Contrasting predictions of Byrne's similarity-attraction hypothesis and Rosenbaum's dissimilarity-repulsion
            hypothesis were tested with 7-, 11-, 15-, and 21-year-olds in Singapore. The study included a control condition of
            no-attitude information and two experimental conditions of similar and dissimilar attitudes. Measures of attraction,
            assumed similarity of attitudes, and accuracy in perceiving the manipulations were taken. The repulsion hypothesis
            was supported with the two younger groups; the attraction hypothesis was supported with the two older groups.
            The repulsion effect emerged because the two younger groups assumed a high level of attitudinal similarity in the
            control condition of no-attitude information and because they inaccurately perceived the manipulated similarity of
            attitudes in the experimental conditions. These results reaffirm the similarity-attraction hypothesis and further
            demonstrate the role of age-related cognitive processes in interpersonal attraction.

One of the empirical regularities in personality and social psychology is the relationship between attitudinal similarity and interpersonal attraction. It has repeatedly been demonstrated that people like those who hold similar attitudes and
opinions (e.g., Byrne, 1961, 1971; Byrne & Griffitt, 1966; Byrne & Nelson, 1965; Clore & Baldridge, 1968; Condon & Crano, 1988; Singh, 1974; Tesser, 1993). Most textbooks of social psychology include a discussion on this similarity-attraction link.

However, Rosenbaum (1986b) proposed a repulsion interpretation for this well-established relationship. According to
Rosenbaum, attitude similarity does not lead to liking, but attitude dissimilarity does indeed lead to repulsion. Support for his hypothesis comes from the finding that a control condition of nonattitudinal information, such as a photograph alone, differs from an experimental condition of dissimilar attitudes plus photograph but not from in experimental condition of similar attitudes plus photograph. Similarly, attraction toward a person described by the individual's personality alone differs when the same personality profile is paired with a political affiliation dissimilar from that of the subject but not when it is paired with a political affiliation similar to that of the subject.

Subsequent studies testing the contrasting predictions of the similarity-attraction and dissimilarity-repulsion hypotheses
have, however, reaffirmed the former and rejected the latter. Smeaton, Byrne, and Murnen (1989), for example, held the number of dissimilar attitudes constant at 8 but varied the number of similar attitudes from 0 to 62. As predicted by the similarity-attraction hypothesis, attraction increased as the number of similar attitudes increased.

Singh and Tan (1992) asked subjects to first judge a stranger, a randomly sampled same-sex university student, in a control condition of no-attitude information. Later, the subjects judged the same student again, knowing that the stranger shared 0.00, 0.50, and 1.00 proportion of similar attitudes with them. Unlike Rosenbaum's (1986b) experiments, this two-stage design contrasted the control condition of no-attitude information with the experimental condition of attitude information alone. Attraction responses in such a control condition of no-attitude information differed significantly from the attraction responses in the experimental conditions, showing that both similar and dissimilar attitudes affected attraction. In an auxiliary between-subjects experiment, attraction response was also higher in the experimental condition of similar attitudes than in the control condition of no-attitude information. These results agree with the similarity-attraction hypothesis but disagree with the dissimilarity-repulsion hypothesis.

The experiments by Singh and Tan (1992), following the suggestion by Byrne, Clore, and Smeaton (1986) and Smeaton et al. (1989), had also taken measures of assumed similarity of attitudes with the stranger in the control condition of no-attitude information. They found that the relative effects of similar attitudes (1.00 proportion of similar attitudes) and dissimilar attitudes (0.00 proportion of similar attitudes) primarily depend on how discrepant their levels are from the assumed similarity value of .73 in the control condition. Such discrepancies for the dissimilar and similar attitudes were .73 and .27, respectively. Naturally, therefore, the effects of dissimilar attitudes on repulsion were nearly three times greater than those of similar attitudes on attraction. One theoretical implication of this finding is that the variables that can affect the level of assumed similarity can also affect the similarity-attraction relationship.

The main purpose of the present study is to show that the cognitive processes of assumed similarity of attitudes (Byrne & Wong, 1962; Hoyle, 1993; Smeaton, Rupp,Vig, & Byrne, 1993) and the accurate perception of the manipulated
similarity of attitudes are crucial factors in the similarity-attraction relationship. When these two factors are considered
together, the lack of similarity effect (Rosenbaum, 1986b) and the greater effectiveness of dissimilar than similar attitudes (Singh & Tan, 1992) in interpersonal attraction follow from the similarity-attraction hypothesis as well.

