Determinants of social comparison choices

Determinants of social comparison choices

, OUHNAL OY EXPEHIMENTAL SOCIAL Determinants of PSYCHOLOGY Social 7, 473489 ( 1971) Comparison Choices CHARLES L. CRUDER' University of Il...

1MB Sizes 0 Downloads 64 Views

, OUHNAL

OY EXPEHIMENTAL

SOCIAL

Determinants

of

PSYCHOLOGY

Social

7, 473489

( 1971)

Comparison

Choices

CHARLES L. CRUDER' University

of Illinois

at Chicago

Circle

Two experiments were conducted to clarify previous, inconsistent findings regarding various hypothesized determinants of social comparison choices. A bogus personality trait scale was administered to subjects, and they were either informed of the range of scores in their group ( R conditions) or not informed (NR conditions). Experiment 1 closely replicated support for a rangeseeking hypothesis and for Festinger’s theory of social comparison processes (1954), provided only minimal support for a positive instance hypothesis, and failed to obtain support for a desirability hypothesis. Experiment 2 was run to complement Experiment 1 in providing an unequivocal test of the positive instance and desirability hypotheses. The results suggested a multifactor model to account for comparison choices, the factors being whether or not a score was (a) an uncertain element (i.e., the highest score), (b) a positive instance, and/or (c) desirable.

The recent attention given to Festinger’s (1954) social comparison theory (Arrowood & Friend, 1969; Latant, 1966; Wheeler, Shaver, Jones, Goethals, Cooper, Robinson, Gruder, & Butzine, 1969) has raised a number of questions open to empirical investigation. The research has typically tested groups of subjects for a personality characteristic under the guise of validating a new assessmentinstrument. Following testing, subjects were told that they had the median score in the group and were given the opportunity to learn another’s score. Festinger’s theory of social comparison processes predicts that in this situation persons prefer to compare themselves with similar others in order to achieve most information regarding the meaning of their score, and therefore will choose to learn scores which are close to their own in rank. Initial inconsistencies in findings have been attributed by investigators to the fact that the conditions under which predictions derived from Festinger’s theory will hold have not been sufficiently specified. In an attempt to clarify some of these inconsistencies, Wheeler et al. ‘The author thanks Ladd Wheeler and Robert S. Wyer, Jr., for their comments on earlier drafts of this report, and Robert J. Duslak and Harmon hoff for their assistance in the collection of the data. 473 @ 1971 by Academic Press, Inc.

critical Look-

474

CHARLES

L.

GRUDER

showed that before persons chose a comparison other who was similar to themselves, it was necessary for them to know the extremes of the dimensions being measured. That is, only when subjects were given information about the range of scores obtained by others in their group before they made their comparison choice did they choose to learn a score ranked next to their own. Otherwise, they tended to choose the highest score. When given the opportunity to learn a second score, subjects tended to choose the lowest score. Pettigrew (1967), in his discussion of Festinger’s theory, also mentioned the assumption that persons must- know the range before they will choose to compare with a similar other. The consistency noticed by Wheeler et al. in subjects’ choices of the most extreme scores led directly to the current experiments. As Arrowood and Friend astutely indicated, while Wheeler et al. predicted that the highest score should be the subjects’ first choice, this prediction does not logically follow from their range-seeking hypothesis. Rationally, if the subject’s only motivation was to locate the limits of the range of scores in his group, the lowest and the highest scores have been equally likely choices. Arrowood and Friend offered two hypotheses to account for the predominant choice of highest score; these were first presented in Thornton and Arrowood (1966)-the positive instance and desirability hypotheses. The former is that the highest score will be preferred as the first choice because it represents a person who possesses the most of the characteristic being measured; he is a positive instance of this characteristic. The person who receives the highest score can potentially serve as the best example of and source of information about persons who possess the characteristic. Following earlier studies, Wheeler et al. left the lowest score undefined; it simply represented the absence of the characteristic, rather than the presence of some other or the opposite characteristic. The desirability hypothesis is simply that a score which represents a positively valued characteristic will be preferred to a score which does not. Wheeler et al. found support for the range seeking and desirability hypotheses but not for the positive instance hypothesis. However, Arrowood and Friend reanalyzed Wheeler et al.‘s data and found patterns consistent with the positive instance and desirability hypotheses. Experiment 1 in the present paper replicated part of Wheeler et al.‘s study in the hope of determining which of the findings were reliable ones. Experiment 2 was designed to complement Experiment 1 in providing an unequivocal test of the positive instance and desirability hypotheses.

