Exaggeration in memory: Systematic distortion of self-evaluative information under reduced...

Available online at www.sciencedirect.com

www.elsevier.com/locate/jesp

Journal of Experimental Social Psychology 44 (2008) 246–259

Exaggeration in memory: Systematic distortion ofself-evaluative information under reduced accessibility

Greg Willard *, Richard H. Gramzow

Northeastern University, Department of Psychology, 125 Nightingale Hall, 360 Huntington Avenue, Boston, MA 02115-5000, USA

Received 13 February 2007Available online 1 May 2007

Abstract

The tendency to exaggerate specific information about oneself can stem from reconstructive memory processes that are distinct frommotivated self-enhancement or self-presentation. While exaggerations sometimes reflect these motives, they also result from attempts toreconstruct one’s past. Three studies examined test scores as they became less accessible in memory. Study 1 provided a real-world illus-tration, demonstrating reduced accessibility and increased exaggeration of SAT scores over time. Two experiments utilized test scoresrandomly assigned in a controlled laboratory setting. Increased exaggeration was observed following distraction (Study 2), and aftera one-week delay (Study 3). Distortions in scores reported were consistent with beliefs about the self, rather than uniformly self-serving.Under reduced accessibility, exaggeration was predicted by beliefs about achievement (Study 1) and subjective perceptions of test per-formance (Study 2). Study 3 manipulated perceived performance. Positive performance feedback caused greater exaggeration underreduced accessibility, whereas negative feedback reduced the tendency to exaggerate.� 2007 Elsevier Inc. All rights reserved.

Keywords: Reconstructive memory; Self-report bias; Self-enhancement

Unlike physical objects, beliefs and memories about theself have a curious tendency to drift upward. Individualsroutinely inflate or elevate their current standing whenevaluating themselves on important attributes. In part, thisupward movement in evaluations of the current self reflectsthe motive to enhance and maintain the positivity of theself-concept (Greenwald, 1980; Sedikides & Strube, 1997).Although it distorts the self, the enhancement motive doesappear to help individuals maintain a healthy sense of opti-mism and persistence in the face of adversity (Taylor &Brown, 1988; cf., Colvin & Block, 1994). Memory for thepast self is often similarly distorted. Ross (1989) has iden-tified implicit theories that aid attempts to reconstructone’s past, but that also appear to contribute to distortionsin autobiographical memory. When there is doubt aboutthe specifics of one’s past, there is a tendency to form loftyrecollections of those specifics (Greenwald, 1980).

0022-1031/$ - see front matter � 2007 Elsevier Inc. All rights reserved.

doi:10.1016/j.jesp.2007.04.012

* Corresponding author. Fax: +1 617 373 8714.E-mail address: [email protected] (G. Willard).

Research on self-evaluation has tended to focus on rel-atively abstract traits and attributes. For example, a sta-tistically impossible majority of individuals ratethemselves as above average on positive and desirableattributes such as ‘‘well-adjusted’’ and ‘‘agreeable’’(Alicke, 1985; Taylor & Brown, 1988). Although there isclear merit to this approach, it is difficult or impossiblefor researchers to verify each participant’s responses.Thus, interpretation of such findings is obscured by aninability to distinguish accurate self-evaluations frominaccurate and overly positive ones (Colvin & Block,1994). Research on autobiographical memory also tendsto focus on participants’ ratings on abstract terms suchas ‘‘pleasant’’ (Walker, Skowronski, & Thompson,2003). Again, this approach yields intriguing findings,including the tendency for events to be perceived as morepleasant over time (Ritchie, Skowronski, & Wood, 2006).However, the inability to access an objective reality makesit difficult to determine whether the bias rests in the initialratings or in their recollection.

mailto:[email protected]

G. Willard, R.H. Gramzow / Journal of Experimental Social Psychology 44 (2008) 246–259 247

Exaggerating specific performance outcomes

An alternative approach is to focus on specific, objec-tive, and verifiable information about the self. This makesit possible to examine the psychological mechanisms under-lying biases in self-evaluation and autobiographical mem-ory, while having direct access to an objectively definedreality. In contrast to ambiguous attribute ratings thatare difficult or impossible to verify (e.g., ‘‘intelligence,’’‘‘generosity’’), it is possible to directly examine the accu-racy of specific self-reports relevant to the attributes ofinterest (e.g., IQ scores, charitable contributions). This isthe approach that we adopt in the current research, as wellas in our previously reported research on exaggeration(Gramzow, Elliot, Asher, & McGregor, 2003; Gramzow& Willard, 2006).

Exaggerating current performance

One primary purpose of our past research was todemonstrate that distinct psychological processes con-tribute to the tendency to exaggerate specific perfor-mance outcomes (Gramzow & Willard, 2006).Specifically, we argued that motivated self-enhancementunderlies the tendency to exaggerate attributes thatreflect current goal pursuits of emotional and motiva-tional significance. We focused on recent college coursegrades and current grade point average (GPA) as rele-vant performance outcomes, and confirmed their self-evaluative, emotional, and motivational significance forstudents currently in college. Consistent with the moti-vated self-enhancement account, global measures of dis-positional self-enhancement predicted the specifictendency to exaggerate course grades and GPA. In amore direct test of the motivated self-enhancementaccount, we then demonstrated that a self-affirmationmanipulation known to temporarily satisfy the self-enhancement motive (e.g., Fein & Spencer, 1997; Gram-zow & Gaertner, 2005) reduced the tendency for stu-dents high in dispositional self-enhancement toexaggerate their GPAs. Therefore, the tendency to exag-gerate specific performance outcomes can, at times,reflect a motivated self-enhancement process.

In subsequent research, we examined the motivationaland performance correlates of GPA exaggeration. Acrossseveral studies, exaggerating one’s performance in collegewas associated with positive emotions, an adaptive motiva-tional orientation, and a pattern of physiological andbehavioral responses suggesting greater composure understress. Moreover, the initial tendency to exaggerate one’sperformance predicted better longitudinal performanceoutcomes across a variety of future college courses (Gram-zow, Willard, & Mendes, 2007; Willard & Gramzow, 2007).Therefore, in this specific context, self-enhancementappeared to be coordinated with positive and functionalmotivational, emotional, and performance correlates (Tay-lor & Brown, 1988).

Exaggerating past performance

Consistent with the notion that distinct psychologicalprocesses contribute to the tendency to exaggerate spe-cific information about the self, not all exaggerationsappear to be driven by self-enhancement concerns. Spe-cific information from one’s past reflecting completedor abandoned goal pursuits becomes less relevant overtime and comes to have an attenuated motivationaland emotional impact. We confirmed that, among stu-dents currently in college, scores on the Scholastic Apti-tude Test (SAT) represent one such goal pursuit. Thetendency to exaggerate past SAT scores was not pre-dicted by dispositional self-enhancement measures, norwas it affected by a self-affirmation manipulation (Gram-zow & Willard, 2006).

Nevertheless, students on average did exaggerate theirSAT scores. So, what process underlies this tendency?One clue to the underlying mechanism was that the mag-nitude of SAT exaggeration was greater for upper-levelstudents than for first-year students (an effect that wasnot influenced by the self-affirmation manipulation). Wereasoned that the specific value of the SAT score shouldbe more accessible in memory among first-year students,because they would have received their SAT scores morerecently, and were likely to have shared that score valuewith others and entered it repeatedly on college applica-tions. This would explain why first-year students wereable to report the value with a high degree of accuracy.By contrast, the specific value of the SAT score shouldbecome less accessible over time for most students(Underwood, 1957). Therefore, upper-level studentswould be more likely to rely on reconstructive memorywhen reporting their SAT scores. We speculated thatoverly positive self-reports of SAT performance mayresult from biases emerging during this reconstructivememory process.

The psychological process contributing to exaggerationof past performance outcomes, such as the SAT, appearsto be largely independent of motivated self-enhancementeffects. However, no previous research offers direct evi-dence for the alternative reconstructive memory account.The purpose of the present research is to examine the influ-ence of reconstructive memory on exaggeration. Thus,while distortions of specific information about the selfsometimes reflect motivational processes, we propose thatsystematic distortions of specific information about the selfalso commonly result from attempts to faithfully recon-struct one’s past.

