The Barnum Effect Revisited: Cognitive and Motivational

Journal of Personality and Social Psychology1985, Vol. 49, No. 5, 1378-1391

Copyright 198S by the American Psychological Association, Inc.0022-35I4/85A00.75

The "Barnum Effect" Revisited: Cognitive and MotivationalFactors in the Acceptance of Personality Descriptions

Joel T. Johnson, Lorraine M. Cain, Toni L. Falke, Jon Hayman,and Edward Perillo

University of California at Davis

Four studies examined factors mediating the so-called "Barnum effect"—the ac-ceptance of character descriptions composed of high base-rate personality traits asuniquely applicable to the self. Departing from previous research, all experimentsexamined the effect in a context in which subjects were not deceived into believingthat the descriptions had been based on personal information supplied by them. InStudy 1 subjects who rated the applicability of personality characteristics to them-selves viewed the characteristics as significantly more applicable than subjects askedto assess their applicability to an acquaintance. In Study 2 subjects estimated thatthey displayed both positive behaviors and their negatively valenced opposites morefrequently than an acquaintance displayed them, indicating that the Barnum effectmight be mediated by the greater availability of evidence to confirm traits in theself. In Studies 3 and 4 the perceived accuracy of paragraph-long personality de-scriptions increased with subjects' familiarity with the applicable individuals. Subjectsrated the descriptions as most accurate for themselves, next most accurate for aclose friend, next most accurate for a moderate friend, and least accurate for acasual acquaintance. In both of the latter studies, however, the influence of familiarityon perceived accuracy was largely confined to positively valenced descriptions. Itwas concluded that, in addition to being influenced by a prior belief in the credibilityof a source, the Barnum effect may be mediated by a combination of cognitive andmotivational factors.

Purveyors of anecdotes demonstrating ourgullibility about ourselves have traditionallyencountered no shortage of supporting mate-rial. The debunker of astrological horoscopes,for example, may point to the conviction withwhich a tenacious Gemini maintains that theGemini birth sign description fits him or herexactly, his or her confidence in the goodnessof the fit undeterred by the comment that te-nacity is purportedly more characteristic ofTaurus than of Gemini. The debunker's fur-ther protestations that Gemini is supposedlyboth excitable and verbose, whereas present

The seminal idea for this research originated in a grad-uate seminar in social cognition taught by Shelley E. Taylorat Harvard University. The research was supported by agrant to Joel T. Johnson from the Academic Senate of theUniversity of California at Davis. We express thanks toJerome Drobny and Frank Miller for their assistance invarious phases of the research and to Lorraine FindleyIrish for her useful comments on the manuscript.

Requests for reprints should be sent to Joel T. Johnson,Department of Psychology, University of California, Davis,California 95616.

company, in contrast, is notably laconic andunflappable, may also prove of no avail, forpresent company may counter that he or sheis in fact both excitable and verbose, and fur-ther has the bona fide memories to prove it.Human credulity—or at least the prevalenceof our belief in it—is also demonstrated by therecurrent theme, in literature and folklore, ofthe rewards that accrue the flatterer for detect-ing our finest hidden qualities. If, according tothat folklore, our credulity is great, it is perhapsgreatest when its object is the self.

But are we in fact relatively more gullibleabout ourselves? Since a classic study by Forer(1949), empirical evidence consistent with thefolklore has been garnered in a series of in-genious experiments (e.g., Handelsman &Snyder, 1982; Snyder& Larson, 1972;Snyder,Larsen, & Bloom, 1976; Snyder, Shenkel, &Lowery, 1977). These experiments all assess aphenomenon which, following Meehl (1956),has been termed the "Barnum effect," afterthe man who purportedly stated that a goodcircus had a "little something for everybody."

1378

THE "BARNUM EFFECT" REVISITED 1379

They have consistently demonstrated that in-dividuals who erroneously believe that certainhigh base rate personality profiles were pre-pared specifically for them usually rate themas extremely accurate descriptions of them-selves. Although given the high base rate va-lidity of the descriptions, the assessment oftheir applicability to the self does not in itselfrepresent erroneous judgment, additional ex-periments have shown that subjects may ratesome positively valenced descriptions as moreapplicable to themselves than to "people ingeneral" (Snyder & Shenkel, 1976). Subjectsare therefore "gullible" in that they believe thedescriptions are unique to them, failing to re-alize that they apply equally well to others.

None of the cited experiments, however,have attempted to test directly the propositionthat, holding constant other relevant variables,individuals are more accepting of the validityof high base rate personality descriptions forthemselves than for others. In all of them, ap-plicability to self has been confounded withthe belief that the description was based onspecific information furnished by the recipient.Higher applicability ratings for the self almostcertainly were influenced by the knowledgethat the descriptions were generated specifi-cally for the self, and not for "people in gen-eral."

The above confound relates to an even morefundamental criticism of all Barnum effectstudies—the suggestion that subject acceptanceof the descriptions may be due less to gullibilityconcerning the self than to the demand char-acteristics (Orne, 1962) inherent in the para-digm. Acceptance of high base rate general-izations might be due, at least in part, to sub-jects' beliefs that the researcher expects themto accept the accuracy of character portraitsostensibly prepared specifically for them. Theexistence of this particular confound or de-mand characteristic in no way impugns theexternal validity of the Barnum effect studies,for the paradigm accurately reflects a fre-quently occurring real world condition—thereceipt of high base-rate feedback from an ex-pert, coupled with the belief that the feedbackwas prepared specifically for the self. However,an investigation of the relative acceptance(perceived applicability to self vs. perceivedapplicability to others) of high base-rate de-scriptions without the erroneous belief that the

descriptions were prepared specifically for theself might, we surmised, provide importantinsight into the specific cognitive and moti-vational antecedents of the effect. The four ex-periments described in this article were de-signed to investigate the Barnum effect in justsuch a deception-free context.

Potential Mediation by Motivational Factors

Several studies using the false feedback par-adigm have reported that subjects accept pos-itive descriptions more readily than negativeones (Collins, Dmitruk, & Ranney, 1977;Snyder & Shenkel, 1976). Although some ev-idence suggests that this difference might be atleast partially due to the higher base rates ofpositive traits in the general population (Snyder& Shenkel, 1976), the results are also consistentwith other studies reporting a positivity biasin self-evaluation. Chambliss (1964), for ex-ample, found that subjects rated positive traitsas significantly more applicable to themselvesthan negative traits, and Wylie (1965) foundthat subjects consistently rated themselves as"above average" on certain key dimensions,further noting that subjects' self-evaluationsexceeded evaluations of them by their peers.Numerous other studies report a related "self-serving" bias in attribution (Riess, Rosenfeld,Melburg, & Tedeschi, 1981; Bradley, 1978;Schlenker & Miller, 1977; Taylor & Koivu-maki, 1976).

All of these experiments provide evidencethat positively valenced descriptions might berated as more applicable to the self than toanother. In parallel fashion, a motivational biaswould cause subjects to rate negative descrip-tions as less applicable to themselves. The lat-ter prediction is also consistent with an expla-nation founded on the Freudian defensemechanism of projection, for to the extent thatindividuals "project," they should find negativetraits relatively typical of another while deny-ing them in themselves.