SOCIAL

COMPARISON

EXPERIMENT

CHOICES

475

1

Experiment 1 was a replication of Wheeler et al’s score variation, in which the dependent variable was which of the other group members’ scores subjects chose to learn. The design was a 2 x 2 factorial with two levels of trait desirability (Pos and Neg). and either informing or not informing subjects of the score range in their group (R and NR). The purpose of the replication was to determine the reliability of Wheeler et al’s results and to reassess their range-seeking hypothesis and the alternatives suggested by Arrowood and Friend (viz., the desirability and positive instance hypotheses).

Subjects in E‘xperiments 1 and 2 were male undergraduates enrolled in Introductory Psychology at the University of Illinois at Chicago Circle who volunteered to participate in order to fulfill a course requirement, One hundred eight subjects participated in Experiment 1, nine at each session. The data of four were not analyzed, three because they misunderstood the instructions regarding the dependent variable and one because he participated twice. Procedure Subjects were seated in booths in which they could see the experimenter but not each other and completed a bogus personality questionnaire in 10 min. They were told that they would perform tasks later in the experimental session in order to allow for the validation of the personality scale. The experimenter then assigned subjects identification letters; although subjects thought they received different letters, all were assigned the same letter. While the experimenter left the room to score the scales, an assistant read an explanation of the personality trait the scale was designed to measure: this constituted the Pos-Neg manipulation, a high score on the scale indicating either a positive (“intellectual flexibility”) or a negative (“intellectual rigidity”) personality trait. he privately gave each subject When the experimenter returned to the room, his score; although subjects thought they had each been given a valid score, they were all assigned the same score. Then the experimenter placed the identification letters in rank order from highest to lowest on a blackboard in full view of all subjects. The subjects’ identification letter was always placed in the middle of the list, the median. When placing the rank order on the blackboard in the R condition, the experimenter wrote the range in which the highest and lowest scores fell next to the letters representing these scores. All subjects were assigned the score 310, the highest score was in the range 550-600, and the lowest score was in the range 25-75. Subjects were then given the opportunity to learn one score of their choosing. They wrote down the letter of the person whose score they most wanted to see. Then they were told that they could learn a second score and wrote down their second choice. After completing a questionnaire while the experimenter was ostensibly srrbjects were debriefed, sworn to secrecy, and dismissed. filling in their choices, The entire experiment lasted about 40 min. the r\-trrme scort’~~ were referred tcr a rrurnbet I,, \Vhe~4(~ of 01,‘s instrrlctions,

476

CHARLES

L.

GRUDER

of times by the words “highest” and “lowest,” in that order. The possibility existed that this order of presentation biased subjects’ responses when they were asked which score they most wanted to learn: perhaps they asked to learn the highest score first simply because that was the one they were used to hearing first. To test for this, the order in which these words were paired was counterbalanced in the NR conditions.

Results

and Discussion

In almost every instance, the results of this experiment replicated those of Wheeler et al’s study. Differences between the replication and the original were nonsignificant except where indicated. Data relevant to Wheeler et al’s range-seeking hypothesis and Festinger’s theory will be presented first, followed by tests of predictions from the positive instance and desirability hypotheses. Finally, the relative validity of these hypotheses will be considered. An item 011 the final questionnaire had subjects rate on a 40-point scale their opinion of the tested personality traits’ desirability. The mean rating of intellectual flexibility was 29.54 and the mean rating of intellectual rigidity was 19.62. These means differed in the intended direction with F( 1,100) = 28.33, p < 601. In NR the differences in comparison choices due to the verbal set provided by the order of presentation of the words “highest” and ‘lowest” were nonsignificant. The proportions of subjects making the predominant choice (the highest score) in the “highest-lowest” and “lowest-highest” orders were 50 and 68.8% in Pos and 50 and 58.8% in Neg, respectively. In light of the lack of differences, subsequent analyses were computed after collapsing over orders. The range-seeking hypothesis states that Range-seeking hypothesis. subjects who do not know the range of scores in their group will choose to learn scores in order to discover the range. The first prediction from the range-seeking hypothesis in Wheeler et al’s study, which was supported by their data, was that subjects in NR attempted to determine the range of scores by choosing the highest score first. The predominant first choice in NR in the present study was also the highest score, 58.8% choosing it in Pos and 54.3%in Neg. Each of these observed proportions differed significantly from the proportion expected by chance (with eight scores to choose from-12.5%).? Also consistent with Wheeler et al’s data, the difference between NR and R in proportions choosing to see the highest score was significant by one-tailed .z, corrected for continuity, within each level of trait desirability, Pos (58.8 vs 27X?, ’ Except transformed by \f'allwr

where otherwise using an arcsin and 1,~ (1953.

specified. in tests on proportions, the proportions werr transformation and thr test used was that described pp. 423-424).

