the impact of attitudes on memory: an affair to remember€¦ · the impact of attitudes on memory:...

26
Psychological Bulletin 1999, Vol. 125, No. 1,64-89 Copyright 1999 by the American Psychological Association, Inc. 0033-2909/99/S3.00 The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University of Michigan Shelly Chaiken New York University Kelly Shaw-Barnes Purdue University Many theories of the effects of attitudes on memory for attitude-relevant information would predict that attitudinally congenial information should be more memorable than uncongenial information. Yet, this meta-analysis showed that this congeniality effect is inconsistent across the experiments in this research literature and small when these effects are aggregated. The tendency of the congeniality effect to decrease over the years spanned by this literature appeared to reflect the weaker methods used in the earlier studies. The effect was stronger in 2 kinds of earlier experiments that may be tinged with artifact: those in which the coding of recall measures was not known to be blind and those that used recognition measures that were not corrected for bias. Nonetheless, several additional characteristics of the studies moderated the congeniality effect and suggested that both attitude structure and motivation to process attitude-relevant information are relevant to understanding the conditions under which people have superior memory for attitudinally congenial or uncongenial information. Do attitudes exert a causal impact on what people learn and remember? Historically, psychologists have proposed a congenial- ity effect on memory—the hypothesis that people have better memory for information that supports, confirms, or reinforces their evaluations of social, political, and personal issues than for infor- mation that undermines or challenges these attitudes. This issue of whether attitudes influence memory has endured since the begin- ning of experimental research in social psychology. It is as central to contemporary research on attitudes and social cognition (see Eagly, 1992; Eagly & Chaiken, 1993) as it was to the "new look" perspective of the late 1940s, which held that psychological states such as expectancies, schemas, values, affect, and attitudes exert selective effects at all stages of information processing (e.g., Bruner & Goodman, 1947; see also Fiske & Taylor, 1991; Olson, Roese, & Zanna, 1996). Moreover, well before the "new look," Alice H. Eagly, Department of Psychology, Northwestern University; Serena Chen, Department of Psychology, University of Michigan; Shelly Chaiken, Department of Psychology, New York University; Kelly Shaw- Barnes, Department of Psychology, Purdue University. Kelly Shaw-Barnes is now at the Department of Psychology, Knox College. This research was supported by National Institute of Mental Health Grant 1-R01-MH48972-01A2. This article was completed while Shelly Chaiken was a Fellow at the Center for Advanced Study in the Behavioral Sciences; she is grateful for financial support provided by National Science Foundation Grant SBR-9601236. Our thanks are extended to Rod Bond and Peter Smith for performing an analysis cited in this article and to Blair T. Johnson and Michael Roloff for comments on a draft of this article. A table showing the effect sizes and study characteristics for each study included in the meta-analysis is available from Alice H. Eagly. Correspondence concerning this article should be addressed to Alice H. Eagly, Department of Psychology, Northwestern University, 2029 Sheri- dan Road, Evanston, Illinois 60208-2710. Electronic mail may be sent to [email protected]. prominent philosopher-scientists such as William James (1890/ 1952) and Sigmund Freud (1938) expressed the idea that attitudes bias information processing in a manner that favors attitudinally congruent information. The first attitude theorists accepted this idea (e.g., Allport, 1935), and early textbook writers featured attitudinal selectivity as a basic principle of social psychology (e.g., Asch, 1952). Early attitude theorists were correct in recognizing the impor- tance of understanding attitudes' effects on memory. Such under- standing should illuminate the broader issue of attitudinal selec- tivity in information processing because selective memory may be mediated at least in part by selective effects at earlier processing stages (e.g., attention, encoding). In addition, research on selective memory should inform theory concerning the persistence of atti- tudes and their resistance to change because both persistence and resistance should be heightened by prior information processing that favors attitudinally congruent material and ignores or refutes incongruent material. A Brief History of Attitude Memory Research In the modal study of the impact of attitudes on memory, participants indicate their attitudes on a questionnaire and subse- quently receive information that is evaluatively consistent or in- consistent with their attitudes. Memory measures, most commonly free recall of the information, are then administered. The history of such research has been far less coherent than early intuitions and theorizing suggested. Some initial experiments seemed to provide strong support for the congeniality effect (e.g., A. L. Edwards, 1941a; Levine & Murphy, 1943). Most influential was Levine and Murphy's study, in which 5 procommunist and 5 anticommunist participants read and then tried to recall both a procommunist message and an anticommunist message on five different learning trials and then attempted to recall these messages on five subse- 64

Upload: others

Post on 18-Oct-2020

8 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

Psychological Bulletin1999, Vol. 125, No. 1,64-89

Copyright 1999 by the American Psychological Association, Inc.0033-2909/99/S3.00

The Impact of Attitudes on Memory: An Affair to Remember

Alice H. EaglyNorthwestern University

Serena ChenUniversity of Michigan

Shelly ChaikenNew York University

Kelly Shaw-BarnesPurdue University

Many theories of the effects of attitudes on memory for attitude-relevant information would predict thatattitudinally congenial information should be more memorable than uncongenial information. Yet, thismeta-analysis showed that this congeniality effect is inconsistent across the experiments in this researchliterature and small when these effects are aggregated. The tendency of the congeniality effect to decreaseover the years spanned by this literature appeared to reflect the weaker methods used in the earlierstudies. The effect was stronger in 2 kinds of earlier experiments that may be tinged with artifact: thosein which the coding of recall measures was not known to be blind and those that used recognitionmeasures that were not corrected for bias. Nonetheless, several additional characteristics of the studiesmoderated the congeniality effect and suggested that both attitude structure and motivation to processattitude-relevant information are relevant to understanding the conditions under which people havesuperior memory for attitudinally congenial or uncongenial information.

Do attitudes exert a causal impact on what people learn andremember? Historically, psychologists have proposed a congenial-ity effect on memory—the hypothesis that people have bettermemory for information that supports, confirms, or reinforces theirevaluations of social, political, and personal issues than for infor-mation that undermines or challenges these attitudes. This issue ofwhether attitudes influence memory has endured since the begin-ning of experimental research in social psychology. It is as centralto contemporary research on attitudes and social cognition (seeEagly, 1992; Eagly & Chaiken, 1993) as it was to the "new look"perspective of the late 1940s, which held that psychological statessuch as expectancies, schemas, values, affect, and attitudes exertselective effects at all stages of information processing (e.g.,Bruner & Goodman, 1947; see also Fiske & Taylor, 1991; Olson,Roese, & Zanna, 1996). Moreover, well before the "new look,"

Alice H. Eagly, Department of Psychology, Northwestern University;Serena Chen, Department of Psychology, University of Michigan; ShellyChaiken, Department of Psychology, New York University; Kelly Shaw-Barnes, Department of Psychology, Purdue University.

Kelly Shaw-Barnes is now at the Department of Psychology, KnoxCollege.

This research was supported by National Institute of Mental HealthGrant 1-R01-MH48972-01A2. This article was completed while ShellyChaiken was a Fellow at the Center for Advanced Study in the BehavioralSciences; she is grateful for financial support provided by National ScienceFoundation Grant SBR-9601236. Our thanks are extended to Rod Bondand Peter Smith for performing an analysis cited in this article and to BlairT. Johnson and Michael Roloff for comments on a draft of this article.

A table showing the effect sizes and study characteristics for each studyincluded in the meta-analysis is available from Alice H. Eagly.

Correspondence concerning this article should be addressed to Alice H.Eagly, Department of Psychology, Northwestern University, 2029 Sheri-dan Road, Evanston, Illinois 60208-2710. Electronic mail may be sent [email protected].

prominent philosopher-scientists such as William James (1890/1952) and Sigmund Freud (1938) expressed the idea that attitudesbias information processing in a manner that favors attitudinallycongruent information. The first attitude theorists accepted thisidea (e.g., Allport, 1935), and early textbook writers featuredattitudinal selectivity as a basic principle of social psychology(e.g., Asch, 1952).

Early attitude theorists were correct in recognizing the impor-tance of understanding attitudes' effects on memory. Such under-standing should illuminate the broader issue of attitudinal selec-tivity in information processing because selective memory may bemediated at least in part by selective effects at earlier processingstages (e.g., attention, encoding). In addition, research on selectivememory should inform theory concerning the persistence of atti-tudes and their resistance to change because both persistence andresistance should be heightened by prior information processingthat favors attitudinally congruent material and ignores or refutesincongruent material.

A Brief History of Attitude Memory Research

In the modal study of the impact of attitudes on memory,participants indicate their attitudes on a questionnaire and subse-quently receive information that is evaluatively consistent or in-consistent with their attitudes. Memory measures, most commonlyfree recall of the information, are then administered. The history ofsuch research has been far less coherent than early intuitions andtheorizing suggested. Some initial experiments seemed to providestrong support for the congeniality effect (e.g., A. L. Edwards,1941a; Levine & Murphy, 1943). Most influential was Levine andMurphy's study, in which 5 procommunist and 5 anticommunistparticipants read and then tried to recall both a procommunistmessage and an anticommunist message on five different learningtrials and then attempted to recall these messages on five subse-

64

Page 2: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 65

quent forgetting trials. On each trial, procommunists recalled moreof the procommunist message than anticommunists, and anticom-munists recalled more of the anticommunist message thanprocommunists.

The conceptual replications of Levine and Murphy's (1943)experiment that were reported during the 1950s produced findingsthat were generally weaker and less consistent within studies (e.g.,Mannello, 1954; Paulson, 1954). Additional findings of this erasuggested that variables such as the utility and plausibility ofinformation could moderate attitudes' impact on memory (Jones &Aneshansel, 1956; Jones & Kohler, 1958). Nonetheless, a state-ment published in the 1962 behavioral sciences report to thePresident's Science Advisory Committee conveyed strong viewsabout the effects of attitudes on information processing (Behav-ioral Sciences Subpanel, 1962):

Individuals engage in selective exposure and selective perception.Those least predisposed to change are least likely to allow themselvesto be exposed to a persuasive communication, and if they are exposed,are most likely to engage in misperception, a kind of motivatedmissing-the-point. If a new piece of information would weaken theexisting structure of their ideas and emotions, it will be shunned,rejected, or quickly forgotten; if it reinforces the structure, it will besought out, quickly accepted, and remembered, (p. 277)

This statement, prepared by a group of eminent behavioral scien-tists, promoted classic ideas about attitude selectivity, includingthe underlying assumption that people are motivated to defendtheir attitudes. This reasoning included the principle that thisdefense could take place at one or more stages of informationprocessing. Initially, people may avoid exposure to uncongenialinformation. Even if they are exposed to uncongenial information,they may nonetheless give it little attention. Moreover, they maydistort its meaning so that its impact is blunted. Even if peoplecorrectly encode uncongenial information, they may dismiss it asinvalid. All of these defensive processes could compromise mem-ory for uncongenial information.

The 1962 report from the President's Science Advisory Com-mittee represented the zenith of social psychologists' belief in theattitude congeniality hypothesis (Behavioral Sciences Subpanel,1962). The nadir appeared only a few years later. Many studiespublished during the 1960s failed to find support for the conge-niality hypothesis (e.g., Brigham & Cook, 1969; Fitzgerald &Ausubel, 1963; Greenwald & Sakumura, 1967; Waly & Cook,1966), and the results of other studies were ambiguous (e.g.,Malpass, 1969; Smith & Jamieson, 1972). The null results inparticular were accompanied by a growing perception that theprevious two decades had produced fragile congeniality effectsthat were less attributable to the true impact of attitudes on mem-ory than to methodological deficiencies, such as a possible ten-dency for congenial information to be more familiar than uncon-genial information. By the mid-1970s, consensus opinion fullyreversed the early belief in the congeniality effect. For example,Greaves (1972) described selective memory research as "unam-biguously inconclusive" (p. 327), and Greenwald (1975) portrayedit as consisting of a few early confirmations of the attitude con-geniality hypothesis and a larger number of more carefully exe-cuted failures to replicate the phenomenon. Other reviewers of theattitudes literature concurred (e.g., Fishbein & Ajzen, 1972), andtreatments of the attitude-memory relation receded in advanced

and elementary textbooks. These doubts were compounded byFreedman and Scare's (1965) questioning of support for anotherimportant attitudinal selectivity hypothesis: selective exposure toinformation. Yet, despite this discouraging environment, experi-ments on attitude memory continued to be published in psycho-logical journals (e.g., Furnham & Proctor, 1990; Hymes, 1986;Read & Rosson, 1982), albeit with decreasing frequency.

The apparent inconsistencies in the findings of experimentstesting the congeniality hypothesis and prominent challenges to theexistence of any relation between attitudes and memory thus raisedserious questions about the validity of the congeniality hypothesis.In an effort to address these issues in a meta-analytic review,Roberts (1985) obtained a small-to-moderate tendency for peopleto remember attitude-congruent information better than attitude-incongruent information, as indicated by his mean effect sizeof 0.36 in the d metric. However, Roberts's incomplete samplingof the literature and inclusion of some studies with inadequateexperimental designs cast doubt on his conclusions. Moreover, heexamined few of the numerous potential moderators of the con-geniality effect, although he did find a larger effect for experi-ments' delayed measures than their immediate measures and forexperiments conducted at earlier than later dates. Given the limi-tations of Roberts's review as well as the inconsistency of hisconclusions with prior claims that the congeniality effect is prob-ably null (Fishbein & Ajzen, 1972; Greenwald, 1975), a compre-hensive quantitative synthesis of attitude memory research is longoverdue. We thus conducted such a meta-analysis and designed itnot only to examine the empirical robustness of the congenialityeffect but also to illuminate a variety of theoretical and method-ological issues relevant to understanding attitudes' effects onmemory.

Theoretical Analysis of the Impact of Attitudes onMemory

Understanding the attitude-memory relation requkes taking intoaccount processes that may mediate this relation and variables thatmay moderate it. In the background of our analysis of underlyingprocesses is traditional reasoning by attitude theorists that peopleare motivated to defend their attitudes and that defense-motivatedprocessing often gives rise to superior memory for congenialinformation. This reasoning suggests that memory for incongruentinformation can be poor because people avoid, distort, or represssuch information (e.g., Festinger, 1957, 1964; Freud, 1938). Yet,neglected in these motivational analyses promoted by early theo-rists are the cognitive mechanisms that may play a role in attitudes'influence on memory. Most notably, to the extent that people'sattitudes and associated beliefs constitute schemas that affect in-formation processing, the poor fit of uncongenial information toattitudinal structures may inhibit its learning and retrieval. Struc-tural aspects of attitudes may play an even larger role as the timeincreases between learning information and responding to memorymeasures. For example, as time goes by, reconstruction processesmay increasingly rely on attitude structure to fill in gaps in mem-ory for previously presented information (Bartlett, 1932). There-fore, in planning our synthesis, we took into account both cogni-tive and motivational mechanisms but developed more complexassumptions about these mechanisms than had earlier attitudetheorists.

Page 3: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

66 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

First, we consider the cognitive mechanisms that may affect theattitude-memory relation. The idea that uncongenial informationhas a poor fit to attitudinal knowledge structures is too simple,because it assumes that these structures are necessarily unipolar, inthe sense that they represent mainly attitude-consistent informa-tion. Attitudinal knowledge structures likely are well supplied withbeliefs supportive of one's attitude (e.g., Feather, 1970), whichshould serve as retrieval cues for remembering congenial informa-tion and thus should promote memory for such information. How-ever, as Pratkanis (1984/1985, 1989) suggested, if people areattitudinally bipolar in the sense that they are at least familiar witharguments that oppose their attitudes, uncongenial informationmay be as strongly linked to their attitudes as is congenial infor-mation. Pratkanis's data suggested that bipolar structures are com-mon for controversial social issues such as abortion, whereasunipolar structures may be common for a variety of other topicssuch as sports and music, for which people have little knowledgeof arguments that oppose their attitudes. Despite the fact thatpeople with bipolar structures ordinarily disagree with most of thestored arguments that oppose their attitudes and agree with argu-ments that support their attitudes (Feather, 1969), their knowledgeof counterattitudinal arguments may foster memory for such in-formation. Given that the stimulus information in typical attitudememory studies pertained to controversial social issues, bipolaritymay have been common in these experiments. As a result, the fitof uncongenial information to attitudinal knowledge structuresmay have been as good as the fit of congenial information, and thecongeniality bias in memory thereby dampened.

To explore the implications of these ideas about bipolarity, weensured that the coding of the studies in our meta-analysis encom-passed the controversiality of attitudinal issues. Controversiality,which commonly induces exposure to both sides of issues, shouldheighten bipolarity, and thus might weaken the tendency for par-ticipants to better remember congenial information. Also, becausethe dramatic development of the mass media, especially television,may have increased exposure to both sides of many issues, asecular trend toward greater bipolarity in attitude structures couldhave contributed to the decrease in the congeniality effect thatRoberts (1985) found over the years spanned by this researchliterature.

Consistent with these arguments about bipolarity and unipolar-ity, Judd and Kulik (1980) maintained that to the extent thatpeople's attitudes exist as bipolar schemas that represent mainlyextreme proattitudinal and extreme counterattitudinal positions,memory may favor extreme attitude-relevant information, as op-posed to moderate information, regardless of its congeniality.Therefore, when feasible, we also calculated effect sizes represent-ing the extent to which more extreme versus more moderateinformation was better remembered.

