Copyright 1986 bychc American Psychological Association, Inc. 0022-3514/86/HJ0.75 Attributions of Personality Based on Physical Appearance, Speech, and Handwriting Rebecca M. Warner University of New Hampshire David B. Sugarman Rhode Island College The effect of facial appearance, speech style, and handwriting on personality attributions was examined. The source consistency hypothesis predicted that an actor will receive consistent attributions across all three types of information. Thedifferential information hypothesis predicted that different personality dimensions are used to differentiate the actors within each type of information. In a 3 X 6 multivariate analysis of variance (MANOVA) design, each judge rated a single actor/information combination on scales of social evaluation, intellectual evaluation, activity, potency, emotionality, and sociability. Pho- tographs of actors were differentiated primarily in terms of positive social and intellectual evaluation; the speech of actors was differentiated primarily along an activity dimension; and the writing of the actors was differentiated primarily along a potency dimension. This study supported the differential information hypothesis and suggested that these three types of information about an actor may lead judges to use different personality dimensions. Person perception studies have shown that observers readily make attributions about the personality traits, abilities, and emotions of other persons based on limited information. Three types of information have been extensively studied: facial ap- pearance, expressive noncontent characteristics of speech (such as pitch, tone, and tempo), and handwriting. Numerous studies have reported that facial features and expression influence at- tributions about the attractiveness, pleasantness, intellectual and social skills, and mental health of the target person (Adams, 1977; Berscheid & Walster, 1974; Bull & Stevens, 1979; Dion, Ber- scheid, & Walster, 1972; Guise, Pollans, & Turkat, 1982; Jones, Hannson, & Phillips, 1978; Unger, Hilderbrand, & Madar, 1982). Observers seem to apply a generally positive stereotype to persons who are physically attractive. There has been less consensus about the kinds of attributions that are made based on noncontent characteristics of speech. Pitch and tempo aifected ratings of "competence" and "benev- olence" (Brown, Strong, & Rencher, 1974). Acoustic parameters such as pitch variation, amplitude variation, and tempo corre- lated with ratings of pleasantness, activity, potency, and various emotions (Scherer, 1974). Attributions are made about sex, age, personality, social class, and ethnicity based on various speech characteristics (Scherer & Giles, 1979). Ratings of emotionality have been linked to speech patterns in other studies (Ostwald, 1965; Williams & Stevens, 1972). Unlike the findings on physical attractiveness, where the results could be summarized as a pos- itive evaluative stereotype ("what is beautiful is good"), there does not seem to be a single principle that describes the attri- butions given to noncontent stylistic characteristics of speech. We gratefully acknowledge the assistance of Christine Barwick, Susan Carroll, Robert Filocco, Kathy Schneider, and Nancy Wysocki in the data collection. Thanks to Ellen Cohn, Rodney Triplet, and an anonymous reviewer for their comments on a draft. Correspondence concerning this article should be addressed to Rebecca Warner, Department of Psychology, University of New Hampshire, Dur- ham, New Hampshire 03824. There has been less interest in the study of handwriting as expressive behavior, possibly because it seems too similar to gra- phology. However, there are studies that document a relation between stylistic features of handwriting (signature size, neatness, slant, etc.) and attributions that are made about the writer, either by the self or by others. Judgments about self-esteem, dominance, potency, and intellectual competence vary as a function of sig- nature size (Aiken & Zweigenhaft, 1978; Zweigenhaft, 1977). Some investigators report that there is a stronger relation between the handwriting style and attributions for female writers than for male writers (Bull & Stevens, 1979; Jorgenson, 1977). The attributions that judges make based on appearance and expressive behavior may not agree with judgments made by cli- nicians or with results from standardized psychological tests or with self-reports of the actors about their internal states. The issue of accuracy in person perception is a complex one (cf. Cronbach, 1955) that is beyond the scope of this article. Here, the issue is whether judges form consistent or consensual attri- butions in response to a particular expressive behavior display such as a speech sample, and not whether their attributions are correct. The fact that actors may engage in impression manage- ment or deception further complicates the problem of obtaining "accurate" attributions about the actor's internal states based on the actor's expressive behaviors (cf. Baron, 1981; Edinger & Patterson, 1983). However, even if there is not an