To support the position just stated, an experiment patterned after Experiment 2 of Singh and Tan (1992) was performed on subjects varying in age. It had three between-group conditions--a control condition of no-attitude information and two experimental conditions of similar and dissimilar attitudes. For such an experiment, the similarity-attraction hypothesis predicts that attraction response of the control condition of no-attitude information should differ from the responses of the experimental conditions of similar and dissimilar attitudes depending on the level of similarity already assumed in the control condition. In contrast, the dissimilarity-repulsion hypothesis, which does not attach any importance to assumed similarity (Rosenbaum, 1986a), predicts that the control condition of no-attitude information should differ from the experimental condition of dissimilar attitudes only.

The main virtue of the present experiment lies in its developmental perspective. Attraction literature shows that people
assume similarity of attitudes in the absence of attitudinal information about the stranger (Byrne & Wong, 1962) and that such assumed similarity can decrease the effects of similar attitudes on attraction but enhance the effects of dissimilar attitudes on repulsion (Hoyle, 1993; Singh & Tan, 1992; Smeaton et al., 1993). Furthermore, 9-year-olds give significantly higher attraction ratings to a stranger than do adults (Byrne & Griffitt, 1966). Such positive evaluations of stimulus persons by young children have also been reported from the developmental studies of how children form impressions (Rholes & Ruble, 1988) and expectations (Rholes & Ruble, 1984) of others. By varying the age of subjects, therefore, it may be possible to obtain evidence seemingly supportive of both the attraction and repulsion hypotheses at different age levels and to account for both kinds of evidence within the same similarity-attraction framework.

Peevers and Secord's (1973) results are directly relevant to the present issue of the role of age in interpersonal attraction. They asked kindergartners, third and seventh graders, high school juniors, and university students to describe in their own words three friends and one disliked peer. One of the dependent variables derived from content analyses was personal involvement, which refers "to the degree to which the describer involved himself in item descriptive of another person" (p. 124). Statements like "He gave me a cookie," "We go bike-riding together," and "He has blue eyes," for example, were coded as egocentric mutual and other oriented, respectively, representing three degrees of personal involvement. It was found that the percentages of egocentric responses decreased but those of other-oriented responses increased with age. From this age trend in personal involvement in describing others, it may be speculated that assumed similarity of attitudes with strangers, also considered as an index of egocentricity, should decrease with age.

If younger children gossip about attitude issues with their peers less than do older children (Gottman & Mettetal, 1986), then awareness of diversity of opinions held about any issue would be lower at younger than older ages. In inferring attitudes of peers, therefore, young children may just project their own attitudes to strangers. Such an egocentricity would result in higher levels of assumed similarity of attitudes at younger than older ages. Avery high assumed similarity and hence no difference between the assumed and the manipulated similarity will render negligible the difference between the attraction responses of the no-attitude-information control condition and of the similar-attitudes experimental condition. This outcome will obviously support the repulsion hypothesis, a possibility suggested by Byrne et al. (1986) in their reply to Rosenbaum (1986b).

On the other hand, teenagers and adults generally gossip about diverse issues with their peers over time (Gottman &
Mettetal, 1986). Having discussed a given issue with numerous people, they become aware of the diversity of opinions held about it. An awareness of diversity in attitudes and opinions held by others plus an age-related decline in positive responses to others but age-related increase in other-oriented responses may lower the level of assumed similarity of attitudes in adolescents and adults. That will clearly widen the gap between the assumed similarity in the control condition of no-attitude information and the manipulated similarity of attitudes in the experimental condition. Consequently, the difference between the attraction responses of the control and experimental conditions will differ just as the similarity-attraction hypothesis posits.

If the foregoing reasoning is correct, then the similarity-attraction relationship should be moderated by the age of the
subjects. In addition to affecting assumed similarity of attitudes in the no-attitude-information control condition, age of
subjects may also be expected to influence the accuracy in perceiving the manipulated similarity of attitudes. Generally,
the manipulated attitudinal information can be expected to be effective only to the extent it is perceived accurately (Duck, 1973) and differs from the assumed similarity of attitudes in the control condition of no-attitude information (Hoyle, 1993; Singh & Tan, 1992). When children assume a high level of similarity between the attitudes of the stranger and their own but fail to perceive the manipulated similarity accurately, the manipulated similarity obviously cannot influence attraction. Thus data that appear to support the repulsion hypothesis would actually be attributable to the cognitive processes of assumed similarity of attitudes and accuracy in perceiving the manipulated similarity.

Despite the potential contributions of a developmental analysis of interpersonal attraction, there is a paucity of
developmental studies in the literature. The one study (Byrne & Griffitt, 1966) that did examine age differences in the
attitudinal similarity-attraction relationship did not find them. The authors noted that "the law of attraction is as strongly
operative by 9 years of age as it is in young adulthood" (p. 702). This conclusion also agrees with the naturalistic
observations that the first time two unacquainted third graders meet, they normally begin by asking one another questions directed at discovering whether they have any common attitudes and orientations (Furman & Childs, 1981). Nevertheless, the lack of age differences in the attitudinal similarity-attraction relationship is surprising, for friendship means the sharing of material goods or overt fun activities to younger children but of private thoughts and feelings to the older ones (Youniss & Volpe, 1978).