(‘orlditir)lr .v I’I

1:

Pos

Neg Highest

First

choice

Subject

Lowest, Highest

Second

rhoice

score

score score

Sttbjert

Lowest

score

Neg

(+ I I) A JI F c J<; 13

.58.S s.s vi. 9 5 H

54.3 s 6 0 2 0

0 0 0 20.6

8.6 ‘2 9 0 23 .!I

0 0 0 5 6

33 .5

C; I I) A H F <: 15 H

17.6 2 0 0 5 .9

11.4 S.6 2 9 3 7

11.1 16.7 27.x - . 9 b

17.6 5 9 17.6 ‘23 r,

5.9 0 8.5 5X.S

11.4 X.6 5.7 46.7

16.7 - , n b 0 16.7

0 17.6 0 17.6

34

3.5

1s

1’7

2::. 5 0 5 .!I 0

11.8

17.6 17.6

p < .032) and Neg (54.3 vs 23.5%, p < .038). The difference between the proportions of Pos and Neg NR subjects choosing the highest score was trivial. Whether comparing NR subjects’ choice of the highest score with chance or with R subjects’ choice of the highest score, NR subjects conformed to the prediction made by the range-seeking hypothesis. Another implication of the range-seeking hypothesis is that NR subjects will expend each of their two choices on an extreme score. In the present study, as in the study by Wheeler et al., large proportions of NR subjects chose one of the extreme scores on their first choice and the other extreme on their second choice: 62.3% of the subjects in Experiment 2 expended both choices on the extreme scores. This proportion differed significantly from the chance expectation of 3.6%” ’ There are 8!/2!6! = 28 combinations of eight things As all combinations of two are equally likely, the proportion make any combination of choices is 46s = 3.6%.

taken two of subjects

at a time. expected to

478

CHARLES

1~. CRUDER

Festinger’s theory. In Wheeler et ~1,‘s study, subjects who were aware of the range of scores (R) behaved as Festinger’s theory predicted, comparing their scores either with others whom they thought possessed a desirable trait (in Pos) or did not possess an undesirable one (in Neg). Data in the present study were also in accord with this prediction: 94.4% of Pos subjects compared upward and 70.5% of Neg subjects compared downward. The differences between each proportion and chance (50%) were significant by one-tailed tests (Pos: p < .OOOl, Neg: p < 04, see footnote 2) but were marginally different from one another ( p = .077 by Fisher’s exact test). Another prediction from Festinger’s theory was that the subjects’ predominant choice in R would be the score of the individual who was most similar to them in rank in the positively valued direction. The current findings replicated Wheclcr et d’s qualified support for this prediction. The score of the most similar individual in the desirable direction was most frequently chosen in Pos (33.3%) and Neg (23.5%). However, the highest score was also frequently chosen in Pos (27.8%) and equally frequently chosen in Neg (23.5%). The predominant second choice in the R conditions of Wheeler et al’s study was the most similar score in the undesirable direction. They interpreted this pattern of first and second choices as a subject’s attempt to calibrate finely his own position. In the present study, second choices in Neg-R replicated those of the earlier study; in POS-R the most similar score in the undesirable direction was tied with two other scores as the second most preferred choice (see Table 1). While 39% of Wheeler et al.‘s subjects expended both choices on the two most similar scores, only 14.3% of the R subjects in the present study did so; the difference between these proportions was significant with p < .036, two-tailed. However, proportions in both studies differed significantly from the chance expectation of 3.6%. As in Wheeler et d’s study, another strong two-choice strategy in R was to choose the highest and lowest scores: 15% made these choices in the earlier study and 28.6% did so in the present study. While these proportions did not differ significantly, it should be pointed out that this strategy was predominant in the present study and subordinate in Wheeler et al’s study. Positive instance hypothesis. Wheeler et al. tested the positive instance hypothesis by holding desirability constant and comparing the proportion of choices of a positive instance (the higher scores) and a negative instance (the lower scores). The prediction from the positive then, was that choices upward by Pos subjects instance hypothesis, (desirable and a positive instance) should be more frequent than choices