isomorphic relation between physical appearance or expressive behavior dis- play and the internal states of the actor, it may be useful to ex- amine how particular expressive behaviors influence attributions made by outside observers. The analysis of multichannel communication has been ad- dressed from two different perspectives which are relevant here. The first perspective involves the study of expressive behavior as an important, and often neglected, means of personality assess- ment (Allport, 1961). According to Hall and Lindzey (1970), Allport and his colleagues collected extensive data on the physical appearance and expressive behaviors of persons. In one study, judges were asked to match the speech of an actor with other 792 ATTRIBUTIONS OF PERSONALITY 793 information about that person such as physical appearance, gait or handwriting, personality characteristics, age, occupation, and so forth. Allport and Cantril (1934) found that matches were made at better than chance levels, and suggested that this was evidence for consistency among the expressive behaviors emitted by an individual actor. Differences in expressive behavior were quite consistent over time for an individual (Allport & Vernon, 1933). Although there was some redundancy in the information that was available across channels, no one channel was an exact replica of another channel in terms of the information about personality that it contained. In the present study, one question that is addressed is whether judges give consistent ratings to actors when each judge receives only one type of information about the actor (facial appearance, voice, or handwriting). The second question focuses on person perception, or the at- tributions that observers make based on the appearance and be- havior of the actor. Many studies of multichannel communication ask how much weight is given to each channel in forming the overall judgment, particularly when the information in different channels is discrepant (e.g., positive verbal content paired with negative tone of voice). Mehrabian and Ferris (1967) found that attributions about positivity of attitude were more closely related to facial expressions than to vocal style or verbal content. Their formula (Overall Affect Judgment = .07 X Verbal + .38 X Vo- cal + .55 X Visual) provides estimates of the relative importance of these three channels. Many other investigators have found evidence for "visual primacy" (see DePaulo, Rosenthal, Eisenstat, Rogers, & Finkelstein, 1978, for a review). There is an alternate way to frame the question about multi- channel communication. Several studies suggest that there may not be a consistent set of weights that describe how observers combine information from several channels. Instead, the relative importance of channels may depend on the attribute being judged (Ekman, Friesen, O'Sullivan, & Scherer, 1980). Specifically, ob- servers might rely on different personality dimensions when they try to "decode" visual communications than when they decode vocal style or verbal content. Zuckerman, Amidon, Bishop, and Pomerantz (1982) reported that the relative importance of face and voice in judging affect varied, depending on the type of affect that was being judged. Tone of voice was a better source of information about dominance and submissiveness, whereas the face provided more information about liking or disliking. Burns and Beier (1973) also found that information channels (vocal and visual) differed with regard to the amount of information they conveyed about various mood states; in particular, anxiety attributions were influenced more by the vocal than the visual channel in their study although most of the other moods they examined (e.g., anger, happiness, sadness) were dominated by visual information. It is also possible that encoders differ in the extent to which they rely on visual versus audio channels as ways of encoding affect (Berman, Shulman, & Marwit, 1976); and that decoders differ in their skill at decoding particular communication channels (Rosenthal, Hall, DiMatteo, Rogers, & Archer, 1979). Situational factors are probably im- portant determinants of the choice of channels that are used to communicate affect (Krauss, Apple, Morency, Wenzel, & Win- ton, 1981). At this point, it may be useful to ask specific questions; for example, when a judge is relying on visual information such as facial appearance, does that judge use different personality or affective dimensions to describe the actor than when the judge is relying on noncontent vocal style, handwriting, or other chan- nels? For instance, do judges rely primarily on an "evaluative" dimension when judging physical appearance, and primarily on a "potency" dimension when judging handwriting? The emphasis here is on the consistency and differentiation of judgments rather than on accuracy (cf. Berman et al., 1976). Rather than predicting overall primacy of the visual channel over other channels such as vocal or verbal, it may be reasonable to predict that the visual channel provides relatively more information about certain di- mensions of affect or