTS

-ll i .L_J

1 1

\% %u \

-

uPRO- DETERRENCE ANTI-DETERRENCE

RESULTS DETAILS RESULTS DET1 1 1

AILS

TIME

-1

1 1 1

-

B71ANn-DETERREfJCE

-

MB-

-PRO-DETERRENCE

RESULTS DETAILS RESULTS DETAILS

+j+2

+1

0

-1

—2

Figure J . Top panel: Attitude changes on capital punishment relative to start of experiment asreported across time by subjects who received prodeterrence study first. Bottom panel: Attitudechanges on capital punishment relative to start of experiment as reported across time by subjectswho received antideterrence study first.

suggested by correlational analyses. Subjects'decisions about whether to accept a study'sfindings at face value or to search for flawsand entertain alternative interpretationsseemed to depend far less on the particularprocedure employed than on whether thestudy's results coincided with their existingbeliefs.

The Normative Issue

It is worth commenting explicitly about thenormative status of our subjects' apparentbiases. First, there can be no real quarrel witha willingness to infer that studies supportingone's theory-based expectations are moreprobative than, or methodologically superiorto, studies that contradict one's expectations.When an "objective truth" is known or

strongly assumed, then studies whose out-comes reflect that truth may reasonably begiven greater credence than studies whoseoutcomes fail to reflect that truth. Hence thephysicist would be "biased," but appropri-ately so, if a new procedure for evaluating thespeed of light were accepted if it gave the"right answer" but rejected if it gave the"wrong answer." The same bias leads mostof us to be skeptical about reports of mirac-ulous virgin births or herbal cures for cancer,and despite the risk that such theory-basedand experience-based skepticism may renderus unable to recognize a miraculous eventwhen it occurs, overall we are surely wellserved by our bias. Our subjects' willingnessto impugn or defend findings as a function oftheir conformity to expectations can, in part,be similarly defended. Only the strength of


their initial convictions in the face of theexisting inconclusive social data and argu-ments can be regarded as "suspect."

Our subjects' main inferential shortcoming,in other words, did not lie in their inclinationto process evidence in a biased manner. Will-ingness to interpret new evidence in the lightof past knowledge and experience is essentialfor any organism to make sense of, andrespond adaptively to, its environment.Rather, their sin lay in their readiness to useevidence already processed in a biased man-ner to bolster the very theory or belief thatinitially "justified" the processing bias. Inso doing, subjects exposed themselves to thefamiliar risk of making their hypotheses un-falsifiable—a serious risk in a domain whereit is clear that at least one party in a disputeholds a false hypothesis—and allowing them-

selves to be encouraged by patterns of datathat they ought to have found troubling.Through such processes laypeople and pro-fessional scientists alike find it all too easyto cling to impressions, beliefs, and theoriesthat have ceased to be compatible with thelatest and best evidence available (Mahoney,1976, 1977).

Polarization: Real or Merely Reported?

Before further pursuing the broader impli-cations of the present demonstration, it isnecessary to consider an important questionraised by our procedure: Did our subjectsreally show change (i.e., polarization) in theirprivate beliefs about the desirability anddeterrent efficacy of capital punishment? Cer-tainly they told us, explicitly, that their

MORE BELIEF THATCAPITAL PUNISHMENTHAS DETERRENT EFFECT

NO CHANGE

MORE DISBELIEF THATCAPITAL PUNISHMENTHAS DETERRENT EFFECT

MORE BELIEF THATCAPITAL PUNISHMENTHAS DETERRENT EFFECT

NO CHANGE

MORE DISBELIEF THATCAPITAL PUNISHMENTHAS DETERRENT EFFECT

1 1 1HI PROPONENTS

I] OPPONENTS

11 In . I_ L_J u

PRO-DETERRENCE ANTI-DETERRENCE

—

-

«

RESULTS DETAILS RESULTS DETAILS

1 1 I

-

m I!! f!

n-

E j 1

1 — '

ANTI-DETERRENCE PRO-DETERRENCE

-

-

-

-

RESULTS DETAILS RESULTS DETAILS1 1 1

+3

+2

+ 1

-1

-2

+3

+2

+ 1

_1

-2

-3

TIME

Figure 2. Top panel: Belief changes on capital punishment's deterrent efficacy relative to start ofexperiment as reported across time by subjects who received prodeterrence study first. Bottompanel: Belief changes on capital punishment's deterrent efficacy relative to start of experimentas reported across time by subjects who received antideterrence study first.