Our theoretical analysis also acknowledges the importance ofattitude strength and, more specifically, processing motivation. Inmany contemporary theories of attitudes, defense of attitudes isregarded as a consequence of their strength, with stronger attitudespredicted to exert greater selective effects on information process-ing (see Eagly & Chaiken, 1998; Petty & Krosnick, 1995). Al-though most conceptions of attitude strength assume more potentselectivity effects for stronger attitudes, some emphasize cognitiveaspects of strength, and others emphasize motivational aspects.Examples of cognitively defined strength variables include attitude

accessibility (Fazio, 1995) and evaluative-cognitive consistency(Chaiken, Pomerantz, & Giner-Sorolla, 1995). The impossibilityof coding studies on indicators of such cognitive aspects of attitudestrength (e.g., response latencies that index attitude accessibility)led us to focus on one important, and codable, motivational indi-cator of attitude strength: value-relevant involvement.

Value-relevant involvement refers to a psychological state inwhich one's attitude toward a particular entity is linked to impor-tant values or significant reference groups (B. T. Johnson & Eagly,1989; termed ego involvement by Sherif & Hovland, 1961). Forexample, issues such as racial integration in the 1950s and 1960sand abortion rights in the 1980s and 1990s would be consideredhigh in value relevance. Because this variable has been shown topredict resistance to influence (see B. T. Johnson & Eagly, 1989;Maio & Olson, 1995) as well as affective and cognitive processesthat mediate resistance (Zuwerink & Devine, 1996), the conge-niality effect may be larger with higher levels of value relevance.Moreover, this prediction coheres with both earlier (Sherif &Hovland, 1961) and contemporary (Chaiken, Giner-Sorolla, &Chen, 1996) theoretical perspectives, which hold that defensemotivation—defined as the conscious or nonconscious goal ofmaintaining, reinforcing, or defending beliefs and attitudes—in-creases to the extent that attitudes implicate important, self-defining values and reference groups.

The meta-analysis also assessed a second form of involvement,one that we believed would dampen the level of defense motiva-tion. Specifically, we assessed outcome-relevant involvement, apsychological state in which one's attitude toward a particularentity is tied to important, near-term goals (B. T. Johnson & Eagly,1989; termed personal relevance by Petty & Wegener, 1998). Forexample, issues such as whether college courses should instituteessay exams or institute a study week prior to exam week would beconsidered relatively high in outcome relevance for student par-ticipants because resolution of these issues would have an impacton students' ability to attain their goals. Research on this variableindicates that heightened outcome relevance ordinarily increasesrelatively objective, open-minded information processing (e.g.,Boninger, Krosnick, Berent, & Fabrigar, 1995; B. T. Johnson,1994; B. T. Johnson & Eagly, 1989; Petty & Cacioppo, 1990),presumably because increased outcome relevance intensifies peo-ple's concerns about the correctness, or accuracy, of their attitudesand other judgments (e.g., Chaiken, Wood, & Eagly, 1996; Petty &Wegener, 1998). This accuracy-oriented, or open-minded, process-ing should dampen the selectivity that follows from defendingone's attitude. Therefore, outcome-relevant involvement shouldreduce the size of the congeniality effect.

A complexity in understanding the impact of heightened moti-vation to defend one's attitudes is that a variety of specific pro-cesses may be elicited by such motivation (Chaiken, Liberman, &Eagly, 1989; Kunda, 1990). Most prior conceptions of defensiveprocessing have assumed relatively passive strategies wherebypeople shun, distort, or forget attitude-incongruent information.Moving away from these somewhat simple assumptions, we con-sidered the possibility of a more active strategy that would entailgreater attention to incongruent information and more elaborationof it (e.g., K. Edwards & Smith, 1996). This conception of attitu-dinal defense suggests that stored beliefs supportive of attitudesenable people not only to remember congenial arguments but alsoto refute challenging information (Biek, Wood, & Chaiken, 1996;

Page 4: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 67

Chaiken, Wood, & Eagly, 1996; Eagly & Chaiken, 1993, 1995,1998; Greenwald, 1968; Wood, Rhodes, & Biek, 1995). Moreover,bipolarity in the form of familiarity with arguments opposed toone's own attitude may further enable refutational elaboration ofuncongenial information. Such active refutational processes wouldreflect an attitudinal bias in processing but may enhance ratherthan reduce memory for uncongenial information, because theyentail careful scrutiny of this information (e.g., Chaiken et al.,1995; Eagly, Kulesa, Brannon, Shaw-Barnes, & Hutson-Comeaux,1996; Zimmerman & Chaiken, 1996). Although it would havebeen desirable to code studies on participants' active versus pas-sive defense strategies, studies did not report information thatwould have enabled such coding. We suspected that, becauseRoberts (1985) obtained an overall congeniality effect, passivestrategies may have predominated in attitude memory experiments.However, in view of the limitations of Roberts's review andevidence of active defense of attitudes on controversial issues (e.g.,K. Edwards & Smith, 1996), we were uncertain about the relativefrequency of active and passive processes. Nonetheless, our pre-dictions that value-relevant involvement would increase the con-geniality bias and that outcome-relevant involvement would de-crease it assume that defensive goals are pursued primarily bypassive processing strategies by which people shun, avoid, anddistort uncongenial information.

In summary, active defense of attitudes might promote goodmemory for uncongenial information, and bipolar attitude struc-tures could reduce the memory advantage of congenial informa-tion. Such theoretical considerations weakened our a priori confi-dence that this meta-analysis would confirm an overallcongeniality bias in memory and suggested that experiments mightsometimes find better memory for uncongenial information. De-spite these possibilities, we framed our meta-analysis in terms ofthe traditional hypothesis that attitudes exert a congeniality bias onmemory and that this bias holds to a greater or lesser extent.

Design of the Present Synthesis

In identifying studies for inclusion in the meta-analysis, wesought only experiments of unambiguous relevance to the hypoth-esis that attitudes exert a causal impact on memory for informationthat people confront in their environments (see Method section fordetails). The sample was thus limited to studies in which partici-pants were experimentally exposed to proattitudinal and counter-attitudinal information. This criterion excluded studies in whichparticipants merely recollected whatever attitude-relevant experi-ence they may have had in the past (e.g., Ross, 1989). Limiting thesample to experiments with appropriate designs also caused us toexclude studies in which participants' attitudes were assessed afterexposure to attitude-relevant messages, because such studies arevulnerable to the alternative causal interpretation that exposure topersuasive communications changes attitudes. In addition, thissynthesis does not encompass research on memory for informationthat is congruent versus incongruent with people's expectancies, aliterature that was quantitatively synthesized by Stangor and Mc-Millan (1992) and Rojahn and Pettigrew (1992). The minimaloverlap between our sample of studies and those of these twosyntheses of research on expectancy-relevant information is theo-retically sensible because the rationale for why attitudes biasmemory in favor of congruent information need not, and typically

has not, posited expectancies as an agent or mediator of sucheffects. Although the evaluation of entities as positive or negativemay well influence people's hopes, desires, and wishes about theattitudinal position advocated by subsequently encountered infor-mation, these evaluative tendencies bear at best a complex rela-tionship to people's expectancies about the attitudinal congenialityof this information (McGuire & McGuire, 1991).

Despite these more stringent criteria for excluding data sets thatdid not provide appropriate tests of the congeniality hypothesis, thepresent synthesis is considerably more comprehensive than Rob-erts's (1985) review. Even though our selection criteria resulted inthe elimination of 10 of Roberts's 38 experiments, our sample ofstudies included 70 experiments. Although this larger numberincluded some experiments published after Roberts concluded hissearch, the increment was mainly due to more thorough searchtechniques.

The coding system for this quantitative synthesis encompassedmany potentially important methodological variables in view ofhints that early attitude memory studies might have been method-ologically flawed (e.g., Greenwald, 1975) and Roberts's (1985)finding that earlier studies yielded a larger congeniality bias. Thesevariables were designed to deal with issues as mundane as mea-surement reliability and validity and as complex as the confound-ing of attitudinal congeniality with other variables such as famil-iarity with the information.

Our methodological focus directed attention to how memorywas assessed. This issue was neglected by earlier researchers, whofocused on explanations for why and how attitudes might influenceinformation processing and generally treated memory as a mono-lithic and unitary construct. Representing some of the complexitiesof memory requires, at a minimum, that measures be classified intomajor types. Thus, like the syntheses of expectancy effects onmemory (Rojahn & Pettigrew, 1992; Stangor & McMillan, 1992),the present synthesis of attitudinal effects on memory classifiedeffect sizes according to whether they were based on recall or oneither of two types of recognition measures—uncorrected recog-nition measures, which are not corrected for guessing and otherbiases, and recognition sensitivity measures, which are corrected(see Shapiro, 1994; Srull, 1984). Although in the primary studiesresearchers virtually never stated hypotheses about the differentialimpact of attitudes on different types of memory measures, someassumptions about the processes that mediate memory for attitude-relevant information could produce differential predictions.

Predictions about the moderating impact of different classes ofmemory measures could follow from Stangor and McMillan's(1992) and Rojahn and Pettigrew's (1992) reports that recognitionsensitivity measures produced greater memory for expectancy-incongruent information, whereas uncorrected recognition mea-sures produced greater memory for congruent information. As theauthors of these two meta-analyses reasoned, this result likelyoccurred at least in part because, when confronted by recognitionmeasures, research participants tended to guess in line with theirexpectancies. To the extent that participants might also tend toguess in line with their attitudes, the congeniality effect should besmaller for recognition sensitivity measures than for uncorrectedrecognition measures. In addition, measures of recognition arethought to assess whether information has been encoded at all,whereas measures of free recall are thought to assess, at least inpart, the ability to retrieve information from memory (e.g., Stull,

Page 5: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

68 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

1984). Therefore, comparing the magnitude of the congenialityeffect on recall and recognition measures may prove informativeconcerning whether attitude memory effects are mediated by en-coding or retrieval processes.

Method

Sample of Studies

We conducted computer-based information searches using the keywordattitude (or functionally similar terms such as opinion, belief, affect, andframe of reference) or selectivity (or functionally similar terms such ascongruent—incongruent and consistent—inconsistent) when paired withmemory (or functionally similar terms such as recall, retention, recogni-tion, forgetting, and remembering). These keywords were searched in thefollowing databases from the beginning point of each database through1995: Psychological Abstracts (PsycLIT), Dissertation Abstracts Interna-tional (Dissertation Abstracts Ondisc), and Educational Resources Infor-mation Center. We searched the Social Sciences Citation Index to locatearticles that cited any of five seminal attitude memory experiments (Green-wald & Sakumura, 1967; Jones & Aneshansel, 1956; Jones & Kohler,1958; Levine & Murphy, 1943; Watson & Hartmann, 1939). Also searchedwere the reference lists of Roberts's (1985) review article and all of theattitude memory studies that we located.

One of the criteria for including studies in the sample was that partici-pants' attitudes were assessed, or a difference in their attitudes could besafely assumed between the congenial and uncongenial conditions. Forexample, in some studies examining memory for favorable or unfavorableinformation pertaining to an in-group or out-group, participants' attitudestoward their own ethnic in-group (e.g., English Canadians) could be safelyassumed to be more positive than their attitudes toward an out-group (e.g.,French Canadians; Dutta, Kanungo, & Freibergs, 1972, Experiment 3).Other criteria were that (a) standardized attitude-relevant, verbal informa-tion was presented to participants; (b) this information was favorable orunfavorable toward the attitude object; (c) memory for this informationwas assessed; and (d) the design varied the evaluative consistency betweenparticipants' attitudes and the attitudinal position of the information, pro-ducing both congenial and uncongenial conditions.

Studies were excluded if the test of congeniality was based on assump-tions about participants' attitudes that were not empirically validated andare arguably invalid given other findings in the literature. For example,Alper and Korchin's (1952) and Clark's (1940) assumption that femalestudents were more profemale and antimale than male students is risky inview of Eagly, Mladinic, and Otto's (1991) finding that male and femalestudents did not differ in their attitudes toward either men or women.Greaves's (1972) assumption that U.S. students had more favorable atti-tudes toward the United States than Sweden is risky in view of Stapf,Stroebe, and Jonas's (1986) evidence that U.S. students' attitudes towardSweden were nearly as favorable as their attitudes toward the United Statesand that their attitudes toward Swedes were slightly more favorable thantheir attitudes toward Americans.

Studies were omitted if they did not ensure that all participants wereexposed to the same stimulus information (e.g., Eberhart & Bauer, 1941;Feather, 1969). They were also excluded if the stimulus information waspresented in incomplete form (e.g., Donohew, 1966; Manis & Dawes,1961) or pictorial form (e.g., Gackenbach & Auerbach, 1985), or if theinformation took a neutral or ambiguous position on the relevant issue(e.g., Clark, 1940; Garber, 1955). Studies were also omitted if the memorymeasure did not assess memory for the stimulus information (e.g., Weiss,1953).

Studies were omitted if the attitude object that was the focus of theattitude-relevant information was not at least moderately compatible withthe attitude object implied by the attitude measure, because, for an ade-quate test of the congeniality hypothesis, these two attitude objects should

be the same or at least moderately closely related. For example, studies byMatthew (1975), Furnham and Singh (1986), and Furnham and Duignan(1989) were excluded because they assessed participants' attitudes towardequal rights for women but presented information consisting of positiveand negative statements about men or women as social groups. Becauseattitudes toward equal rights for women have been shown to be unrelatedor minimally related to attitudes toward women and men as social groups(Eagly & Mladinic, 1989; Swim & Cohen, 1997), the attitude measure wasnot compatible with the information.

Studies of the consistency of stimuli with expectancies or stereotypeswere excluded and carefully distinguished from attitude memory studies,which examined the consistency of stimuli with attitudes. Studies ofmemory for information that is congruent versus incongruent with expec-tancies are designed to compare reactions to information that is expectedversus unexpected rather than consistent versus inconsistent with a priorattitude (see Rojahn & Pettigrew, 1992; Stangor & McMillan, 1992). Apossible area of overlap between these two literatures consists of studiesexamining evaluative expectancies; for example, Reutzel and Hollings-worth (1991) presented information inducing a positive or negative attitudetoward a fictitious country before presenting a passage that containedfavorable or unfavorable information about the country. Such studies wereexcluded if they presented evaluatively consistent stimulus informationwhose content was highly redundant with the information that establishedthe expectancy and presented evaluatively inconsistent information whosecontent was highly contradictory to this information (e.g., Ikegami, 1986;Reutzel & Hollingsworth, 1991). In contrast, a study establishing evalua-tive expectancies would be included if it presented stimulus informationthat was merely evaluatively consistent or inconsistent with the valence ofthe expectancy but was not highly consistent or inconsistent in terms of itsspecific content.

Studies were excluded if attitude was assessed only as a postmeasure(e.g., Berent, Krosnick, & Boninger, 1993; Brigham & Cook, 1969; Doob,1953; Pratkanis, Syak, & Gamble, 1987; Upmeyer & Layer, 1974), becausethe stimulus information may have affected the attitude and, moreover,such studies cannot be distinguished from persuasion studies that includeda measure of memory for message content (e.g., Leippe & Elkin, 1987).However, if attitude was operationalized as agreement with individualattitude-relevant statements (e.g., Greenwald & Sakumura, 1967), studieswere included even though attitude was assessed immediately after thepresentation of the stimuli. This inclusion rule recognizes that an agree-ment rating following a statement is less likely to be confounded withpersuasion than a rating following a longer message.1 Finally, studies wereomitted if the congeniality effect size, its direction, and significance wereall unknown (e.g., Gormly, 1974).

In summary, these somewhat stringent selection criteria narrowed thesample to studies with designs appropriate to examine the impact of theattitudinal congeniality of verbal attitude-relevant information on memoryfor the information.2 The resulting sample consisted of 60 documentsreporting 70 studies.

Variables Coded From Each Study

The following general information was coded from each report: (a) yearof publication; (b) publication form (journal article; dissertation, master'sthesis, or other unpublished document); (c) participant population (highschool or junior high school students; college students; mixed or other, e.g.,

1 Subsequent analyses of the congeniality effect showed that thesestudies in which agreement was assessed immediately after presentation ofeach attitude-relevant statement did not differ systematically from the otherstudies included in the meta-analysis.

2 A list of the studies excluded by these criteria is available from AliceH. Eagly.

Page 6: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 69

visitors to Ontario Science Center)3; and (d) nation where study wasconducted (United States; other).

To represent the experimental design, coders distinguished betweenthree methods of defining whether information was attitudinally congenialor uncongenial: (a) sidedness, defined by information on participants' ownside versus the other side of the issue (e.g., Levine & Murphy, 1943;Watson & Hartmann, 1939); (b) agreeability, defined by participants' ratedagreement with the stimuli (e.g., Eiser & Monk, 1978; Judd & Kulik, 1980;Postman & Murphy, 1943); and (c) balance, defined by the sidedness andthe agreeability of the information—that is, congenial information con-sisted of agreeable information on participants' own side of the issue anddisagreeable information on the other side, and uncongenial informationconsisted of agreeable information on the other side and disagreeableinformation on participants' own side (Callahan & Olson, 1981; Olson,1980; Waly & Cook, 1966). Agreeability was determined by the partici-pants' own judgments or by ratings from independent judges. With balancedefinitions, the agreeability of the information was sometimes defined asits plausibility or effectiveness. The findings of studies using the balancedefinition often could be decomposed to provide a test of the congenialityhypothesis according to either the sidedness or the agreeability definitionor both definitions (e.g., Olson, 1980). In such cases, the sidedness defi-nition considered only agreeable information on participants' own side anddisagreeable information on the other side; the agreeability definitionconsidered all agreeable versus disagreeable information, regardless of itsside.