Potential Mediation by Cognitive Factors

Despite the research indicating a positivitybias, a different line of reasoning might leadus to speculate that the Barnum effect mayalso reflect a simple cognitive error in adjust-ment of a judgment—a failure to make suffi-

1380 JOHNSON, CAIN, FALKE, HAYMAN, PERILLO

cient allowance for the greater availability ofself knowledge versus knowledge of others.Simply put, we have a larger memory storeabout ourselves than about any other person,and are consequently relatively more likely tofind available behavioral evidence confirmingany common trait in ourselves. But while useof the "availability heuristic" (Kahneman &Tversky, 1973) to assess the presence of traitsin our own behavior may result in fairly ac-curate judgments, use of this heuristic to assessa trait in another may commonly lead us tounderestimate its prevalence. The cause of in-accuracy may lie in our failure to consider thebase rate of the population of available in-stances from which confirming evidence is tobe drawn, and subsequently to upgrade ourestimate of prevalence or applicability to com-pensate for this relatively scanty data baseconcerning the other. From this perspective, aself-other difference in perceived applicabilityof given traits is another example of failure touse population base rates in deriving judg-ments (Kahneman & Tversky, 1982). In thegeneral sense that it posits an error in the pro-cess of inference that is independent of moti-vational factors, this explanation may betermed a cognitive one.

Is there research evidence consistent withthis cognitive account? Using similar formats,Nisbett, Caputo, Legant, and Maracek (1973)and Goldberg (1981) found that, at least undercertain circumstances, individuals rate com-mon personality traits as less descriptive ofthemselves than of others while making sig-nificantly greater situational attributions forthemselves. Monson, Tanke, and Lund (1980),in contrast, found that when subjects weresimply asked to endorse or reject the appli-cability of personality traits (with no oppor-tunity for situational attributions), they as-cribed a significantly greater number of traitsto themselves than to acquaintances. None ofthese experiments, however, directly investi-gated whether individuals, in estimating therelative frequency of certain common behav-iors in themselves and others, adjust suffi-ciently for the differential availability of con-firming evidence.

An experiment by Funder (1980) providesmore directly relevant evidence. His subjectstended to rate themselves higher than did theirpeers on internal dispositional characteristics

(e.g., "is introspective"), whereas their peerstended to rate them higher on traits especiallyvisible to an external observer (e.g., "is phys-ically attractive"). These results imply that in-dividuals in fact may fail to adjust for theirinsufficient knowledge of others' dispositionaltraits. The further implication is that individ-uals may rate Barnum style descriptions, whichfocus largely on internal states, needs, and as-pirations (e.g., Forer, 1949), as less typical ofacquaintances than of themselves.

Study 1 investigates these implications di-rectly by assessing whether Forer's original listof personality traits is judged as more descrip-tive of the self than of an acquaintance. Studies2,3, and 4 assess more precisely the underlyingquestions of whether the mediation of self-other differences is primarily motivational,primarily cognitive, or best explained by amodel including both cognitive and motiva-tional factors, (e.g., Johnson & Judd, 1983).

Study 1

Method

Subjects. Subjects were 96 undergraduate volunteersfrom an introductory psychology class at the University ofCalifornia, Davis.

Design and procedure. Each subject read a list of 13personality characteristics originally taken by Forer (1949)from a newstand astrology book, and used in numeroussubsequent experiments (e.g., Snyder & Larson, 1972;Snyder et al. 1976). Subject were randomly assigned to oneof two conditions. Subjects in Condition 1 were asked toassess the applicability of the characteristics to themselvesunder the following instructions:

How common are certain personality traits in thegeneral population?

Listed below are a number of different personalitytraits. How many of these traits are characteristicof you?

First consider how accurately each of the traits de-scribes you. Then indicate how accurate the descriptionis by circling the appropriate number on the scale im-mediately below the trait. Please mark all scales. Evenif you believe you do not have enough information tomake a choice, use your best guess.

Subjects in Condition 2 were asked to assess the appli-cability of the same personality traits to an acquaintanceunder the following instructions:

How common are certain personality traits in thegeneral population?

Listed below are a number of different personalitytraits. How many of these traits are characteristic ofsomeone whom you know?

Please think of an acquaintance of yours. By an ac-quaintance we mean someone whom you know, some-


times see, and with whom you sometimes speak, but donot know well enough to consider a close friend. Wewould like you to think of such a person now. Considerhow accurately each of the traits describes that person.Then indicate how accurate the description is by circlingthe appropriate number on the scale immediately belowthe trait. Please mark all scales. Even if you believe youdo not have enough information to make a choice, useyour best guess.

The "best guess" instruction explicitly informs subjectsthat lack of information should not compel a judgmentthat the description is inaccurate. Under a normativemodel, a subject should use his or her own personality asa basis for estimating the prevalence of extremely highbase-rate characteristics in a little known other, and a "bestguess" strategy should involve the basic assumption thatthe unknown other possesses the trait to the same approx-imate degree as the self. We hypothesized, however, thateven subjects explicitly instructed to use their "best guess"still would not guess that the relevant personality traitswere fully replicated in others, and would systematicallyunderestimate the applicability of the descriptions to theiracquaintances.

Experimental materials. The thirteen personalitycharacteristics were drawn verbatim (with slight gram-matical modifications) from Forer (1949), and are the clas-sic Barnum effect statements utilized in numerous otherpublished studies (i.e., Snyder & Larson, 1972).

The trait list used in Condition 1 was changed to thethird-person singular for Condition 2. After assessing theaccuracy of each of the traits on the list, all subjects wererequested to "rate the overall accuracy of this list of char-acteristics as a description of yourself" (Condition 1) or"your acquaintance" (Condition 2). The dependent mea-sure for each accuracy assessment was a nine-point scalewith endpoints marked not at all accurate (1) and extremelyaccurate (9). We believed that these endpoint labels might,in this particular context, be less ambiguous to subjectsthan the totally reject to totally accept labels previouslyused in much of the Barnum effect research.

Results

To assess the significance of differences inself-other accuracy ratings, t tests were per-formed on each of the 13 traits on the list, onthe overall accuracy rating of the list, and onthe mean accuracy rating averaging across the13 traits. Results of the t tests, together withthe means for self-accuracy rating conditionand other-accuracy rating condition, are shownin Table 1. For each of the 13 traits, meanaccuracy ratings for the self were higher thanmean accuracy ratings for the acquaintance.For 7 of the traits the difference was significantat the .05 level or beyond, and for all but twoof the remaining traits the difference was mar-ginally significant at the .10 level. The differ-ence between the overall mean accuracy ratingfor traits in the self-rating condition versus

traits in the other-rating condition (averagingacross the 13 accuracy ratings of each subject)was highly significant, £(88) = 5.21, p< .001.'In addition, subjects in the self-rating conditionviewed the list as a whole as significantly moreaccurate than subjects in the other-rating con-dition, ;(93) = 2.38, p < .02.