attitudes and beliefs did change after eachnew piece of evidence was presented, andfrom the beginning to the end of the experi-ment. Moreover, they did show a willingnessto report a shift in their attitudes in the direc-tion of findings that were contrary to theirbeliefs, at least until those findings wereexposed to methodological scrutiny and pos-sible alternative interpretations. Nevertheless,it could be argued that subjects were notreporting real shifts in attitudes but insteadwere merely reporting what they believed tobe a rational or appropriate response to eachincrement in the available evidence. Althoughwe believe that it remains an impressivedemonstration of assimilation biases to showthat contending factions both believe the samedata to justify their position "objectively,"the potential limitations of the present mea-sures should be kept in mind in evaluating therelationship of this study to prior polarizationresearch. As noted earlier (see Footnote 1)our intended strategy of assessing directchanges from our initial selection measures ofattitudes and beliefs, rather than asking sub-jects to report such changes within the experi-ment, was neither feasible nor appropriate,given the necessity of selecting subjects withstrong and consistent initial views on thisissue. Potentially such methodological prob-lems could be overcome in subsequent re-search through the use of less extreme samplesor, perhaps more convincingly, by seeingwhether biased assimilation of mixed evidencewill make subjects more willing to act ontheir already extreme beliefs.

Belief Perseverance and Attribution Processes

The present results importantly extendthe growing body of research on the persever-ance of impressions and beliefs. Two of thepresent authors and their colleagues have nowamassed a number of studies showing that,once formed, impressions about the self(Ross et al., 1975; Jennings, Lepper, & Ross,Note 2; Lepper, Ross, & Lau, Note 3), beliefsabout other people (Ross et al., 1975), ortheories about functional relationships be-tween variables (Anderson, Lepper, & Ross,Note 4) can survive the total discrediting of

the evidence that first gave rise to suchbeliefs. In essence, these prior studies demon-strate that beliefs can survive the completesubtraction of the critical formative evidenceon which they were initially based. In a com-plementary fashion, the present study showsthat strongly entrenched beliefs can also sur-vive the addition of nonsupportive evidence.

These findings pose some fundamental ques-tions for traditional attribution models. Tothe extent that beliefs and impressions can beshown to persevere in the face of subsequentchallenging data, we need a "top down"rather than-—or perhaps in conjunction with—a "bottom up" approach (cf. Bobrow &Norman, 1975) to the question of how indi-viduals extract meaning from their socialenvironment. Instead of viewing people asimpartial, data-driven processors, the presentresearch suggests our models must take intoaccount the ways in which intuitive scientistsassess the relevance, reliability, representative-ness, and implications of any given sample ofdata or behavior within the framework of thehypotheses or implicit theories they bringto the situation (Lepper, 1977). In everydaylife, as well as in the course of scientificcontroversies (cf. Kuhn, 1970), the mereavailability of contradictory evidence rarelyseems sufficient to cause us to abandon ourprior beliefs or theories.

Social Science Research and Social Policy

We conclude this article, as we began it,by considering the important links betweensocial policy, public attitudes and beliefsabout such policy, and the role of the socialscientist. If our study demonstrates anything,it surely demonstrates that social scientistscan not expect rationality, enlightenment,and consensus about policy to emerge fromtheir attempts to furnish "objective" dataabout burning social issues. If people of oppos-ing views can each find support for thoseviews in the same body of evidence, it is smallwonder that social science research, dealingwith complex and emotional social issues andforced to rely upon inconclusive designs,measures, and modes of analysis, will fre-quently fuel rather than calm the fires ofdebate.


Reference Notes

1. Kinder, D. R., & Kiewiet, D. R. Economic griev-ances and political behavior: The role of personaldiscontents and symbolic judgments. Paper pre-sented at the meeting of the American Psycho-logical Association, San Francisco, August 1977.

2. Jennings, D. L., Lepper, M. R., & Ross, L. Per-sistence of impressions of personal persuasiveness:Perseverance of erroneous self assessments outsidethe debriefing paradigm. Unpublished manuscript,Stanford University, 1978.

3. Lepper, M. R., Ross, L., & Lau, R. R. Persistenceof inaccurate and discredited personal impres-sions: A field demonstration of attributionalperseverance. Unpublished manuscript, StanfordUniversity, 1979.