SOCML

COXIl’AI~ISON

CHOICES

479

downward by Neg subjects (desirable but not a positive instance). Wheeler et al. pointed out that because their range-seeking hypothesis made the same prediction in NR, they tested the positive instance hypothesis only in R. In R, Festinger’s social comparison theory makes the competing prediction that there should be no Pos-Neg difference in the proportion of choices in the desirable direction (i.e., upward in Pos and downward in Neg). As reported in the discussion of predictions from Festinger’s theory, Wheeler ef ul. found no Pos-Neg effect on choice of scores in the desirable direction, while this effect was marginal in the present study (‘p = .077), with 94.4% of Pos subjects and 70.5% of Neg subjects choosing scores in the desirable direction. Assuming that the highest possible score is the most positive instance, Wheeler et al. considered, within R only, choices of the highest score in Pos and the lowest score in Neg. Once again, they found a nonsignificant trend, 41% in Pos and 31% in Neg choosing the most desirable score. The same result was obtained in Experiment 1, 28 vs 18%. These findings indicate that, controlling for desirability and providing the range of scores, subjects were not more likely to choose a score which was a positive instance than one which was a negative instance. In their reanalyses of Wheeler et al.‘s data, Arrowood and Friend combined R and NR data and found a significant Pos-Neg main effect on choice of the best-off other (viz., most desirable score). However, there was also a marginal trend for a Pos-Neg X R-NR interaction ( F( 1, M ) = 2.972, p < .lO) a The Pos-Neg simple effect was significant only in NR (p < .0006, using Fisher’s exact test), a finding ,predicted by the range-seeking hypothesis. The data from Experiment 1 replicated the Pos-Neg main effect (F( ICC ) = 4.168, p < .05, see footnote 4). The trend for a Pos-Neg x R-NR interaction was weaker than in Wheeler et uZ.‘s study, although it was in the same direction; the PosNeg difference once again was significant only in NR (p = .002 using Fisher’s exact test). Thus, it appears that Arrowood and Friend’s claim for support of the positive instance hypothesis is based on a difference which may also be accounted for by the range-seeking hypothesis. Desirahilit!g hypothesis. Arrowood and Friend used the desirability hypothesis to predict that a greater proportion of Pos than Neg subjccts would choose the highest score. The basis for this prediction was that the highest score was a positive instance in both Pos and Neg, but it was desirable only in Pos. They found evidence for this hypothesis in Wheeler et c~Z.‘s R and NR conditions combined, although this main ‘The author thanks John Arrowood for clarification of variance, which is described in Kempthorne ( 1952, and Cochran ( 1967, pp. 327429).

of this procedure for analysis pp. 153-158) and in Snedecor

480

CHARLES

L.

GRUDER

effect was not replicated in Experiment 1. Although there was not a statistically significant Pos-Neg X R-NR interaction on choice of highest score in Wheeler et al.‘s data, the Pos-Neg simple effect was significant in R ( p = .024 by Fisher’s exact test), but not in NR. The lack of a Pos-Neg effect in NR is consistent with the range-seeking hypothesis, and the Pos-Neg effect in R is not inconsistent with Festinger’s theory. The pattern in Wheeler et al’s data on which Arrowood and Friend based a second test of the desirability hypothesis also failed to replicate in Experiment 1. Arrowood and Friend observed a Pos-Neg main effect, indicating that a greater proportion of Neg than Pos subjects chose to learn the lowest score first. As in the case of choices of the highest score, the Pos-Neg x R-NR interaction was not significant in Wheeler et aZ.‘s data, but once again the Pos-Neg main effect was primarily due to the significant simple effect in R (p = .018 by Fisher’s exact test). Combination of positive insta,nce and desirability hypotheses. Following an analysis by Thornton and Arrowood, Wheeler et al. tested to see whether Neg subjects were more likely than Pos subjects to make their two choices in opposite directions. This was expected by the desirability and positive instance hypotheses because by choosing upward on their first choice, Pos subjects satisfied motivations instigated by both sources, while Neg subjects satisfied only the motivation instigated by the availability of a positive instance by so choosing. Neg subjects should have been more likely to choose downward on their second choice after choosing upward on their first choice, or vice versa if they chose downward first. The trend in R and NR was nonsignificantly in the predicted direction both in Wheeler et al.‘s data and in the data from the present study, where 78.8% of Neg subjects and 73.1% of POS subjects made their two choices in opposite directions. Arrowood and Friend compared choices of the highest and lowest Collapsing over the R-NR variable, scores with chance expectations. they found that in Neg the highest and lowest scores were chosen more than expected, while in Pos only the highest score was overchosen. These findings were consistent with the positive instance and desirability hypotheses. This finding was replicated in the present study. In Neg, the highest and lowest scores were overchosen (p < .OOOl and p < .04, onctailed, see footnote 2), while in Pos only the highest score was chosen at greater than the chance level (p < .OOOl, one-tailed, see footnote 2). It might be noted that the finding on the choice of the lowest score in Neg was weaker than in the earlier study, perhaps indicating the relative weakness of the desirability hypothesis. An analysis which has not yet been done is to compare the choices of highest and lowest scores across the various conditions. Specifically,