Autobiographical memory, accessibility, and exaggeration

High accessibility: Accuracy or motivated distortion

Memory for specific self-relevant information can varyin cognitive accessibility across and within individuals.When the information is high in cognitive accessibility

248 G. Willard, R.H. Gramzow / Journal of Experimental Social Psychology 44 (2008) 246–259

(because of recent exposure or rehearsal), participantsnaturally will recall that information accurately. Any ten-dency to exaggerate, therefore, must reflect deliberate dis-tortion (e.g., due to self-enhancement or impressionmanagement concerns). And, individuals do purposivelyexaggerate at times. For example, participants admit lyingto interaction partners as an impression managementstrategy (Feldman, Forrest, & Happ, 2002; see also e.g.,Schlenker, 1975; Shepperd, 1993). In the context of amock job interview, 90% of lies told by interviewees werejudged to be strategic attempts at impression manage-ment, with the majority of these being attempts at self-promotion and enhancement (Weiss & Feldman, 2006).It is important to emphasize, however, that a salientexternal audience (i.e., an impression managementmotive) is not a necessary condition to motivate distortedself-reports. Self-enhancement is also a primary intrapsy-chic self-evaluative motive (Sedikides & Strube, 1997).As such, self-reports of specific information are oftenbiased in the direction of one’s goals and desired standingon the attribute in question (Willard & Gramzow, submit-ted for publication).

Low accessibility: Reconstructive memory

Specific self-relevant information is not always high inaccessibility. Accessibility of information naturallydecreases over time, and as a result of distraction andinterference (Underwood, 1957). As we noted earlier,research on autobiographical memory indicates that indi-viduals draw upon personally relevant information (i.e.,implicit theories and beliefs about the self) in order toreconstruct specific details from their past (Ross, 1989).In many cases, this process results in memory distortions.For example, when evaluating the self at a global level,individuals tend to use aspects of the current self toreconstruct the past self. While the current and past selfare somewhat consistent, individuals tend to overlookgradual changes and, thus, to overestimate the consis-tency of their own attitudes and behaviors over time.This overestimation has several implications. For exam-ple, it appears to frequently result in biased social com-parison processes among older individuals, who tend tolook down upon the relative incompetence, recklessness,and lack of motivation of the younger generations (Ross,1989).

There is a great deal of evidence that individuals drawfrom broad, generalized conceptions of the self when eval-uating the self on more specific dimensions (Kihlstrom &Cantor, 1983; Markus, 1977; Ross, 1989). Research onautobiographical memory is consistent with the idea thatindividuals use generalized conceptions of the self in orderto reconstruct more specific details from their past. Forexample, individuals draw upon their global levels of self-esteem in order to remember the tone of evaluative feed-back (Story, 1998) or the positivity of everyday events(Christensen, Wood, & Barrett, 2003). When attempting

to recall specific and objective information about the selfthat has faded in memory, we propose that individualsdraw upon more abstract beliefs or conceptions relevantto that domain (e.g., beliefs about the overall quality oftheir performance).

But, why would reduced accessibility produce self-serv-ing distortions, rather than random errors? Within the con-text of autobiographical memory, there is a great deal ofevidence that the beliefs and knowledge structures likelyto influence reconstructive memory, in general, tend toreflect a self-positivity bias (Greenwald, 1980). At a globallevel, the self typically is strongly associated in memorywith desirable traits, as well as positive affective and evalu-ative concepts (Greenwald & Farnham, 2000). Individualstend to interpret ambiguous feedback in a self-favoringlight, and to ignore and quickly forget negative self-rele-vant information or threatening feedback. Likewise, indi-viduals tend to perceive the self as the focus of events, asresponsible for desired outcomes, as free of responsibilityfor undesired outcomes, and as constantly improving(e.g., Dunning, Heath, & Suls, 2004; Sedikides & Green,2000; Taylor & Brown, 1988; Ross, 1989; Wilson & Ross,2001). We anticipate that the beliefs or conceptions rele-vant to reconstructing specific information about the selfare likely in many instances to be influenced by these posi-tive biases.

In summary, when specific information about the self ishigh in accessibility (e.g., because of recent exposure), inac-curate self-reports reflect motivated self-enhancement orimpression management concerns. In contrast, underreduced accessibility (e.g., following a distraction), thereis a greater reliance on reconstructive memory, and self-reports of a specific attribute are increasingly influencedby more global beliefs. Both processes contribute to exag-gerated self-reports of performance outcomes, but throughvery different mechanisms.

Overview of the present research

In the present research, we focus on self-reports of testscores as a context within which to examine the tendencyto exaggerate specific information about the self. The firststudy examined the well-established tendency for college stu-dents to exaggerate their SAT scores. Although college stu-dents clearly tend to exaggerate their SAT scores, it isunclear from past research to what extent these misreportedscores reflect errors in reconstructive memory versus impres-sion management strategies stemming from self-enhance-ment concerns (Bahrick, Hall, & Dunlosky, 1993;Shepperd, 1993). Findings from our own past research wereinconsistent with a motivated self-enhancement account, butprovided no direct support for the reconstructive memoryaccount (Gramzow & Willard, 2006). The purpose of Study1 was to examine directly the role of reconstructive memoryin the tendency to exaggerate SAT scores.

Although SAT scores have inherent psychologicalmeaning because of their real-world significance, there

1 The relationship between memory accessibility and time is typicallycurvilinear. While accessibility tends to decline over time, it plateaus aftera certain period, a phenomenon known as the ‘‘Ebbinghaus curve’’(Underwood, 1957). Consistent with a reconstructive memory account,accuracy when reporting SAT scores tends to decline for about two years,and then to stabilize (see also Bahrick et al., 1993; Gramzow & Willard,2006). Thus, we compare students who had recently learned their SATscores (first-year college students) with those who had learned their SATscores two or more years earlier (upper-level college students).

2 The use of difference scores has been criticized on the grounds thatanalyses are sometimes confounded by one of the difference scorecomponents (Cronbach & Furby, 1970). In the present case, the SATexaggeration index is potentially confounded by actual SAT performance.However, controlling for actual SAT score values did not alter any of theresults reported.


are obvious methodological disadvantages to relying onexisting test scores. The purpose of the subsequent twostudies was to examine the proposed role of reconstructivememory in the tendency to exaggerate test scores, in anexperimental context. Participants in Studies 2 and 3 tooka novel test and received a (randomly assigned) test score.Study 2 manipulated the accessibility of the test score witha distractor task. We predicted that exaggeration wouldincrease under reduced accessibility, and that more generalbeliefs about performance on the test would predict thistendency to exaggerate under reduced accessibility. Study3 manipulated beliefs about performance on the test. Wepredicted that participants who were led to believe thatthey had done well on the test would show a heightenedtendency to exaggerate, that participants led to believe thatthey had performed poorly would show a reduced tendencyto exaggerate, and that these effects would intensify withdecreasing accessibility.

Study 1: SAT exaggeration

Study 1 provides a direct test of the proposed role ofreconstructive memory in the tendency for college studentsto exaggerate their SAT scores. First, as an index of acces-sibility in memory, we measured the amount of time thatstudents took to report their SAT scores (see e.g., Fazio,1986). Consistent with the notion that SAT scores wouldbecome less accessible over time, we predicted that upper-level students would have longer response times (RTs)when reporting their scores than would first-year students.Second, we examined the accuracy of participants’ self-reported SAT scores. Consistent with the notion that a pre-valent default tendency to perceive the self positively wouldinfluence reconstructive memory under reduced accessibil-ity, we predicted that upper-level students would tend toexaggerate their SAT scores more than first-year students(Gramzow & Willard, 2006). Finally, we administered theAchievement scale from the Personality Research Form(Jackson, 1984) as an individual-difference measure of gen-eral beliefs about the self and achievement. Consistent withthe notion that more general beliefs about the self influencethe reconstructive memory process under reduced accessi-bility, we predicted that the achievement measure wouldpredict SAT exaggeration among upper-level students(low accessibility) but not among first-year students (highaccessibility).