Other features of the design that were coded included (a) independentvariables contributing to the test of congeniality (participant attitude only;attitudinal position of information only; both participant attitude and po-sition of information; mixed); (b) position of participants' attitudes (pro,con, or one side; both pro and con; one side but differing in extremity;unselected for attitudinal position or mixed); (c) definition of attitudegroups (extreme groups; median or other attitudinal split, e.g., pro vs. con;only one attitude group; continuous variation of attitude; mixed or other);(d) position of stimulus information (pro, con, or one side; both pro andcon); and (e) error term for calculating congeniality effect size (withinsubjects; between subjects; mixed or other, e.g., correlational design).

In addition, the following characteristics of the attitude measure werecoded; (a) inclusion of measure (included; attitude position merely as-sumed); (b) reliability (single item; multiple item with unknown or lowreliability; multiple item with high reliability, >.80; no measure, unclear,unknown, or mixed); (c) timing (pretest in prior session; pretest withinsession; simultaneous with presentation of stimulus information; no mea-sure; unclear, unknown, or mixed); and (d) compatibility of assessedattitude with the attitude object described by the stimulus information(high; moderate).

Several characteristics of the attitudinal issue were assessed. The valuerelevance of the issue was judged to be high if it was judged to be linkedto important values or significant reference groups (e.g., morally or ethi-cally significant issues such as abortion, U.S. involvement in Vietnam);otherwise, it was judged to be low. The outcome relevance of the issue wasjudged to be high if it had the potential for impact on participants'relatively near-term outcomes (e.g.,-tuition increase, use of essay exams);otherwise, it was judged to be low. The coders judged that the controver-siality of the issue was high or low by rendering a subjective assessment ofthe extent to which both pro- and anti-information were readily available inthe participants' environment. Similarly, the coders judged that familiaritywith the issue was high or low by rendering a subjective assessment of howwidely discussed the issue was in the participants' environment. Whenstudies used multiple issues, one or more of these characteristics of theattitudinal issue sometimes had to be coded as "mixed." Also, in a fewinstances there was no basis for judging one or more of these characteris-tics because the coders lacked sufficient information about how the par-ticular issue was regarded by participants and presented in their environ-ment. Preliminary, more detailed coding of these four characteristics of the

attitudinal issues was simplified because it did not yield superior predictionof the effect sizes. The coders also noted whether the participants wereselected as activists on the attitudinal issue.

The following aspects of the stimulus information were coded: (a) modeof presentation (written paced or timed; written unpaced; audio; video;live; other or mixed); (b) form (words or paired associates; statements;passages; other or mixed); (c) source (written by these or other researchers;taken or condensed from other sources such as newspaper articles; un-known, other, or mixed); (d) length (number of words, if known); (e)evidence of attitudinal position (pretesting or participants' ratings; noevidence); (f) attitudinal extremity (high; medium; mixed or unknown);and (g) greater familiarity with the congenial than the uncongenial stimulusinformation (no; unlikely; likely). This familiarity confound was coded noif the researcher tested and failed to confirm the hypothesis that partici-pants were more familiar with congenial information. Greater familiaritywith the congenial information was coded unlikely if (a) design featuresmade a familiarity confound unlikely (e.g., novel issue, adjectives asstimuli); (b) the attitudinal issue was widely discussed and highly contro-versial, suggesting that participants were familiar with both sides; or (c) theauthor made plausible on some other basis a claim that familiarity was notconfounded with congeniality. Finally, this familiarity confound was codedlikely in the absence of features resulting in coding as no or unlikely.

The following contextual aspects of the experiment were also coded:(a) setting for exposure to stimulus information (laboratory; class; otheror unknown); (b) setting for memory measurement (laboratory; class;other or unknown); (c) social setting (alone, with or without experi-menter; with other participants; unknown or mixed); (d) set for respond-ing to stimulus information (intentional learning, i.e., learn or memo-rize; less intentional, i.e., give opinions or attitudes, judge for argumenteffectiveness, give other judgments; unclear, unknown, or mixed); and(e) activation of attitudes during exposure to stimulus information(high; medium; low; mixed or no basis for judgment). Activation wascoded by the following rules: high if a premeasure of attitude wasadministered just before the presentation of the stimuli or if the attitudewas especially salient in the cover story (e.g., Love & Greenwald,1978); medium if the experimental instructions mentioned an interest inparticipants' attitudes (e.g., Furnham & Proctor, 1990); and low ifattitudes or agreement was not mentioned or if a distracting or mis-leading set was imposed (e.g., Levine & Murphy, 1943).

Characteristics of the memory measurement included the time intervalbetween the message and the memory measure, which was represented bycoding (a) timing of memory measure (immediate, defined as withinsession; delayed, defined as in subsequent session; both immediate anddelayed); (b) interpolated activity between message and immediate mem-ory measure (present; absent; mixed); (c) time interval for immediatemeasure, in minutes; and (d) time interval for delayed measure, in days.

Several different types of memory measures appeared in this literature:free recall, cued recall, uncorrected recognition (hit rate; false-alarm rate;miss rate; correct rejection rate; forced choice or multiple choice), recog-

3 The code "mixed" was used to describe the source of the participantpopulation when the study had participants from two or more of the othercategories (e.g., both high school and college students). This mixed codealso appears for many other variables for this same reason—namely, thatthe study encompassed two or more of the coded categories. An additionalcomplexity in the use of the mixed code stems from our procedure ofattaching the coded variables to individual effect sizes, which representedthe study as a whole or only a portion of a study's data (e.g., intentionallearning set only, less intentional set only; see next subsection, Computa-tion and Analysis of Effect Sizes). In such cases, the mixed code might beused for the whole study (e.g., combining across the sets) but not for theeffect sizes representing a portion of the data (e.g., only the intentionallearning set).

Page 7: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

70 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

nition sensitivity (d' and similar measures), or mixed. Recognition sensi-tivity (d1) is a measure of recognition memory derived from statisticaldecision theory (Green & Swets, 1966). It is ordinarily computed bysubtracting the standard score corresponding to the false-alarm rate (i.e.,proportion of foils selected) from the standard score corresponding to thehit rate (i.e., proportion of correct items selected; Srull, 1984, p. 31).Higher sensitivity scores thus reflect greater ability to accurately distin-guish between the correct items and the foil items. Separate analyses on thevarious types of uncorrected recognition measures were potentially infor-mative, but there were too few instances of each type to permit meaningfulanalyses. Measures of distortion of the stimulus information were excludedbecause they were few in number and very heterogeneous in their design.Also, it was often not clear what the congeniality hypothesis would predictabout the direction of distortion on the measure (e.g., A. L. Edwards,1941b).

Other aspects of the memory measure that were coded are the following:(a) blindness of coding of measure requiring content analysis (blind; notblind; no information; mixed); (b) reliability of coding (two coders withhigh reliability, >.80; two coders with low or unknown reliability; onecoder; unclear, unknown, or mixed); (c) number of trials assessing memorythat entered into effect size; (d) number of exposures to stimulus informa-tion before memory was assessed; and (e) order of memory measure insequence of measures (first or only; second; third or higher; mixed orunknown).

All coding was performed independently by two teams, each consistingof two of the authors, with a satisfactory mean agreement across thevariables: K = .80 (91% agreement) for variables requiring categoricaljudgments and r = 1.00 for variables estimated on a continuous scale. Thekappas ranged from .35 (91% agreement, for selection of activists) to 1.00(100% agreement, for several variables). Disagreements were resolved bydiscussion.

Computation and Analysis of Effect Sizes

The effect size calculated was g, which was the difference betweenmemory for attitudinally congenial and uncongenial information, di-vided by the relevant standard deviation. When congeniality was abetween-subjects variable, this standard deviation was computed sepa-rately within the congenial and uncongenial conditions and pooled.When congeniality was a within-subjects variable, the standard devia-tion was the standard deviation of the differences between the pairedobservations, because the pooled standard deviation could not be esti-mated from the statistics provided. All standard deviations were esti-mated, whenever possible, only from the portion of each study's dataentering into the effect size. The sign of the effect size was positivewhen congenial information was more memorable than uncongenialinformation and negative when uncongenial information was morememorable. In addition, we also computed extremity effect sizes when-ever possible. The sign of these effect sizes was positive when extremeinformation was more memorable than moderate information and neg-ative when moderate information was more memorable.

To reduce the potential for knowledge of a study's congenialityfinding to bias the coding of its study characteristics, we calculated theeffect sizes after the coding was completed. These effect sizes werecalculated independently by two of the authors with the aid of acomputer program (B. T. Johnson, 1993), and the statistical significanceand direction of the congeniality comparisons were also recorded.Whenever possible, separate effect sizes were calculated for differenttypes of measures of memory. In addition, whenever possible, separateeffect sizes were calculated within levels of the following independentvariables, when they were manipulated or varied in an experimentaldesign: (a) delay of memory measurement; (b) definition of attitudinalcongeniality; (c) set for responding to stimulus information; (d) in-volvement, commitment, or importance; (e) learning trials (first trial vs.

later trials); and (f) any individual difference variable (e.g., high vs. lowdogmatism). These within-level congeniality effect sizes were usedonly in models examining the impact of the independent variable thatproduced these levels. For example, in a model examining the impact ofmeasurement delay the separate immediate and delayed effect sizeswere used, whereas in other models the effect sizes were averaged overdelay of measurement. This strategy, recommended by Cooper (1989),took advantage of within-study manipulations of moderator variableswhile preserving insofar as possible the independence of the effectsizes. Moreover, to fully satisfy the independence assumption of meta-analytic statistics, we also computed a single study-level effect size,averaging all separately calculated effect sizes for each study (see B. T.Johnson & Eagly, in press).

The computation of g was based on (a) means and standard deviations orerror terms for 167 of the effect sizes; (b) Fs, ts, or zs for 48; (c)correlations or chi-squares for 23; (d) proportions or frequencies for 14;and (e) mixed statistical sources for 14. Thus, a total of 271 effect sizeswere computed.

Following Hedges and Olkin's (1985) procedures, the gs were convertedto ds by correcting them for bias. All mean ds (and models) were computedwith each effect size weighted by the reciprocal of its variance, a procedurethat gives more weight to effect sizes that are more reliably estimated.4 Todetermine whether each set of ds shared a common effect size (i.e., wereconsistent across the studies), we calculated a homogeneity statistic Q,which has an approximate chi-square distribution with k — 1 degrees offreedom, where k is the number of effect sizes.

In the absence of homogeneity, we accounted for variability in hetero-geneous effect sizes by calculating fixed-effect categorical and continuousmodels that related the effect sizes to the attributes of the studies (Hedges& Olkin, 1985). Calculation of a categorical model provides a between-classes effect, QB, and a test of the homogeneity of the effect sizes withineach class, Qw, The continuous models are least-squares simple linear andmultiple regressions. Each such model yields a test of the significance ofeach predictor as well as a test of model specification, <2E, which evaluateswhether significant systematic variation remains unexplained in the regres-sion model.

As a supplementary analysis, we attained homogeneity among the effectsizes by sequentially removing the effect sizes that reduced the homoge-neity statistic by the largest amount (see Hedges & Olkin, 1985). Studiesyielding these outlying effect sizes were then examined after the fact todetermine if they appeared to differ methodologically from the otherstudies. In addition, the central tendency of the effect sizes was recomputedafter removal of these effect sizes.

Results

Characteristics of the Studies

Before summarizing the tests of the congeniality hypothesisreported in attitude memory research, we display the characteris-tics of the studies in the sample so that the overall meta-analyticfindings can be interpreted in relation to the studies' typicalmethods. As shown by the central tendencies of the characteristicslisted in Table 1, the majority of studies were published in earlierdecades and as journal articles. Typically, the participants werecollege students, and the study was conducted in the United States.

4 In this calculation, numbers of participants were used in all instances.In some studies, the sample sizes in the statistical analyses used a differentmetric (e.g., number of participants X number of items, as in Judd & Kulik,1980). Because these numbers were often not comparable with numbers ofparticipants, the sample sizes were always based on numbers of partici-pants.

Page 8: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 71

The most common experimental design implemented a sidednessdefinition of attitudinal congeniality by presenting participantswith information on their own side versus the opposite side of theissue through varying both participant attitude and the attitudinalposition of the information. The design usually included groupsformed on the basis of their attitudes, either extreme groups orgroups divided at the median or split on another basis. The designtypically included both pro and con information, and the error termfor calculating the congeniality effect was commonly based onbetween-subjects variability only, within-subjects variability only,or mixed sources. Commonly, an attitude measure was included,although it was not known to have high reliability. The attitudemeasure, typically administered as a pretest in a prior session,usually assessed an attitude that was highly compatible with theattitude object described by the stimulus information.

The attitudinal issues used in the research were typically high invalue relevance and high or low in outcome relevance. The issueswere somewhat more often high than low in controversiality andsomewhat more often low than high in familiarity to participants.The participants were rarely selected as activists.

The stimulus information was most often written (paced orunpaced), but other modes of presentation were not uncommon.Passages were the most frequent form of stimulus information,with statements also quite common. These stimuli were usuallywritten by researchers, and their length was substantial. Althoughsometimes there was empirical evidence of the attitudinal positionof this stimulus information, often there was no such evidence. Theattitudinal extremity of the stimulus information was generallyunknown. In the typical study, it was unlikely that the participantswere more familiar with the congenial stimulus information thanthe uncongenial information.

The contextual characteristics of the experiment included a labo-ratory setting for both exposure to the stimuli and measurement ofmemory. Participants usually took part with other participants andsomewhat less often took part alone. When it was possible to discernthe set that the participants were given for responding to the stimulusinformation, most often it was a less intentional set (i.e., give judg-ments) rather than an intentional learning set (i.e., learn or memorize).Generally, the activation of participants' attitudes during exposure tothe stimulus information was high or low.

The measurement of memory typically consisted of only an imme-diate measure, although a substantial minority of studies includedboth immediate and delayed measures. Typically, there was no inter-polated activity between the message and the immediate memorymeasure, which followed within 1 min. The delayed measure, ifpresent, typically occurred 1 day later. The modal type of memorymeasure was free recall, but other types of measures occurred. If themeasure required coding, coders were most frequently not known tobe blind to participants' experimental condition. Although often twocoders established high reliability, the coding reliability was com-monly unclear, unknown, or mixed. Ordinarily, the test of memorywas based on only one assessment trial and one exposure to thestimuli. Typically, the memory measure was the first in a sequence ofmeasures or the only measure.

Summary of Study-Level Effect Sizes

The summary of the study-level effect sizes given in Table 2allows one to determine whether, on the whole, congenial infor-

mation was more memorable than uncongenial information. In thistable (and subsequent tables), an overall congeniality effect issuggested by a positive mean effect size that differed significantlyfrom the no-difference value of 0.00 (as indicated by a confidenceinterval that does not include 0.00). The mean weighted effect sizeof 0.23 averaged across all 65 studies that yielded study-leveleffect sizes differed from 0.00, indicating an overall congenialityeffect.5 The mean unweighted effect size of 0.33 also differedfrom 0.00. However, the median effect size of 0.10 was quitesmall, and the proportion of study-level comparisons favoringcongeniality (41 of 68 comparisons, yielding a proportion of .60)did not differ from .50, the proportion expected under the nullhypothesis (p < .12 by sign test).6

As also shown in Table 2, the effect sizes were not homoge-neous, and homogeneity was not attained until 17 of them wereremoved, which is a substantial proportion of the effect sizes (.26).The removal of these outlying effect sizes produced a considerablysmaller weighted mean of 0.08, which nonetheless differed signif-icantly from 0.00. The reason that the removal of these effect sizessubstantially reduced the mean effect size can be understood frominspecting the stem-and-leaf plot given in Figure 1. Although thegreat majority of the effect sizes in the full sample were locatedbetween —0.5 and 0.8, the data set includes quite a few largepositive effect sizes. There was one extremely large positive effectsize, which is a value of 8.74 from an experiment conducted inJapan (Kitano, 1970). Elimination of this one effect size from thecalculation of the mean weighted study-level effect size changed itonly slightly, d+ = 0.22, CI = 0.18/0.26, g(63) = 670.83, p <.001. The next largest effect size, a value of 2.20, was from thefrequently cited Levine and Murphy (1943) study. Of the 17 effectsizes whose removal was required to attain homogeneity, 14 werepositive.7 Only the Kitano (1970) study was excluded from thesubsequent tests of models; its extremely large magnitude suggeststhat the experiment or its reporting had unique attributes notdiscernible from the article.

Given that only 60% of the effect sizes favored congeniality, itis also interesting to aggregate the absolute value of the effect

5 The five studies for which effect sizes could not be calculated wereDupree (1992), Janis and Rausch (1970), Judd and Kulik (1980), Wallston(1970), and Watson and Hartman (1939).

6 For three studies (Dupree, 1992; Janis & Rausch, 1970; Wallston,1970), the congeniality effect was known to be nonsignificant; the effect'sdirection was known, but statistical information was not sufficient to allowcalculation of an effect size. When effect sizes of 0.00 were assigned tothese studies, the estimates of central tendency became the following: Mweighted d = 0.22 (confidence interval, or CI = 0.18/0.26); M unweightedd = 0.33 (CI = 0.05/0.62); median d = 0.08.

7 The studies removed in addition to Kitano (1970) and Levine andMurphy (1943), listed in the order in which they were removed, along withtheir effect sizes, are the following: Schmidt (1969), 1.37; Gallon (1966/1967), -1.24; Jones and Kohler (1958, Study 2), 1.72; A. L. Edwards(1941a), 1.04; Cacioppo and Petty (1979, Study 2), -0.55; Cacioppo andPetty (1979, Study 1), -0.82; Kanungo and Dutta (1966), 0.46; Jones andKohler (1958, Study 1), 0.84; Dutta et al. (1972, Study 2), 0.80; Dutta etal. (1972, Study 1), 0.79; Zimmerman and Chaiken (1996, Study 1), 0.67;Dutta et al. (1972, Study 3), 0.82; J. T. Johnson and Judd (1983, Study 1),-0.41; Roberts (1984), 0.53; and Howard-Pitney, Borgida, and Omoto(1986), 0.53.