personality (such as a general evaluative dimension); but that other channels might provide more infor- mation about other personality dimensions, such as potency or dominance. The purpose of the present study is to see whether judges who are provided with only one information channel (face, voice, handwriting) can consistently differentiate the actors along several personality dimensions (social evaluation, intellectual evaluation, potency, activity, sociability, and emotionality). It is expected that different dimensions will be used for each channel. Specif- ically, based on the studies just cited, it is predicted that there will be consistency in the judges' attributions about both social and intellectual evaluation given to photographs. Based on the existing research on handwriting and personality attributions, it is reasonable to predict that judges will give consistent ratings of potency and possibly intellectual evaluation based on the handwriting samples. The existing data on attributions based on voice does not permit clear-cut predictions, but judgments of emotionality and activity might be made based on noncontent vocal style. In the present study, three types of information (photograph, speech sample, or handwriting) were provided about 6 actors. Each judge saw only one actor/information combination (e.g., only Person 1's handwriting or Person 3's speech). The judges rated the target on bipolar adjectives that were combined into scales measuring various kinds of attributions about the person- ality or ability of the target. Three hypotheses are proposed. The first hypothesis could be termed the source consistency or actor consistency hypothesis. This hypothesis predicts that judges will produce similar personality attributions for each actor regardless of the information condition; that is, if Actor 1 "s facial appearance is rated as high on social evaluation (warmth, cheerfulness, op- timism, friendliness, attractiveness), then Actor 1's speech and handwriting should also receive high ratings on social evaluation. If consistent information about personality is extracted by the judges from each of these three types of information, this might be taken to imply that the physical appearance and behavior display emitted by the actor contain a consistent message about personality that is similar across all three channels (as suggested by Allport, 1961). A second hypothesis can be termed thedifferential information hypothesis. This suggests that the pattern of judgments about the actors would be different for each type of information. It is proposed that judges might find it easy to form high consensus judgments about a certain personality dimension for one type of information and difficult to form high consensus judgments using this dimension with other types of information; for ex- ample, judges might form high consensus judgments about ac- 794 REBECCA M. WARNER AND DAVIDB. SUGARMAN tivity only when they are reacting to speech, and not when they are evaluating photographs and writing samples. There could also be artifactual differences in the mean ratings given to each type of information, for instance, a tendency to give higher ratings of intellectual competence for writing samples than for facial appearance. Note that this is not the same as the differential information hypothesis. Method Materials The stimuli were generated in the following manner. All the members of one section of the introductory psychology class were brought into the lab (N- 40). Each person was allowed time to become familiar with a passage taken from an art history textbook, and then was tape recorded while reading the passage out loud. This paragraph was selected in order to standardize the speech content, and it was neutral in emotional content. Next each person posed for four color slides (head and shoulders only). The one slide that had the most natural looking smile and the best overall technical quality was selected for use. This selection procedure made it possible to standardize the facial expression; the rwnsmiling photographs varied so much that choosing nonsmiling photographs would have resulted in much variability in expressions. It should be noted that past research suggests that smiling faces convey more information about emotionality than nonsmiling faces; the results of this study cannot safely be generalized to nonsmiling faces. Finally each person was asked to copy the same paragraph that was used for the speech sample onto an unlined sheet of paper using a normal style of handwriting. The only constraint placed on this task was that the entire paragraph had to fit on one side of an 8 X 11 in. page. Thus there were three types of information for each actor: a head and shoulders photograph, a tape recorded speech sample, and a handwriting sample. Fromthe 40 actors available, only the 26 women were retained for use as actors. 