Method

Participants

Participants were 740 undergraduate students (194male, 546 female) who completed a computerized surveyin partial fulfillment of a psychology course option. Agesranged from 18 to 24 (M = 18.73). Participant age, sex,and ethnicity did not influence the results in any of thethree studies reported. The majority of students were inthe first year (n = 540), however the sample also con-

tained a substantial number of upper-level students(n = 200).1

Self-reported SAT scores

Each participant was asked to report the total score that heor she received on the SAT: ‘‘What was your combined scoreon the math and verbal sections of the SAT? If you are unsure,please provide your best estimate.’’ At the time data were col-lected, possible combined SAT scores ranged from 400 to1600. The computer recorded the amount of time that eachparticipant took to report his or her SAT score. With consent(requested at the end of the survey), we obtained each partici-pant’s actual SAT score from the university registrar.

Achievement scale

We examined participant’s general beliefs about them-selves in the achievement domain using the Achievementscale from the Personality Research Form (Jackson,1984). This measure is comprised of 10 positively phraseditems (e.g., ‘‘I enjoy work more than play’’) and 10 nega-tively phrased items (e.g., ‘‘I really do not enjoy hardwork’’). Participants responded to each item using a scaleranging from 1 (strongly disagree) to 7 (strongly agree).The negatively phrased items were reverse-scored and aver-aged with the positive items to create a single index ofachievement (Cronbach’s a = .80).

Results

SAT exaggeration

Overall, participants’ self-reported SAT scores(M = 1182, SD = 132) were significantly higher than theiractual SAT scores (M = 1161, SD = 137), t(739) = 6.87,p < .001. This pattern can also be expressed in terms of per-centages. Approximately half of the participants were accu-rate when reporting their SAT scores (47%). Of those whowere inaccurate, more than twice as many exaggerated(37% of total sample) than underreported (16%).

To index SAT exaggeration for subsequent analyses, wecomputed a difference score by subtracting each student’sactual SAT score from his or her self-reported SAT score.A value of zero on this index indicates an accurate self-reported score, positive values indicate exaggeration, andnegative values indicate underreporting.2 Values on the

Table 1Study 1: Patterns of SAT score self-reports as a function of class standing

Class SAT exaggerationindex

Percentages

n MD SD Exaggerated(%)

Accurate(%)

Underreported(%)

First-year 540 9.94 51.06 30 56 14Upper-level 200 39.43 85.32 55 24 21


exaggeration index greater than 3 SD from the M weredeemed to be statistical outliers (n = 15). These values wereWinsorized in order to reduce their potential influence: thelowest extreme value was replaced by the nearest accept-able value plus 1% of that value, and each subsequentextreme value was replaced by the previous Winsorizedvalue plus an additional 1% of that value. Consistent withthe significant difference between self-reported and actualSAT scores reported above, the mean of the SAT exagger-ation index following this procedure remained significantlyhigher than zero (MD = 17.92, SD = 63.51), t(739) = 7.67,p < .001.

0

5

10

15

20

25

30

35

40

45

50

55

60

Low HighAchievement

SAT

exag

gera

tion

First-YearUpper-Level

Class Standing

β = .25, p = .015

β = -.01, p = .734

Class standing, accessibility, and exaggeration

SAT score accessibility. Consistent with the predictionthat specific SAT scores would be less accessible forupper-level students, the average amount of time taken toreport SAT scores was significantly longer for upper-levelstudents (M = 12.35 s, SD = 7.11) than for first-year stu-dents (M = 9.78 s, SD = 5.36), t(738) = 5.69, p < .001.3

SAT exaggeration. Table 1 displays mean SAT exagger-ation difference scores as a function of class standing. Con-sistent with previous research, upper-level studentsexaggerated significantly more than did first-year students,t(738) = 5.73, p = .001. The enhanced tendency for upper-level students to exaggerate is also apparent in the patternof percentages displayed in Table 1. The majority of first-year students were accurate, whereas the majority ofupper-level students exaggerated.

Fig. 1. SAT exaggeration as a function of class standing and achievement(Study 1).

Achievement as a predictor of exaggerationWe used multiple regression to examine the prediction

that achievement-related beliefs about the self would pre-dict SAT exaggeration among upper-level students (lowaccessibility), but that this effect would be less evidentamong first-year students (high accessibility). Class stand-ing (0 = first-year, 1 = upper-level) was included as a cate-gorical predictor of SAT exaggeration, achievement wasincluded as a mean-centered continuous predictor, andthe Class Standing · Achievement term was included. Aspredicted, the interaction term in this model was signifi-cant, indicating that the association between achievementand SAT exaggeration was significantly more positive forupper-level students than for first-year students,B = 21.38, SE = 8.92, t(728) = 2.40, p = .017.4 Fig. 1 dis-

3 As with most RT measures, the RT distributions in each study werepositively skewed. Analyses involving RT are based on logarithmictransformations, in order to correct for skewness (Fazio, 1986). We reportraw RT Ms for ease of interpretation.

4 We do not report standardized regression coefficients (bs) forinteraction terms involving categorical variables. These values have anambiguous interpretation and are influenced by the proportion ofparticipants in each category, among other things. Indeed, the value ofb can change simply as a result of switching the dummy-coding procedure.Importantly, the raw regression coefficients and tests of significance arenot similarly affected (Aiken & West, 1991; Cohen, Cohen, West, & Aiken,2003).

plays predicted values of SAT exaggeration at high andlow levels of achievement (i.e., 1 SD above and below themean). Achievement was a positive and significant predic-tor of SAT exaggeration among upper-level students,B = 20.29, SE = 8.33, b = .25, t(728) = 2.44, p = .015, butwas not associated with SAT exaggeration among first-yearstudents, B = �1.08, SE = 3.19, b = �.01, t(728) = �0.34,p = .734.

Discussion

The primary goal of Study 1 was to provide a directtest of the proposed role of reconstructive memory inthe tendency to exaggerate test scores. We focused onthe real-world tendency for college students to exaggeratetheir SAT scores. Compared to first-year students, upper-level students took significantly longer when reportingtheir SAT scores. In addition, whereas the majority offirst-year students (56%) reported their SAT scores accu-rately, the majority of upper-level students (76%) werenot accurate. These findings suggest that SAT scoresbecome less accessible in memory over time. Consistentwith previous research, the pattern of inaccuracy on eitherside of the actual SAT score was not symmetrical. There


was a substantially greater tendency to exaggerate than tounderreport, and this tendency increased among upper-level (low accessibility) students, relative to first-year (highaccessibility) students. This pattern is consistent with thehypothesis that a pervasive tendency to perceive the selfpositively biases the reconstructive memory process, yield-ing exaggerated self-reports under reduced accessibility.

Study 1 also employed an individual-difference measureof beliefs about the self and achievement. As predicted,this measure was a significant positive predictor of exag-geration among upper-level students (low accessibility),but not among first-year students (high accessibility). Thisfinding supports the hypothesis that general beliefs aboutthe self in a given domain (e.g., achievement) influencereconstructive memory for a specific performance out-come within that domain (e.g., SAT scores), as timepasses and the specific information becomes less accessiblein memory.

Although Study 1 provides a clear demonstration of anincreasing pattern of exaggeration under reduced accessi-bility, in a real-world context, the correlational nature ofthe study comes with inherent limitations. For example, itremains possible that the motivation to exaggerate forself-enhancement or self-presentational reasons increasesover time. It is unclear at present why these motiveswould develop a more potent influence over time. Fur-thermore, previous research demonstrated that a self-affir-mation manipulation known to reduce the influence ofmotivated self-enhancement has no impact on patternsof increasing SAT exaggeration over time (Gramzow &Willard, 2006). Nevertheless, it is without question thatfirst-year and upper-level students differ not only in tem-poral distance from the SAT experience, but in otherways that could potentially influence the tendency toexaggerate.