Page 9: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

72 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

Table 1Summary of Study Characteristics

Variable and class Value Variable and class Value

General information

Median publication yearPublication form

Journal articleDissertation, master's thesis, or other unpublished

documentParticipant population

High school or junior high school studentsCollege studentsMixed or other

Nation where study was conductedUnited StatesOther

Attitudinal issue

1977

50

20

4606

5515

Experimental design

Definition of attitude congenialitySidedness 52Agreeability 6Sidedness and agreeability 5Balance3 7

Independent variables contributing to the test of congenialityParticipant attitude only 10Attitudinal position of information only 15Both participant attitude and position of information 40Mixed 5

Position of participants' attitudesPro, con, or one side 4Both pro and con 40One side but differing in extremity 14Unselected for attitudinal position or mixed 12

Definition of attitude groupsExtreme groups 24Median or other attitudinal split 5Only one attitude group 8Continuous variation of attitude 12Mixed or other 21

Position of stimulus informationPro, con, or one side 10Both pro and con 60

Error term for calculating congeniality effect sizeWithin subjects 20Between subjects 23Mixed or other 27

Attitude measurement

Inclusion of measureIncluded 60Attitude position merely assumed 10

ReliabilitySingle item 24Multiple item with unknown or low reliability 22Multiple item with high reliability 10No measure, or unclear, unknown, or mixed 14

TimingPretest in prior session 31Pretest within session 13Simultaneous with presentation of stimulus information 6No measure, or unclear, unknown, or mixed 20

Compatibility of assessed attitude with attitude objectdescribed by the stimulus informationHigh 62Moderate 8

Value relevanceHighLowMixed or no basis for judgment

Outcome relevanceHighLowMixed or no basis for judgment

ControversialityHighLowMixed or no basis for judgment

FamiliarityHighLowMixed or no basis for judgment

Selection of activists as participantsYesNo

471013

32317

392110

223810

268

Stimulus information

Mode of presentationWritten paced or timed 21Written unpaced 20Audio 14Video 6Live 3Other or mixed 6

FormWords or paired associates 4Statements 25Passages 34Other or mixed 7

SourceWritten by these or other researchers 41Taken or condensed from other sources 12Unknown, other, or mixed 17

Median length, in words 424Evidence of attitudinal position

Pretesting or participants' ratings 34No evidence 36

Attitudinal extremityHigh 5Medium 4Mixed or unknown 61

Greater familiarity with congenial information thanuncongenial informationNo 13Unlikely 43Likely 14

Context of experiment

Setting for exposure to stimulus informationLaboratoryClassOther or unknown

Setting for memory measurementLaboratoryClassOther or unknown

5299

5210

Page 10: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 73

Table 1 (continued)

Variable and class Value Variable and class Value

Context of experiment (continued)

Social settingAlone, with or without experimenterWith other participantsUnknown or mixed

Set for responding to stimulus informationIntentional learningLess intentionalUnclear, unknown, or mixed

Activation of attitudes during exposure to stimulusinformation

HighMediumLowMixed or no basis for judgment

25396

133324

306

259

Memory measurement

Timing of memory measureImmediate 46Delayed 5Both immediate and delayed 19

Interpolated activity between message and immediatememory measure

Present 13Absent 32Mixed 25

Median time interval for immediate measure, in min 1

Memory measurement (continued)

Median time interval for delayed measure, in days 1Type of measure

Free recall 38Cued recall 6Unconnected recognition 17Recognition sensitivity 3Mixed 6

Blindness of coding of measures requiring content analysisBlind 17Not blind 1No information 20Mixed 5

Reliability of codingTwo or more coders with high reliability 19Two or more coders with low or unknown reliability 3One coder 6Unclear, unknown, or mixed 15

Median number of trials assessing memory that entered intostudy-level effect size 1

Median number of exposures to stimulus information beforememory assessed 1

Order of memory measure in sequence of measuresFirst or only 35Second 14Third or higher 10

Mixed or unknown 11

Note. For categorical variables, numbers in the table represent frequency of studies in each class. Medians of continuous variables are based on studiesfor which information was available on each variable." All of the studies yielding effect sizes by the balance definition of congeniality also yielded sidedness effect sizes, and one also yielded an agreeabilityeffect size.

sizes. This analysis produced a mean weighted effect size of 0.38(CI = 0.34/0.42),8 which should be compared with the mean of thesigned effect sizes, 0.23. This absolute-value mean thus shows thatthe impact of attitudes on memory can be considered more sub-stantial if the direction of attitudes' effect on memory is ignored,because effects occurred in both the congeniality and the counter-congeniality direction. The analysis of the absolute-value effectsizes showed that they were heterogeneous, (2(64) = 617.78, p <.001, and nine outlying effect sizes (14% of the sample) wereremoved to produce homogeneity.

Only five studies allowed the calculation of extremity effectsizes, and these yielded a mean weighted d of 0.24 (CI = 0.06/0.42), with positive values assigned to superior memory for ex-treme information. Thus, extreme information had a small advan-tage over more moderate information. Also, these five effect sizesapproached homogeneity, Q(4) = 8.12, p < .10.

Effects of Type of Memory Measure

Classification of the effect sizes by the type of memory measureshould help interpret the overall tests of the congeniality hypoth-esis. Because a preliminary analysis established that free-recall andcued-recall measures did not produce different results, these twotypes of measures were combined into a single recall category.Another preliminary analysis divided these recall measures ac-cording to whether the relevant document (a) stated that the coding

of recall was blind to participants' attitudes (or no coding wasrequired) or (b) failed to mention that the coding was blind.Although the effect sizes were significantly larger when the codingwas not known to be blind, a subsequent analysis showed that thiselevation of the effect sizes occurred for the small subset of studiespublished prior to 1960 and not for the studies published subse-quently. Therefore, recall measures were divided into two groups:(a) pre-1960 measures for which coding was not known to be blindand (b) all other recall measures, regardless of whether coding wasknown to be blind. Recognition measures were also divided intotwo groups: (a) uncorrected recognition measures and (b) recog-nition sensitivity (d') measures, which are controlled for bias.

The resulting analysis, shown in Table 3, displays the effectsizes for the two types of recall measures and the two types ofrecognition measures. The congeniality effect was very large forthe pre-1960 recall measures that were not known to be blind andquite small but significant for the remaining recall measures.9

8 With the Kitano (1970) study removed, this mean weighted effect sizebecame 0.37 (CI = 0.33/0.41), and the set of effect sizes was still heter-ogeneous, G(63) = 452.03, p < .001.

9 These studies from before 1960 whose recall measures were not knownto be blind are the following: Jones and Aneshansel (1956), d = -0.03;Jones and Kohler (1958), Study 1, d = 0.84; Jones and Kohler (1958),Study 2, d = 1.72; and Levine and Murphy (1943), d = 2.20. Because thisanalysis was based on study-level effect sizes, it failed to represent the

Page 11: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

74 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

Table 2Summary of Study-Level Effect Sizes

Criterion Value

Known effect sizes

Mean weighted d (d+)a

95% CI for d+

Homogeneity (Q) of ds comprisingMean unweighted d95% CI for mean unweighted dMedian d

650.23

0.18/0.27842.82***

0.330.03/0.62

0.10

Known effect sizes excluding outliers

No. (proportion) of removed outliers

Mean weighted d (d+)95% CI for d+

Homogeneity (Q) of ds comprising d+

17 (.26)48

0.080.03/0.12

65.37

Note. Effect sizes are positive for superior memory for congenial infor-mation and negative for superior memory for uncongenial information, n =sample size; CI = confidence interval; d = effect size; d+ = meanweighted effect size; Q = homogeneity of ds.* Effect sizes were weighted by the reciprocal of the variance. b Signifi-cance indicates rejection of the hypothesis of homogeneity.***p < .001.

Nonblind coding may well have taken place in the earlier studies(and was in fact implied for Jones & Kohler, 1958, Study 1), andcoders' expectancies may have biased findings toward the hypoth-esized congeniality effect. In addition, the congeniality effect wassignificant for the uncorrected recognition measures and absent forthe recognition sensitivity measures. Uncorrected recognition mea-sures, which were not controlled for guessing and other biases,may also have been biased toward congeniality.

Effects of Publication Year

As shown in Table 4, the congeniality effect weakened duringthe years over which these experiments were conducted: The effectof publication year on the 64 study-level effect sizes was substan-tial. In view of the strong effects of type of memory measurealready presented, it is important to determine whether this seculartrend might plausibly be attributed to measurement biases thatfavored the congeniality effect in the earlier studies. Calculation ofcontinuous models within various subsets of the effect sizes al-lowed exploration of this issue.

interaction effect produced by Jones and Aneshansel, who predicted acongeniality effect in their control condition (i.e., learning set) and areversal of the effect in their experimental condition (i.e., counterarguingset). In line with this prediction, the effect sizes for Jones and Aneshansel'sexperiment were 0.75 in their control condition and -0.82 in their exper-imental condition. When 0.75 was substituted for the study-level effect sizeof —0.03, the mean weighted effect size for the first category of Table 3(recall from pre-1960 studies with coding not known to be blind) be-came 1.36, CI = 1.19/1.53, Gw.(3) = 49.42, p < .001. The one pre-1960study using a recall measure for which the absence of blind coding was notan issue (Postman & Murphy, 1943, which did not require coding) pro-duced a small effect size of 0.12.

When the recall measures and the recognition measures wereanalyzed separately, the relation of the effect sizes to publicationyear remained significant in both analyses (see Table 4). Toexamine whether this weakening of the congeniality effect overtime on the recall measures reflected the large effect sizes from thepre-1960 studies that were suspected to have nonblind coding offree recall, we removed these four effect sizes from the set of recalleffect sizes and recalculated the continuous model for the recallmeasures. Consistent with the suspicion that this coding problemwas critical to the shrinkage of the congeniality effect, this modelwas weaker and only marginally significant. For the recognitionmeasures, the decline in the strength of the congeniality effectcould have reflected the use of recognition sensitivity measures inthe more recent studies, because these measures yielded only verysmall effect sizes. Indeed, removing these presumably less-biasedestimates of the congeniality effect (leaving only the uncorrectedrecognition measures) reduced the magnitude of the effect ofpublication year, although it remained significant.

Another possibility is that the overall weakening of the conge-

8.7

2.22.12.01.91.81.71.61.51.41.31.21.11.00.90.80.70.60.50.40.30.20.10.0-0.0-0.1-0.2-0.3-0.4-0.5-0.6-0.7-0.8-0.9-1.0-1.1-1.2

4

0

2

7

4

0249379033755662524580024566013389999887633322187210925615

2

4

Figure 1. Stem-and-leaf plot of effect sizes (ds).

Page 12: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 75

Table 3Categorical Model for Type of Memory Measure

Type of measureBetween-classes

effect (QB)Mean weightedeffect size (</,+) 95% CI

Homogeneity withineach class (Qw.f

127.12***Recall from pre-1960 studies with

coding not known to be blindAll other recallUncorrected recognitionRecognition sensitivity

440215

1.06,,0.12.0.26C

-0.05a

0.89/1.230.07/0.180.19/0.33

-0.19/0.08

116.32***145.08***293.91***

1.54

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. Differencesbetween means that do not have a subscript in common are significant at the .05 level or beyond by ScheffS post hoc contrasts. CI = confidence interval.a Significance indicates rejection of the hypothesis of homogeneity.***p < .001.

niality effect might be due in part to increasing bipolarity inparticipants' attitude structures, a variable that we argued may berelated to the controversially of the issues used in attitude memoryexperiments. However, the correlation between year of publicationand issues' controversiality proved to be nonsignificant, r(54) =.20, p < .15.10 Moreover, the magnitude of the relation betweenpublication year and the congeniality effect sizes was not reducedwhen controversiality was controlled by entering it along withpublication year in a multiple regression model.

Complementing these analyses of the congeniality effect sizes isan analysis on the absolute value of the congeniality effect sizes.This model was also significant, b = —0.01, /3 = —.36, SE(b) =0.0013, p < .001, indicating that the effect sizes also becamesmaller in their absolute value as years passed. Thus, the disap-pearance of the congeniality effect was not accompanied by agreater trend toward countercongeniality. Rather, as Figure 2shows, the effects tended to gravitate toward null findings.

Given the results of these initial analyses involving type ofmemory measure and publication year, the remaining moderatoranalyses were calculated separately within the recognition mea-sures and the recall measures and also on overall memory, whichignored the distinction between types of memory measures. Theset of recall measures (and the set of overall memory measures)used in these subsequent analyses omitted the four effect sizesfrom the pre-1960 studies that did not note blind coding because(a) these measures were suspected to be contaminated by artifactand (b) their inclusion or exclusion did have substantial impact onthe findings that other moderators produced. However, the set ofrecognition measures used in the remaining analyses was com-plete. The measures based on recognition sensitivity were notsufficiently numerous to analyze separately, and their inclusion orexclusion had little impact on the findings that other moderatorsproduced.

Effects of Characteristics of Attitudinal Issue

The congeniality effect was significantly related to the valuerelevance and the outcome relevance of the attitudinal issues.These models showed trends in the predicted directions of valuerelevance increasing the congeniality effect and outcome relevancedecreasing it. As shown in Table 5, the congeniality effect waslarger when the attitudinal issue was high in value relevancecompared with low in value relevance. This model was significantfor both recognition and recall. As shown in Table 6, the conge-

niality effect was weaker when the attitudinal issue was high inoutcome relevance compared with low in outcome relevance,although this model was nonsignificant on the recognitionmeasures.

As shown in Table 7, the controversiality of the attitudinal issuealso produced a significant overall model in the predicted directionof high controversiality dampening the congeniality effect. Yet,when controversiality was examined separately within the recalland recognition measures, only the recall model was significant.

Effects of Timing of Memory Measure

As shown in Table 8, the congeniality effect was larger for themeasures that were delayed, defined as administered in a sessionsubsequent to the stimulus exposure, than it was for measures thatwere immediate, defined as administered in the same session asstimulus exposure. Although this model confirmed the effect ob-tained by Roberts (1985), it was significant only on the recognitionmeasures and not on the recall measures. Moreover, parallel ten-dencies for the congeniality effect to increase with delay were notobtained for models examining (a) the presence versus absence ofinterpolated activity preceding the immediate measure and (b) thelength of the time interval before measurement within the stimulusexposure session for the immediate measures or between theexposure and delayed sessions for the delayed measures.

Effects of Nation Where Study Was Conducted

As shown in Table 9, the congeniality effect was stronger instudies conducted in nations other than the United States than instudies conducted in the United States. This difference was sig-nificant for both the recall and recognition measures. Interpretationof this unpredicted nations effect should take into account the factthat the foreign studies included several conducted in Canada,whose effect sizes differed from those of the U.S. studies and didnot differ from those of studies conducted in other non-U.S.nations.

Other Moderators

Table 10 displays the relation of the effect sizes to the error termused for calculating the congeniality effect size. In studies that pro-

10 This calculation omitted studies for which controversiality was mixedor impossible to code.

Page 13: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

76 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

Table 4Continuous Models for Publication Year

Type of measure

Overall memoryAll recallRecall omitting pre-1960 studies with

coding not known to be blindAll recognitionUncorrected recognition

6444

402621

-.01***-.01***

-.oot-.01***-.01***

-.35-.36

-.14-.37-.29

Note. Models are weighted least-squares simple linear regressions calcu-lated with weights equal to the reciprocal of the variance for each effectsize. Effect sizes are positive for superior memory for congenial informa-tion and negative for superior memory for uncongenial information, b =unstandardized regression coefficient, ft = standardized regression coeffi-cient,tp < -10 (marginally significant). ***p < .001.

duced a within-subjects error term or an error term that was mixed (orthe design was correlational), the congeniality effect was larger thanit was in studies that produced a between-subjects error term. How-ever, this model remained significant only for the recall measures.

Table 11 presents the relation between publication form and thesize of the congeniality effect. Consistent with the claim that apublication bias favored the congeniality effect, the congenialityeffect was present in journal articles but absent in the unpublishedliterature. However, this difference between published and unpub-lished studies was significant only for the recognition measures.

Analyses treating other study characteristics as independentvariables either did not yield significant models or were uninter-pretable because of missing data or small sample sizes withincategories. For example, participants' set for responding to attitu-dinal information did not significantly affect the size of the con-geniality effect. In this regard, an initially more detailed classifi-cation of instructional sets, which our coding system implemented,produced no gains of prediction over our final classification of setsinto two categories: those produced by instructions to learn inten-tionally (e.g., learn, memorize) versus less intentionally (e.g., giveopinions, judge for argument effectiveness). Other variables yield-ing nonsignificant or uninterpretable models included experimen-tal design characteristics such as number of exposure trials, num-ber of memory trials, and definition of attitude congeniality;contextual characteristics such as activation of attitudes duringstimulus exposure; and stimulus characteristics such as the famil-iarity of the attitudinal issue, the relative familiarity of the conge-nial and uncongenial information, and the form, length, attitudinalextremity, and mode of stimulus presentation.