SOCIAL

COMPARISON

481

CHOICES

the positive instance and desirability hypotheses jointly predict an interaction between choice of highest and lowest scores and Pos-Neg. The highest score should be the predominant choice in Pos (positive instance and desirable), while the highest and lowest scores should be equally preferred in Neg (highest score is a positive instance and lowest score is desirable). On the other hand, Wheeler et aZ.‘s rangeseeking hypothesis does not predict this interaction in NR; it predicts only a main effect, more subjects choosing the highest score than the lowest score. In R, Festinger’s social comparison theory predicts that subjects will predominantly choose the most similar score in the desirable direction. The only implication of this prediction for choices of the highest and lowest scores is that they should be less frequent in R than in NR; Pos-R and Neg-R should not differ in the frequency of choices on the extreme scores. In sum, the positive instance and desirability hypotheses predict an interaction between Pos-Neg and whether the highest, lowest, or other scores were chosen, while the range-seeking and social comparison hypotheses predict an interaction between R-NR and choice of scores. The interaction between Pos-Neg and score choice was significant in Wheeler et al.‘s data [x’( 2) = 14.723, p < .005], but not in those of Experiment 1 ]x’( 2) = 1.013, see Table l] .5 In light of the close replication of other findings, this failure to replicate casts doubt on the validity of the positive instance and desirability hypotheses. This doubt is even greater in light of the significant interaction between R-NR and score choice in both the earlier [x’( 2) = 33.596. ‘p < .005] and later studies [x2(2) = 24.841, p < .005]. This interaction was predicted by the range-seeking and social comparison hypotheses, and not by the positive instance and desirability hypotheses. It is evident in Table 1 that the interaction in Experiment 1 was due to the greater proportion of subjects in R than in NR who chose the middle six scores, and the greater proportion in NR than in R who chose the highest score. These patterns support Wheeler et aZ.‘s interpretation of their data. The range-seeking and social comparison hypotheses often made predictions inconsistent with those of the positive instance and desirability hypotheses. The evidence bearing on the former two hypotheses were uniformly supporting. The evidence which supported the positive instance hypothesis is also explicable by the range-seeking and social comparison hypotheses, and that bearing on the desirability hypothesis was minimal and not replicated. The final, new analysis reported above favored the range-seeking and social comparison interpretation. HOWi A procedure (1962.

pp.

629-632)

for

partition was

used

of x2 described by in analyses reported

Sutcliffe in this

(1957) paper.

and

Winter

482

CHARLES

L.

GRUDER

ever, as will become clear in Experiment necessarily mutually exclusive. EXPERIMENT

2, these hypotheses

are not

2

The design of Wheeler et aZ.‘s experiment, partially replicated in Experiment 1, was not ideally suited to a test of the positive instance and desirability hypotheses. In testing both, it was necessary to assume that the highest score in Pos and the lowest score in Neg were equally desirable. Similarly, the highest score in Neg and the lowest score in Pos were assumed to be equally undesirable. However, the trait descriptions were unidimensional with high scores being the positive instance and low scores left undefined. There was no evidence that subjects in Pos and Neg perceived low scores consistently with high scores in Neg and Pos, respectively. In fact, because low scores were undefined, it is reasonable to argue that the above assumptions were poor ones. One purpose of Experiment 2 was to remedy this flaw. All subjects were run in NR. Both the highest and lowest scores were made positive instances for two groups of subjects. For one group the highest score was Pos and the lowest score was Neg, while for the second group the desirability of extreme scores was the reverse. In Wheeler et nl.‘s experiment and in Experiment 1, the single positive instance was always the highest score. To counterbalance this manipulation, two additional groups of subjects were run with only the lowest score as the positive instance. For one group the lowest score was desirable (Pos), while for the other it was undesirable (Neg), It was thought that the four conditions described, considered with the two NR conditions from Experiment 1, would enable a more valid assessment not only of the positive instance and desirability hypotheses, but of the relative importance of these and the range-seeking hypothesis in accounting for choices. Procedure The procedure for the treatment in which only the lowest score was the positive instance was identical to the NR treatment of Experiment 1, except that the lowest score rather than the highest score was ostensibly assigned to the subject possessing the most of the trait being measured. In the treatment in which both the highest and the lowest scores were positive instances, both the desirable and the undesirable traits were described to each subject. Under the assumption that social comparison choices are determined by both positive instance and desirability, the prediction where only the lowest score was a positive instance was that the lowest score would be the most frequent choice when it is desirable, while the lowest and highest scores would be somewhat more equally preferred when the lowest score was undesirable. This prediction is the opposite of that made by the same hypotheses in Experiment 1 because of the change