In addition, the use of preexisting test scores has nota-ble disadvantages. A student’s actual SAT score mightreflect the importance that the student places on that formof achievement, which could influence the accessibility ofthat score. There is also the issue of ceiling and flooreffects. Students with high SAT scores are unable to exag-gerate as much as students with lower scores, whereaslow-scoring students are unable to underreport as muchas high-scoring students. While statistically controllingfor actual SAT scores did not alter the pattern of findingsin Study 1 (see Footnote 2), an ideal context in which toexamine our predictions would be one in which actual testperformance did not constrain the magnitude by which agiven participant could exaggerate or underreport his orher score.

Finally, the measure of achievement orientation that weemployed in Study 1 did not assess beliefs about perfor-mance on the SAT, per se. We reasoned that a general mea-sure of beliefs about the self and achievement would guidereconstructive memory for a specific score on an achieve-ment test. A measure (or manipulation) targeting perceivedperformance on the specific test, however, would more

directly address the influence of reconstructive memoryon exaggeration. In order to address these issues, twoexperiments examined recall of test scores randomlyassigned within a controlled laboratory setting. The firstexperiment included a manipulation of accessibility, andthe second included a manipulation of perceivedperformance.

Study 2: Accessibility manipulation

Each participant completed a test that ostensibly mea-sured a novel domain described as ‘‘integrative ability’’and then received a score on the test (in reality, test scoreswere randomly assigned). Study 2 manipulated the accessi-bility of the test score with a distractor task. There werethree main objectives to this study. The first was to preventany association between the actual test score and the exag-geration index. This was accomplished by randomly assign-ing each participant a test score, independent of theirperceived or actual performance. The second was to exper-imentally manipulate the accessibility of the test score. Par-ticipants in the control condition were asked to recall theirscores on the test immediately after receiving them,whereas participants in the distractor condition first com-pleted a distractor task, in order to reduce the accessibilityof the score. We measured the amount of time that eachstudent took to report his or her test score as an index ofaccessibility. The third purpose of Study 2 was to measureperceived performance on the test, and to use this domain-specific belief as a predictor of exaggeration, under rela-tively high (control condition) and low (distractor condi-tion) levels of accessibility.

Study 2 tested two key predictions. The first predictionwas that participants in the distractor condition wouldexaggerate their scores more than control participants.This prediction was based on the hypothesis that partici-pants would tend to perceive ambiguous self-relevantfeedback positively (e.g., Greenwald, 1980; Taylor &Brown, 1988), and that this default self-positivity biaswould influence reconstructive memory under reducedaccessibility (i.e., in the distractor condition). The secondprediction was that perceived test performance would be apredictor of exaggeration in the distractor condition, butnot in the control condition. This prediction was basedon the hypothesis that individual differences in perceivedperformance would influence reconstructive memory forthe specific performance outcome under reducedaccessibility.

Method

Participants

Participants were 151 undergraduate students who com-pleted the experiment in partial fulfillment of a psychologycourse option. Between one and four students participatedin each session (seated at individual cubicles). Participantswho spent less than half the available time on the test


(n = 19) were removed from analyses, resulting in a finalvalid sample of 132 (61 male, 71 female, ages 18–25).5

Procedure

Novel test. Participants completed a bogus ability testentitled the Integrative Ability Inventory (IAI). The IAIwas administered in order to provide a plausible basisfor giving each participant a specific test score. To intro-duce the IAI as a test of ability in a novel domain, weprovided instructions that were adapted from the (no-longer-administered) logical reasoning portion of theGraduate Record Examination (GRE: Educational Test-ing Service, 1998):

The following items measure integrative ability. Noknowledge of formal logic or mathematics is requiredfor solving these problems. Integrative ability refers toyour general capability and capacity to understandstructured sets of relationships, deduce new informa-tion, analyze and evaluate arguments, draw sound infer-ences, and identify plausible causal explanations.

By suggesting that the IAI was unrelated to mathemat-ics and logic ability, participants presumably would havelimited preconceptions about their own abilities. Eachitem provided participants with a logical argument or sit-uation and set of conditions. Participants responded toeach question (e.g., ‘‘Which of the following, if true, moststrengthens the above argument?’’) by selecting one of fivemultiple-choice responses, labeled ‘‘a’’ through ‘‘e.’’ Inorder to make actual performance on the IAI as ambigu-ous as possible, the test consisted of 10 easily solubleitems, with solution norms at or above .85 (i.e., at least85% of former test-takers chose the credited response)and 10 difficult items, with solution norms at or below.25. Easy and difficult items were administered in alternat-ing pairs.

Each participant was given a maximum of 25 min tocomplete the IAI. On average, participants spent over20 min working on the test (M = 22.83; SD = 3.98),with 47% (n = 62) failing to complete the entire test inthe allotted time. Following the test, a progress barwas displayed for 7 s as participants waited, ostensibly,for the computer to calculate their score. In reality,each participant was randomly assigned a score rangingfrom 603 to 723. The score was centered on the com-puter monitor in large type, and participants lookedat their scores for as long as they wished before press-ing a key to continue.

5 None of these participants performed significantly above chance on thetest (i.e., 20% correct). This suggests that they guessed, rather thanengaged in their performance. We excluded these participants primarilybecause we considered the score unlikely to hold the same psychologicalmeaning (i.e., that the score was a valid indicator of their ability andperformance). Similar results in each experiment were obtained fromanalyses retaining these participants.

Distractor task manipulation. After receiving a test score,each participant was randomly assigned to either the dis-tractor condition or the control condition. Participants inthe control condition proceeded immediately to the sectionin which they were asked to report their scores on the IAI(see below). Participants in the distractor condition firstcompleted a lexical decision task in which they indicatedwhether or not strings of letters appearing on the monitorformed actual words. This task took about 10 min tocomplete.

Self-reported IAI scores. Immediately after participantsreceived their scores (control condition) or after completingthe lexical decision task (distractor condition), the com-puter prompted each participant to indicate the exact scorethat he or she received on the IAI: ‘‘What was your scoreon the Integrative Ability Inventory? (possible scores rangefrom 100 to 900).’’

Post-test measures. Finally, participants responded topost-test measures of confidence, importance, and per-ceived performance, in counterbalanced order: ‘‘How con-fident are you that the score you indicated is accurate?,’’using a scale ranging from 1 (not at all confident) to 10(extremely confident); ‘‘How important was it for you todo well on the IAI?,’’ using a scale from 1 (not at all) to7 (very important); ‘‘Overall, how well do you think youdid on the IAI?,’’ using a scale from 1 (very poorly) to 7(very well); and, ‘‘Based on your score, in what percentilewould you estimate your Integrative Ability to be relativeto other students your own age and gender?,’’ using a scalefrom 1 (better than 10%) to 10 (better than 100%). The lat-ter two responses were standardized and averaged to createan index of perceived performance (Cronbach’s a = .59).

Results

IAI exaggeration

Overall, participants’ self-reported IAI scores(M = 675.49, SD = 59.87) were significantly higher thanthe scores they actually received (M = 663.57,SD = 50.88), t(131) = 4.62, p < .001. This pattern can alsobe expressed in terms of percentages. Although the major-ity of participants were accurate (66%), there was a cleartendency among those who were inaccurate to exaggerate(28% of total sample) rather than to underreport (6%).

To index the accuracy of self-reported IAI scores, wecreated a difference score by subtracting the score that eachparticipant was randomly assigned from his or her self-reported score. A value of zero indicates that the reportedscore was accurate, positive values indicate exaggeration,and negative values indicate underreporting. As in Study1, values greater than 3 SD from the M on this measure(n = 5) were considered to be statistical outliers and Wins-orized in order to reduce their potential influence. Consis-tent with the significant difference between self-reportedand actual IAI scores reported above, the mean of theIAI exaggeration index following this procedure remained

10

15

20

25

30

IAI e

xagg

erat

ion

ControlDistractor

Condition

β = .33, p = .006

Table 2Study 2: Patterns of IAI score self-reports as a function of distractorcondition

Condition IAI exaggerationindex

Percentages

n MD SD Exaggerated(%)

Accurate(%)

Underreported(%)

Control 66 5.74 17.88 14 83 3Distractor 66 16.96 31.63 42 49 9


significantly higher than zero (MD = 11.35, SD = 26.21),t(131) = 4.98, p < .001.6

Distractor manipulation

Manipulation checks. Participants took significantlylonger to report their scores in the distractor condition(M = 11.21 s, SD = 5.16) than did participants in the con-trol condition (M = 8.16 s, SD = 2.81), t(130) = 4.54,p < .001. Participants in the distractor condition alsoreported significantly less confidence in the accuracy oftheir self-reported IAI scores (M = 6.15, SD = 2.90) rela-tive to participants in the control condition (M = 7.14,SD = 2.49), t(130) = 2.09, p = .039. These results suggestthat the distractor manipulation was effective in reducingthe accessibility of the specific test score values.