A careful look at Byrne and Griffitt's (1966) method discloses at least two weaknesses that may have reduced the power of their developmental investigation. One is that they used different sets of attitude issues with their two age groups to ensure the relevance of the issues to the particular group. However, using different sets of attitude issues with different age groups confounded attitude issues with age of subjects. Another, more important weakness is that the manipulations of attitudinal similarity did not include the extreme levels of 0.00 and 1.00 proportion of similar attitudes. Singh (1974, Table 1, p. 299) shows that an interaction effect in attraction function is more likely when these extreme levels are included in the design than when they are excluded.

It must be noted that Byrne and Griffitt's (1966) study was not primarily concerned with the issue of age differences in the similarity-attraction and dissimilarity-repulsion processes. It did not consider the different levels of assumed similarity of attitudes in the two age groups, nor did it include a control condition of no-attitude information. Without such a baseline, it is impossible to judge whether the similarity effect they demonstrated was based on attraction, repulsion, or both.

On several grounds, the experiment reported here can be regarded as a markedly improved test of the hypothesis of age differences in responding to attitudinal information. First, the age groups represented a wider range (7 to 21 years) than those of Byrne and Griffitt (1966; 9 to 20 years). Second, subjects in all four age groups were administered the same attitude survey consisting of issues relevant to all. Third, similarity manipulations were at the two extreme levels of 0.00 and 1.00. Fourth, the control condition of no-attitude information served as a base for comparison with the experimental conditions of similar and dissimilar attitudes. It also assessed the subjects' assumed similarity of attitudes. Finally, accuracy in perceiving the manipulated similarity was measured to explore the role of an additional cognitive process in the similarity-attraction relationship.

Subjects
Subjects were 180 males and 180 females from four age groups. Each age group had 45 males and 45 females who were randomly assigned to one of three conditions, each consisting of 15 males and 15 females. The three groups of children were from the primary and secondary schools of Singapore. The subjects in the adult group were from the National University of Singapore. Mean ages of the four groups of subjects were 6.90, 11.20, 14.94, and 20.50 years; respective standard deviations were 0.34, 0.66, 0.66, and 1.96. These groups of subjects will be referred to as 7-, 11-, 15-, and 21-year-olds throughout this article.

Design
The design was a 4 x 2 x 3 factorial, with age and sex of subjects as the first two classification factors and attitudinal
information (no-attitude information, all dissimilar attitudes, and all similar attitudes) as a third treatment factor. As in
previous studies (Byrne, 1971; Singh, 1973), sex of the subjects did not produce any difference in attraction responses, and the dissimilarity-repulsion hypothesis (Rosenbaum, 1986a, 1986b) does not make differential predictions for men and women. Hence sex was dropped as a factor from all the subsequent analyses. That resulted in 30 subjects in each cell of the 4 x 3 analysis.

Experimental Materials
Attitude survey. Attitudes were operationalized as responses to a 23-item attitude survey. It included 8 items from Byrne (1971), 4 from Singh and Tan (1992), and 11 new items. The new items were generated by interviewing five young children. The wording of all items was simplified to make the statements understandable to children. A pretest of item relevance with at least five subjects from each of the four age groups resulted in an elimination of eight items. Thus the final attitude survey consisted of 15 issues: watching TV, smoking, sports, sharing of housework, reading books, going to parties, working mothers, homework, fast food, fighting, money, pets, one-child families, boys and girls, and drama serials on the local channel. Each issue was expressed by six statements, without the neutral statement, just as in other attraction studies (Byrne, 1971; Drigotas, 1993; Rosenbaum, 1986b; Singh & Tan, 1992; Smeaton et al., 1989). The 6 points of each scale were represented by six equidistant boxes and their corresponding statements in six different rows. All 360 subjects participated in the attitude survey 2 weeks prior to the experimental session.

Experimental booklet. A specifically tailored experimental booklet was prepared for each subject. Attitude similarity
between the subject and the stranger was manipulated via a bogus survey form allegedly completed by the stranger. In the similarity condition, the simulated responses of the stranger to all 15 items of the attitude survey were on the same side and just one box away from the subjects' own responses. In the dissimilarity condition, the simulated responses were three boxes away from the subjects' own responses and on the other side of the scale. This method of constant discrepancy in manipulating attitudinal dissimilarity and similarity has been widely adopted (e.g., Byrne, 1971; Drigotas, 1993; Rosenbaum, 1986b; Singh & Tan, 1992; Smeaton et al., 1989). There was, of course, no such manipulation for the subjects assigned to the no-attitude-information control condition.