SOCIAL

COMPARISON

CHOICES

483

of the positive instance from the highest to the lowest score.When both extreme scores were positive instances, however, this factor was constant, and the desirability hypothesis predicted more choices of the most desirable score than the most

tIndesirablescore,regardlessof which washighestand which lowest. Seventy subjects participated score as the positive instance, positive instances.

in Experiment and 35 being

2, 35 being exposed to the lowest exposed to both extreme scores as

Results

Subjects who were exposed to two positive instances rated flexibility (hi = 28.89) as significantly more desirable than rigidity (A4 = 13.14) on a 40-point scale on the postexperimental questionnaire (F( I,33) = 41.85, p < .OOl). Of the subjects exposed only to the lowest score as a positive instance, flexibility was rated more favorably (M = 21.78) than rigidity (Al = 16.53, I?( 1,33) = 2.71, p < .ll). While this latter difl’erence was statistically marginal due to high variability, the previous success of the same manipulation of trait desirability in the study by Wheeler et al. and in Experiment 1 allows confidence in it here. In Table 2 are the percentages of subjects who chose to see each score on their first and second choices. Considering first columns I and II, it is apparent that first choices did not follow the predictions based on the combination of the positive instance and desirability hypotheses, nor are the data opposite to those of the NR conditions in Experiment 1. The data in columns I and II indicate a preference for the most desirable score as a first comparison choice. Where the lowest score was desirable (column I), it was the predominant choice, 66.7% choosing it; this percentage differed significantly (p < ,001, see footnote 2) from the chance expectation ( 12.5%), and from 22.2% [x2( 1) = 5.56, p < .02], the percentage who chose the next most frequently chosen score (the highest score). On the other hand, when the lowest score was undesirable (column II), the highest score was the predominant choice, 58.8% choosing it; this percentage differed significantly (p < .OOOl, see footnote 2) from chance, and marginally from 17.6% ]x’( I) = 2.74, p < .lO], the percentage who chose the next most frequently chosen score (the most similar score in the desirable direction). Consistent with Wheeler et aZ.‘s range-seeking hypothesis, 77.8% in I and 52.9% in II expended both choices on extreme scores. The difference between these proportions was not statistically significant. Turning to the conditions in which both extreme scores were positive instances (Columns III and IV), it is apparent that only where the highest score was desirable and the lowest score undesirable (column III) was the desirability hypothesis unequivocally supported. Both scores were positive instances, but the most desirable (and highest) score was

(!ondiliolt Highest and lowest wore:: are positive instnllc-es Lowest posit.ive

Choice

1:anlc Highest

First

Second

choire

choice

score

Subject

Lowest

score

Highest

score

Subject

Lowest, N

score

(; I I) A H F C! I+: B C+ I I) A H F c 14: H

wore is instance

I Lowest score is Pos

II Lowest, srore is Neg

22.2 0 0 5.5

58.S 0

0 0 .5 6 66.7

3

9

17.6 5 .!) 0 0 11.X x2.5 5.9 0

17.6

III Highest score is Pas Lowest, score is Neg

I\. Highest scclre is Neg Lowest, score is POS

5x.s 17.6 5.9 11,s

“‘2.2 0 0 16.7

0 0 (I 5 0

0 11.1 11.1 3X

1l.S 0 0 17.6

27.5 5.6 0 0

11.8 5 0 0 52.9

11.1

17

1s

9

5 6 11 .l :3x .9

the predominant choice, 58.8% differing significantly (p < .OOOl, see footnote 2) from chance, and marginally (p < .lO) from the 17.6% who chose the next most frequent score (the second highest score). On the other hand, where the highest score was undesirable and the lowest score was desirable (column IV), there was a nonsignificant difference between the number choosing the highest and lowest scores (22.2 and 38.9%, respectively). However, the most desirable (lowest) score was preferred more than would be expected by chance (p < .004, see footnote 2); preference for the least desirable (highest) score, though, did not differ from the chance expectation.