IAI exaggeration. Table 2 displays the mean IAI exag-geration difference scores in each experimental condition.As predicted, the degree of exaggeration was significantlygreater in the distractor condition than in the control con-dition, t(130) = 2.51, p = .013, indicating that the interven-ing distraction increased the tendency to exaggerate. Theincrease in exaggeration within the distractor condition isclearly apparent in the pattern of percentages displayedin Table 2. The vast majority of students in the control con-dition were accurate (83%). Among the control partici-pants who were inaccurate, there was a modest tendencyto provide exaggerated IAI scores rather than to underre-port. By contrast, the majority of participants in the dis-tractor condition (51%) provided inaccurate IAI scores.Moreover, the pattern of inaccuracy among participantsin this condition was heavily skewed, with four times asmany exaggerating as underreporting.

Participants in the control and distractor conditions didnot differ in the amount of time they spent looking at theirIAI scores, t(130) = 0.84, p = .404. There was also no differ-ence in the amount of time spent on the test, t(130) = 0.04,p = .966, nor was there a difference between conditions interms of the number of IAI questions answered correctly,t(130) = 0.07, p = .947. Thus, neither length of exposure tothe score, nor effort, nor actual performance on the test couldaccount for the difference in levels of exaggeration betweenthe two conditions. These patterns were to be expected, ofcourse, because these variables were measured prior to therandom assignment to condition. Participants in the distrac-tor and control conditions also did not significantly differ onthe post-test measures of perceived IAI test performance,t(130) = 1.22, p = .226, estimated percentile relative to oth-ers, t(130) = 0.87, p = .387, or importance, t(130) = 1.27,p = .205.

6 IAI exaggeration was not associated with the actual numerical valuesof the (randomly assigned) IAI scores in either experiment. Therefore,there was no need to control for the actual scores received on the IAI insubsequent analyses. Additionally, the IAI exaggeration index wasunrelated to the amount of time participants spent taking the test orlooking at their score, number of items attempted, or number of itemscorrect, or (all rs < j.05j).

Perceived performance as a predictor of exaggeration

We used multiple regression to examine the predictionthat perceived performance on the IAI would be associatedwith IAI exaggeration among participants in the distractorcondition (low accessibility), but that this effect would notexist among participants in the control condition (highaccessibility). Distractor task condition (0 = control,1 = distractor) was included as a categorical predictor ofIAI exaggeration, perceived performance on the IAI wasincluded as a mean-centered continuous predictor, andthe Distractor Task · Perceived Performance term was alsoincluded. As predicted, the interaction term in this modelwas significant, B = 10.70, SE = 4.41, t(128) = 2.43,p = .017, indicating that the association between perceivedperformance and IAI exaggeration was significantly morepositive among participants in the distractor conditionthan among control participants. Fig. 2 displays the pre-dicted values of IAI exaggeration at high and low levelsof perceived performance (i.e., at 1 SD above and belowthe mean). Perceived performance was a positive and sig-nificant predictor of IAI exaggeration in the distractor con-dition, B = 8.79, SE = 3.09, b = .33, t(128) = 2.82,p = .006, but was not associated with IAI exaggeration inthe control condition, B = �1.97, SE = 3.14, b = �.08,t(128) = �0.63, p = .533.

0

5

Low

Perceived Performance

β = -.08, p = .533

High

Fig. 2. IAI exaggeration as a function of distractor condition andperceived performance (Study 2).


Discussion

On average, participants in Study 2 exaggerated the ran-domly assigned scores that they received on a novel test. Amanipulation that reduced the accessibility of the specifictest score (as indicated by longer RTs and reduced confi-dence) increased the tendency to exaggerate. Specifically,participants who completed a distractor task after receivingtheir scores (low accessibility) were more likely to exagger-ate than those who reported their score immediately afterreceiving it (high accessibility). Reduced accessibility,therefore, resulted in a systematic pattern of distortionwhen individuals attempt to recall specific self-evaluativeinformation, rather than a balanced pattern of under-and over-reporting. Combined with results from Study 1,these findings provide additional support for the hypothe-sis that a default tendency to perceive the self (and one’sperformance) positively biases the reconstructive memoryprocess, yielding exaggerated self-reports under reducedaccessibility.

Results from subjective perceptions of test performancefurther support a reconstructive memory account. Perceivedperformance was a significant positive predictor of exagger-ation, but only under reduced accessibility (i.e., in the dis-tractor condition). These results suggest that participantsin the distractor condition who felt that they had performedwell on the test tended to reconstruct higher scores, resultingin a greater overall tendency to exaggerate in that condition.Taken together, these findings suggest that the distractortask caused a reduction in the accessibility of the specific testscore, and that memory reconstruction of that specific scoreunder reduced accessibility was guided by subjective percep-tions of performance on the test.

We observed in Study 1 that the accessibility and accu-racy of SAT scores decreased over time, with substantiallygreater increases in exaggeration than in underreporting.As discussed previously, Study 1 was correlational, andmethodological limitations allow for alternative interpreta-tions of the observed pattern of results. Study 2 reduced thepotential for such confounds by randomly assigning scoresto participants, and by experimentally manipulating acces-sibility. However, it is possible that the manipulationemployed in Study 2 had other, unintended effects. Forexample, completing the distractor task may have affectedself-control as a result of ego-depletion (Muraven & Bau-meister, 2000). If so, participants in this condition maynot have been able to restrain their motivation to self-enhance. Alternatively, this manipulation may haveelevated concerns about performance, thereby alteringparticipants’ self-evaluative motives (e.g., by increasingthe motivation to enhance or protect the self). It is note-worthy, therefore, that participants in the two conditionsdid not differ on subsequent measures that would havebeen influenced by self-enhancement or self-protectionmotives, such as perceived test performance, ratings ofperformance relative to others, and importance ascribedto the test. Nevertheless, it remains to be demonstrated that

reduced accessibility as a function solely of a time delay(such as that observed in Study 1) increases the tendencyto exaggerate scores on a novel test.

We designed Study 2 in order to render performance onthe IAI as ambiguous as possible. The test consisted of anequal proportion of easy and difficult items. More impor-tantly, we provided participants with no information aboutthe performance implications of their test scores. Thismethod allowed us to control for individual differences inscores, while examining individual differences in perceivedperformance as a predictor of distortions in self-reportsunder reduced accessibility. However, such vacuous feed-back probably is uncommon outside the laboratory. In mostcases, information is available about the quality of perfor-mance represented by a given test score (i.e., a percentileranking relative to others). The majority of participants inStudy 2 (71%) inferred that their test performance was aboveaverage, despite the ambiguity of their performance and thescore. Regression analyses indicated that exaggeration underreduced accessibility was guided by the subjective percep-tions of test performance. It is possible that this finding couldbe attributable to a motivated self-report bias (i.e., if respon-dents who exaggerated their objective numerical scores alsoexaggerated their subjective performance on the self-reportmeasures). However, there is no explanation for why thisrelationship differed as a function of the accessibility of thescore. Nevertheless, it remains to be demonstrated defini-tively that exaggerated self-reports of specific tests scoresunder reduced accessibility are caused by more generalbeliefs about test performance.