Attraction measure. The attraction of the subject toward the stranger was measured by a modified version of Byrne's
(1971) Interpersonal Judgment Scale (IJS). The modified IJS consisted of four 7-point items that dealt with general
knowledge, intelligence, personal feelings, and enjoyment of company. Each rating scale was represented by seven
equidistant rectangular boxes numbered 1 (very low) to 7 (very high). Subjects were required to shade one of the seven boxes of each scale to indicate their responses. These simplifications and the removal of the items of morality and adjustment from the original IJS were intended to enable subjects from the younger age groups to comprehend the task requirements. A pretest of the scale with children of the two younger groups indicated that they understood well the items of the modified IJS, and their ratings of a best friend, a neighbor they did not know very well, and someone they could not get along with well were appropriately ordered.

Responses to the last two items--namely, personal feelings and enjoyment of company--were summed to yield a measure of attraction. The attraction score ranged from 2 (very low) to 14 (very high). The coefficients of internal consistency reliability of the attraction measure for the 7-, 11-, 15-, and 21-year-olds were .55, .74, .89, and .93, respectively. Although the coefficient of reliability for the 7-year olds is lower than what is normally acceptable, a uniform effect of the manipulations of attitudinal information, F(2, 87) = 3.62, p < .05, was present on ratings of both the personal feelings and enjoyment of company, F(2, 87) = 0.30. No interaction between the attitudinal information and two types of ratings indicates that both items of personal feelings and enjoyment of company measured the same construct. Accordingly, the modified IJS was adjudged as a suitable measure of attraction for all four age groups.

Procedure
The 7-year-olds were run in groups of six at a time to facilitate understanding of the task. Other subjects were run in groups of approximately 30 at a time.

In the no-attitude-information control condition, subjects received an instruction sheet and the IJS. They were told (a) that a person of their own age and sex was selected for them, (b) that the person was not of the same institution as the subjects, and (c) that subjects did not know the stranger personally. The task was to form an opinion of the stranger and judge that person by shading an appropriate box of each of the four items of the IJS. After completing the IJS, the subjects completed the 154 tem attitude survey in a way they thought the stranger would respond. A comparison of the responses to this survey with those already given during the first survey yielded an index of assumed similarity of attitudes between the subject and the stranger.

In each experimental session, a nearly equal number of subjects from the similarity and dissimilarity conditions were run simultaneously to counter possible effects of experimenters' expectancy (Rosenthal, 1966). Subjects received an instruction sheet, a simulated bogus attitude survey, and the IJS. They were asked to judge the person on the basis of responses to the attitude survey. To make the manipulations convincing, the portion containing the respondent's background information was cut away from the bogus attitude survey. Subjects were told that this was done to preserve anonymity of the stranger. The remaining information about the stranger was similar to that of the control condition. After completing the IJS, subjects were handed a blank attitude survey and asked to recall how the stranger had filled in the survey. This was intended to find out age differences in accurate perception of the manipulations (Singh & Tan, 1992).

In both the control and experimental conditions, necessary precautions were taken to ensure that subjects of the younger groups comprehended the task. The instructions were elaborated orally to promote understanding of the task. Furthermore, an introduction of the IJS was made prior to the data collection. An example on "looks" was given by the experimenter to illustrate the different levels of response along the scale. After data collection, subjects were informed of the study's purpose and debriefed.

Attraction and Repulsion Hypotheses
According to the similarity-attraction hypothesis, both similar and dissimilar attitudes affect attraction. Hence the
control condition of no-attitude information should differ from both the experimental conditions of similar and dissimilar
attitudes. In contrast, the dissimilarity-repulsion hypothesis regards similar attitudes as irrelevant for relationship
development but dissimilar ones as potent causes of repulsion. Therefore, it predicts that the control condition of no-attitude information should differ from only the experimental condition of dissimilar attitudes.

Results yielded evidence seemingly supporting each hypothesis, but at different age levels. The similarity hypothesis was supported with the 15- and 21-year-olds; the repulsion hypothesis was supported with the two younger groups. Thus the Age of Subjects x Attitudinal Information effect was statistically significant, F(6, 348) = 7.12, p < .001.

It can be seen that the experimental condition of dissimilar attitudes yielded lower mean attraction scores than the control condition of no-attitude information across all four age groups, F(1, 348) = 11.81, 194.40, 453.75, and 163.35, ps < .001, for 7-, 11-, 15-, and 21-year-olds, respectively. These differences, detected by planned comparisons, are as predicted by both the attraction and repulsion hypotheses.

Planned comparisons between the means of the control condition of no-attitude information and the experimental condition of similar attitudes provided apparently critical tests between the two hypotheses. The two means differed at the level of the 15 - and 21-year-olds, F(1,348) = 7.35 and 60.00, respectively, ps <. 001, but not at the level of the 7-and 11-year-olds, F(1,348) = 0.12 and 1.35, respectively, ns. Hence responses of the two younger groups are consistent with the dissimilarity-repulsion hypothesis; those of the two older groups are consistent with the similarity-attraction hypothesis.