Further analyses investigated the extent towhich the self-other difference in judged ac-curacy may have been influenced by motiva-tional factors. A separate group of 23 under-graduate judges subsequently rated each of the13 personality trait descriptions, as well as theoverall list, on a 9-point scale with endpoint 1labeled extremely positive and endpoint 9 la-beled extremely negative. Although the judgesviewed the overall list as slightly positive (M =4.87), they rated 8 of the descriptive statements(numbers 2, 3, 5, 6, 7, 9, 10, and 12 in Table1) as at least slightly on the negative side ofthe neutral point, and only 5 statements(numbers 1, 4, 8, 11, and 13) on the positiveside of the scale. As revealed by Table 1, therewas a significant self-other difference on 4 ofthe 5 positive statements, but on only 3 of the8 negative ones. As also indicated by Table 1,however, separate ; tests performed on themean ratings of both positive and negativestatements revealed a significant self-other dif-ference for each set.

A final 2 X 2 (Person Rated X StatementValence) analysis of variance (ANOVA), inwhich mean rating of positive-negative state-ments was a within-subjects repeated measuresvariable, indicated main effects for both state-ment valence, F( 1, 88) = 51.60, p < .001, andperson rated, F(l, 88) = 28.17, p < .001, butno interaction between these variables, F(l,88) < 1. Although positive statements wererated as more accurate descriptions of bothself and other, both positive and negative state-ments were seen as more applicable to the self.In sum, although a higher proportion of in-dividual positive statements indicated signifi-cant self-other differences, there was strongevidence that individuals, even when expresslyinstructed to adjust for insufficient informa-tion by use of their best guess, systematically

1 Because of missing responses on some dependent mea-sures, data from 6 subjects were discarded from the analysison mean accuracy ratings.


Table 1Mean Accuracy Ratings of Thirteen Personality Traits: Study 1

Personality trait

1. Need to be liked'2. Tendency toward self-criticism11

3. Unused capacity15

4. Compensation for personalityweakness'

5. Sexual adjustment problems'"6. Disciplined outside, insecure inside1"7. Serious self-doubts'"8. Preference for change*9. Pride in independent thinking6

10. Frankness found unwiseb

1 1 . Extraversion and introversion*12. Unrealistic aspirations'"13. Security as major goal"

Overall accuracy rating of listMean accuracy rating of 1 3 statementsMean accuracy rating of 5 positive

statementsMean accuracy ratings of 8 negative

statements

Self-ratingcondition

5.947.005.78

5.593.584.395.736.716.374.986.884.636.205.655.69

6.30

5.31

Other-ratingcondition

4.984.304.28

4.453.173.554.465.615.614.785.113.705.394.724.53

5.06

4.20

t

2.197.443.47

2.99.88

1.732.772.811.75.46

3.471.821.882.385.21

4.80

4.38

df

94*94***94***

94**939493**93"939291***929393*88***

88***

88***

Note. Higher numbers indicate ratings of greater accuracy.* Rated as positive by judges.b Rated as negative by judges.*/?<.05. **p<.01. ***;>< .001.

underestimate the relative prevalence of bothpositive and negative personality traits inothers.

Study 2

Study 2 was designed to examine more pre-cisely the mediators of the self-other distinc-tion in judged applicability of personalitytraits. It used a series of bipolar (positive-neg-ative) trait adjectives to investigate the influ-ence of valence in the context of a different setof stimulus materials. It was also designed toassess directly the cognitive explanation for theself-other distinction—that is, to investigatespecifically whether individuals underestimate,relative to themselves, the number of instancesin which others engage in high base-rate be-haviors.

In examining the latter question, the studyalso attempted to shed light on certain currentissues surrounding trait attribution. Monsonet al. (1980) cogently argue that the greatersituationality ascribed to the self in Nisbett et

al. (1973) might have been the artifact of aresponse format in which subjects were askedeither to select one of two bipolar traits as de-scriptive or to choose the "depends on the sit-uation" option. Subjects who wished to attrib-ute both traits in a bipolar pair to themselveswere in effect coerced into situational attri-butions. When Monson and associates pre-sented the traits separately, giving subjects theoption of endorsing them both, subjects as-cribed a significantly greater number of traitsto themselves than to acquaintances. As notedby Goldberg (1981), however, these results mayhave been partially due to uncertainty—sub-jects may have rejected a trait as descriptive ofanother simply because they lacked definitiveinformation about the target.

Study 2 directly examined whether individ-uals would view high base-rate behaviors asless frequent in others than in themselves whenspecifically instructed to adjust for inadequateinformation by using their best guess. In ad-dition, it investigated the extent to which at-tribution to the self is influenced by trait pos-itivity.


MethodOverview. Subjects were randomly assigned to one of

two rating groups—bipolar or single trait. Bipolar groupsubjects rated both themselves and an acquaintance on aseries of 9-point scales. One endpoint represented a pos-itively valenced trait and the other represented a negativelyvalenced antonym. Single trait group subjects estimatedthe frequency with which both they and an acquaintancedisplayed each of the individual traits which appeared asendpoints on the bipolar scales.

Subjects. Subjects were 69 undergraduate students atthe University of California at Davis. They received extracredit in an introductory psychology course for their par-ticipation.

Experimental materials. Traits used in the study wereselected from an initial group produced by five under-graduate students. The students had been instructed togenerate trait pairs with positive and negative antonyms ateach pole and to list only traits with a high rate of prevalenceamong their peers. Of the original pairs so generated, 24were subsequently rated by a separate group of under-graduate judges as markedly differing from one another invalence.2 These 24 trait pairs were as follows, with the morepositively rated trait appearing as the first in each pair:affectionate-reserved; sophisticated-naive; modest-vain;cautious-impulsive; cheerful-sad; clear-thinking-confused;brave-fearful; decisive-hesitant; enthusiastic-apathetic;friendly-disagreeable; generous-selfish; self-reliant-de-pendent; flexible-stubborn; good-tempered-irritable; op-timistic-pessimistic; active-passive; calm-excitable; re-laxed-tense; punctual-tardy; attentive-distractible; out-going-shy; rugged-delicate; mature-childish; trusting-suspicious.

Design and procedure. Subjects in both groups wereinformed that the experimenters were investigating "per-ceptions of the prevalence of certain personality traits,"and were requested to rate both themselves and an ac-quaintance on the personality scales. An acquaintance wasdefined as "someone whom you know, see occasionally,and speak with occasionally, but do not know well enoughto consider a friend." All subjects were instructed to "useyour best guess" if they did not have enough informationto make a choice.

Bipolar group subjects (n = 34) rated both themselvesand an acquaintance on each of the trait pairs, with thepositive and negative antonym in each pair appearing re-spectively as the positive and negative endpoint on a 9-point scale. Between-subjects conditions were order of pre-sentation of traits (two different random orders) and orderof individual rated (self rated first vs. acquaintance ratedfirst). Subjects were randomly assigned to condition.

Single trait group subjects (n = 35) estimated the fre-quency with which both they and an acquaintance displayedeach of the individual traits. Estimates were made on 9-point scales with endpoints marked "frequently" and "in-frequently." As in the bipolar group, between-subjects con-ditions were order of presentation of trait and order ofindividual rated, with subjects randomly assigned to con-dition.

Results

Because subjects in the bipolar group andthe single trait group used different rating

scales, their data were separately analyzed. Aninitial series of 2 X 2 X 2 (Order of Traits XOrder of Person Rated (Self First/Acquain-tance First) X Person Rated (Self/Acquain-tance) ANOVAS were performed on each of the24 bipolar scales and on each of the 48 singletrait rating-scales. Order of traits and order ofperson rated were between-subjects variables;person rated was a within-subjects repeatedmeasures variable.