1 Six actors were chosen by random selection from this group of 26. Randomselection was used, rather than systematic choice of models, to represent levels of some previously de- termined factor such as "physical attractiveness," because the intent of this study was to examine an ecologically valid selection of stimuli (cf. Brunswik, 1956) rather than to rigorously control the type of information available to the judges. Thus, the actors do not represent extremes. 2 A 6 X3 factorial design was set up, in which 18 stimuli were included3 types of information for 6 actors. Multivariate analysis of variance (MAN- OVA) and discriminant analysis were used to evaluate whether the set of ratings received by each actor, by each type of information, and by each actor by information combination were significantly different. Questionnaire Development The dependent variables were six summated scales based on a ques- tionnaire consisting of bipolar adjective ratings and Likert type items. There were six dependent measures: scales measuring social evaluation, intellectual evaluation, potency, emotionality, activity, and sociability. The first three scales were created in an instrument development study reported in detail in the Appendix. The factor structure of the bipolar adjective ratings in both the instrument development study and the main study are presented in Table 1. These three scales are sums of bipolar adjective rating items that were grouped according to factor analysis re- sults, and the Cronbach alphas for these scales ranged from .776 to .706. The second set of three scales were taken from the Buss and Plomin Emotionality Activity Sociability Impulsivity temperament rating system (EASI; 1975), and the Cronbach alphas for these scales when applied by our judges ranged from .82 to .65. Table 1 Varimax-Rotated Factor Structures for the 13 Bipolar Adjectives Questionnaire development Measure 1 II HI Questionnaire confirmation II III Attractive Cheerful Friendly Optimistic Warm Knowledgahle Competent Responsible Intelligent Sensible Strong Sturdy Dominant .219 .016 .137 .164 .150 .707 .558 .738 ,719 .677 .252 .079 .064 ,395 .704 .650 ,629 .748 .093 .271 .113 .196 .117 .039 .038 -.001 .084 .180 -.096 .122 -.187 .140 .243 .050 .106 .063 .647 m .755 .359 JOO .714 m .709 .038 .160 .134 .098 .032 .091 .226 .151 .158 .062 .013 .148 .118 .646 .630 .605 .560 .607 .136 .038 .143 .148 .250 .053 .264 .023 .076 .206 .048 .092 .038 .686 m .589 Note. Underscored numbers denote factor loadings greater than .30. The ratings questionnaire consisted of the 13 bipolar adjectives that were chosen during the instrument development phase, 9-filler items (ad- ditional bipolar adjectives that were not included in the analysis), and the Buss and Plomin (1975) EASI temperament scales (four scales, each containing five items, assessing emotionality, activity, sociability, and im- pulsivity). The Buss scales were chosen because they had previously been used to rate other persons (rather than exclusively for self-rating), because the factor structure and reliability of the scales was extensively docu- mented, and because of the content areas they covered. Judges Judges were recruited from the introductory psychology subject pool. A total of 404 judges participated in the study; 65%of themwere women. Judges were asked whether they had any previous familiarity or contact with the actors they were rating. Any judges who were familiar with the actors or who had missing values on the adjective rating scales were elim- inated from subsequent analyses, leaving an Nof 382. Initial analyses 1 In choosing actors, the researchers deliberately avoided the strategy of trying to obtain actors who would be very diverse with respect to age, sex, social class or regional background, ethnicity, or educational level. If the models differed in sex or age, for instance, differences in the per- sonality ratings they received could be attributable to sex or age stereo- types, and there were not enough models to systematically examine these kinds of differences. Sex of actor was a variable that was deliberately excluded from the design because it would have greatly complicated the analysis, and because it was not feasible to include a large enough sample of actors to examine both sex stereotype effects on ratings and variability of ratings received by individual actors within each sex. We wished to examine individual differences in the more usual sense. Thus, the selection of models was restricted to Caucasian female college students. 2 The mean attractiveness ratings received by the 6 female actors ranged from 1.75 to 2.31 (on a scale from 0 to 4); whereas this difference was statistically significant, F(5, 381) = 2.732, p = .019, it is clear that this randomsampling procedure resulted in selection of average attractiveness. This would tend to result in smaller correlations between attractiveness and other ratings than would be obtained if we had deliberately selected actors who were very high and very low on attractiveness. However, the use of extreme groups to test for the presence of relations generally results in overestimation of the strength of relations (cf. Feldt, 1961). ATTRIBUTIONS OF PERSONALITY 795 indicated that there were no significant differences in the ratings given by male and female judges, therefore