On the surface level, these age differences suggest that interpersonal evaluation undergoes developmental changes in which only repulsion is relevant at first, whereas both attraction and repulsion become operative as the individual matures (Rosenbaum, 1986a). Such an interpretation would obviously be at variance with the recent finding that adults tend to initially include similar others and then exclude dissimilar others in their inclusion/exclusion decisions on group membership (Drigotas, 1993). Before taking the developmental interpretation seriously, therefore, it is necessary to examine whether the lack of evidence for the similarity-attraction hypothesis with the two younger groups of subjects can be accounted for by age differences in assumed similarity of attitudes in the control condition of no-attitude information and in the accuracy of perceiving similarity in the experimental conditions of similar and dissimilar attitudes.

Age Differences in Assumed and Perceived Similarity of Attitudes
The repulsion interpretation of the lack of similarity effect in the two younger groups will be bolstered if they are shown to be similar to the older groups with respect to (a) assumed similarity of attitudes in the control condition and (b) perception of manipulated similarity in the experimental conditions. Results from these measures yielded unambiguous evidence against the repulsion hypothesis.

Three aspects of the age trends displayed in Figure 2 are of immediate interest. First, there is an age difference in assumed similarity of attitudes in the control condition of no-attitude information, F(3,348) = 6.84, p < .001. The mean assumed similarity of attitudes were .86, .86, .82, and .74 for the 7-, 11-, 15-, and 21-year-olds, respectively. Multiple comparisons by the Newman-Keuls test further revealed that mean assumed similarity of the 21-year-olds differed significantly (p <.05) from all the three younger groups, which did not differ among themselves.

Second, age of subjects played an important role in perceiving the manipulated similarity of attitudes as well. The mean
perceived similarity in the condition of 0.00 similar attitudes (i.e., all dissimilar attitudes) were .44, .21, .07, and .02 for
the 7-, 11-, 15-, and 21-year-olds, respectively. The corresponding means in the condition of 1.00 similar attitudes were .86, .89, .96, and .98. The age difference was statistically significant in both the dissimilar and similar conditions, F(3, 348) = 77.57 and 6.65, ps < .001. In the condition of dissimilar attitudes, the two younger groups differed from each other as well as from the two older groups, which did not differ. In the condition of similar attitudes, the two younger groups perceived the manipulated similarity less accurately than did the two older groups.

Third, and the most important, the difference between the level of assumed similarity obtained in the control condition  of no-attitude information and the level of perceived similarity in the experimental condition of dissimilar attitudes was
uniformly present across all the four age groups, F(1, 348) = 188.70, 441.43, 606.97, and 553.89, ps < .0001, for 7-, 11-, 15-, and 21-year-olds, respectively. However, the picture was different with comparisons involving the experimental condition of similar attitudes. The two younger groups did not perceive the manipulated similarity as any different from their assumed similarity, F(1,348) = 0.00, and 1.31, ns, for 7- and 11-year-olds, respectively. However, the two older groups did perceive the manipulated similarity to be closer to 1.00. This produced a significant difference between their assumed and perceived similarity, F(1,348) = 20.31 and 63.09, ps < .0001, for 15- and 21-year-olds, respectively. Accordingly, the absence of a similarity effect in the two younger groups of subjects can be accounted for, as was hypothesized, by their inaccurate perception of the manipulated similarity.

The foregoing results are theoretically important in three ways. First, the finding of .74 assumed similarity of attitudes in
the 21-year-old university students is identical to that found by Byrne and Wong (1962) in the United States and by Singh and Tan (1992) in Singapore. This convergence of findings ensures that the experimental task invoked similar process across cultures. Second, there is an age trend in assumed similarity of attitudes: The assumed similarity of attitudes decreases with age. It is not surprising, therefore, to find the similarity effect getting stronger as age advances. Finally, the 15-year-olds assumed nearly the same level of similarity of attitudes with the stranger as did the two younger groups of subjects. Nevertheless, they were more accurate, as compared to the two younger groups, in perceiving the manipulated similarity of attitudes. Consequently, the difference between their assumed and perceived similarity widened, and the attraction responses of the corresponding conditions differed. This indicates that both level of assumed similarity and accurate perception of manipulated similarity play essential roles in engendering the similarity-attraction effect.

Further Evidence Against the Repulsion Hypothesis
In the overall analysis of variance of attraction measure, the effect of subjects' age was statistically significant, F(3, 348) = 40.87, p < .0001. As already noted, the Age of Subjects x Condition of Attitudinal Information effect was also significant. Hence the nature of the age effect was further examined.