SOCIAL

COMl’ARISON

CHOICES

485

An additional finding, which was consistent with the range-seeking hypothesis, was that 52.9% of the subjects in III and 50% of those in IV expended both choices on the extreme scores. These were each different from the chance expectation. DISCUSSION

It was apparent from the results of the present experiments that none of the three hypotheses alone, range-seeking, positive instance, or desirability, was sufficient to account for the observed social comparison choices under NR conditions. The positive instance and desirability hypotheses were consistent with some findings, but not with others. Given this empirical observation, it was necessary to identify the factors which could account for the deviations from predictions. Data from Wheeler et al.‘s study and from Experiment 1 indicated that subjects, in their desire to establish the range of scores in their group, tended to choose the highest score first. Wheeler et al. suggested that this tendency may have been a result of their subjects estimating the lowest score to be zero, leaving only the highest score unknown. However, Arrowood and Friend reported that data gathered for purposes of a study in progress revealed that only 3 of 287 subjects estimated the obtained score of the lowest ranking person in a group of seven to be zero. Wheeler (personal communication) subsequently suggested that while subjects may not estimate the lowest score to be zero, they may nevertheless be more certain of their estimates of the lowest score than of the highest score. This follows from the fact that tests of various kinds often have zero as the lowest possible score. This convention might cause zero to be an end anchor for a subject’s estimate of the lowest score obtained in the group. Volkmann (1951) has speculated that an anchor may cause the judge to feel more certain of estimates he makes in the vicinity of the anchor. Data from a study to be described below were consistent with this argument in that the lowest and modal estimate of the lowest possible score was zero, with 24 of 61 subjects making it. Implications of this analysis become apparent when the range-seeking hypothesis is viewed in light of Jones and Gerard’s (1967) hypothesis that persons act so as to reduce uncertainty. Subjects who are uncertain of the range of scores in the group are motivated to learn it. Other things equal, if they are more uncertain of the highest score than the lowest score, they should be more’ likely to choose to learn the highest score when given one choice. Data were gathered in thrx NH condition from 61 additional subjects in an attempt to assessthis uncertainty reduction hypothesis; both Pos and Nrg conditions were run. After subjects learned their own score

486

CHARLES

L.

GRUDER

and the rank order within their group, they completed a questionnaire in which the following measures of uncertainty were obtained: estimates of the ranges in which the highest and lowest actual scores fell and the exact score of each member of the group. In addition, subjects rated how likely it was that each of their estimates was accurate; they used an 11-point scale, 0 indicating low likelihood of accuracy, and 10 indicating high likelihood. Following this, subjects were allowed to make two social comparison choices in the same manner as in Experiments 1 and 2. One aspect of the uncertainty reduction hypothesis was that subjects would be less certain of the highest score than the lowest score. Specifically, subjects were expected to estimate a wider range for the highest score, and to judge these estimates and estimates of the highest score as less likely to be accurate. After a square-root transformation was performed on the range estimates in order to reduce the nonhomogeneity of the within-cell variances, an analysis of variance showed that the estimated range of the highest score (A4 = 90.02) was significantly wider than the estimated range of the lowest score (M = 76.86, F( 1,57) = 4.73, p < .05). Th ere were no differences on either set of accuracy ratings. Another analysis of interest was whether any of these measures of uncertainty showed a significant correlation with social comparison score choices. The fact that none did may at first seem quite damaging to the uncertainty reduction hypothesis. However, it must be remembered that this search for an additional causa1 factor was necessitated by the failure of the desirability and positive instance factors to account for observed data. Uncertainty is conceived in the present scheme as a third determinant of social comparison choices, which operates in conjunction with the other determinants. What follows is speculation as to how these three factors might be jointly operating to determine the pattern of social comparison choices observed in Tables 1 and 2. A follow-up experiment in which certainty is manipulated might reveal its individual effects in the present context, as is has been shown to affect social comparison choices in other contexts (Gerard, 1963; Gerard & Rabbie, 1961; Rabbie, 1964). The findings in these experiments suggested the following multifactor model to account for social comparison choices when the range of scores on the characteristic being measured was unknown to the chooser. In each experimental condition the presence or absence of each of three attribiites of the extreme scores was identified: positive instance, desir;ll)lc~. and highest ( uilcertnin) . Each attribute was weighted equally (Table 3). Using a siinplc majority rule for thca pr~~scncc of these