Differences in self-other ratings variedmarkedly according to group. Bipolar groupsubjects saw a significant difference betweenthemselves and others on only 2 of the 24 bi-polar scales, viewing themselves as significantlymore toward the hesitant pole of the hesitant-decisive scale, F(\, 30) = 4.69, p < .05, andsignificantly more toward the tense pole of therelaxed-tense scale, F(l, 30) = 8.07, p < .01.Single trait group subjects, in contrast, ratedthemselves as displaying 42 of the 48 traitsmore frequently than their acquaintances. For13 of these traits—flexible, excitable, affec-tionate, modest, brave, apathetic, generous,self-reliant, dependent, tense, attentive, ma-ture, and impulsive—differences were signifi-cant at the .05 level or beyond. Eight of these13 traits had been previously rated as the pos-itive trait in the pair. Subjects rated their ac-quaintances as displaying only one trait, tar-diness, significantly more frequently thanthemselves.

To provide a more meaningful data sum-mary, and in particular to examine the inter-action of the effect for person rated with traitvalence, additional ANOVAS were performedthat collapsed across the ratings of individualtraits. A 2 X 2 X 2 (Order of Traits X Orderof Person Rated X Person Rated) repeatedmeasures ANOVA on the mean trait ratings ofBipolar Group subjects indicated that, al-though both self and acquaintance were ratedcloser to the positive pole of the scale, self wasnot rated more positively than acquaintance,F(l, 30) < 1, (M = 3.79 for self vs. M = 3.86for acquaintance, with lower numbers indi-

2 Nineteen judges separately rated the valence of eachtrait on a 9-point scale ranging from extremely positive (1)to extremely negative (9). Twenty-two of the 24 pairs dif-fered significantly in valence at the .05 level or beyond,and the means of the remaining two trait pairs—rugged-delicate and cautious-impulsive—were in the predicteddirection.


eating more positive ratings). A 2 X 2 X 2 X2 (Order of Traits X Order of Person Rated XPerson Rated X Trait Valence) ANOVA was alsoperformed on the mean ratings of single traitgroup subjects. Order of traits and order ofperson rated were between-subjects variablesin this analysis, while person rated and traitvalence were within-subject repeated measuresvariables. In a result parallel to that of Study1, analysis indicated highly significant maineffects for trait valence, F(l, 31) = 135.55,p < .001, and person rated, F(l, 31) = 34.75,p < .001, but no significant Trait Valence XPerson Rated interaction, F(l, 31) < 1. Sub-jects, in other words, demonstrated an equiv-alent positivity bias in rating both themselves(M = 6.72 for frequency of positive behaviorsvs. M = 4.31 for frequency of negative behav-iors) and others (M = 6.07 for frequency ofpositive behaviors vs. M = 3.91 for frequencyof negative behaviors). Although the rated ab-solute frequency for both positive and negativebehaviors was higher for the self than for theother, the mean relative frequency, or ratio ofpositive behaviors to negative behaviors, wasvirtually identical (1.56 for self vs. 1.55 forother). Neither the main effect for trait valencenor the main effect for person rated interactedsignificantly with either of the between-subjectsvariables related to order of rating. These re-sults are fully consistent with the analysis ofthe bipolar group indicating the absence of agreater positivity bias on self ratings.

Discussion

The results of Study 2 are consistent with apurely cognitive interpretation of self-otherdifferences in acceptance of the applicabilityof prevalent traits in that they indicate that thedifferences may be primarily due to inadequateadjustment for the lesser availability of knowl-edge concerning the other. Although the datacoincide with previous research indicating astrong positivity bias in person evaluation (e.g.,Sears, 1983;Bruner&Taguiri, 1954), they alsoindicate that the bias is equally strong for eval-uations of self and other. The mean ratios offrequency of positive behaviors to frequencyof negative behaviors indicate that subjectsrated both themselves and others as engagingin positive behaviors approximately 61% of thetime, a percentage close to the "golden section"identified in previous research. The results ofAdams-Webber (1979) and Rigdon and Epting

(1982) indicate that subjects confronted witha dichotomous choice select the positive roleas descriptive of others approximately 62% ofthe time, perhaps because it is in this propor-tion that negative events are maximally strikingas figure against a general background of pos-itive events. Despite these equivalent ratios,however, and in accord with the cognitive ex-planation for the Barnum effect, subjects sys-tematically underestimated the frequency withwhich their acquaintances, relative to them-selves, engaged in both positive and negativebehaviors. Study 3 investigated whether thissymmetry in relative underestimation of pos-itive and negative behaviors would extend tothe evaluation of paragraph-long, Baraum-style descriptions, as opposed to single adjec-tive traits, or whether, in the context of longerdescriptions, the self-other distinction wouldinteract with description valence. It also ex-amined certain additional issues not consid-ered by the initial studies.

Study 3

If a purely cognitive mediation of theBarnum effect implies similar self-other dif-ferences in the acceptance of positive and neg-ative descriptions, mediation by motivationalfactors alone implies opposite self-other effectsfor descriptions of opposing valence—althoughpositive descriptions would be perceived asmore applicable to the self, negative descrip-tions would be seen as less applicable. Put an-other way, predictions based on a motivationalexplanation dovetail with those based on acognitive explanation for positive descriptions,but diverge for negative descriptions. In thelatter case, motivational factors should compelgreater rejection of negative descriptions asapplicable to the self, but a cognitive expla-nation based on the differential availability ofconfirming evidence would predict theirgreater perceived applicability.

In addition to examining these differingpredictions, Study 3 expanded the number ofpersons rated to self, close friend, moderatefriend, and casual acquaintance. In this regardthe experiment tested the prediction, sup-ported by both cognitive and motivational ex-planations, of a systematic relation betweendegree of familiarity and perceived accuracy.

Finally, Snyder et al. (1977) noted no gendereffects in previous Barnum effect research. A


secondary purpose of Study 3 was to examinethe interaction of the self/other effect with sexof subject under our particular paradigm.

Method

Overview. In a latin square design (Cochran & Cox, 1957),subjects rated the applicability of various paragraph-longdescriptions to themselves, a close friend, a moderate friend,and a casual acquaintance. Each subject rated 8 paragraphsin 4 sets, with a set including a paragraph related to apositive trait—for example, affectionate—and one relatedto a negatively valenced antonym—for example, reserved.Each subject rated the applicability of each of the para-graphs in a set to a different person—that is, one set wasjudged for accuracy with regard to the self, one set foraccuracy with regard to a close friend, and so forth. Pairingof set with person rated was completely counterbalancedbetween subjects.

Subjects. Subjects were 64 undergraduates at the Uni-versity of California, Davis. They received extra credit inan introductory psychology course for their participation.