If the dissimilarity-repulsion hypothesis has any merit, then the age effect should be in the condition of dissimilar attitudes but not in the condition of similar attitudes. In other words, the factorial plot of the age effect across the dissimilar and similar levels of attitudinal information should yield a total convergence at the level of similar attitudes. This would happen because the age effect will be in the repulsion produced by dissimilar attitudes but not in the attraction based on "irrelevant" attitudinal similarity. The pattern of results in this study does not meet this requirement of the repulsion hypothesis. Tests of the simple effects of age at the two levels of attitudinal information further showed that the age differences are present in both the dissimilarity and similarity conditions, F(3, 232) = 35.28 and 4.34, respectively, ps < .01. On this basis also, the repulsion hypothesis can be rejected.

Three other aspects of the age differences also deserve attention. First, the attraction curve of the7-year-olds had the highest elevation and the shallowest slope, which indicates that the 7-year-olds are qualitatively different from the remaining three age groups of this study. In statistical analyses, the slope of the attraction function of the 7-year-olds differed significantly from that of the 15- and 21-year-olds, F(1,232) = 26.84 and 16.72, ps < .001, respectively, but only marginally from that of the 11-year-olds, F(1,232) = 3.07, p< .10. As the two younger groups differed strongly in intercept, F(1,232) = 45.41, p < .0001, and the marginally significant difference in their slopes had the expected trend, the original interpretation made from the pattern in graph about the 7-year-olds was accepted without any
qualification.

Second, the attraction curves for 15- and 21-year-olds have nearly similar elevation and slope, F(1, 232) = 3.21 and
1.89, ns. Perhaps the relationship between proportion of similar attitudes and interpersonal attraction attains stability
beginning at about 15 years of age.

Finally, the slope of the attraction curve for the 11-year-olds is statistically shallower than that of the 15- and
21-year-olds, F(1, 232) = 11.75 and 5.46, ps < .01, respectively. This difference implies that the attitudinal information is less important for the 11-year-olds than for the 15-year-olds in forming interpersonal relationships.

There are three key findings of the present research. First, it is possible to obtain evidence apparently supportive of both the similarity-attraction and dissimilarity-repulsion hypotheses using the same method. Comparisons of attraction
response from the control condition of no-attitude information with attraction responses from the experimental conditions of similar attitudes and of dissimilar attitudes seemingly supported the dissimilarity-repulsion hypothesis with the 7-and 11-year-olds and the similarity-attraction hypothesis with the 15-and 21-year-olds. This could be taken to mean that the evidence suggesting that attitudinal similarity is irrelevant for attraction but that attitudinal dissimilarity does lead to repulsion is not confined to only the methods used by Rosenbaum (1986b).

Second, the similarity-attraction hypothesis can actually account for the evidence that ostensibly supports the
dissimilarity-repulsion hypothesis. An earlier explanation of the weak or no-similarity effect considered only assumed
similarity of attitudes in the condition of no-attitude information (Byrne et al., 1986; Hoyle, 1993; Singh & Tan, 1992;
Sineaton et al., 1989). Although the present results confirm the significance of assumed similarity of attitudes, they further indicate that an accurate perception of the manipulated similarity is also necessary and no less important. The repulsion effect emerged in the two younger groups not because they were unaffected by attitudinal similarity but because they assumed a high level of attitudinal similarity and because they inaccurately perceived the manipulated similarity as lower than what it was. In these two cases, therefore, the control and experimental conditions basically made what Byrne et al. (1986, p. 1168) call a comparison of "similarity with similarity." In fact, when the assumed and perceived similarity in the two conditions differed (as with the two older groups), the well-known similarity-attraction relationship reemerged.

Third, age plays a much more important role in interpersonal attraction than accorded to it previously (Byrne & Griffitt, 1966). Not only assumed similarity but also accuracy of attitudinal similarity manipulations varied as a function of age. Whereas the former decreased with age, the latter increased. As a result, the similarity effect was stronger with the older than with the younger subjects. This confirms the prediction that the age of the subjects is a moderator of the
similarity-attraction relationship. No less important, the present findings of developmental differences in assumed
similarity and in accurate perception of the manipulations are novel additions to the literature.

What is the relationship between age and interpersonal attraction? Results presented in this article identify two age trends. One is that the 7-year-olds are indeed positive in their interpersonal orientation: They believe that others are very similar to them. This positive bias perhaps prevents them from seeing others accurately. That is why they did not perceive a similar stranger as more similar than was assumed without any attitudinal cues. Such a positive orientation may have also distorted their perception of the dissimilar stranger who was not perceived as having totally dissimilar attitudes. These results portray the 7-year-olds as egocentric: They excessively rely on what they assume about others and show relatively little awareness of the external information available about them. This result provides an independent confirmation of the finding that interpersonal judgments by children are biased toward the maintenance of a positive view of others (Rholes & Ruble, 1988).