sex was not included as a factor in the analyses. Procedure A set of 6 female actors was chosen randomly from the same pool of 26 actors that was used in the instrument development study. There were 18 conditions: 6 actors each contributed three types of information (pho- tograph, speech sample, writing sample). In each condition, judges rated only one of the actor by stimulus combinations in order to avoid possible carryover effects. In the slide condition, a group of judges was shown a color slide of one actor and asked to fill out the rating questionnaire. The slide was displayed during the entire rating period. In the tape condition, the tape (which was about 1 min in duration) was played once before the judges began rating, and played a second time about 5min into the rating period. In the handwriting condition, each judge was given a photocopy of one actor's handwriting and asked to rate the writer. Judges in both the tape and handwriting conditions were informed that the content of the message (a paragraph from an art history textbook) was provided to the actor, and were instructed to ignore content and focus on the style of the speech or writing. Results After preliminary data reduction through the formation of summated scales (as reported in the Appendix), there were six dependent variables. Three were person perception variables de- rived from our own bipolar adjective ratings (social evaluation, intellectual evaluation, and potency). Three were Buss temper- ament scales (Emotionality, Activity, and Sociability). These scales are not orthogonal; correlations between these scales ranged from .383 to +.688. Because there were six correlated depen- dent variables, MANOVA was used to evaluate the predictive use- fulness of these six personality dimensions. Multivariate Analysis of Variance A 6 X 3 MANOVA was performed using the six personality scales (social evaluation, intellectual evaluation, potency, emo- tionality, activity, and sociability) as dependent variables. The factors were the six levels of the actor identity factor and the three levels of information type (photograph, speech, and hand- writing). Thus, each of the 18 cells in the design corresponded to an actor/information stimulus combination (e.g., Actor 1's speech, or Actor 4's handwriting). Type of Information was treated as a fixed factor, and Identity of Actor was treated as a random factor, thus, the significance test for the main effect of information type had to use the interaction component of the model as the error term. Because of unequal within-cell , the computational procedures for nonorthogonal analysis of variance (ANOVA) with regression-type partitioning of sums of squares were used. Each effect was tested controlling for all other effects. There did not appear to be serious violations of the assumptions of multivariate normality. Each individual dependent variable had an approximately normal distribution, and the within-cell correlation matrix was reasonably homogeneous across-treatment conditions. The pooled within-cell correlation matrix among the six dependent variables is shown in Table 2. Generally the cor- relations were low, except that the Buss sociability scale correlated .503 with the social evaluation scale and .541 with the Buss ac- tivity scale. For the overall MANOVA, all three of the effects were significant. For the Actor X Information Type interaction, A = .634, ap- proximate F(60, 1886) = 2.860, p <.001. This substantial (and disordinal) interaction makes it difficult to interpret the main effects. However, both the main effects were also significant. For the differences in the ratings received by the 6 actors, A =.833, approximate f\W, 1438) =2.25, p <.001; for the differences in the ratings given to the three different types of information, A = .0206, approximate F(\2, 10) =4.971, p =.008. (This extraordinarily small lambda was obtained when the interaction term was used as the error term; if the within-cell error term is used, A =.463, approximate^12, 718) = 28.12,;) <.001. The main effect for information is significant using either error term.) Interpreting the significant interaction in light of the hy- potheses, it appears the pattern of judgments differs from one information type to another. To obtain more information about possible differences across-information types, a follow-up analysis of the simple main effects of the actor was done separately for each of the three information conditions. This included a dis- criminant function analysis to evaluate the relative contribution of each of the six dependent variables to the discrimination among actors. The results of the discriminant analyses that were done sep- arately for each information type are shown in Table 3. For all three information types, the first two dimensions of the five di- mensional solution are shown. Standard procedures for testing the significance of these roots were used (Nie &Hull, 1981). For all three of these information types, the set of Roots 1-5 was significant (p <.05). The set containing Roots 2-5was significant (p =.033) for the slide condition, and