Experimental materials. Paragraphs were generated byundergraduate writers to describe prevalent behaviorsamong their peer group that were characteristic of partic-ular trait dimensions. The positivity or negativity of thebehaviors described was moderated to meet the high base-rate restriction. Twenty-three undergraduate judges sub-sequently rated the applicability of each of the paragraphdescriptions to the average person on 9-point scales, withhigher numbers indicating greater accuracy. Analysis in-dicated no significant systematic main effect for Valenceon the rated prevalence of the description, (M = 5.30 forpositive paragraphs vs. M = 5.42 for negative paragraphs),f\\,22)< 1.

In constructing the paragraph descriptions, certainguidelines were consistently followed: The statements werevague and relatively short, and the descriptions were fre-quently hedged with qualifiers, for example, "sometimes"or "occasionally."3 Paragraph pairs involved the trait di-mensions of affectionate/reserved; resolute/irresolute;generous/selfish; and careful/rash. To exemplify the stim-ulus materials, the affectionate/reserved paragraph pair isshown in its entirety:

(Affectionate paragraph): When this individual is witha person who can be trusted, warm feelings for that per-son may be expressed. Best friends may be told howmuch their company is appreciated. When others sharetheir inner thoughts and emotions with this individual,he or she may find it easy to do the same. With friendsit may be quite easy for this person to speak what is onhis or her mind and to share deep feelings that are notshared with strangers. In the company of such friendsthe person feels honest and true to his or her emotions.At such times this person finds it easy to be open anddemonstrative. People may sometimes remark on howtalkative and affectionate this individual seems to be onthese occasions.

(Reserved paragraph): This individual may feel thereis a time and a place for emotions, and may occasionallyshow embarrassment about displaying affection in public.Sometimes the person may prefer to keep some distancefrom others. In particular, there are certain tasks thatthis person would prefer to do alone. When friends havehurt this person, he or she may experience difficulty in

showing them that he or she still cares. People who don'tknow this person well may be inclined to say he or shesometimes appears cold and indifferent to strangers. Thisperson may find it difficult to express genuine feelingsfor acquaintances that he or she does not know well andmay sometimes be uncomfortable about showing affec-tion for others.

Design and procedure. Each subject read all four setsof stimulus paragraphs. Subjects were randomly assignedto two orders of presentation of paragraphs such that onehalf read the four positive paragraphs first and one halfread the four negative paragraphs first. There were 12 malesin each of the order conditions, 19 females in the positiveparagraphs first condition, and 21 females in the negativeparagraphs first condition. Each subject rated all four in-dividuals, but only one person (the same individual) oneach of the two paragraphs in the four paragraph sets. Theperson rated was completely counterbalanced for eachparagraph set, so that one quarter of all subjects ratedthemselves on the affectionate-reserved paragraph set, onequarter rated a close friend on this set, one quarter a mod-erate friend, and one quarter a casual acquaintance. Underthis procedure, each subject rated each of the 4 individualsand each paragraph set, but only one paragraph set/indi-vidual combination. Order of rating both individuals andparagraphs was also completely counterbalanced betweensubjects, with the constraint that either the four negativeor the four positive paragraphs be rated first. Subjects ratedeach paragraph immediately after reading it.

Subjects were randomly assigned to condition. They weredebriefed after completing the task.

Instructions. Subjects were requested to read severalparagraphs and to decide "how accurately each of theseparagraphs describes someone whom you know." Beforereading the paragraphs, all subjects were requested to thinkof a close friend (a person whom "you know very well"and "see frequently"); a moderate friend (a person whom"you know and see occasionally and also consider afriend"); and a casual acquaintance ("someone you knowin passing but not very well . . . either someone you donot see often or do not spend more than a few minuteswith when you see him or her"). The instructions stressedthat subjects should think of the same close friend wheneverrating the applicability of a paragraph to a close friend,and the same moderate friend when rating the applicabilityof a paragraph to a moderate friend, etc.

Dependent measure. The dependent measure for allparagraphs was the rating on a 9-point scale of how ac-curately the paragraph described the designated individualwith ranges from not at all accurately (1) to extremelyaccurately (9). Subjects were again instructed to "use yourbest guess" in the event of uncertainty.

Results

A 2 X 2 X 4 (Order of Paragraph Presen-tation X Paragraph Valence X Person Rated)

3 The descriptions were subsequently rated for valenceby undergraduate judges using 9-point scales. Three of thefour pairs differed significantly in valence at the .05 levelor beyond and mean ratings for the remaining pair, care-ful-rash, were in the predicted direction, F ( l , 18) = 1.39,p < .25.


Table 2Mean Accuracy Judgments of Personality Descriptions for Four Different Individuals: Study 3

Rating

Rating for positive descriptionsRating for negative descriptions

M

Self

6.974.775.87

Closefriend

6.394.615.50

Moderatefriend

5.094.804.95

Casualacquaintance

4.734.764.75

Note. Higher numbers indicate ratings of greater accuracy.

ANOVA was performed on paragraph accuracyjudgments, collapsing across individual para-graphs. Order of paragraph presentation wasa between-subjects variable designatingwhether subjects rated the four positive para-graphs or the four negative paragraphs first.Paragraph valence (positive/negative) and per-son rated (self/close friend/moderate friend/casual acquaintance) were within-subjectsvariables.

The analysis indicated a highly significantmain effect for paragraph valence, with sub-jects rating positive paragraphs as generallymore applicable than negative paragraphs, F( 1,62) = 17.04, p < .001. There was also a highlysignificant main effect for person rated, F(3,186) = 13.52, p < .001. The precise nature ofthe latter effect, however, was more meaning-fully investigated by an a priori contrast testingwhether perceived applicability varied system-atically with degree of acquaintance. When thiscomparison (3, 1, — 1, — 3) was performed onthe means shown in the final row of Table 2,the results were highly significant, F( 1, 186) =39.40, p < .001, with the contrast accountingfor 97% of the variance among the means.Subjects perceived the character descriptionsas most applicable to themselves, next mostapplicable to a close friend, next most appli-cable to a moderate friend, and least applicableto a casual acquaintance. Separate analyses ofthe four individual pairs of stimulus para-graphs, although revealing some variation fromthe general pattern, indicated that applicabilityratings for self and close friend contrasted sig-nificantly (at least at the .05 level) with appli-cability ratings for moderate friend and casualacquaintance (1, 1, — 1, -1 comparison).

There was, however a significant PersonRated X Valence interaction, P(3, 186) = 5.41,p < .002. As may be noted by an examinationof the means in Table 2, this pattern was con-

fined to ratings of positive paragraphs. Al-though the contrast was significant when per-formed on judgments of positive paragraphsalone, F( 1, 186) = 52.55, p < .001, accountingfor 96% of the variance among the means, itdid not attain significance on ratings of neg-ative paragraphs, F(l, 186) < 1. Degree of ac-quaintance was not systematically related tothe latter ratings.

Neither the effect for valence nor the effectfor person rated interacted significantly withorder of paragraph presentation (positiveparagraphs first/negative paragraphs first),both interaction Fs < 1, and there was also nomain effect for order, F < 1. An additionalANOVA adding gender as a between-subjectsblocking variable also indicated no significantmain effect or interactions involving sex ofsubject.

Discussion

Study 3 provided strong evidence of a sys-tematic relation between degree of acquain-tance and judged accuracy of personality de-scriptions. Results, however, also indicated apronounced valence-related asymmetry. Al-though positive descriptions displayed the pre-dicted relation, negative descriptions did not.Study 4 was undertaken to investigate whetherthis asymmetry would replicate with differentsubjects and different specific stimulus mate-rials.