4. Anderson, C. A., Lepper, M. R., & Ross, L. Theoryperseverance: The persistent influence of dis-credited information. Unpublished manuscript,Stanford University, 1979.

References

Abelson, R. P. Modes of resolution of belief dilemmas.Conflict Resolution, 1959, 3, 343-352.

Abelson, R. P. Are attitudes necessary? In B. T.King & E. McGuinnies (Eds.), Attitudes, conflict,and social change. New York: Academic Press,1972.

Abelson, R. P. Script processing in attitude formationand decision making. In J. S. Carroll and J. W.Payne (Eds.), Cognition and social behavior.Hillsdale, N.J.: Erlbaum, 1976.

Allport, G. W. The nature of prejudice. Reading,Mass.: Addison-Wesley, 1954.

Asch, S. Forming impressions of personality. Journalof Abnormal and Social Psychology, 1946, 41, 258-290.

Bacon, F. The new organon and related writings.New York: Liberal Arts Press, 1960. (Originallypublished, 1620.)

Bobrow, D. G., & Norman, D. A. Some principlesof memory schemata. In D. G. Bobrow & A.Collins (Eds.), Representation and understanding:Studies in cognitive science. New York: AcademicPress, 1975.

Bruner, J. S., & Potter, M. C. Interference in visualrecognition. Science, 1964, 144, 424-425.

Campbell, D. T. On reforms as experiments. Amer-ican Psychologist, 1969, 24, 409-429.

Chapman, L., & Chapman, J. The genesis of popularbut erroneous psychodiagnostic observations.Journal of Abnormal Psychology, 1967, 72, 193-204.

Chapman, L., & Chapman, J. Illusory correlation asan obstacle to the use of valid psychodiagnostic

signs. Journal of Abnormal Psychology, 1969, 74,271-280.

Furman v. Georgia, 408 U.S. 238, 1972.Jones, E. E., & Goethals, G. R. Order effects in

impression formation. Attribution context and thenature of the entity. In E. E. Jones et al. (Eds.),Attribution: Perceiving the causes of behavior,Morristown, N.J.: General Learning Press, 1971.

Katz, E. The two-step flow of communication: Anup-to-date report on an hypothesis. Public OpinionQuarterly, 1957, 21, 61-68.

Kuhn, T. S. The structure of scientific revolutions(2nd ed.). Chicago: University of Chicago Press,1970.

Lepper, M. R. Cognition and social behavior. Science,1977, 196, 765-766.

Luchins, A. S. Mechanization in problem solving:The effect of Einstellung. Psychological Mono-graphs, 1942, 54, 1-95.

Luchins, A. S. Experimental attempts to minimizethe impact of first impressions. In C. I. Ho viand(Ed.), The order of presentation in persuasion.New Haven, Conn.: Yale University Press, 1957.

Mahoney, M. J. Scientist as subject: The psycho-logical imperative. Cambridge, Mass.: Ballinger,1976.

Mahoney, M. J. Publication prejudices: An experi-mental study of confirmatory bias in the peerreview system. Cognitive Therapy and Research,1977, 7, 161-17S.

Merton, R. K. The self-fulfilling prophecy. AntiochReview, 1948, 8, 193-210.

Nisbett, R., & Ross, L. Human inference: Strategiesand shortcomings of social judgment. EnglewoodCliffs, N.J.: Prentice-Hall, in press.

Rosenhan, D. L. On being sane in insane places.Science, 1973, 179, 250-258.

Ross, L. The intuitive psychologist and his short-comings: Distortions in the attribution process. InL. Berkowitz (Ed.), Advances in experimentalsocial psychology (Vol. 10). New York: AcademicPress, 1977.

Ross, L., Lepper, M. R., & Hubbard, M. Perseverancein self-perception and social perception: Biasedattributional processes in the debriefing paradigm.Journal of Personality and Social Psychology, 1975,32, 880-892.

Sarat, A., & Vidmar, N. Public opinion, the deathpenalty, and the Eighth Amendment: Testing theMarshall hypothesis. Wisconsin Law Review, 1976,171-197.

Sellin, T. (Ed.). The death penalty. New York:Harper & Row, 1967.

Snyder, M., Tanke, E. D., & Berscheid, E. Socialperception and interpersonal behavior: On the self-fulfilling nature of social stereotypes. Journal ofPersonality and Social Psychology, 1977, 35, 656-666.

Received February 16, 1979 •

biased assimilation and attitude polarization: the effects ... · biased assimilation and attitude...

Documents