Multiple Regression Analysis

Although most of the significant predictors of the effect sizeswere not highly confounded, concerns about their lack of completeindependence as well as our desire to determine the adequacy of amodel containing several predictors led us to estimate multipleregression models that examined the simultaneous impact of thevariables that proved to be significant on a univariate basis. Forthese analyses, the categorical variables were dummy coded. Themodel appearing in Table 12 therefore entered type of memorymeasure (recognition vs. recall), publication year, value relevance.

outcome relevance, controversiality, timing of memory measure-ment, error term for calculating the effect size, publication form,and nation where study was conducted. Several studies appearedmore than once in this analysis because they manipulated or variedsome of these study characteristics on a within-study basis (spe-cifically, type of memory measure, value relevance, controversial-ity, and timing of memory measurement); the effect sizes used inthis multiple regression analysis were calculated within the result-ing levels of these variables.

As shown in Table 12, the significant predictors in this multipleregression model were type of memory measure, publication year,error term, and nation where study was conducted. Of the remain-ing predictors, value relevance, outcome relevance, timing ofmemory measurement, and publication form were rendered non-significant. Although controversiality was marginal in this multi-ple regression equation, the direction of its relation to congenialitywas opposite to its direction hi the categorical model (see Table 7).The failure of value relevance to remain significant was accountedfor largely by its covariation with controversiality, r(S5) = .44,p < .001. When controversiality was removed from the regressionequation, value relevance showed a significant positive relation tocongeniality, b = 0.12, (3 = .10, SE(b) = 0.06, p < .05. In contrastto the tendency for issues' value relevance to be correlated withtheir controversiality, their value relevance was uncorrelated withtheir outcome relevance, r(85) = .04, ns.

The model shown in Table 12 was moderately successful inaccounting for variability in the magnitude of the effect sizes.However, the test of model specification (<2E) showed that signif-icant systematic variation remained unexplained, and the relativelylarge Birge ratio (RB; see Hedges, 1994) showed that the between-studies variation not accounted for by the multiple regressionmodel was much larger than what would be expected on the basisof the within-study sampling variance.

Individual Differences in the Strength of the CongenialityEffect

As shown in Table 13, several studies examined the congenialityeffect separately within groups of participants who were classifiedby measures of personality or attitude structure. Unfortunately, theimpact of these variables cannot be examined meta-analyticallybecause multiple representations of these variables could not befound in the attitude memory literature. Nonetheless, the effectsizes computed for these studies suggest that individual differencesin the extent to which people exhibit a congeniality bias areexplained in part by personality and attitudinal characteristics.Although the difference between the effect sizes for the high andlow levels of the individual difference variable was significant(one-tailed) for only two studies by RosenthaPs (1991, p. 65)method for comparing two effect sizes (Kleck & Wheaton, 1967;Zimmerman & Chaiken, 1996, Study 1), the difference was sig-nificant in these studies as well as a third study with the morepowerful statistical tests and more complex memory measuressometimes appearing in the reports.11 Specifically, Kleck and

11 For example, Zimmerman and Chaiken (1996) adjusted their recallmeasure for intrusions, whereas we did not because such adjustments werenot implemented in other studies in this literature.

Page 14: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 77

2.5

2.0

1.5

1.0

.5

0.0

-.5

-1.0

-1.5

-2.0

-2.5

1940 1950 19901960 1970 1980

Year of Publication

Figure 2. Scatter plot of relation between publication year and effect sizes.

2000

Wheaton (1967) found that the congeniality effect was signifi-cantly larger for participants scoring high compared with low ondogmatism. Olson (1980) reported that for the sidedness definitionof congeniality (but not for other definitions), repressers showed asignificantly stronger congeniality effect than sensitizers. Finally,Zimmerman and Chaiken (1996) conducted two studies in whichthe congeniality effect was significantly larger among participantswhose attitudes were lower rather than higher in evaluative-cognitive consistency, defined as the degree of consistency be-tween one's overall evaluation of an attitude object and the eval-uative content of one's beliefs about it. In addition, their Study 2revealed that the congeniality effect was significantly largeramong participants whose attitudes were higher rather than lowerin evaluative-affective consistency, defined as the degree of con-sistency between one's overall evaluation of an attitude object and

the evaluative content of the affective reactions it elicits. More-over, in Study 2, the tendency of both of these structural variablesto moderate the congeniality effect was more pronounced amongparticipants whose defense motivation had been experimentallyheightened.

Discussion

Although memory was generally slightly better for attitudinallycongenial information than for uncongenial information, the find-ings produced by attitude memory studies proved to be veryheterogeneous, and, as expected, the congeniality effect was muchstronger in the beginning years of this research literature. Toexplain why the congeniality effect progressively eroded and,more generally, why it was considerably stronger under some

Table 5Categorical Model for Value Relevance of Attitudinal Issue

Type of measure and classof value relevance

Overall memoryHighLow

RecallHighLow

RecognitionHighLow

Between-classeseffect (gB)

23.95***

13.04***

9.34**

n

5816

3510

236

Mean weightedeffect size (4+)

0.20-0.07

0.17-0.06

0.22-0.11

95% CI

0.15/0.24-0.17/0.03

0.11/0.23-0.17/0.05

0.16/0.29-0.31/0.09

Homogeneity withineach class (2w.)a

384.81***73.39***

95.55***41.54***

288.10***31.65***

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. CIconfidence interval.a Significance indicates rejection of the hypothesis of homogeneity.**p < .01. ***p < .001.

Page 15: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

78 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

Table 6Categorical Model for Outcome Relevance of Attitudinal Issue

Type of measure and class ofoutcome relevance

Overall memoryHighLow

RecallHighLow

RecognitionHighLow

Between-classeseffect «2B)

4.61*

18.01***

2.64

n

3830

2418

1421

Mean weightedeffect size {di+)

0.120.21

0.020.25

0.240.14

95% CI

0.06/0.170.14/0.27

-0.05/0.090.17/0.33

0.16/0.320.04/0.23

Homogeneity withineach class (Qw,)"

166.34***290.38***

64.93***62.82***

85.66***224.11***

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. CIconfidence interval.a Significance indicates rejection of the hypothesis of homogeneity.* />< .05. ***/>< .001.

circumstances than others, we must engage both methodologicaland theoretical principles. We invoke these principles by firstconsidering the overall consistency and magnitude of the conge-niality effect and then discussing its moderators.

Consistency and Strength of the Congeniality Effect

The inconsistency in the magnitude of the effect sizes wasshown by the rejection of the hypothesis of homogeneity and bythe removal of a somewhat large proportion of the effect sizes(26%) before a homogeneous set was attained (see Hedges, 1987).Confirming this judgment about the inconsistency of the effectsizes, the analysis of the direction of the effects indicated that 60%of the study-level comparisons favored congeniality and 40%favored countercongeniality.

Although this heterogeneity of the effect sizes focuses attentionon moderator variables rather than the central tendency, it is clearthat the congeniality effect must be described as generally small,although reliable in the set of studies. Nearly half of the study-leveleffect sizes (31 of 65) fell in the range from 0.20 to -0.20 (seeFigure 1) and thus were quite small. As shown in Table 2, thevalues of the measures of central tendency ranged from a high

of 0.33 for the unweighted mean to a low of 0.08 for the weightedmean with outliers excluded. These values are comparable to theresults of Roberts's (1985) earlier meta-analysis on a smallersubset of the attitude memory literature, which produced an un-weighted mean of 0.36.

Confirming the small magnitude of the average congenialityeffect size are Cohen's (1977) benchmarks for the d metric bywhich 0.20 can be described as small, 0.50 as medium, and 0.80 aslarge. Moreover, transforming our mean weighted effect sizeof 0.23 into the metric of Rosenthal and Rubin's (1982) binomialeffect-size display suggests that above-average memory was man-ifested by approximately 55.5% of participants who received con-genial information and 44.5% of the participants who receiveduncongenial information. This difference of approximately 11%suggests a modest overall tendency toward congeniality.

Although many meta-analyses in experimental social psychol-ogy have produced larger mean effect sizes than the presentmeta-analysis, attitude memory studies depart from typical meth-ods in social psychology because these studies generally measuredrather than manipulated one key independent variable: partici-pants' attitudes. However, given evidence of comparable magni-

Table 7Categorical Model for Controversiality of Attitudinal Issue

Type of measure and classof Controversiality

Overall memoryHighLow

RecallHighLow

RecognitionHighLow

Between-classeseffect (QB)

8.41**

9.87**

0.39

n

3828

2416

1412

Mean weightedeffect size (dl+)

0.100.21

0.050.21

0.180.22

95% CI

0.04/0.150.16/0.27

-0.03/0.120.14/0.29

0.09/0.270.13/0.30

Homogeneity withineach class (£>w.)a

136.12***316.16***

36.36198.85***

94.71***217.31***

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. The lowcategory includes all studies for which Controversiality was not coded as high. CI = confidence interval.2 Significance indicates rejection of the hypothesis of homogeneity,tp < .10 (marginally significant). **p < .01. *** p < .001.

Page 16: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 79

Table 8Categorical Model for Timing of Memory Measure

Type of measure and class of timingof memory measurement

Overall memoryImmediateDelayed

RecallImmediateDelayed

RecognitionImmediateDelayed

Between-classeseffect (GB)

16.61***

0.00

30.75***

n

6119

3810

239

Mean weightedeffect size (di+)

0.110.31

0.130.14

0.080.47

95% CI

0.07/0.160.23/0.40

0.07/0.190.02/0.26

0.01/0.160.36/0.59

Homogeneity withineach class (Qw.)a

251.31***193.29***

134.06***11.37

116.25***166.69***

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. CIconfidence interval." Significance indicates rejection of the hypothesis of homogeneity.***/> < -001.

tudes of the effects of dispositional variables and manipulatedsituational variables (Funder & Ozer, 1983), reliance on an indi-vidual differences variable in attitude memory studies cannot beheld responsible for the small size of the effects they produced.

Effects of Type of Memory Measure

The search for moderators that would account for variability inthe effect sizes began with classifying them according to type ofmemory measure. These analyses raised questions about the pres-ence of artifacts in early experiments that used recall measures thatrequired coding of participants' responses yet presented no evi-dence that the coders were blind to the experimental conditions.These experiments yielded very large effect sizes that differedfrom those yielded by other early experiments. In the absence ofevidence that coding of free recall in these early studies was blindto participants' attitudes, it is possible that coders were aware ofparticipants' attitudinal positions and of the researchers' predictionof congeniality. Because these studies predated social psycholo-gists' attention to demand characteristics and experimenter expec-tancy effects (see Rosenthal & Rosnow, 1969), researchers might

not have been vigilant about the potential for artifact from non-blind coding. Consistent with a growth of methodological sophis-tication, classifying the post-1960 studies according to whetherblind coding was explicitly mentioned in the report did not pro-duce a significant model. Presumably in later studies, the omissionof information about the blindness of coding of free recall usuallyreflected incomplete reporting of the method, not a failure toimplement blind coding. When the pre-1960 experiments whosecoding was not known to be blind were omitted, the remainingrecall measures produced a very small congeniality effect of 0.12.

For the recognition measures, the size of the congeniality effectdepended on whether studies assessed memory by recognitionsensitivity, which is corrected for bias, or by measures that are notcorrected. Whereas the recognition sensitivity measures producedan effect that did not differ from 0.00, the uncorrected recognitionmeasures produced a larger effect size of 0.26. These uncorrectedrecognition measures differed widely in their specific format andincluded hit rates, correct rejection rates, and forced-choice ormultiple-choice measures. As a general rule, such measures areheld to be influenced by various response biases. For example,

Table 9Categorical Model for Nation Where Study Was Conducted

Type of measure and class of nationwhere study was conducted

Overall memoryUnited StatesOther

RecallUnited StatesOther

RecognitionUnited StatesOther

Between-classeseffect (QB) n

107.19***4614

59.96***2911

95.81***233

Mean weightedeffect size (di+)

0.040.56

-0.020.43

0.101.18

95% CI

-0.00/0.090.47/0.65

-0.08/0.050.34/0.53

0.04/0.160.97/1.38

Homogeneity withineach class (2W. )a

263.90***79.34***

64.87***20.25**

199.22***17.37***

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. The othernations and number of studies for each nation are the following: Canada, 7; Great Britain, 2; Germany, 2; New Zealand, 2; India, 1. CI = confidenceinterval.a Significance indicates rejection of the hypothesis of homogeneity.**/>< .01. **V < .001.

Page 17: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

80 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

Table 10Categorical Model for the Error Term for Calculating Congeniality Effect Size

Type of measure andclass of error term

Overall memoryWithin subjectsBetween subjectsMixed or other

RecallWithin subjectsBetween subjectsMixed or other

RecognitionWithin subjectsBetween subjectsMixed or other

Between-classeseffect (QB)

10.03**

29.19***

0.22

n

211821

161014

8117

Mean weightedeffect size (di+)

0.25,0.08b

0.18.

0.25a

-0.13b

0.1 6a

0.210.180.21

95% CI

0.17/0.330.01/0.150.11/0.24

0.16/0.33-0.24/-0.02

0.07/0.24

0.05/0.380.10/0.270.10/0.32

Homogeneity withineach class (Gw.)a

48.91***270.87***120.62***

47.37***29.92***39.60***

4.78225.56***

81.85***

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. Differencesbetween means that do not have a subscript in common are significant at the .05 level or beyond by a priori contrasts. CI = confidence interval.a Significance indicates rejection of the hypothesis of homogeneity.**/><.01. ***/><.001.

hit-rate measures would favor yea-sayers, who have a generaltendency to select many responses as correct. In contrast, recog-nition sensitivity measures remove this type of bias by subtractingthe standard score corresponding to the false-alarm rate from thestandard score corresponding to the hit rate.

In the introduction, we noted that a guessing strategy might haveincreased the magnitude of the congeniality effect for measuresthat did not correct for this bias. Guessing would have this impactif it were more often consistent than inconsistent with participants'attitudes. Because the uncorrected recognition measures favoredcongeniality to a greater extent than the recognition sensitivitymeasures, d+ = 0.26 versus —0.05, our meta-analysis substanti-ates a tendency for guessing to have been consistent with partic-ipants' own attitudes. However, this finding should be comparedwith analogous results in meta-analyses of memory for informa-tion that is congruent versus incongruent with expectancies. Spe-cifically, in Stangor and McMillan's (1992) meta-analysis, theuncorrected recognition measures produced much better memoryfor expectancy-congruent information, d+ = 0.63, whereas therecognition sensitivity measures produced considerably bettermemory for expectancy-incongruent information, d+ = -0.44.The corresponding d+ values from Rojahn and Pettigrew's (1992)meta-analysis of this same literature were 0.64 and —0.59. Thesefindings suggest that an expectancy-consistent guessing strategywas manifested on the uncorrected measures, even though prefer-ential encoding of the expectancy-inconsistent information wasdemonstrated by the recognition sensitivity measures.

Although these findings concerning the impact of information'sconsistency with expectancies seem understandable in view of theinherent reasonableness of guessing in line with one's expectan-cies, guessing in line with one's attitudes seems less likely. Sup-porting this reasoning, the difference in the effects produced by theuncorrected recognition measures and the recognition sensitivitymeasures was far smaller in the present synthesis than in themeta-analyses by Stangor and McMillan (1992) and Rojahn andPettigrew (1992). Nonetheless, this small tendency for the uncor-rected recognition measures to yield a stronger congeniality effectthan the recognition sensitivity measures invites interpretation.

Perhaps wishful thinking caused participants to slightly favorguesses in line with their attitudes (McGuire & McGuire, 1991), orparticipants' use of their prior attitude as a memory heuristic mighthave produced some tendency toward attitude-consistent guessing.Alternatively, the subjective experience of recognizing informa-tion that is compatible with one's preferred attitudinal positionmight be attributed to its truth value. These possibilities are worthyof investigation.

The classification of effect sizes by type of memory measurepotentially allowed us to examine the extent to which encoding orretrieval processes mediate attitudes' effects on memory. Whereasfree-recall measures are held to assess the ability to retrieveinformation from memory, recognition measures are held to assesswhether information has been encoded at all (e.g., Srull, 1984).Because the mean effect sizes for recall (excluding the earlystudies whose coding was not known to be blind) and recognitionsensitivity did not differ from one another, neither encoding norretrieval processes are especially implicated by this analysis.

Finally, with respect to the distinction between explicit or directand implicit or indirect measures of memory (e.g., Jacoby, Toth,Lindsay, & Debner, 1992; Schacter, 1987), it is important to notethat in every study included in the synthesis, memory was assessedby direct measures—that is, by asking participants to consciouslyremember attitude-relevant information that had previously beenpresented. In no case was memory assessed by indirect measuresthat left participants unaware that their memory was being as-sessed and that thus evaluated the effects of prior learning onsubsequent responding in the absence of conscious recollection ofthis learning. Both implicit and explicit memory measures shouldbe used in subsequent work not only because of the significance ofthis implicit-explicit distinction for understanding underlying cog-nitive processes, but also because of the relevance of the distinc-tion to ecological validity. Thus, the attitude-relevant informationthat comes to mind in everyday situations is most germane tounderstanding attitudes' impact on memory. This is not to suggestthat deliberate attempts to retrieve and ponder what one canrecollect about an attitudinal issue are unimportant. To balance thepicture, however, investigators ought to consider the role that prior

Page 18: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 81

Table 11Categorical Model for Publication Form

Type of measure and classof publication form

Overall memoryJournal articleDissertation or unpublished document15

RecallJournal articleDissertation or unpublished document

RecognitionJournal articleDissertation or unpublished document

Between-classeseffect (GB)

28.96***

0.24

63.70***

n

4218

2713

179

Mean weightedeffect size (di+)

0.24-0.01

0.130.10

0.37-0.17

95% CI

0.19/0.29-0.08/0.07

0.07/0.200.02/0.19

0.30/0.44-0.27/-0.06

Homogeneity withineach class (2w/)

a

308.15***113.31***

122.21***22.62*

169.16***79.56***

Note. Effect sizes are positive for superior memory for congenial information and negative for superior memory for uncongenial information. CIconfidence interval.' Significance indicates rejection of the hypothesis of homogeneity. b Neither of the two master's theses in the sample yielded an effect size.* p < .05. ***p < .001.

exposure to attitude-relevant information plays when people do notconsciously reflect on their attitudes or knowledge.