fell just short of conven- tional significance levels for the handwriting condition (p =.083) and the tape recorded speech condition (p = .064). To simplify interpretation of these results, the following rule was adopted: A particular dependent variable was viewed as being a useful source of discriminating information only if its beta weight for the stan- dardized canonical discriminating function and its structure coefficient (correlation with the discriminating function) both exceeded .40 in absolute value and had the same sign. In addition, the univariate F ratio for each variable was also examined to see whether it was consistent with the beta weight and structure coef- ficient. Finally, only those dimensions that were significant in the dimension reduction analysis will be interpreted. Conse- quently, Dimensions 1 and 2 can be interpreted for the slide condition, but only Dimension 1 will be interpreted for the handwriting and tape conditions. For the scales that met these conditions the standardized beta weights and structure coeffi- cients are underscored in Table 3. For the slide condition, the first dimension along which the actors were discriminated by the judges seems to be best char- acterized in terms of social evaluation, although there were other scales that had moderately strong correlations with this dimen- sion. The social evaluation scale involves attributions of warmth, friendliness, optimism, cheerfulness, and attractiveness. There was a second orthogonal dimension along which the judges dif- ferentiated the slides; the scales which had the highest correlations with this second dimension were intellectual evaluation and ac- tivity. Taken together, these two scales suggest a pattern of com- petence and "quickness," possibly both mental and physical quickness. For the handwriting condition, it was clear that only one of the six scales was a useful discriminator among the actors: 796 REBECCA M. WARNER AND DAVIDB. SUGARMAN Table 2 Within-Cell Correlation Matrix of the Six Scales Scale Scale 1. Social Evaluation 2. Intellectual Evaluation 3. Potency 4. Activity 5. Sociability 6. Emotionality Mean .732" .242 .139 .371 .503 -.160 2.36 .604 .265 .203 .005 -.107 2.46 .785 .378 .222 -.330 1.89 .866 .541 -.061 2.09 .911 -.274 2.35 .655 1.78 Note. Both means and standard deviations are simple univariate results. * Standard deviations are on the diagonal. the potency scale. For the tape recorded speech condition, the basis on which the judges apparently differentiated the six tape recorded speech samples was primarily activity and social eval- uation. A few other inferences can be tentatively drawn. For instance, although the sociability scale had a significant univariate F ratio for the differences among actors in the slide condition, it appears that this significant result may have occurred because sociability is correlated with two other scales that were more effective pre- dictors (the social evaluation and activity scales). The sociability scale could be seen as redundant with these other two scales, and it appears to have been less effective in describing perceived dif- ferences among the actors. In general, the results of the discrim- inant analysis were fairly consistent with the univariate F ratios that are also reported in Table 3. The only variables that appeared less useful within the context of the multivariate model than might have been expected from the univariate F ratios were so- ciability (in the slide condition) and intellectual evaluation (in the tape condition). Another consistent finding was that the emotionality scale was not a useful discriminator in any of the information conditions. This reflects the generally lower level of consensus among judges in their preceptions of emotionality. A large beta weight, structure coefficient, and univariate F ratio for any trait implies that disagreements among judges were relatively small, and that actors received consistently different ratings on that trait variable. Thus, when a scale such as social Table 3 Simple Effects Analysis of Actors Within Each Level of the Information Factor Dimension 1 Dimension 2 Handwriting condition Tape condition Univariate Scale Social Evaluation Intellectual Evaluation Potency Activity Sociability Emotionality B -.816 .348 -.256 -.177 -.112 -.029 re B Slide condition -.885 -.040 .050 -.707 -.359 -.564 -.665 .190 .743 -.686 .219 .132 re -.273 -.672 .296 -.454 -.047 -.049 F 8.89 3.03 2.10 4.66 4.87 .45 P <.001 .011 .065 <.001 <.001 .810 Social Evaluation Intellectual Evaluation Potency Activity Sociability Emotionality .312 -.222 -.887 -.165 .199 -.363 .232 -.376 -.801 -.299 .169 -.141 -.469 -.429 .429 -.524 1.01 1 .257 -.240 -.558 .192 -.091 .514 -.009 1.19 2.04 3.71 1.81 1.50 .27 .314 .073 .003 .110 .190 .928 Social Evaluation Intellectual Evaluation Potency Activity Sociability Emotionality .460 -.177 -.426 1.069 -.498 .022 .500 .034 -.123 .771 .210 .179 -.120 .929 -.408 .041 -.310 .066 -.103 .792 -.253 -.141 -.452 .203 3.94 3.13 .53 7.92 1.99 .75 .002 .009 .755 <.001 .079 .585 Note. B = Standardized beta coefficient; re = Canonical