Study 3: Perceived performance manipulation

The purpose of Study 3 was to provide further supportfor a reconstructive memory account for exaggerationunder reduced accessibility by examining the effects of per-ceived performance on distortions in self-reported testscores. Whereas Study 2 provided participants with nofeedback about their scores, Study 3 removed the ambigu-ity surrounding the meaning of the test score that each par-ticipant received by providing explicit performancefeedback. We examined memory for this test score at rela-tively high and low levels of accessibility by asking partic-ipants to recall the scores they received at the end of thelaboratory session (Time 1) and again one week later (Time2). Thus, we sought to extend previous findings by replicat-ing the pattern of increased exaggeration over time(observed for SAT scores in Study 1), while controllingfor individual differences in actual score values (as in Study2). We hypothesized that, as memory for the specific testscore degraded, individuals would increasingly rely on theirsubjective perceptions of performance on the test (whichwere directly manipulated) in order to approximate a spe-cific score value. We tested this hypothesis by manipulatingperformance feedback, using a 3 (Feedback: Positive, No-Feedback, Negative) 2 (Time: Time 1, Time 2) mixed facto-rial design, with time as a within-participants factor. As in

-25-20-15-10

-505

101520253035404550556065

IAI e

xagg

erat

ion

Time 1Time 2


Study 2, each participant completed the IAI and received arandomly assigned score. The same set of score values wasused in all three conditions, allowing us to manipulate eval-uative feedback independently of actual score values. Wepredicted an increasing influence of this manipulation atTime 2, such that distortions in scores reported underreduced accessibility would become increasingly consistentwith the evaluative feedback received.

Method

Participants

Participants were 275 undergraduate students who com-pleted the experiment in partial fulfillment of a psychologycourse option. Participants were excluded from analyses ifthey spent less than half of the allotted time on the IAI(n = 7, see Footnote 5) or failed to respond at Time 2(n = 7). These criteria resulted in a valid sample of 261 par-ticipants (90 male, 171 female, ages 18–22).

Procedure

The procedure was similar to Study 2. Each participantwas randomly assigned to one of three feedback conditions(described below). Each participant took the IAI, receiveda randomly assigned score, and completed a distractortask. Each participant was then asked to report the scorethat he or she received on the IAI, and to complete thesame post-measures as in Study 2 (confidence, perceivedperformance, and importance). Each participant was con-tacted one week following the experiment and asked torecall his or her IAI score. Each participant againresponded to questions assessing confidence in the accu-racy of his or her self-reported score, perceived perfor-mance, and importance.

Feedback manipulation. Evaluative feedback was manip-ulated as each participant received his or her IAI score. Thethree feedback conditions did not differ in the actual scorevalues assigned to participants. Each participant in the no-feedback condition was shown a numerical score with noadditional information (as in Study 2). The remaining par-ticipants were given additional evaluative feedback in theform of a percentile ranking. We did not provide these par-ticipants with a specific percentile because we were con-cerned that it could become confused with the test scoreitself. Rather, a scale from 0 to 100 was displayed on thescreen, with labels ranging from ‘‘poor’’ to ‘‘excellent.’’The numeric score was placed in a box above this scale.In the positive feedback condition, an arrow pointed fromthe participant’s score to a point just above the 80th per-centile, next to the ‘‘excellent’’ label. In the negative feed-back condition, this arrow pointed near the 50thpercentile, just below the ‘‘average’’ label.7

7 Half of the participants in the positive and negative feedbackconditions viewed their percentile ranking (the feedback manipulation)before viewing their numerical score, whereas the other half first viewedtheir score. This counterbalancing had no effect on any of the results.

Results

IAI exaggeration indexes

Two difference scores were computed to index exagger-ation: one based on the self-reported test score collectedduring the experiment (Time 1) and one from the follow-up contact (Time 2). As in the Studies 1 and 2, valuesgreater than three SD from the M at Time 1 (n = 5) orTime 2 (n = 6) were considered to be statistical outliersand Winsorized in order to reduce their influence.

Accessibility and manipulation checks

Accessibility. Consistent with the prediction that IAIscores would become less accessible over time, participantsreported significantly less confidence in the accuracy oftheir self-reported IAI scores at Time 2 (M = 5.92,SD = 2.63) than at Time 1 (M = 6.38, SD = 2.80),t(260) = 2.55, p = .011. We did not measure RTs at Time 2.

Feedback manipulation. As predicted, the feedbackmanipulation had a significant impact on perceived perfor-mance, F(2,258) = 33.05, p < .001. Pairwise contrasts indi-cated that participants in the positive feedback condition(M = 4.73, SD = 1.26) perceived that they had done betteron the test than did participants in the no-feedback condi-tion (M = 3.94, SD = 1.38), F(1, 258) = 12.12, p = .001.Likewise, participants in the negative feedback condition(M = 3.25, SD = 1.34) reported perceiving that they haddone worse than did participants in the no-feedback condi-tion, F(1, 258) = 9.28, p = .003.

Performance feedback and exaggeration over time

Mean levels of IAI exaggeration as a function of feed-back condition and time are displayed in Fig. 3. The centralhypothesis guiding Study 3 was that distortions in self-reported test scores would be consistent with perceived per-formance on that test (i.e., explicit performance feedback),and that these distortions would increase as the accessibil-ity of the specific test score declined over time. We testedthis prediction with a 3 (Feedback: Positive, No-Feedback,

-35-30

Positive Feedback Negative FeedbackNo Feedback

Fig. 3. IAI exaggeration as a function of time and feedback condition(Study 3).


Negative) · 2 (Time: Time 1, Time 2) mixed ANOVA, withtime as a within-participants factor. The main effect fortime was not significant, F < 1. The main effect for feed-back condition was significant, F(2,258) = 30.14,p < .001. As predicted, this feedback effect was qualifiedby a significant Time · Feedback interaction,F(2, 258) = 28.17, p < .001. We examined the effects of thefeedback manipulation separately within each time periodin order to assist interpretation of the pattern contributingto the Time · Feedback interaction.

Time 1. The feedback manipulation did not have a sig-nificant effect on IAI exaggeration at Time 1,F(2, 258) = 2.28, p = .105. Descriptively, there was a ten-dency for participants who received positive feedback(MD = 21.31, SD = 45.12) to exaggerate more than controlparticipants (MD = 15.04, SD = 25.86), who in turn exag-gerated more than participants who received negative feed-back (MD = 10.41, SD = 32.37). Pairwise contrastsindicated that the positive and negative feedback condi-tions differed significantly, F(1, 258) = 4.53, p = .034,although neither differed from the control condition(Fs < 1). This trend is also apparent in the percentages dis-played in Table 3. Additionally, there was a moderate inci-dence of underreporting in the negative feedbackcondition, whereas there was little or no underreportingin the positive and no-feedback conditions.

Time 2. By Time 2, the relatively weak patterns sug-gested at Time 1 were more pronounced. The effect of feed-back on IAI exaggeration was highly significant,F(2, 258) = 36.90, p < .001. Pairwise contrasts indicatedthat participants in the positive feedback condition(MD = 57.15, SD = 68.25) exaggerated significantly morethan participants in the no-feedback condition(MD = 23.80, SD = 37.06), F(1, 258) = 7.616, p = .006.Likewise, participants in the negative feedback condition(MD = �26.02, SD = 83.75) exaggerated significantly lessthan participants in the no feedback condition,F(1, 258) = 16.85, p < .001. Indeed, by Time 2, the meanIAI difference score within the negative feedback conditionwas significantly less than zero, t(103) = �3.17, p = .002.

The change in exaggeration over time is clearly apparentin the pattern of percentages across conditions (Table 3).At Time 2, the percentage of participants who were accu-

Table 3Study 3: Patterns of IAI score self-reports as a function of feedbackcondition and time

Feedback Time 1

Exaggerated (%) Accurate (%) Underreported (%)

Positive 37 60 3None 34 66 0Negative 38 48 14

Time 2

Positive 69 22 9None 58 28 14Negative 38 21 41

rate was low and relatively equivalent across the three con-ditions (ranging from 21% to 28%). However, the patternof inaccuracy was directly in line with our predictions. Par-ticipants in the positive feedback condition showed thegreatest asymmetry in their pattern of inaccuracy, with amajority of these participants exaggerating and very fewunderreporting. The control condition is marked by a sim-ilar, but weaker, pattern of increasing exaggeration. Bycontrast, there was no longer a systematic tendency toexaggerate among the negative feedback participants; thepattern of inaccuracy was fairly symmetrical, with similarpercentages exaggerating and underreporting.