Study 4Method

The design, procedure, instruction set, and dependentmeasures in Study 4 were all identical to those used inStudy 3. The sets of stimulus paragraphs used, however,involved the bipolar trait dimensions of trusting-suspicious,self-reliant-dependent, stubborn-flexible, and cautious-impulsive. Paragraphs were composed by undergraduatesunder guidelines and constraints identical to those of Study


Table 3Mean Accuracy Judgments of Personality Descriptions for Four Different Individuals: Study 4

Rating

Rating for positive descriptionsRating for negative descriptions

M

Self

6.354.335.34

Closefriend

5.984.265.12

Moderatefriend

5.394.394.89

Casualacquaintance

5.183.684.43

Note. Higher numbers indicate ratings of greater accuracy.

3 and entailed descriptions of high base rate characteristicswith either a moderately positive or moderately negativevalence.4 As in Study 3, the paragraphs were subsequentlyrated by undergraduate judges on 9-point scales for theirapplicability to the "average person." Consistent with theStudy 3 results, analysis revealed no significant systematicmain effect for valence on perceived prevalence (M forpositive paragraphs = 4.91 vs. M for negative paragraphs =5.ll,F(\,22)< 1.

Paragraphs averaged two to three sentences longer thanthose used in Study 3. Subjects in the experiment were 80undergraduates at the University of California at Davis,who received extra credit in an introductory psychologycourse for their participation. There were 28 females and12 males in the positive paragraphs first condition, and 26females and 14 males in the negative paragraphs first con-dition.

Results

A 2 X 2 X 4 (Order of Paragraph Presen-tation X Paragraph Valence X Person Rated)ANOVA on paragraph accuracy judgmentsagain indicated a highly significant main effectfor paragraph valence, with subjects ratingpositive paragraphs as more applicable, F(l,78) = 50.10, p < .001. The main effect forperson rated was also highly significant, F(3,234) = 9.27, p < .001, as was the a priori con-trast (3, 1, -1, -3), F(l, 234) = 26.76, p <.001, which accounted for 96% of the varianceamong the means. As indicated by the meansin the final row of Table 3 (which collapseacross the valence variable) the pattern of theeffect directly parallels the pattern of Study 3,although the applicability ratings are generallyslightly lower. Subjects rated the descriptionsas most applicable to themselves, next mostapplicable to a close friend, next most appli-cable to a moderate friend, and least applicableto a casual acquaintance. Analysis of individ-ual paragraph sets, although revealing somevariation from the general pattern, indicatedthat for three of the four pairs applicabilityratings for self and close friend contrasted sig-nificantly (at the .05 level for two of the pairs

and at the .08 level for the third) with appli-cability ratings for moderate friend and casualacquaintance. The exception to this generalpattern was the cautious-impulsive paragraphset, in which accuracy ratings for self and closefriend did not differ significantly from accuracyratings for moderate friend and casual ac-quaintance.

In another parallel to Study 3, however, thisrank order of means occurs for ratings of pos-itive paragraphs alone. The accuracy ratingsof positive paragraphs show a progressive in-crement with degree of acquaintance, with thea priori contrast highly significant, F(l, 234) =17.12, p < .001, accounting for 97% of thevariance among the means. Although the Per-son Rated X Valence interaction does not at-tain significance F(3, 234) = 1.14, the patternof accuracy ratings for negative paragraphs isstrikingly different. With negative paragraphs,accuracy ratings for self, close friend, andmoderate friend do not differ significantly fromone another (all Fs < 1), but accuracy ratingsfor these 3 individuals in the aggregate differsignificantly from accuracy ratings for casualacquaintance, F(l, 234) = 4.86, p < .05 withthis comparison accounting for 98% of thevariance among the means. As in Study 3, nei-ther the effect for paragraph valence nor theeffect for person rated interacted significantlywith order of paragraph presentation; there was

4 The Study 3 paragraphs in the careful-rash set, whichdid not differ significantly in valence ratings by the separategroup of judges, were also the only paragraphs that showedno significant main effect for positivity on subjects' judg-ments of accuracy. The cautious-impulsive set in Study 4embodied descriptions on a similar trait dimension thatwere rated by undergraduate judges as differing significantlyin valence (at the .01 level). Study 4 investigated whetherthis significant difference in valence would be associatedwith a significant effect for positivity on perceived accuracy.The undergraduate judges also found that the paragraphsin the other three pairs used in Study 4 also differed sig-nificantly in valence at the .05 level or beyond.


also no significant main effect for order. Anadditional ANOVA with gender as a between-subjects variable indicated an absence of a sig-nificant main effect and the absence of anysignificant interaction between sex of subjectand paragraph valence or person rated.

Discussion

In sum, the results of Study 4 generally rep-licated those of Study 3, with the slightly lowerapplicability ratings perhaps explained by theobservation of Sundberg (1955) that longerdescriptions are generally perceived as less ac-curate. In both experiments a highly significanteffect for person rated assumed the same pat-tern—the more familiar the individual rated,the more likely was a description to be judgedapplicable to that person. In both experiments,however, this effect was due largely to the pat-tern of accuracy ratings for positive paragraphsalone.

General Discussion

This series of experiments investigatedwhether the Barnum effect—the acceptance ofhigh base rate descriptions as uniquely appli-cable to the self—would occur when subjectshad not been deceived into believing that thedescriptions had been prepared specifically forthem. All four of the studies showed a strongself-other distinction under these circum-stances, supporting the proposition that the ef-fect is not mediated exclusively by the credi-bility awarded the alleged source—for exam-ple, psychologist, astrologer, or computer—ofthe description. The results, it would seem, arealso difficult to explain in terms of demandcharacteristics. Absent the belief that the per-sonality summaries were prepared specificallyfor the subject, a strong motivation to confirmthe expectations of the experimenter by ac-cepting the descriptions which he or she pre-pared does not explain the greater perceivedapplicability to the self. Rather, the studiessupply evidence that self-other differences inperceived applicability may be mediated byboth cognitive and motivational factors.

Study 1 examined accuracy ratings of thepersonality descriptions originally prepared byForer (1949) and subsequently used as stimulusmaterials in several later Barnum effect studies.Subjects judging the applicability of these

statements to themselves rated them signifi-cantly higher in accuracy than subjects judgingtheir applicability to an acquaintance and thisdifference prevailed for both positive and neg-ative statements. Study 2 provides evidencesuggesting that the effect may be mediated bya failure to adjust sufficiently for the relativelack of confirming evidence of high base-ratecharacteristics in the other. Subjects rating thefrequency with which dispositional character-istics were displayed underestimated the fre-quency with which an acquaintance, relativeto the self, manifested both positive and neg-ative traits.

The data of Study 2, although not inconsis-tent with Nisbett et al. (1973) and Goldberg(1981), also provide some explanation for thefindings of Monson et al. (1980) that individ-uals asked simply to accept or reject traits finda greater number applicable to themselves thanto others. Our results indicate that subjectsrating themselves readily find more evidenceconfirming the existence of individual traits.When traits are rated on bipolar scales, how-ever, relatively strong evidence confirming theexistence of a trait may be effectively cancelledby relatively strong evidence confirming theexistence of its polar opposite, with the endresult that bipolar ratings of the self do notdiffer significantly from bipolar ratings of anacquaintance.