Another, and more important, developmental difference appeared in the relative effectiveness of attitudinal information. The attraction function of the 7-year-olds had the shallowest slope, which implies that the attitudinal information was of low importance to them. This interpretation was also buttressed by the fact that their data did not fit to the well-known
straight-line relationship between proportion of similar attitudes and attraction. The steepness of the attraction function
increased slightly from 7 to 11 years but strongly from 11 to 15 years. There was no difference between 15 and 21 years at all. These age differences in the slope of the attraction curve suggest that the effectiveness of attitudinal information, at least in Singapore, steadily increases up to the age of 15 years and stabilizes thereafter. It is natural, therefore, that the attitudes of the stranger were among the most preferred categories of information sought by university students in the acquaintance process (Singh & Tan, 1992).

If subjects aged 15 and older indeed value similarity of attitudes in relationship formation, then attraction response in the condition of similar attitudes should always be higher than in the control condition of no-attitude information. Contrary to this requirement, neither Rosenbaum (1986b) nor Hoyle (1993) found such a difference with their subjects from a university population. Two reasons may account for the lack of similarity effect with American adults of these studies. First, the manipulated proportion of attitudinal similarity was about .80 and not 1.00, which is necessary to yield the similarity effect (Singh, 1974). Both the experiments of Singh and Tan (1992) and the present experiment obtained similarity effect because the manipulated proportion of similar attitudes was at the extreme level of 1.00.

Second, and no less important, even if the manipulated proportion of similarity was 1.00 in Experiment 2 of Rosenbaum (1986b), the comparison was made between a control condition of nonattitudinal information, such as personality traits, and an experimental condition of personality traits and a similar political affiliation. According to Byrne et al. (1986), the so-called no-attitude control condition of Rosenbaum was actually a nonattitudinal-information condition because it included other information whose impact is unknown. To obtain the effects of similar attitudes, the comparison must be between no-attitude and similar-attitude conditions. When such comparisons are in fact made, as in Singh and Tan (1992) and in the present study, the similarity-attraction relation emerges rather unambiguously. Although Hoyle (1993) did not find a statistically significant difference between attraction responses of his control condition of no-attitude information and experimental condition of .75 similar attitudes, the pattern of differences between assumed and perceived similarity led him also to conclude for similarity-attraction and against the dissimilarity-repulsion hypothesis.

From the differences between the results of studies mentioned above, it may be concluded that both similar and dissimilar attitudes affect attraction but that the similarity effect is detectable only when comparisons are made between a control condition of no-attitude information and an experimental condition of all similar attitudes. This conclusion has two obvious implications. One is that it is possible to experimentally create an adequate control condition of no-attitude information contrary to the reservations expressed by Rosenbaum (1986b) as well as by Byrne et al. (1986). As long as subjects are informed that the stranger is randomly picked from their own in-group, such as a university or class, they readily assume similarity of attitudes and judge the stranger on the basis of this assumed similarity. This strategy is attested by the fact that adult subjects of both Singh and Tan (1992) and Hoyle (1993, p. 313 n. 2) rendered sensible judgments in a no-attitude control condition. Results of the present study further extend the adequacy of such a control condition from university adults to primary school children.

Another implication is that it may be relatively easier to obtain evidence for the dissimilarity-repulsion hypothesis than for the similarity-attraction hypothesis. Both the possibilities of (a) a high level of assumed similarity of attitudes with strangers from one's in-group and (b) rarely encountering a stranger with all similar attitudes in real life may make the
similarity-attraction relationship rather elusive. On the contrary, dissimilarity of attitudes is more pervasive in everyday
life. When dissimilarity is encountered, it usually differs markedly from the level of assumed similarity. Hence the repulsion effects of dissimilar attitudes generally appear much stronger than the attraction effects of similar attitudes (Singh & Tan, 1992). Similar attitudes have greater impact on attraction relative to dissimilar ones on repulsion only when the initial level of assumed similarity is rather low (Smeaton et al., 1993). From this vantage, one of the contributions of the present study lies in showing that an extreme manipulation of attitudinal similarity is necessary to yield the minimal attraction effect relative to already assumed similarity and that the likelihood of finding out repulsion effects is higher, as dissimilarities of attitudes, opinions, and values are facts of real life.

"The debate about 'similarity and attraction,'" note Duck and Barnes (1992), "has been the inverted intellectual Titanic
of the last thirty years: Everyone thinks it should sink but it doesn't" (p. 199). This article further documents that the
relationship between attitudinal similarity and interpersonal attraction may never sink if the cognitive processes of
assumed similarity (Hoyle, 1993; Singh &Tan, 1992; Smeaton et al., 1989, 1993) and accuracy in perceiving the
manipulations (Duck, 1973) are also accorded the role they deserve in interpersonal evaluations. This is what the present research coupled with an earlier one (Singh & Tan, 1992) tried to demonstrate. Results clearly dispel the doubts that Rosenbaum (1986a, 1986b) raised about the causal connection between attitudinal similarity and interpersonal attraction. Therefore, the present authors agree with Byrne et al. (1986), who noted that "attitude similarity is as potent today as when Aristotle (1932) observed its effect on friendship" (p. 1167).