Discussion

Study 3 provides direct evidence in support of thehypothesis that the effects of reduced accessibility on pat-terns of self-reported test scores are driven by more generalbeliefs about performance that guide reconstructive mem-ory. Participants in the ambiguous no-feedback controlcondition tended to exaggerate their scores after complet-ing a distractor task (replicating Study 2), and this ten-dency increased following a more substantial time delay.Combined with results from the first two studies, thesefindings provide additional support for the hypothesis thata default tendency to perceive the self (and one’s perfor-mance) positively biases the reconstructive memory pro-cess, yielding exaggerated self-reports under reducedaccessibility.

The crucial feature of Study 3, however, was the manip-ulation of feedback about the meaning of the scorereceived. Positive feedback about the score increased thetendency to exaggerate under reduced accessibility,whereas negative feedback decreased exaggeration.Although the overall effect of feedback at Time 1 was notsignificant, the specific comparison between positive andnegative feedback was significant. Given that score recallat Time 1 came immediately after a distractor task, thispattern is analogous to the association between perceivedperformance and exaggeration that we observed in the dis-tractor condition in Study 2. After a one-week delay, thepattern was far more dramatic. Participants who receivedpositive feedback were over-reporting their scores bynearly 60 points, whereas participants who received nega-tive feedback were under-reporting by 26 points. It is clear,then, that participants relied upon their more generalbeliefs about the quality of their test performance whenattempting to recall the specific score value under reducedaccessibility.

It should be noted that some participants in Study 3receiving negative feedback did, in fact, exaggerate theirscores (38%), and that this proportion is comparable, forexample, to the proportion of first-year students who exag-gerated their relatively highly accessible SAT scores inStudy 1 (31%). Additionally, unlike the other two condi-tions, the greater tendency to exaggerate than to underre-port in the negative feedback condition at Time 1 cannot


be attributed to reconstructive memory. Thus, similar toStudies 1 and 2, some deliberate or motivated distortionsmust have occurred.

The important point with respect to our hypotheses isthat a deliberate bias account cannot explain the funda-mental pattern of results—decreased exaggeration andgreater underreporting in the negative feedback conditionrelative to the other two conditions, either at Time 1 orat Time 2—whereas a reconstructive memory accountcan. Thus, although some degree of motivated misrep-orting certainly occurred, the findings from Study 3 alsostrongly support a reconstructive memory account. If dis-tortions in self-reported test scores resulted solely frommotivated self-report bias, we would expect greater exag-geration among those receiving negative feedback, whopresumably would be motivated to exaggerate for defensiveor self-presentational reasons, and less exaggeration amongparticipants receiving positive feedback, who believed thattheir actual score is an indicator of good performance.

Participants in the present studies received randomlyassigned scores on a test within a novel domain, and laterreported those scores in private. In contrast to the SAT(where scores are a determinant of a major life outcome,and often are informally interpreted as an index of intelli-gence), participants should be better able discount the testitself, or the domain ostensibly measured. Thus, partici-pants who received negative feedback were likely able tonegate the self-evaluative impact of their ostensibly poorperformance. We believe that there was relatively littlemotivation to distort test scores under these minimal cir-cumstances. Past research suggests that the prevalence ofmotivated distortions would have been much greater hadthe domain been of stronger current relevance or practicalsignificance to participants (Gramzow & Willard, 2006), orif the participants had an audience to potentially impresswhen reporting their scores (e.g., Feldman et al., 2002;Weiss & Feldman, 2006). Future research should explorethese possibilities, and further examine the motivationalaspects of exaggeration.

General discussion

Memory rises—as can be witnessed in the fundamentaltendency for individuals to reconstruct an exaggerated self.The present research demonstrated this phenomenon whilesystematically exploring the underlying mechanisms. Threestudies provided converging evidence for the unique influ-ence of reconstructive memory on the tendency to exagger-ate specific information about the self. The basic patternobserved across the three studies was that reduced accessi-bility of a specific test score was associated with systematicdistortion in self-reports of that score. Although overallinaccuracy increased under reduced accessibility, increasesin exaggeration under most circumstances far exceededany increases in underreporting. This asymmetric patternof distortion held for preexisting scores on an establishedtest of real-world significance (the SAT), as well as for ran-

domly assigned scores on a novel test of ambiguous signif-icance (the IAI). The pattern of increased exaggerationheld regardless of whether accessibility was reduced overa substantial passage of time (Studies 1 and 3) or via a dis-traction manipulated in the laboratory (Study 2). Althoughwe acknowledge that deliberate and motivated misrep-orting also occurs, results from each of the present studiesindicate that reconstructive processes are an independentand substantial source of bias in autobiographical memory.

Sources of distortion in reconstructive memory

It is clear that the reconstructive memory processresulted in a tendency to exaggerate test scores as theybecame less accessible in memory. But, what psychologicalprocesses influenced the reconstructive memory processtoward exaggeration? We hypothesized that, if specificinformation is not readily accessible in memory, individu-als draw upon general beliefs about their standing withinthe domain relevant to the specific information in question.Study 1 demonstrated that beliefs about the self andachievement predicted the tendency to exaggerate SATscores among students for whom the specific SAT scoreswere less accessible. Studies 2 and 3 demonstrated that per-ceived performance on a novel test taken within an exper-iment predicted the tendency to exaggerate scores on thattest among participants for whom the specific test scoreswere less accessible. Study 3 provided the clearest evidencefor the influence of relevant beliefs on reconstructive mem-ory by demonstrating that changes over time in memoryfor test scores were increasingly shaped by evaluative feed-back initially associated with those scores. Following a dis-tractor task, there was a trend suggesting that participantsled to believe that they had performed very well on the testshowed a heightened tendency to exaggerate their testscores, whereas participants led to believed that they hadperformed poorly showed a reduced tendency to exagger-ate and an increased tendency to underreport their scores.Indeed, the specific comparison between the positive andnegative feedback conditions was significant at Time 1.More importantly, these initial tendencies substantiallyincreased following a further reduction in score accessibil-ity over time.

It is important to emphasize that, unlike any previousresearch, these patterns cannot be attributed to the scoresthemselves, because the specific, randomly assigned, valueswere identical across feedback conditions. Moreover, thesepatterns do not appear to be the result of other well-estab-lished sources of bias, such as deliberate misreporting,ignorance, or information neglect. Presumably, partici-pants who believed they did extremely well on the testwould be less inclined to exaggerate for defensive or self-presentational reasons than would participants who didpoorly. Memory distortions resulting from ignorance orinformation neglect also would be characterized by a linkbetween poor performance and exaggeration. We observedthe opposite pattern.


Broader implications and future research

Although the present research focused exclusively onbias in self-reported test scores, we believe that thefindings have broader theoretical and methodologicalimplications for theory and research on self-report bias,self-evaluation, and autobiographical memory. The dis-tinction between motivated distortion and more subtlereconstructive memory processes as sources of biasapplies to many domains of self-evaluation. There arenumerous domains in which people tend to systemati-cally misreport specific self-relevant information (e.g.,prosocial or socially desirable behavior). Future researchshould address the degree to which (or conditions underwhich) biased self-reports in these domains reflect themotivation to distort the self, or result from more subtlebiases in reconstructive memory (see also Gramzow &Willard, 2006).

One issue that the current manuscript did not directlyaddress was the extent to which biased self-evaluation, asrepresented by the tendency to exaggerate, is potentiallyadaptive or maladaptive. As noted in the Introduction,our previous research suggests that exaggeration of attri-butes reflecting current and ongoing goal pursuits isassociated with positive emotion, an adaptive motiva-tional orientation, enhanced performance, and greatercomposure under stress (Gramzow et al., 2007; Willard& Gramzow, 2007). The present studies each includedmeasures of motivation, emotion, self-esteem, and subjec-tive well-being. We found no association between any ofthese measures and SAT exaggeration or IAI exaggera-tion, and, thus, did not include these analyses in the’Results’ sections. This lack of association is consistentwith our model. As we have argued here, and elsewhere,some biases resulting from reconstructive memory pro-cesses have different psychological implications than dothose directly resulting from motivational processes(Gramzow & Willard, 2006). Future research should fur-ther examine the short- and long-term psychologicalimplications of biased self-evaluation resulting fromreconstructive memory distortion.