These data regarding systematic self-otherdifferences in the estimated prevalence of highbase-rate behaviors indicate that any adjust-ment subjects make for the differential avail-ability of confirming evidence is inadequate.One explanation for the inadequacy is thatsubjects ignore self-other differences in thebase rates of instances from which confirmingevidence is to be drawn. Under this account,the relative dearth of available evidence ofcommon traits in another may have led themto believe that these traits were in fact moreprevalent in themselves. A different explana-tion, however, would attribute the self-otherdifference to subjects' conservatism in com-mitting themselves on the existence of traitsin another. Although explicitly instructed touse their "best guess" in the absence of infor-mation sufficient to make a choice, subjectsmight nevertheless have experienced some re-luctance to guess, instead estimating that thebehaviors of which they had little pertinentknowledge were relatively infrequent.


Note, however, that these two different ex-planations have an underlying affinity. In cer-tain contexts a judgmental conservatism maybe well-justified, but not in the context of ourparticular paradigm. In that paradigm, both afirm belief that high base-rate behaviors aremore prevalent in the self (Explanation 1) anda reluctance to ascribe those behaviors to alittle-known other (Explanation 2) may beviewed as variations of the same basic failureto draw the normatively appropriate inference,or as two different versions of the same cog-nitive "error." The appropriate inference isthat, just as human beings might be thoughtquite likely to have two eyes, two arms, andtwo legs, they are also reasonably likely, in theface of little tangible evidence to the contrary,to manifest traits with an extremely high baserate. Apart from motivational influence, wewould expect to find little systematic under-estimation of the physical traits—for exampleheight and weight—of little-known others, orof certain demographic characteristics—forexample age or amount of education. The rea-son for the difference between the two cate-gories of traits may be that, although there ismuch tangible and objective evidence as to thedistribution of physical and demographictraits, there is less evidence of the commonalityof inner states or psychological variables. Boththe outright denial of the commonality of thelatter characteristics, as well as the reluctanceto admit their prevalence without strong sup-porting information, may be products of afundamental failure to appreciate a generallyhigh degree of similarity between our ac-quaintances and ourselves.

The findings of Studies 1 and 2, indicatinginsufficient adjustment for lesser knowledge ofboth positive and negative traits in others, sug-gested that descriptions of positive and negativevalence, separately judged, might both be ratedas more applicable to the self than to others.The results of Studies 3 and 4 did indicatehighly significant main effects for person rated,and furthermore indicated a systematic rela-tion between degree of familiarity and per-ceived accuracy of the description. In both ex-periments, however, this main effect for fa-miliarity was largely confined to positivedescriptions, indicating the influence of mo-tivational factors. This particular asymmetryis in fact quite consistent with Snyder andShenkel (1976), who found that subjects rated

positive feedback as more applicable to them-selves than to "people in general," but saw nosignificant difference between the applicabilityof negative feedback to themselves and a gen-eralized other.

The observed relation between perceivedaccuracy and degree of familiarity is reminis-cent of the results of Taylor and Koivumaki(1976), who found that a positivity bias in at-tribution increased with degree of acquain-tance. These results might be taken to implythat the Barnum effect is largely motivationalin origin—a product of the desire to view selfand friends in a relatively positive light. Buthow does a motivational explanation accountfor the asymmetrical nature of the effect? In-dividuals motivated to rate mildly positive de-scriptions as most applicable to themselvesought, it would seem, also be motivated to ratemildly negative descriptions as least applicableto themselves, but none of the present studiesprovide any indication of the latter tendency.One reasonable explanation for the asymmetryinvolves the assumption that, while people aremotivated to be particularly kind in evaluatingthemselves and their friends, they are not mo-tivated to be particularly unkind in evaluatinglittle-known others. But this account, althoughplausible, does not fully explain why, if we aremotivated to view ourselves and our friends ina positive light, the negative paragraphs in bothof the latter two studies were rated at medium-level accuracy (between 4 and 5 on 9-pointscales) as descriptions of both self and friends.It may be that a nonmotivational factor—per-sonal knowledge that the descriptions were infact applicable—inhibited a relatively strongmotivation to reject the descriptions as accu-rate portraits of the self.

Our results would appear to indicate thatthe Barnum effect may be influenced both bymotivational factors and by cognitive factorsrelated to our failure to adjust for the lesseravailability of confirming evidence regardingthe little-known other. Because both theamount of available information and the desireto view an individual positively may increasewith degree of acquaintance, cognitive andmotivational influences may converge for pos-itive descriptions, producing a systematic pos-itive relation between familiarity and judgedaccuracy. With negative descriptions, however,cognitive and motivational influences mayproduce opposing tendencies. Both the store


of knowledge concerning the prevalence ofnegative characteristics and the motivation toreject their applicability may increase with de-gree of acquaintance. These countervailing in-fluences may in effect nullify one another, re-sulting in similar medium-level accuracy rat-ings for descriptions of both self and other.However, increasing the motivation to rejectthe applicability of negative descriptions to theself (by, for example, presenting extremelynegative descriptions) might well result in lessperceived accuracy for the self than for others,confirming the predictions of a purely moti-vational model.

The results of Studies 3 and 4, which oth-erwise are remarkably parallel, diverge slightlyin that the Study 4 ratings of the applicabilityof negative descriptions to a casual acquain-tance are lower than applicability ratings forthe other 3 persons in the aggregate; in Study3 the perceived applicability of negative de-scriptions does not vary significantly with per-son rated. One plausible explanation for thisdiscrepancy is that both personal knowledgeof, and inferences concerning, personality traitsin casual acquaintances may vary with theparticular trait. In instances in which personalknowledge of a trait is particularly scanty, anegative description may be rated as less ap-plicable to a casual acquaintance than to theself or friends, even though perceived appli-cability to the latter may have been reducedby a motivational bias. The present experi-ments, which were designed to examine theBarnum effect with the original Forer stimulusmaterials and other high base rate descriptionsresembling astrological character sketches, didnot systematically explore sources of variationrelated to category of trait. One goal of futureresearch might be to investigate the influenceof variations along the external-internal traitdimension proposed by Funder (1980).

A more basic issue involves the question ofwhy the pattern of self-other rating differenceswas influenced by valence in Studies 3 and 4when neither Study 1 nor Study 2 gave anyclear indication of such an influence. The mostobvious explanation relates to fundamentaldifferences in the stimulus materials used. Itmay be that the motivation to deny the appli-cability of negative traits to the self becomesa significant factor only when those negativetraits are presented in the form of a relativelylengthy description, which carries with it the

implication of being a fairly complete charactersketch. It may be one thing, for example, toacknowledge frequent stubbornness in the self,potentially attributable to a particular categoryof situations, and yet quite another to accedeto the accuracy of the character sketch of astubborn person, with its concomittant impli-cation that stubbornness is a central trait at-tributable to dispositional factors.