Effects of Publication Year

Consistent with past speculations (e.g., Greenwald, 1975), thecongeniality effect decreased in magnitude over the decades dur-ing which attitude memory research was conducted. Because thiserosion could have been mediated by changes in the way thatmemory was assessed, the continuous model for publication yearwas also calculated within types of measures. When the possiblyartifactual effect sizes from the early studies whose coding was not

Table 12Multiple Regression Model

Predictor or outcome

Type of memory measure"Publication yearValue relevance of attitudinal issueb

Outcome relevance of attitudinal issue0

Controversiality of attitudinal issued

Timing of memory measurement'Error term for calculating congeniality effect sizef

Publication formg

Nation where study was conducted11

Additive constantMultiple RQjRj

b

0.17***-0.01***

0.06-0.01

o.iot0.080.14**0.070.52***

16.66.61

300.803.91

ft

0.19-0.29

0.05-0.01

0.120.070.140.080.48

******

Note. This model is a weighted least-squares multiple regression calcu-lated with weights equal to the reciprocal of the variance for each effectsize. Effect sizes are positive for superior memory for congenial informa-tion and negative for superior memory for uncongenial information, b =unstandardized regression coefficient; j3 = standardized regression coef-ficient; n = 87.a 1 = recognition, 0 = recall. b 1 = high, 0 = low. c 1 = high, 0 =low. d 1 = high, 0 = low, mixed, or unknown. ° 1 = delayed, 0 =immediate. f 1 = within, mixed, or other, 0 = between. 6 1 = journalarticle, 0 = dissertation or unpublished document. h 1 = other, 0 =United States. ' Significance indicates model not correctly specified.j Birge ratio (see Hedges, 1994).t / > < .10 (marginally significant). ** p < .01. ***/><.001.

known to be blind were removed from the set of recall measures,the relation of the effect sizes to publication year became smallerand only marginally significant, because these deleted effect sizeswere especially large. Similarly, measurement artifact might ac-count for the tendency for effect sizes based on recognition mea-sures to decrease over time. To examine this possibility, weremoved the recognition sensitivity measures, which producedsmaller effect sizes; these more valid measures tended to occur inmore recent studies. If the temporal weakening of the congenialityeffect stems in part from earlier studies' somewhat inflated uncor-rected recognition measures, removal of the later studies' recog-nition sensitivity measures would attenuate the secular trend. In-deed, the trend did weaken, although it remained significant: Ingeneral, methodological weaknesses in the measurement of mem-ory apparently accounted for a portion of the decline of thecongeniality effect over time, although the recall measures pro-vided stronger evidence for this conclusion than the recognitionmeasures.

In contrast to this success in showing that measurement artifactlikely mediated the weakening of the congeniality effect, there wasno evidence that the decline over time was due to a tendency forrecent studies to feature more controversial issues. In addition, thesecular trend was not accomplished mainly by the addition ofcountercongeniality effects to the set of effect sizes. On the con-trary, since the late 1960s, large positive effects and large negativeeffects became somewhat rarer, as the effects converged on a nullvalue. This tendency was informed by the finding that a continuousmodel for year of publication showed as strong a secular trendwhen calculated on the absolute value of the effect sizes as it didwhen calculated on the signed effect sizes. Consistent with Green-wald's (1975) argument, one reasonable interpretation of thesetrends is that the true mean value of the congeniality effect in thisresearch literature is null; as methods became more refined, ex-periments approached this outcome.

Effects of Characteristics of Attitudinal Issue

The several variables representing the characteristics of theattitudinal issue produced mixed results. The larger congenialityeffect for highly value-relevant issues is consistent with our hy-pothesis that value relevance enhances defense motivation, if it is

Page 19: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

82 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

Table 13Individual Differences in the Strength of the Congeniality Effect

Variable and studyEffect size (d)

at high level of variableEffect size (d)

at low level of variable

DogmatismKleck and Wheaton (1967)

CommitmentPomerantz, Chaiken, and Tordesillas (1995)

ActivismEagly, Kulesa, Brannon, Shaw-Barnes, and

Hutson-Comeaux (1996, main study)Repression sensitization

Olson (1980)Evaluative-cognitive consistency

Zimmerman and Chaiken (1996, Study 1)Zimmerman and Chaiken (1996, Study 2)

Evaluative-affective consistencyZimmerman and Chaiken (1996, Study 2)

0.49

-0.23

-0.12

0.24

0.12-0.26

-0.04

-0.16

-0.08

0.01

0.10

0.820.10

-0.16

Note. Effect sizes are positive for superior memory for congenial information and negative for superiormemory for uncongenial information.

also assumed that defense motivation is often served by passiveinformation-processing strategies that would enhance congenialityeffects. However, the modest size of value relevance's moderationof the congeniality effect is not incompatible with our argumentthat heightened defense motivation is also capable of producingactive processing that would tend to improve memory for uncon-genial information. This reasoning about active attitudinal defenseis compatible with the categorical model examining moderation bythe controversiality of the attitudinal issue. Consistent with theargument that highly controversial issues are associated with bi-polar attitude structures and that bipolarity facilitates storage ofuncongenial information and fosters active counterarguing of it,the congeniality effect was smaller for controversial issues, al-though this moderation became nonsignificant when the recogni-tion measures were examined separately and was not preserved inthe multiple regression model.

Complicating the interpretation of both value relevance andcontroversiality is the covariation of these two variables and theirconsequent tendency to suppress each other's effects; it is notsurprising that more controversial issues such as abortion and thewar in Vietnam tended to be value relevant. Consequently, in themultiple regression models, the effect of value relevance wasweakened by the inclusion of controversiality in the equation.Issue controversiality warrants further evaluation as an antecedentof bipolarity, because its validity as an index of bipolarity restsmainly on Pratkanis's (1984/1985, 1989) research (see introduc-tion). Moreover, controversiality no doubt represents only one ofthe multiple determinants of the extent to which people attend to,learn, and store information on both sides of attitudinal issues.These effects might have been clarified in this meta-analysis if ithad been possible to obtain direct measures of the bipolarity versusunipolarity of attitude structures or of the tendency to use activeversus passive processing. Because our associated primary re-search on attitude memory using the highly controversial issues ofabortion and gays in the military (Eagly et al, 1996) found thatparticipants engaged in thoughtful counterarguing of uncongenialmessages and remembered them to the extent that they engaged inthis active process, assessment of active versus passive processing

is critical in future research. However, this subsequent researchfailed to find effects of bipolarity using direct assessments of theextent to which participants had prior knowledge of proattitudinalversus counterattitudinal arguments.

Results for the impact of outcome relevance were equivocal.Consistent with the assumption that this form of involvementwould foster accuracy-oriented processing and therefore woulddampen the tendency to better recall congenial information, highoutcome relevance was associated with a weaker congenialityeffect. However, this effect proved to be nonsignificant on therecognition measures as well as in the multiple regression model.

Effects of Timing of Memory Measure

Confirming Roberts's (1985) finding, the expected tendencywas obtained for the congeniality effect to be larger on delayedmeasures, defined as measures administered in a session subse-quent to the stimulus exposure. However, our confidence in anyoverall tendency for congeniality to increase with delay was less-ened by the failure of the congeniality effect to increase with thenumber of minutes that memory measurement was delayed withinthe stimulus exposure session or with the number of days thatelapsed before the delayed measurement session. Also, the conge-niality bias did not increase with the presence of interpolatedactivity before memory was assessed in the immediate session.Our confidence concerning the effects of delay was also lessenedby the nonsignificance in the multiple regression model of theimmediate versus delayed timing of memory measurement, de-fined as occurring in the stimulus exposure session versus asubsequent session, and by the lack of moderation of the conge-niality effect by a 2-week delay of measurement in our associatedprimary research (Eagly et al., 1996). Finally, the failure of themodel for immediate versus delayed timing of memory measure-ment to be significant on the recall measures raises questions aboutRoberts's (1985) argument that a greater congeniality effect ondelayed memory measures implicated retrieval (or reconstructive)processes, as opposed to encoding processes, as the more impor-tant determinant of attitude congeniality effects. Because encoding

Page 20: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 83

processes should be better assessed by recognition measures andretrieval processes by recall measures, Roberts's argument sug-gests that the delay effect that he and we observed on an overallbasis ought to have held more strongly for recall than recognitionmeasures—indeed, the reverse of what we found. Whether theattitude congeniality effect is more pronounced as time goes byand whether such a tendency interacts with type of memorymeasure remain worthy issues for research, because of their im-portance in explicating the cognitive processes that underlie atti-tude memory effects.

Effects of Nation Where Study Was Conducted

Difficult to interpret is the greater magnitude of the congenialityeffect in studies conducted in nations other than the United States.Our inability to identify any study characteristics that explainedthe impact of this variable on the effect sizes suggested that theforeign studies represented methodologies that were not discern-ibly different from those of the U.S. studies. Awkward from thepoint of view of cultural explanations of this result is the tendencyfor the Canadian studies to cluster with the other foreign studiesrather than the American studies. Although the United States hasbeen described as more individualist in cultural orientation thanCanada, the two nations are regarded as resembling one anothermore than either resembles any other nation (Lipset, 1990). Inaddition, of the foreign nations in our synthesis—Canada, Ger-many, Great Britain, India, and New Zealand—only India wouldtypically be labeled collectivist by cross-cultural psychologists(e.g., Triandis, 1995).

Our quantitative synthesis is not the first to produce strongerfindings in studies conducted in foreign nations. In Bond andSmith's (1996) meta-analysis, which produced smaller conformityeffects in individualist than collectivist nations, U.S. studies pro-duced a smaller average conformity effect than non-U.S. studies,including Canada (R. Bond, personal communication, March 31,1997). In addition, Wood, Lundgren, Ouellette, Busceme, andBlackstone's (1994) synthesis of minority influence experimentsestablished that the European initiators of this research paradigm(e.g., S. Moscovici, G. Mugny) produced stronger evidence forhypothesized minority influence effects than did other researchers,most of whom conducted their research in the United States.However, effects of nation cannot generally be accounted for bythe principle that effects are larger where research paradigmsoriginate because investigators in the United States initiated re-search on conformity and the attitude-memory relation. Althoughneither we nor the authors of these other two meta-analyses iden-tified methodological characteristics that accounted for observednations effects, the question of whether research practices differbetween nations warrants attention.

Because the broad distinction between individualism and col-lectivism (Triandis, 1995) proved useful in accounting for cross-national effects in Bond and Smith's (1996) synthesis of confor-mity studies, we scored the countries represented in the attitudememory literature on individualism-collectivism using Hofstede's(1980) values. Although the tendency was significant for thecongeniality effect to be smaller in more individualist countries,adding this individualism-versus-collectivism variable to the mul-tiple regression model only slightly reduced the prediction of the

effect sizes by the nation variable. Moreover, individualism versuscollectivism was not itself significant in the model.

A different cultural explanation invokes possible differences injournalistic and other media practices to explain the tendencies forthe congeniality effect to be weaker in the United States as well asto decline over the decades of attitude memory research. Distinc-tive in the United States in recent decades is the emphasis placedon balanced media coverage of controversial social issues, a styleoften known as objective journalism (DeFieur & Dennis, 1991).This style includes an emphasis on giving both sides of a contro-versy an opportunity to present their arguments and framing storiesin an emotionally detached manner. Although objective journalismmet with criticism in the 1960s in particular (Schudson, 1995), thestyle continues to predominate. In fact, the Society of ProfessionalJournalists includes in its Code of Ethics the statement that, "Newsreports should be free of opinion or bias and represent all sides ofan issue" (Applegate, 1996, p. 115). To the extent that the newsand other media coverage of social and political issues are morebalanced in the United States than in foreign countries, especiallyin recent years, both the nations effect and the year-of-publicationeffect might be explained at least in part in terms of an increasingtendency for U.S. citizens' attitudes to be more bipolar in structure,because of frequent exposure to both pro and con sides of contro-versial attitudinal issues. Unfortunately, quantitative informationsuitable for evaluating this argument was not available. Addressingthese issues systematically would require comparative studies ofmedia coverage and its impact on attitude structure.

Other Moderators

The larger congeniality effect associated with studies yieldingan error term that was based on within-subjects or mixed variabil-ity, compared with between-subjects variability, lends itself to twopossible interpretations. One interpretation is that within-subjectsdesigns provided a more precise error term because participantsserved as their own controls. Indeed, larger effects have beenfound in other meta-analyses for within-subjects than between-subjects designs (e.g., Eagly, Ashmore, Makhijani, & Longo,1991). A second interpretation is that exposure to both congenialand uncongenial information may favor congeniality because pre-sentation of only one side of an issue may produce a demand topay attention, regardless of congeniality, whereas presentation ofboth sides of an issue allows congenial and uncongenial informa-tion to compete for attention and processing capacity. Because ourassociated primary research (Eagly et al., 1996) obtained a nullcongeniality effect for a within-subjects manipulation of conge-niality as well as between-subjects manipulations, we favor thefirst interpretation, which implicates statistical precision.

The tendency for the congeniality effect to be larger in journalarticles than in dissertations and other unpublished documents istypical of meta-analytic findings (e.g., Lipsey & Wilson, 1993).Research on publication bias has consistently shown that studieswith positive, statistically significant results are more likely to bepublished (see review by Begg, 1994). This bias arises from thedampening of authors' intentions to publish when their findingsturn out to be null or contrary to established hypotheses (Green-wald, 1975; Rotton, Foos, Van Meek, & Levitt, 1995). Evidencefor publication bias in research literatures underscores the desir-ability of including unpublished studies in quantitative syntheses.

Page 21: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

84 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

This meta-analysis appropriately included as many such studies ascould be located: The 20 studies (29% of the total) from disserta-tions, master's theses, or other unpublished documents produced amodest downward correction of the mean effect size.

Notable Nonconfirmations of Hypotheses

A number of potential moderators of attitude-memory effectsdid not prove to be potent. In particular, the absence of a relationbetween the congeniality bias and the extent to which congenialinformation was more familiar than uncongenial information co-heres with the null findings of the few experiments that explicitlyexamined this potential confound (e.g., Greenwald & Sakumura,1967; Zimmerman & Chaiken, 1996). Because it seems obviousthat prior familiarity with congenial information should enhanceits memorability, an opposing tendency for novel information to bebetter remembered may have conspired to produce our null findingfor the familiarity variable. If this reasoning is correct, subsequentresearchers should take the familiarity of their congenial anduncongenial information into account and strive to equate the twotypes of information on this dimension.

This synthesis also obtained little evidence that the congenialityeffect receded to the extent that various cues promoted partici-pants' processing of attitudinally incongruent information or, moregenerally, promoted or suppressed processing of all attitude-relevant information. There was some evidence that outcome rel-evance, the motivational variable we thought would enhance ob-jective information processing (B. T. Johnson & Eagly, 1989),often attenuated the congeniality effect. Nonetheless, despite widevariation in the experimental instructions that participants re-ceived, these did not covary with the effect sizes, regardless ofwhether we coded them in a detailed way or divided them accord-ing to whether they explicitly instructed participants to learn ormemorize information. Because many of the instructional sets thatdid not favor intentional learning (e.g., judging the effectiveness ofpersuasive arguments) could easily have induced participants tofocus on uncongenial information to the same extent as congenialinformation, the failure to document a smaller congeniality effectfor studies featuring intentional learning instructions did not, inretrospect, surprise us. Rather than dismissing the potential mod-erating role of instructional sets or motivational goals, we urgegreater attention to them in subsequent attitude memory research.Particularly encouraging are studies suggesting that enhanced de-fense motivation, whether manipulated (Zimmerman & Chaiken,1996) or assessed through individual differences such as dogma-tism (Kleck & Wheaton, 1967; Rokeach, 1960), may increaseattitudes' impact on memory and that this increase may promotegreater memory for incongruent information when active ratherthan passive strategies for defending attitudes are used (e.g., bysensitizers vs. repressers, Olson, 1980; see also Zimmerman &Chaiken, 1996).

Conclusion

When viewing the results of this meta-analysis solely in terms ofthe magnitude of the attitude congeniality effect it obtained, psy-chologists might make the mistake of concluding, in line withinformal characterizations of this literature (e.g., Fishbein &Ajzen, 1975; Greenwald, 1975), that attitudes exert little, if any,

impact on memory for attitude-relevant information. Nonetheless,with respect to the issue of effect magnitude, our synthesis went farbeyond the mere recognition that the congeniality effect is gener-ally small because it revealed that this effect grew smaller (a) whenmethodologically suspect data sets were removed, (b) when studiesusing more accurate recognition-sensitivity measures of recogni-tion memory were isolated, and (c) when research progressed fromearly to contemporary times. When these moderating conditionsare taken into account, the tendency for attitudes to bias memoryin a congenial direction warrants description as extremely small oreven as null. Moreover, our associated experiments that weredesigned to model prototypical attitude memory experiments ob-tained null congeniality effects under a fairly wide range of con-ditions (Eagly et al., 1996). However, these findings do not supporta broader verdict that attitudes exert little or no influence onmemory for attitude-relevant information.