Several theorists have proposed that the motivation andtendency to view and present the self positively is unique toWestern culture, and have presented evidence suggestingthat individuals in East Asian cultures tend to self-criticize(e.g., Kitayama et al., 1997). Thus, it is possible that thetendency for reconstructive memory to contribute to exag-gerated performance is unique to Western culture as well.In other words, it is questionable whether subtle self-enhancing distortions in reconstructive memory, such asthose observed in the present research, extend to non-Wes-tern cultures. It is entirely possible that the tendency to self-criticize produces self-effacing distortions in reconstructivememory. We are unaware of any research examining cul-tural differences in exaggeration. It is an interesting ques-tion for future research whether broad cultural influences,

such as individualism–collectivism, are influential in recon-structive memory for specific self-evaluative information.

Conclusion

The primary contribution of the present research is inexplicating and demonstrating a subtle and presumablyunintentional process through which specific informationabout the self becomes inflated or elevated. We observedthat the passage of time and memory interference causedby distraction reduce the accessibility of specific self-eval-uative information and, in the absence of explicit negativefeedback, produce distortions in reconstructive memorythat are largely favorable to the self. Although previousresearch on self-evaluation and self-report bias hasfocused primarily on motivated self-enhancement orintentional self-presentational motives, the presentresearch suggest that accounting for accessibility (e.g.,temporal distance) and the context in which informationwas learned (e.g., positive or negative evaluative valence)may be equally important in determining the source ofself-serving biases.

References

Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and

interpreting interactions. Newbury Park, CA: Sage Publications.Alicke, M. D. (1985). Global self-evaluation as determined by the

desirability and controllability of trait adjectives. Journal of Person-

ality and Social Psychology, 49, 1621–1630.Bahrick, H. P., Hall, L. K., & Dunlosky, J. (1993). Reconstructive

processing of memory content for high versus low test scores andgrades. Applied Cognitive Psychology, 7, 1–10.

Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple

regression/correlation analysis for the behavioral science (3rd ed.).Mahwah, NJ: Erlbaum.

Colvin, C. R., & Block, J. (1994). Do positive illusions foster mentalhealth? An examination of the Taylor and Brown formulation.Psychological Bulletin, 116, 3–20.

Christensen, T. C., Wood, J. V., & Barrett, L. F. (2003). Rememberingeveryday experience through the prism of self-esteem. Personality and

Social Psychology Bulletin, 29, 51–62.Cronbach, L. J., & Furby, L. (1970). How should we measure ‘‘change’’:

Or should we? Psychological Bulletin, 74, 68–80.Dunning, D., Heath, C., & Suls, J. M. (2004). Flawed self-assessment:

Implications for health, education, and the workplace. Psychological

Science in the Public Interest, 5, 69–106.Educational Testing Service. (1998). GRE practicing to take the general

test big book. Princeton, NJ: Educational Testing Service.Fazio, R. H. (1986). How do attitudes guide behavior? In R. M.

Sorrentino & E. T. Higgins (Eds.), The handbook of motivation and

cognition: Foundations of social behavior (pp. 204–243). New York:Guilford.

Fein, S., & Spencer, S. J. (1997). Prejudice as self-image maintenance:Affirming the self through derogating others. Journal of Personality

and Social Psychology, 73, 31–44.Feldman, R. S., Forrest, J. A., & Happ, B. R. (2002). Self-presentation

and verbal deception: do self-presenters lie more? Basic and Applied

Social Psychology, 24, 163–170.Gramzow, R. H., Elliot, A. J., Asher, E., & McGregor, H. A. (2003). Self-

evaluation bias and academic performance: Some ways and somereasons why. Journal of Research in Personality, 37, 41–61.


Gramzow, R. H., & Gaertner, L. (2005). Self-esteem and favoritismtoward novel in-groups: The self as an evaluative base. Journal of

Personality and Social Psychology, 88, 801–815.Gramzow, R. H., & Willard, G. (2006). Exaggerating current and past

performance: Motivated self-enhancement versus reconstructive mem-ory. Personality and Social Psychology Bulletin, 32, 1114–1125.

Gramzow, R.H., & Willard, G., & Mendes, W.B. (2007). Big tales andcool heads: Academic exaggeration is related to cardiac vagalreactivity and emotional equanimity, submitted for publication.

Greenwald, A. G. (1980). The totalitarian ego: Fabrication and revision ofpersonal history. American Psychologist, 35, 603–618.

Greenwald, A. G., & Farnham, S. D. (2000). Using the implicitassociation test to measure self-esteem and self-concept. Journal of

Personality and Social Psychology, 79, 1022–1038.Jackson, D. N. (1984). Personality research form manual. Port Huron, MI:

Research Psychologists Press.Kihlstrom, J. F., & Cantor, N. (1983). Mental representations of the self.

Advances in Experimental Social Psychology, 21, 159–187.Kitayama, S., Markus, H. R., Matsumoto, H., & Norasakkunkit, V.

(1997). Individual and collective processes in the construction of theself: Self-enhancement in the United States and self-criticism in Japan.Journal of Personality and Social Psychology, 72, 1245–1267.

Markus, H. R. (1977). Self-schemata and processing information aboutthe self. Journal of Personality and Social Psychology, 35, 63–78.

Muraven, M. R., & Baumeister, R. F. (2000). Self-regulation anddepletion of limited resources: Does self-control resemble a muscle?Psychological Bulletin, 126, 247–259.

Ritchie, T. D., Skowronski, J. J., & Wood, S. E. (2006). Event self-importance, event rehearsal, and the fading affect bias in autobio-graphical memory. Self and Identity, 5, 172–195.

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

Schlenker, B. R. (1975). Self-presentation: Managing the impression ofconsistency when reality interferes with self-enhancement. Journal of

Personality and Social Psychology, 32, 1030–1037.Sedikides, C., & Green, J. D. (2000). On the self protective nature of

inconsistency negativity management: Using the person memoryparadigm to examine self referent memory. Journal of Personality

and Social Psychology, 79, 906–922.Sedikides, C., & Strube, M. J. (1997). Self evaluation: To thine own self be

good, to thine own self be sure, to thine own self be true, and to thine ownself be better. Advances in Experimental Social Psychology, 29, 209–269.

Shepperd, J. A. (1993). Student derogation of the scholastic aptitude test:Biases in perceptions and presentations of college board scores. Basic

and Applied Social Psychology, 14, 455–473.Story, A. L. (1998). Self-esteem and memory for favorable and unfavor-

able personality feedback. Personality and Social Psychology Bulletin,

24, 51–64.Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being: A social

psychological perspective on mental health. Psychological Bulletin,

103, 193–210.Underwood, B. J. (1957). Interference and forgetting. Psychological

Review, 64, 49–60.Walker, W. R., Skowronski, J. J., & Thompson, C. P. (2003). Life is

pleasant—and memory helps to keep it that way! Review of General

Psychology, 7, 203–210.Weiss, B., & Feldman, R. S. (2006). Looking good and lying to do it:

Deception as an impression management strategy in job interviews.Journal of Applied Social Psychology, 36, 1070–1086.

Willard, G. & Gramzow, R.H. (2007). Academic exaggeration: Affective,motivational, and performance correlates, submitted for publication.

Wilson, A. E., & Ross, M. (2001). From chump to champ: People’sappraisals of their earlier and present selves. Journal of Personality and

Social Psychology, 80, 572–584.

Date post:	25-Oct-2016
Category:	Documents
Upload:	greg-willard
View:	212 times
Download:	0 times

Exaggeration in memory: Systematic distortion of self-evaluative information under reduced...

Documents