score

Lowest

SCOW

Highest

Score

score

Desirable Positive

Uncertain

1 inst,ance 2

1

1

I Lowest score is Pos

Lowest

1

instance

Positive

Uncertain T)esirable

1

illslance

2

1 1

II Lowest score is Neg -___~-

is positive

1

Positive

Positive

1

installre

instance 3

1 1

III Highest score is Pos Lowest score is Neg Uncertain Iksirable

SCORICS

1

I

are

Ilesirable Positive

Uncertain Positive

1 instance 2

2

1 instance

Highest score is Neg Lowest score is Pos

IV

and lowest scores positive instances

Condition

TABLE 3 OF EXTREME

Highest,

ATTRIBUTES

1

1

score

instance 3

Positive

0

1 1

Unrertain Desirable

\ Highest score is Pos

Highest -.~-___

1

VI

in$txllce

I)esirable

Positive

Ullrertaill

1

2 1

1

instnllce

Highest score is Xeg

is positive

1

2

0 R g

8 E z z?

F

m 8

488

CHAHIXS

1,. GRUDEll

attributes, it was possible to predict the predominant first choices in the NR conditions of Experiment 1 and in Experiment 2. For example, it can be seen from column I of Table 3 that the lowest score would have been predicted as the predominant choice because it was both a positive instance and desirable, while the highest score was merely a more uncertain element. In column II, on the other hand, the lowest score was only a positive instance, while the highest score was both more uncertain and desirable; therefore, the latter should have been preferred. Following this procedure for each of the other conditions, and comparing the predictions with the data from the NR conditions in Table 1 and from Table 2, it is apparent that these three attributes were accurate predictors of social comparison choices when the range was unknown. Although, from the logic of this argument, it would seem that this simple model should have predicted the relative magnitudes of the preferences, it did not. Perhaps if there were some rationale for differential weighting of these three attributes, this sophistication would enable more precise prediction. Another possibility, of course, is that as yet unidentified factors play roles in determining social comparison choices in the situation studies. REFERENCES ARROWOOD,

parison FESTINGER,

A. J., & FHIEND, R. Other factors determining the choice of a comother. Journal of Experimental Social Psychology, 1969, 5, 233-239. L. A theory of social comparison processes. Human Relations, 1954, 7,

117-140. H. B. Emotional uncertanity and social comparison. JozlmaZ of Abnormal and Social Psychology, 1963, 66, 568-573. GERAHD, H. B., & RABBIE, J. M. Fear and social comparison. Journal of Abnormal and Social Psychology, 1961, 62, 586-592. JONES, E. E., & GERARD, H. B. Foundations of social psychology. New York: Wiley, 1967. KEMPTHORNE, 0. The design and analysis of experiments. New York: Wiley, 1952. LATAN& B. (Ed.) Studies in social comparison. Journal of Experimental Social Psychology, 1966, Suppl. 1. PETTIGREW, T. F. Social evaluation theory: Convergences and appIications. Nebraska Symposium on Motizjafion. Lincoln: Univ. of Nebraska Press, 1967. Pp. 241318. RABBIE, J. M. Differential preference for companionship under threat. Journal of Abnormal and Social Psychology, 1964, 67, 643-648. SNEDECOR, G. W., & COCHRAN, W. Statistical methods. (6th ed.) Ames: Iowa State Univ. Press, 1967. SUTCLIFFE, J. P. A general method of analysis of frequency data for multiple classification designs. Psychological Bulletin, 1957, 54, 134-137. THORNTON, D. A., & ARROWOOD, A. J. Self-evaluation, self-enhancement, and the locus of social comparison. Journal of Experimental Social Psychology, 1966. SuppI. 1, 40-48. GERARD,

SOCIAI,

CORIPARISON

CHOlCES

489

VOLKMANN, J. Scales of judgment and their implications for social psychology. In J. H. Rohrer and M. Sherif (Eds.), Social psychology at the crossroads. New York: Harper, 1951. Pp. 27s’294. WALKER, H., & LEV, J. Statistical inference. New York: Holt, 1953. WHEELER, L., SHAVER, K. G., JONES, R. A., GOETHALS, C. R., COOPER, J., ROBINSON, J. E., GHUDER, C. L., & BUTZINE, K. W. Factors determining choice of a comparison other. Journal of Experimental Social Psychology, 1969, 5, 219-232. WINEW, B. J. Statistical principles in experimental design. New York: McGraw-Hill, 1962. (Received

July

10,

1970)