If attitudes have no effect on memory, empirical efforts topredict variation in the size of the attitude congeniality effect oughtto have been unsuccessful. Yet, this synthesis detected reliablemoderator variables, above and beyond those moderators that areinterpretable in terms of measurement artifact contaminating aportion of the studies. Moderators such as value relevance andcontroversiality produced theoretically meaningful variation inattitudes' effects on memory, and one not easily interpreted mod-erator—the nation in which studies were conducted—raised newquestions about the mechanisms by which attitudes affect memory.Moreover, a great deal of systematic variation remained unex-plained, in large part because many theoretically important mod-erators of the attitude-memory relation could not be assessedmeta-analytically. These considerations suggest the value of re-newed attention to this research area.

Investigators' attention could usefully be directed to motiva-tional and structural moderators that this synthesis could not rep-resent. Regarding motivational moderators, explicit attention tofactors that motivate defensive processing of attitude-relevant in-formation is called for. Moreover, it is also critical for research toascertain the extent to which defensive processing is active in thesense that people attend to uncongenial information in order tocounterargue it. Such processing would still be in the defense ofattitudes but could manifest itself, at least on immediate measuresof memory, in the absence of a congeniality effect (Eagly et al.,1996) or even, under some circumstances, a reversal of the con-geniality effect (Zimmerman & Chaiken, 1996).

Regarding attitude structure as a moderator of attitudes' effectson memory, possessing a coherent set of attitude-congruent beliefsshould enable people to challenge uncongenial information, atleast when they are sufficiently motivated to defend their attitudes.Also, despite the inability of this synthesis to produce a strong testof the hypothesis that a bipolar attitude structure dampens thecongeniality effect, experimental tests of this hypothesis may lendit support and thus advance understanding of how attitude structurecan foster memory for both uncongenial and congenial informa-tion. In this regard, our auxiliary synthesis of the handful of studiesthat contrasted participants' memory for extreme versus moderateinformation revealed that extreme information was better remem-bered, a finding compatible with our bipolarity logic and Judd andKulik's (1980) conceptualization of attitudes as bipolar schemas.

Additional aspects of attitude structure should be addressed inprimary research. In particular, Zimmerman and Chaiken's (1996)

Page 22: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 85

research points to the importance of taking into account whetherattitudes are based mainly on cognitions (higher evaluative-cognitive consistency) or mainly on affect (higher evaluative-affective consistency). In fact, their findings that cognitively basedattitudes produced an incongruency effect on immediate memorywhereas affectively based attitudes produced a congruency effectwere replicated in a recent study conducted after the cutoff of ourliterature search (Duckworth & Chaiken, 1998). In addition, thisstudy found that having participants focus on their cognitionstoward the target issue produced an incongruency effect, whereashaving them focus on their feelings produced a congruency effect.These studies indicate that the direction of attitudes' impact onmemory depends on particular aspects of available and accessibleintra-attitudinal structure. It is clear that other important concep-tions of attitude structure and attitude strength that have not beenexamined in the attitude memory literature (e.g., accessibility,working knowledge) deserve attention.

Despite the importance of the moderators that this meta-analysisuncovered and the likely importance of moderators that are yet tobe investigated, it is extremely unlikely that subsequent researchwill change the central tendency in this research literature bydemonstrating a strong and consistent tendency for attitudes to biasmemory in a congenial direction. Insofar as new research is able toclarify factors that promote better memory for information that iscongenial, uncongenial, or extreme, concentration on the classiccongeniality effect should decline, and the general power of atti-tudes to influence memory in a variety of ways should be betterappreciated. Yet, the one-sided focus of past researchers on pro-ducing a congeniality effect is understandable in view of initialtheorizing about attitude memory and early findings that seemedhighly supportive of this theorizing, especially Levine and Mur-phy's (1943) dramatic results. In view of the subsequent literatureand the circumstantial evidence provided by this meta-analysis,Levine and Murphy's findings were very likely in error. Unfortu-nately, these findings fostered the expectation of a strong androbust congeniality effect and set the tone of several decades ofexperimentation.

In addition to recognizing that attitudes have different kinds ofimpact on memory and examining the motivational and structuralvariables that produce these varying outcomes, investigatorsshould address the processes that underlie these outcomes. Pro-cesses such as attention and elaboration that may mediate memoryeffects have not been directly examined in prior research butemerged as critical mediators of memory for uncongenial mes-sages in our associated primary research (Eagly et al., 1996). Astheorizing about attitude memory effects grows more sophisti-cated, studies will need to incorporate measures of potential me-diating processes and manipulations designed to influence theseprocesses.

Questions concerning the measures used to assess memory andthe timing of these measurements, although not resolved by thissynthesis, emerged as important. In particular, more research usingrecognition-sensitivity measures is called for given that this classof measures, culled from only a few studies, produced null effects.Also, memory measures that are implicit as well as explicit shouldbe used in subsequent work, because of the significance of thisimplicit-explicit distinction for understanding underlying cogni-tive processes. Primary research that explicitly contrasts the direc-tion and magnitude of attitude memory effects across different

types of memory measures should help disentangle the role ofencoding and retrieval processes.

In closing, we reiterate the major implication of our findings—namely, that attitudes exert a variety of effects on memory. Thisquantitative synthesis has shed light on some of the factors andmechanisms that produce these effects. Consistent with Eagly andWood's (1994) view that meta-analyses are generally a way stationwhere analysts sift and evaluate evidence as they proceed towardmore thorough understanding of a particular relation, our empiricalfindings and theoretical analysis show that this research areacontinues to provide exciting opportunities for developing empir-ical and theoretical knowledge about the functioning of attitudes.

References

References marked with an asterisk indicate studies included in the meta-analysis.

Allport, G. W. (1935). Attitudes. In C. Murchison (Ed.), Handbook ofsocial psychology (pp. 798-844). Worcester, MA: Clark UniversityPress.

Alper, T. G., & Korchin, S. J. (1952). Memory for socially relevantmaterial. Journal of Abnormal and Social Psychology, 47, 25-27.

*Ambler, R. S. (1973). The relationship of ego-involvement to messageperception and retention (Unpublished doctoral dissertation, Ohio StateUniversity, 1972). Dissertation Abstracts International, 33, 6482A.

Applegate, E. (1996). Print and broadcast journalism: A critical exami-nation. Westport, CT: Praeger.

Asch, S. E. (1952). Social psychology. Englewood Cliffs, NJ: PrenticeHall.

Bartlett, F. C. (1932). Remembering: A study in experimental and socialpsychology. Cambridge, England: Cambridge University Press.

Begg, C. B. (1994). Publication bias. In H. Cooper & L. V. Hedges (Eds.),The handbook of research synthesis (pp. 399-409). New York: RussellSage Foundation.

Behavioral Science Subpanel of President's Science Advisory Committee.(1962). Strengthening the behavioral sciences. Behavioral Science, 7,275-288.

Berent, M. K., Krosnick, J. A., & Boninger, D. S. (1993). Attitude impor-tance and memory for attitude-relevant information. Unpublished manu-script, Ohio State University.

*Berger, W. G. (1970). Some correlates of attitude change, retention of theattitude message, and evaluation of the communicator (Unpublisheddoctoral dissertation, Michigan State University, 1969). DissertationAbstracts International, 31, 1515B.

Biek, M., Wood, W., & Chaiken, S. (1996). Working knowledge, cognitiveprocessing, and attitudes: On the determinants of bias. Personality andSocial Psychology Bulletin, 22, 547-556.

Bond, R., & Smith, P. B. (1996). Culture and conformity: A meta-analysisof studies using Asch's (1952b, 1956) line judgment task. PsychologicalBulletin, 119, 111-137.

Boninger, D., Krosnick, J. A., Berent, M., & Fabrigar, L. R. (1995). Thecauses and consequences of attitude importance. In R. E. Petty & J. A.Krosnick (Eds.), Attitude strength: Antecedents and consequences (pp.159-190). Mahwah, NJ: Erlbaum.

*Bothwell, R. K., & Brigham, J. C. (1983). Selective evaluation and recallduring the 1980 Reagan-Carter debate. Journal of Applied Social Psy-chology, 13, 427-442.

Brigham, J. C., & Cook, S. W. (1969). The influence of attitude on therecall of controversial material: A failure to confirm. Journal of Exper-imental Social Psychology, 5, 240-243.

*Briody, D. M. (1971). The effects of ego-involvement and initial attitudeon the physiological responsivity to and content retention of a repeated

Page 23: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

86 EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

message (Unpublished doctoral dissertation, University of Utah, 1971).Dissertation Abstracts International, 32, 3461A.

Bruner, J. S., & Goodman, C. C. (1947). Value and need as organizingfactors in perception. Journal of Abnormal and Social Psychology, 42,33-44.

*Buss, A. H., & Brock, T. C. (1963). Repression and guilt in relation toaggression. Journal of Abnormal and Social Psychology, 66, 345—350.

*Cacioppo, J. T., & Petty, R. E. (1979). Effects of message repetition andposition on cognitive response, recall, and persuasion. Journal of Per-sonality and Social Psychology, 37, 97-109.

*Callahan, M. J., & Olson, J. M. (1981, June). The effects of attitudes onlong term memory in an intentional learning paradigm. Paper presentedat the meeting of the Canadian Psychological Association, Toronto,Ontario, Canada.

Chaiken, S., Giner-Sorolla, R., & Chen, S. (1996). Beyond accuracy:Defense and impression motives in heuristic and systematic informationprocessing. In P. M. Gollwitzer & J. A. Bargh (Eds.), The psychology ofaction: Linking cognition and motivation to behavior (pp. 553-578).New York: Guilford Press.

Chaiken, S., Liberman, A., & Eagly, A, H. (1989). Heuristic and systematicprocessing within and beyond the persuasion context. In J. S. Uleman &J. A. Bargh (Eds.), Unintended thought (pp. 212-252). New York:Guilford Press.

Chaiken, S., Pomerantz, E. M., & Giner-Sorolla, R. (1995). Structuralconsistency and attitude strength. In R. E. Petty & ). A. Krosnick (Eds.),Attitude strength: Antecedents and consequences (pp. 387-412). Hills-dale, NJ: Erlbaum.

Chaiken, S., Wood, W., & Eagly, A. H. (1996). Principles of persuasion.In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Hand-book of basic principles (pp. 702-742). New York: Guilford Press.

Clark, K. B. (1940). Some factors influencing the remembering of prosematerial. Archives of Psychology (Columbia University), No. 253 73, pp.5-73.

Cohen, J. (1977). Statistical power analysis for the behavioral sciences(Rev. ed.). New York: Academic Press.

Cooper, H. M. (1989). Integrating research: A guide for literature reviews(2nd ed.). Newbury Park, CA: Sage.

DeFleur, M. L., & Dennis, E. E. (1991). Understanding mass communi-cation (4th ed.). Boston: Houghton Mifflin.

Donohew, L. (1966). Decoder behavior on incongruent political material:A pilot study. Journal of Communication, 16, 133-142.

Doob, L. W. (1953). Effects of initial serial position and attitude uponrecall under conditions of low motivation. Journal of Abnormal andSocial Psychology, 48, 199-205.

*Drake, R. A. (1991). Processing persuasive arguments: Recall and rec-ognition as a function of agreement and manipulated activation asym-metry. Brain and Cognition, 15, 83—94.

Duckworth, K., & Chaiken, S. (1998). [The effects of availability andaccessibility of affective and cognitive components of attitude on mem-ory for attitude-relevant information]. Unpublished raw data, New YorkUniversity.

*Dupree, D. M. (1992). Sources of error in the recall of controversialmaterial. Unpublished master's thesis, Emory University.

*Dutta, S., Kanungo, R. N., & Freibergs, V. (1972). Retention of affectivematerial: Effects of intensity of affect on retrieval. Journal of Person-ality and Social Psychology, 23, 64-80.

Eagly, A. H. (1992). Uneven progress: Social psychology and the study ofattitudes. Journal of Personality and Social Psychology, 63, 693-710.

Eagly, A. H., Ashmore, R. D., Makhijani, M. G., & Longo, L. C. (1991).What is beautiful is good, but. . . : A meta-analytic review of researchon the physical attractiveness stereotype. Psychological Bulletin, 110,109-128.

Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes. FortWorth, TX: Harcourt, Brace, Jovanovich.

Eagly, A. H., & Chaiken, S. (1995). Attitude strength, attitude structure,and resistance to change. In R. E. Petty & J. A. Krosnick (Eds.), Attitudestrength: Antecedents and consequences (pp. 413-432). Mahwah, NJ:Erlbaum.

Eagly, A. H., & Chaiken, S. (1998). Attitude structure and function. In D.Gilbert, S. Fiske, & G. Lindzey (Eds.), The handbook of social psychol-ogy (4th ed., Vol. 1, pp. 269-322). New York: McGraw-Hill.

*Eagly, A. H., Kulesa, P., Brannon, L., Shaw-Barnes, K., & Hutson-Comeaux, S. (1996). Why counterattitudinal messages are as memora-ble as proattitudinal messages: The importance of active defense againstattack. Unpublished manuscript. (Preliminary experiment and Experi-ment 1 included in meta-analysis)

Eagly, A. H., & Mladinic, A. (1989). Gender stereotypes and attitudestoward women and men. Personality and Social Psychology Bulletin, 15,543-558.

Eagly, A. H., Mladinic, A., & Otto, S. (1991). Are women evaluated morefavorably than men? An analysis of attitudes, beliefs, and emotions.Psychology of Women Quarterly, 15, 203—216.

Eagly, A. H., & Wood, W. (1994). Using research syntheses to plan futureresearch. In H. Cooper & L. V. Hedges (Eds.), The handbook of researchsynthesis (pp. 485-500). New York: Russell Sage Foundation.

Eberhart, J. C., & Bauer, R. (1941). An analysis of the influences on recallof a controversial event: The Chicago Tribune and the Republic Steelstrike. Journal of Social Psychology, 14, 211-228.

*Edwards, A. L. (1941a). Political frames of reference as a factor influ-encing recognition. Journal of Abnormal and Social Psychology, 36,34-50.

Edwards, A. L. (1941b). Rationalization in recognition as a result of apolitical frame of reference. Journal of Abnormal and Social Psychol-ogy, 36, 224-235. (Same study as 1941a)

Edwards, K., & Smith, E. E. (1996). A disconfirmation bias in the evalu-ation of arguments. Journal of Personality and Social Psychology, 71,5-24.

*Eiser, J. R., & Monk, A. F. (1978). Is the recognition of attitude state-ments affected by one's own opinion? European Journal of SocialPsychology, 8, 529-533.

Fazio, R. H. (1995). Attitudes as object-evaluation associations: Determi-nants, consequences, and correlates of attitude accessibility. In R. E.Petty & J. A. Krosnick (Eds.), Attitude strength: Antecedents and con-sequences (pp. 247-282). Mahwah, NJ: Erlbaum.

Feather, N. T. (1969). Attitude and selective recall. Journal of Personalityand Social Psychology, 12, 310-319.

Feather, N. T. (1970). Balancing and positivity effects in social recall.Journal of Personality, 38, 602-628.

Festinger, L. (1957). A theory of cognitive dissonance. Evanston, IL: Row,Peterson.

Festinger, L. (1964). Conflict, decision, and dissonance. Stanford, CA:Stanford University Press.

Fishbein, M., & Ajzen, I. (1972). Attitudes and opinions. Annual Review ofPsychology, 23, 487-544.

Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior:An introduction to theory and research. Reading, MA: Addison-Wesley.

Fiske, S. T., & Taylor, S. E. (1991). Social cognition (2nd ed.). Reading,MA: Addison-Wesley.

*Fitzgerald, D., & Ausubel, D. P. (1963). Cognitive versus affective factorsin the learning and retention of controversial material. Journal of Edu-cational Psychology, 54, 73—84.

*Fortner, B. H., & Henk, W. A. (1991). Effects of issue-related attitude onreaders' comprehension and judgments of unbiased text. Reading Re-search and Instruction, 30, 1-16.

Freedman, J. L., & Sears, D. O. (1965). Selective exposure. In L. Berko-witz (Ed.), Advances in experimental social psychology (Vol. 2, pp.57-97). San Diego, CA: Academic Press.

Freud, S. (1938). The psychopathology of everyday life. In A. A. Brill

Page 24: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 87

(Ed.), The basic writings of Sigmund Freud (pp. 35-178). New York:Random House, Modern Library Edition.

Funder, D. C., & Ozer, D. J. (1983). Behavior as a function of the situation.Journal of Personality and Social Psychology, 44, 107-112.

Furnham, A., & Duignan, S. (1989). The selective recall of attitudeconsistent information: A study concerning sex differences. Psycholo-gia: An International Journal of Psychology in the Orient, 32, 112-119.

*Furnham, A., & Proctor, E. (1990). Memory for information about nu-clear power: A test of the selective recall hypothesis. Current Psychol-ogy: Research and Reviews, 8, 287-297.

Furnham, A., & Singh, A. (1986). Memory for information about sexdifferences. Sex Roles, 15, 479-486.

Gackenbach, J. I., & Auerbach, S. M. (1985). Sex-role attitudes andperceptual learning. Journal of Social Psychology, 125, 233-243.

*Gailon, A. K. (1967). The effect of attitude on learning and recall(Unpublished doctoral dissertation, University of California, Berkeley,1966). Dissertation Abstracts International, 28, 962.

Garber, R. B. (1955). Influence of cognitive and affective factors inlearning and retaining attitudinal materials. Journal of Abnormal andSocial Psychology, 51, 384-389.

*George, J. M. (1976). The effects of attitude and commitment on retention(Unpublished doctoral dissertation, Duke University, 1976). Disserta-tion Abstracts International, 37, 1033B.

*Giner-Sorolla, R., & Chaiken, S. (1994). The causes of hostile mediajudgments. Journal of Experimental Social Psychology, 30, 165-180.