Aristotle. (1932). The rhetoric. New York: Appleton-Century-Crofts.

Byrne, D. (1961). Interpersonal attraction and attitude similarity. Journal of Abnormal and Social Psychology, 62,
713-715.

Byrne, D. (1971). The attraction paradigm. New York: Academic Press.

Byrne, D., Clore, G. L., & Smeaton, G. (1986). The attraction hypothesis: Do similar attitudes affect anything? Journal of Personality and Social Psychology, 51, 1167-1170.

Byrne, D., & Griffith D. (1966). A developmental investigation of the law of attraction. Journal of Personality and Social Psychology, 4, 699702.

Byrne, D., & Nelson, D. (1965). Attraction as a linear function of proportion of positive reinforcements. Journal of
Personality and Social Psychology, 1, 659-663.

Byrne, D., & Wong, T.J. (1962). Racial prejudice, interpersonal attraction, and assumed similarity of attitudes. Journal of Abnormal and Social Psychology, 65, 246-253.

Clore, G. L., & Baldridge, B. (1968). Interpersonal attraction: The role of agreement and topic interest. Journal of
Personality and Social Psychology, 9, 340-346.

Condon, J. W., & Crano, W. D. (1988). Inferred evaluation and the relation between attitude similarity and interpersonal attraction. Journal of Personality and Social Psychology, 54, 789-797.

Drigotas, S. M. (1993). Similarity revisited: A comparison of similarity-attraction versus dissimilarity-repulsion. British
Journal of SociaI Psychology, 32, 365-377.

Duck, S. (1973). Similarity and perceived similarity of personal constructs as influences on friendship choice. British
Journal of Social and Clinical Psychology, 12, 1-6.

Duck, S., & Barnes, M. K. (1992). Disagreeing about agreement: Reconciling differences about similarity.
Communicates Monographs, 59, 199-208.

Furman, W., & Childs, M. K. (1981). A temporal perspective on children's friendships. Paper presented at the biennial meeting of the Society for Research in Child Development, Boston, MA.

Gottman, J., & Mettetal, G. (1986). Speculations about social and affective development: Friendship and acquaintanceship through adolescence. In J. Gottman & J. G. Parker (Eds.), Conversations of friends: Speculations on affective development (pp. 192-237). New York: Cambridge University Press.

Hoyle, R. H. (1993). Interpersonal attraction in the absence of explicit attitudinal information. Social Cognition, 11,
309-320.

Peevers, B. H., & Secord, P. F. (1973). Developmental changes in attribution of descriptive concepts to persons. Journal of Personality and Social Psychology, 27, 120-128.

Rholes, W. S., & Ruble, D. (1984). Children's understanding of dispositional characteristics of others. Child Development, 55, 550-560.

Rholes, W. S., & Ruble, D. (1988). Children's impressions of others: The effects of temporal separation of behavioral
information. Child Development, 59, 872-878.

Rosenbaum, M. E. (1986a). Comment on a proposed two-stage theory of relationship formation: First repulsion; then
attraction. Journal of Personality and Social Psychology, 51, 1171-1172.

Rosenbaum, M. E. (1986b). The repulsion hypothesis: On the nondevelopment of relationship. Journal of Personality and Social Psychology, 51, 1156-1166.

Rosenthal, R. (1966). Experimenter effects in behavioral research. New York: Appleton-Century-Crofts.

Singh, R. (1973). Attraction as a function of similarity in attitudes and personality characteristics. Journal of Social
Psychology, 91, 87-95.

Singh, R. (1974). Reinforcement and attraction: Specifying the effects of affective states. Journal of Research in
Personality, 8, 294-305.

Singh, R., & Tan, L.S.C. (1992). Attitudes and attraction: A test of the similarity-attraction and dissimilarity-repulsion
hypotheses. British Journal of Social Psychology, 31, 227-238.

Smeaton, G., Byrne, D., & Murnen, S. K. (1989). The repulsion hypothesis revisited: Similarity irrelevance or
dissimilarity bias?Journal of Personality and Social Psychology, 56, 54-59.

Smeaton, G., Rupp, D., Vig, C., & Byrne, D. (1993). The mediating role of similarity assumptions on the effects of
attitude similarity and dissimilarity on attraction and repulsion. Manuscript submitted for publication.

Tesser, A. (1993). The importance of heritability in psychological research: The case of attitudes. Psychological Review, 100, 129-142.

Youniss, J., & Volpe, J. (1978). A relational analysis of children's friendships. In W. Damon (Ed.), New directions for
child development (Vol. 1, pp. 1-22). San Francisco:Jossey-Bass.