Despite the valence-related asymmetry, themain effect for acquaintance assumed the pre-dicted pattern in both Studies 3 and 4. Giventhe clear real-world analogues to the stimulusmaterials, the conclusion to be drawn is ap-parent. When personality descriptions—forexample, the character sketches which fre-quently appear in books and newspaper col-umns on astrology—consist of a mix of com-mon positive and negative traits, those de-scriptions will tend to be regarded assignificantly more accurate when the persondescribed is well known to the perceiver andwill be seen as most accurate when the persondescribed is the actual perceiver. The resultsalso indicate that this effect is not wholly con-tingent on any prior credibility accorded thesource of the description. Rather, in accordwith the results of Snyder et al. (1976), it islikely that the source gains additional credi-bility as a function of the perceived accuracyof the personality sketch.

It should be emphasized that, although theperception of applicability of high base ratecharacteristics to the self is not, in and of itself,inaccurate, the data do provide substantial ev-idence of one departure from normative ac-curacy—our subjects failed to adjust their ac-curacy judgments upward to compensate fortheir relative lack of available data on friendsand acquaintances. The error, it would appear,lies in our failure to appreciate that, just as wemay have substantial evidence of Geminicharacteristics in ourselves (and, for that mat-ter, substantial evidence of the characteristicsof each of the signs of the Zodiac), both AuntHarriett and the postman may also have ampledata bases indicating that they also frequentlybehave as typical Geminis.

While our failure to adjust for insufficientknowledge, and concomitant reluctance to as-sume the existence of high base-rate charac-teristics in everyone, may reflect an inferentialerror, that error may well have cultural foun-dations. Subjects' estimations that they per-


form certain common behaviors more fre-quently than their acquaintances would appearto reflect the implicit assumption that our ev-eryday behavior is not fully replicated, evenon the limited dimension of frequency, in theexperience of little known others. Althoughthis assumption may be the product of a desirefor uniqueness (Snyder & Fromkin, 1980), itmay also reflect the expectations of a Westernculture, which emphasizes the differences be-tween individuals and minimizes their simi-larities.

ReferencesAdams-Webber, J. (1979). Personal construct theory: Con-

cepts and applications. New York: Wiley.Bradley, G. W. (1978). Self-serving biases in the attribution

process: A reexamination of the fact or fiction question.Journal of Personality and Social Psychology, 36, 56-71.

Bruner, J. S., & Tagiuri, R. (1954). The perception of peo-ple. In G. Lindzey (Ed.), Handbook of social psychology(Vol. 2, pp. 634-654). Reading, MA: Addison-Wesley.

Chambliss, W. (1964). The negative self: An empirical as-sessment of a theoretical assumption. Sociological In-quiry, 34, 108-112.

Cochran, W. D., & Cox, G. M. (1957). Experimental de-signs. New York: Wiley.

Collins, R. W, Dmitruk, V. M., & Ranney, J. T. (1977).Personal validation: Some empirical and ethical consid-erations. Journal of Consulting and Clinical Psychology,45, 70-77.

Forer, B. R. (1949). The fallacy of personal validation: Aclassroom demonstration of gullibility. Journal of Ab-normal and Social Psychology, 44, 118-123.

Funder, D. C. (1980). Ourselves as others see us: Self-otheragreement and discrepancy in personality ratings. Jour-nal of Personality, 48, 473-493.

Goldberg, L. R. (1981). Unconfounding situational attri-butions from uncertain, neutral, and ambiguous ones:A psychometric analysis of descriptions of oneself andvarious types of others. Journal of Personality and SocialPsychology, 41, 517-552.

Handelsman, M. M., & Snyder, C. R. (1982). Is rejectedfeedback really rejected?: Effects of informativeness onreactions to positive and negative personality feedback.Journal of Personality, 50, 168-179.

Johnson, J. T, & Judd, C. M. (1983). Overlooking theincongruent: Categorization biases in the identificationof political statements. Journal of Personality and SocialPsychology, 45, 978-996.

Kahneman, D., & Tversky, A. (1973). On the psychologyof prediction. Psychological Review, 80, 237-251.

Kahneman, D., & Tversky, A. (1982). Evidential impactof base rates. In D. Kahneman, P. Slovic, & A. Tversky(Eds.), Judgment under uncertainty: Heuristics andbiases (pp. 153-160). New York: Cambridge UniversityPress.

Meehl, P. E. (1956). Wanted—A good cookbook. AmericanPsychologist, 11,262-212.

Monson, T. C., Tanke, E. D., & Lund, J. (1980). Deter-minants of social perception in a naturalistic setting.Journal of Research in Personality, 14, 104-120.

Nisbett, R. E., Caputo, C., Legant, P., & Maracek, J. (1973).Behavior as seen by the actor and as seen by the observer.Journal of Personality and Social Psychology, 27, 154-164.

Orne, M. T. (1962). On the social psychology of the psy-chological experiment: With particular reference to de-mand characteristics and their implications. AmericanPsychologist, 17, 776-783.

Riess, M., Rosenfeld, P., Melburg, V, & Tedeschi, J. T.(1981). Self-serving attributions: Biased private percep-tions and distorted public descriptions. Journal of Per-sonality and Social Psychology, 41, 224-231.

Rigdon, M. A., & Epting, F. R. (1982). A test of the goldensection hypothesis with elicited constructs. Journal ofPersonality and Social Psychology, 43, 1080-1087.

Sears, D. O. (1983). The person-positivity bias. Journal ofPersonality and Social Psychology, 44, 233-250.

Schlenker, B. R., & Miller, R. S. (1977). Egocentrism ingroups: Self-serving biases or logical information pro-cessing? Journal of Personality and Social Psychology,35, 755-764.

Snyder, C. R., & Fromkin, H. L. (1980). Uniqueness: Thehuman pursuit of difference. New York: Plenum Press.

Snyder, C. R., Larsen, D., & Bloom, L. J. (1976). Accep-tance of general personality interpretations prior to andafter receipt of diagnostic feedback supposedly basedon psychological, graphological. and astrological assess-ment procedures. Journal of Clinical Psychology, 32,258-265.

Snyder, C. R., & Larson, G. R. (1972). A further look atstudent acceptance of general personality interpretations.Journal of Consulting and Clinical Psychology, 38, 384-388.

Snyder, C. R., & Shenkel, R. J. (1976). Effects of "favor-ability," modality, and relevance upon acceptance ofgeneral personality interpretations prior to and after re-ceiving diagnostic feedback. Journal of Consulting andClinical Psychology, 44, 34-41.

Snyder, C. R., Shenkel, R. J., & Lowery, C. R. (1977).Acceptance of personality interpretations: The "Barnumeffect" and beyond. Journal of Consulting and ClinicalPsychology, 45, 104-114.

Sundberg, N. M. (1955). The acceptability of "fake" versus"bona fide" personality test interpretations. Journal ofAbnormal and Social Psychology, 50, 145-147.

Taylor, S. E., & Koivumaki, J. H. (1976). The perceptionof self and others: Acquaintanceship, affect, and actor-observer differences. Journal of Personality and SocialPsychology, 33, 403-408.

Wylie, R. C. (1965). Self-ratings, level of ideal-self ratings,and defensiveness. Psychological Reports, 16, 135-150.

Received February 13, 1984Revision received June 29, 1984

Documents

The Barnum Effect Revisited: Cognitive and Motivational