*Giner-Sorolla, R., & Chaiken, S. (1996). Selective use of heuristic andsystematic processing under defense motivation. Unpublished manu-script, New York University.

Gormly, A. V. (1974). Recall of attitudinal and value belief statements ininterpersonal judgment tasks. Bulletin of the Psychonomic Society, 4(2-A), 102-104.

Greaves, G. (1972). Conceptual system functioning and selective recall ofinformation. Journal of Personality and Social Psychology, 21, 327-332.

Green, D. M., & Swets, J. A. (1966). Signal detection theory and psycho-physics. New York: Wiley.

Greenwald, A. G. (1968). Cognitive learning, cognitive response to per-suasion, and attitude change. In A. G. Greenwald, T. C. Brock, & T. M.Ostrom (Eds.), Psychological foundations of attitudes (pp. 147—170).New York: Academic Press.

Greenwald, A. G. (1975). Consequences of prejudice against the nullhypothesis. Psychological Bulletin, 82, 1-20.

*Greenwald, A. G., & Sakumura, J. S. (1967). Attitude and selectivelearning: Where are the phenomena of yesteryear? Journal of Person-ality and Social Psychology, 7, 387-397.

Hedges, L. V. (1987). How hard is hard science, how soft is soft science?The empirical cumulativeness of research. American Psychologist, 42,443-455.

Hedges, L. V. (1994). Fixed effects models. In H. Cooper & L. V. Hedges(Eds.), The handbook of research synthesis (pp. 285-299). New York:Russell Sage Foundation.

Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis.Orlando, FL: Academic Press.

*Henk, W. A., & Holmes, B. C. (1988). Effects of content-related attitudeon the comprehension and retention of expository text. Reading Psy-chology: An International Quarterly, 9, 203-225.

Hofstede, G. (1980). Culture's consequences: International differences inwork-related values. Newbury Park, CA: Sage.

*Howard, J. W. (1978). Social categorization: Biasing of memory foringroup and outgroup information (Unpublished doctoral dissertation,University of Oregon, 1977). Dissertation Abstracts International, 39,443B.

*Howard, J. W., & Rothbart, M. (1980). Social categorization and memory

for in-group and out-group behavior. Journal of Personality and SocialPsychology, 38, 301-310.

*Howard-Pitney, B., Borgida, E., & Omoto, A. M. (1986). Personal in-volvement: An examination of processing differences. Social Cogni-tion, 4, 39-57.

*Hymes, R. W. (1986). Political attitudes as social categories: A new lookat selective memory. Journal of Personality and Social Psychology, 51,233-241.

Ikegami, T. (1986). The role of affect in person memory: The influence ofpositive and negative affect upon recognition memory. Japanese Psy-chological Research, 28, 154-159.

Jacoby, L. L., Torn, J. P., Lindsay, D. S., & Debner, J. A. (1992). Lecturesfor a layperson: Methods for revealing unconscious processes. In R. F.Bornstein & T. S. Pittman (Eds.), Perception without awareness (pp.81-120). New York: Guilford Press.

James, W. (1952). The principles of psychology. Chicago: EncyclopaediaBritannica. (Original work published 1890)

*Jamieson, B. D. (1970). Attitude, plausibility, and the learning and recallof controversial material. Journal of Psychology, 76, 169-174.

*Janis, I. L., & Rausch, C. N. (1970). Selective interest in communicationsthat could arouse decisional conflict: A field study of participants in thedraft-resistance movement. Journal of Personality and Social Psychol-ogy, 14, 46-54.

Johnson, B. T. (1993). DSTAT 1.10: Software for the meta-analytic reviewof research literatures. Hillsdale, NJ: Erlbaum.

Johnson, B. T. (1994). Effects of outcome-relevant involvement and priorinformation on persuasion. Journal of Experimental Social Psychol-ogy, 30, 556-579.

Johnson, B. T., & Eagly, A. H. (1989). The effects of involvement onpersuasion: A meta-analysis. Psychological Bulletin, 106, 290-314.

Johnson, B. T., & Eagly, A. H. (in press). Quantitative synthesis of socialpsychological research. In H. T. Reis & C. M. Judd (Eds.), Handbook ofresearch methods in social psychology. New York: Cambridge Univer-sity Press.

*Johnson, J. T., & Judd, C. M. (1983). Overlooking the incongruent:Categorization biases in the identification of political statements. Jour-nal of Personality and Social Psychology, 45, 978-996.

*Jones, E. E., & Aneshansel, J. (1956). The learning and utilization ofcontravaluant material. Journal of Abnormal and Social Psychology, 53,27-33.

*Jones, E. E., & Kohler, R. (1958). The effects of plausibility on thelearning of controversial statements. Journal of Abnormal and SocialPsychology, 57, 315-320.

*Judd, C. M., & Kulik, J. A. (1980). Schematic effects of social attitudeson information processing and recall. Journal of Personality and SocialPsychology, 38, 569-578.

*Kanungo, R. N., & Dutta, S. (1966). Retention of affective material:Frame of reference or intensity? Journal of Personality and SocialPsychology, 4, 27-35.

*Kitano, E. (1970). A socio-psychological study on memory trace: Theeffect of attitude upon memory and its transformation. Japanese Journalof Psychology, 40, 310-318.

*Kleck, R. E., & Wheaton, J. (1967). Dogmatism and responses to opinion-consistent and opinion-inconsistent information. Journal of Personalityand Social Psychology, 5, 249-252.

Kunda, Z. (1990). The case for motivated reasoning. Psychological Bul-letin, 108, 480-498.

Leippe, M. R., & Elkin, R. A. (1987). When motives clash: Issue involve-ment and response involvement as determinants of persuasion. Journalof Personality and Social Psychology, 52, 269-278.

*Levine, J. M., & Murphy, G. (1943). The learning and forgetting ofcontroversial material. Journal of Abnormal and Social Psychology, 38,507-517.

Page 25: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

EAGLY, CHEN, CHAIKEN, AND SHAW-BARNES

Lipset, S. M. (1990). Continental divide: The values and institutions of theUnited States and Canada. New York: Routledge.

Lipsey, M. W., & Wilson, D. B. (1993). The efficacy of psychological,educational, and behavioral treatment: Confirmation from meta-analysis.American Psychologist, 48, 1181-1209.

*Love, R. E., & Greenwald, A. G. (1978). Cognitive responses to persua-sion as mediators of opinion change. Journal of Social Psychology, 104,231-241.

Maio, G. R., & Olson, J. M. (1995). Involvement and persuasion: Evidencefor different types of involvement. Canadian Journal of BehaviouralScience, 27, 64-78.

*Malpass, R. S. (1969). Effects of attitude on learning and memory: Theinfluence of instruction-induced sets. Journal of Experimental SocialPsychology, 5, 441-453.

Manis, M., & Dawes, R. M. (1961). Cloze score as a function of attitude.Psychological Reports, 9, 79-84.

*Mannello, G. (1954). Attitude as a conditioner of the acquisition of newfacts among eighth grade pupils. Journal of Genetic Psychology, 85,85-103.

Matthew, R. L. (1975). The selective retention of favorable, neutral, andunfavorable statements about men and women measured by signaldetection. Unpublished doctoral dissertation, University of North Da-kota.

McGuire, W. J., & McGuire, C. V. (1991). The content, structure, andoperation of thought systems. In R. S. Wyer, Jr., & T. K. Srull (Eds.),Advances in social cognition (Vol. 4, pp. 1-78). Hillsdale, NJ: Erlbaum.

*Olson, J. M. (1980). Selective recall: Attitudes, schemata, and memory(Unpublished doctoral dissertation, University of Waterloo, Canada,1980). Dissertation Abstracts International, 41, 1163B.

Olson, J. M., Roese, N. J., & Zanna, M. P. (1996). Expectancies. In E. T.Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook ofbasic principles (pp. 211-238). New York: Guilford Press.

*Paulson, S. F. (1954). The effects of the prestige of the speaker andacknowledgment of opposing arguments on audience retention and shiftof opinion. Speech Monographs, 21, 261-211.

Petty, R. E., & Cacioppo, J. T. (1990). Involvement and persuasion:Tradition versus integration. Psychological Bulletin, 107, 367-374.

Petty, R. E., & Krosnick, J. A. (Eds.). (1995). Attitude strength: Anteced-ents and consequences. Mahwah, NJ: Erlbaum.

Petty, R. E., & Wegener, D. T. (1998). Attitude change: Multiple roles forpersuasion variables. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), Thehandbook of social psychology (4th ed., Vol. 1, pp. 323-390). NewYork: McGraw-Hill.

*Pomerantz, E. M., Chaiken, S., & Tordesillas, R. S. (1995). Attitudestrength and resistance processes. Journal of Personality and SocialPsychology, 69, 408-419.

*Postman, L., & Murphy, G. (1943). The factor of attitude in associativememory. Journal of Experimental Psychology, 33, 228—238.

*Pratkanis, A. R. (1985). Attitudes and memory: The heuristic and sche-matic functions of attitudes (Unpublished doctoral dissertation, OhioState University, 1984). Dissertation Abstracts International, 45,3657B.

Pratkanis, A. R. (1989). The cognitive representation of attitudes. In A. R.Pratkanis, S. J. Breckler, & A. G. Greenwald (Eds.), Attitude structureand Junction (pp. 71-98). Hillsdale, NJ: Erlbaum.

Pratkanis, A. R., Syak, P. C., & Gamble, E. C. (1987, May). The role oftechnical knowledge in attitudinal learning. Paper presented at themeeting of the Midwestern Psychological Association, Chicago.

*Read, S. J., & Rosson, M. B. (1982). Rewriting history: The biasingeffects of attitudes on memory. Social Cognition, 1, 240-255.

Reutzel, D. R., & Hollingsworth, P. M. (1991). Investigating topic-relatedattitude: Effect on reading and remembering text. Journal of EducationalResearch, 84, 334-344.

*Roberts, J. V. (1984). Public opinion and capital punishment: The effects

of attitudes upon memory. Canadian Journal of Criminology, 26, 283-291.

Roberts, J. V. (1985). The attitude-memory relationship after 40 years: Ameta-analysis of the literature. Basic and Applied Social Psychology, 6,221-241.

Rojahn, K., & Pettigrew, T. F. (1992). Memory for schema-relevantinformation: A meta-analytic review. British Journal of Social Psychol-ogy, 31, 81-109.

Rokeach, M. (1960). The open and closed mind: Investigations into thenature of belief systems and personality systems. New York: BasicBooks.

Rosenthal, R. (1991). Meta-analytic procedures for social research (Rev.ed.). Newbury Park, CA: Sage.

Rosenthal, R., & Rosnow, R. L. (Eds.). (1969). Artifact in behavioralresearch. New York: Academic Press.

Rosenthal, R., & Rubin, D. B. (1982). A simple, general purpose display ofmagnitude of experimental effect. Journal of Educational Psychol-ogy, 74, 708-712.

Ross, M. (1989). Relation of implicit theories to the construction ofpersonal histories. Psychological Review, 96, 341-357.

Rotton, J., Foos, P. W., Van Meek, L., & Levitt, M. (1995). Publicationpractices and the file drawer problem: A survey of published authors.Journal of Social Behavior and Personality, 10, 1-13.

*Russell, D., & Jones, W. H. (1980). When superstition fails: Reactions todisconfirmation of paranormal beliefs. Personality and Social Psychol-ogy Bulletin, 6, 83-88.

Schacter, D. L. (1987). Implicit memory: History and current status.Journal of Experimental Psychology: Learning, Memory, and Cogni-tion, 13, 501-518.

*Schmidt, L. R. (1969). Selektives Behalten bewertender Aussagen tiber"den" Vater bei Kindern unter besonderer Beriicksichtigung ihrer Ein-stellungen [Children's selective retention for value judgments abouttheir fathers according to their attitudes toward them]. Zeitschrift furExperimentelle und Angewandte Psychologic, 16, 138—156.

Schudson, M. (1995). The power of news. Cambridge, MA: HarvardUniversity Press.

Shapiro, M. A. (1994). Signal detection measures of recognition memory.In A. Lang (Ed.), Measuring psychological responses to media messages(pp. 133-148). Hillsdale, NJ: Erlbaum.

*Sheppard, B. H. (1981). Opinions and remembering revisited: The case ofthe disappearing phenomenon. Unpublished manuscript.

Sherif, M., & Hovland, C. I. (1961). Social judgment: Assimilation andcontrast effects in communication and attitude change. New Haven, CT:Yale University Press.

*Smith, S. S., & Jamieson, B. D. (1972). Effects of attitude and egoinvolvement on the learning and retention of controversial material.Journal of Personality and Social Psychology, 22, 303-310.

*Spiro, R. J., & Sherif, C. W. (1975). Consistency and relativity inselective recall with differing ego-involvement. British Journal of Socialand Clinical Psychology, 14, 351-361.

Srull, T. K. (1984). Methodological techniques for the study of personmemory and social cognition. In R. S. Wyer & T. K. Srull (Eds.),Handbook of social cognition (pp. 73-150). Hillsdale, NJ: Erlbaum.

Stangor, C., & McMillan, D. (1992). Memory for expectancy-congruentand expectancy-incongruent information: A review of the social andsocial developmental literatures. Psychological Bulletin, 111, 42-61.

Stapf, K. H., Stroebe, W., & Jonas, K. (1986). Amerikaner tiber Deutsch-land und die Deutschen: Urteile und Vorurteile [American views ofGermany and the Germans: Opinions and prejudices]. Opladen, Ger-many: Westdeutscher Verlag.

Swim, J. K., & Cohen, L. L. (1997). Overt, covert, and subtle sexism: Acomparison between the Attitudes Toward Women and Modern SexismScales. Psychology of Women Quarterly, 21, 103-118.

Page 26: The Impact of Attitudes on Memory: An Affair to Remember€¦ · The Impact of Attitudes on Memory: An Affair to Remember Alice H. Eagly Northwestern University Serena Chen University

IMPACT OF ATTITUDES ON MEMORY 89

Triandis, H. C. (1995). Individualism and collectivism. Boulder, CO:Westview Press.

*Upmeyer, A., Krolage, 3., Etzel, G., Lilli, W., & Klump, H. (1976).Accentuation of information in real competing groups. European Jour-nal of Social Psychology, 6, 95-97.

Upmeyer, A., & Layer, H. (1974). Accentuation and attitude in socialjudgment. European Journal of Social Psychology, 4, 469-488.

*Wallston, B. S. (1970). Factors influencing selective learning: An inte-grative approach. Unpublished master's thesis, University of Wiscon-sin.

*Waly, P., & Cook, S. W. (1966). Attitude as a determinant of learning andmemory: A failure to confirm. Journal of Personality and Social Psy-chology, 4, 280-288.

*Watson, W. S., & Hartmann, G. W. (1939). The rigidity of a basicattitudinal frame. Journal of Abnormal and Social Psychology, 34,314-335.

Weiss, W. (1953). A "sleeper" effect in opinion change. Journal of Ab-normal and Social Psychology, 48, 173—180.

*WeIden, T. A. (1968). The effect of attitudes and retention of messageorder in controversial material (Unpublished doctoral dissertation, Mich-igan State University, 1961). Dissertation Abstracts International, 22,4442.

*Weldon, D. E., & Malpass, R. S. (1981). Effects of attitudinal, cognitive,

and situational variables on recall of biased communications. Journal ofPersonality and Social Psychology, 40, 39-52.

*Wickey, J. (1970). Cognitive factors in the immediate and delayed recallof attitudinal materials (Unpublished doctoral dissertation, Wayne StateUniversity, 1967). Dissertation Abstracts International, 31, 2270B.

Wood, W., Lundgren, S., Ouellette, J. A., Busceme, S., & Blackstone, T.(1994). Minority influence: A meta-analytic review of social influenceprocesses. Psychological Bulletin, 115, 323-345.

Wood, W. L., Rhodes, N., & Biek, M. (1995). Working knowledge andattitude strength: An information-processing analysis. In R. E. Petty &J. A. Krosnick (Eds.), Attitude strength: Antecedents and consequences(pp. 283-313). Mahwah, NJ: Erlbaum.

*Zimmerman, J., & Chaiken, S. (1996). Attitude structure, processinggoals, and selective memory. Unpublished manuscript.

Zuwerink, J. R., & Devine, P. G. (1996). Attitude importance and resis-tance to persuasion: It's not just the thought that counts. Journal ofPersonality and Social Psychology, 70, 931-944.

Received August 15, 1997Revision received July 16, 1998

Accepted July 21, 1998

New Editors Appointed, 2000-2005

The Publications and Communications Board of the American Psychological Associa-tion announces the appointment of three new editors for 6-year terms beginning in 2000.

As of January 1,1999, manuscripts should be directed as follows:

• For Experimental and Clinical Psychophannacology, submit manuscripts toWarren K. Bickel, PhD, Department of Psychiatry, University of Vermont, 38Fletcher Place, Burlington, VT 05401-1419.

• For the Journal of Counseling Psychology, submit manuscripts to Jo-Ida C.Hansen, PhD, Department of Psychology, University of Minnesota, 75 EastRiver Road, Minneapolis, MN 55455-0344.

• For the Journal of Experimental Psychology: Human Perception and Perfor-mance, submit manuscripts to David A. Rosenbaum, PhD, Department of Psy-chology, Pennsylvania State University, 642 Moore Building, University Park,.PA 16802-3104.

Manuscript submission patterns make the precise date of completion of the 1999 volumesuncertain. Current editors, Charles R. Schuster, PhD; Clara E. Hill, PhD; and Thomas H.Carr, PhD, respectively, will receive and consider manuscripts through December 31,1998. Should 1999 volumes be completed before that date, manuscripts will be redirectedto the new editors for consideration in 2000 volumes.