structure coefficient. Underscored numbers denote dependent variables with both beta and structure coefficient greater than .40, only for significant dimensions. ATTRIBUTIONS OF PERSONALITY 797 evaluation is a useful predictor within the discriminant analysis, this suggests that judges showed better agreement in their as- signment of ratings of social evaluation than in their assignment of ratings of other variables such as emotionality. An alternative method of assessing agreement among judges would be to treat each scale separately in a univariate ANOVA and use the intraclass correlation as an index of reliability of the assessments made by each individual judge. This univariate analysis provides a picture of the relative importance of variables that is generally similar to the picture that is obtained from the multivariate analysis, apart from the exceptions noted earlier. Discussion Two hypotheses were discussed in the Introduction. This study does not provide strong support for the source consistency hy- pothesis; the significant and disordinal interaction between in- formation type and actor makes it difficult to interpret the sig- nificant differences in the ratings received by actors. This study provides clearer support for the differential information hypoth- esis, which stated that each information type or communication channel provides information about different personality di- mensions. Airport's statement (1961) that there would be some redundance across channels is not being disputed here. However, these results suggest that each communication channel (facial appearance, speech, handwriting) provides information about a different collection of personality dimensions, and that a partic- ular attribute (such as potency) can sometimes be judged more consistently from one channel (such as handwriting) than from other channels. Specifically, actors were differentiated along a potency dimension much more clearly based on the handwriting samples than based on speech or facial appearance. Actors were differentiated along an activity dimension more consistently based on their speech than based on other communication channels. For other personality attributes such as social and intellectual evaluation, judges were able to make the most consistent differ- entiation among actors based on the facial appearances. The results obtained here are basically in agreement with ear- lier research. For instance, the handwriting research just cited has consistently found correlations between "potency" type variables (status, self-esteem, etc.) and various features of hand- writing such as signature size, although some authors feel that this correlation may exist only for female writers and not for males (Jorgenson, 1977). In this study, when judges were given the opportunity to make attributions using six personality scales, potency was the only scale that yielded consensus among the judges of handwriting samples. Another point of agreement between this study and earlier findings is the importance of the evaluative dimensions when judging physical appearance. Existing research on physical at- tractiveness suggests that "what is beautiful is good" (Dion et al., 1972). Judges invoke a generally positive personality stereo- type when they make attributions about the personality or abilities of physically attractive persons. In the present study, both of the evaluative scales (Social and intellectual evaluation) showed greater consensus among judges of slides than judges of tapes or handwriting. However, the slides also elicited fairly consistent attributions about activity and sociability characteristics that were only moderately correlated with the evaluative scales. Existing research did not point to a simple description of at- tributions based on tape recorded speech; in the present study, the judges of tapes showed the highest level of agreement when judging activity, and lower levels of agreement when judging social evaluation, sociability, and intellectual evaluation. Given earlier research linking speech to judgments of anxiety or emotionality, the inability of the judges to agree on the emotionality ratings of tape recorded speech in this study seems anomalous. One potential explanation is that actors read an emotionally neutral paragraph from a textbook probably limited the amount of in- formation about emotionality that was available from speech. There was wide variability in the emotionality ratings given, but judges did not agree which voices belonged to more or less emo- tional speakers. The lack of consensus among judges in rating emotionality from speech in this study may represent a meth- odological artifact rather than a general lack of emotional in- formation in speech. This study has several limitations that restrict the types of conclusions that can be drawn. First, there were only 6 female actors. The results may be partly due to peculiarities of these particular actors, and certainly cannot be generalized to males until the findings have been replicated using male actors. Second, the selection of adjectives and rating scales does not cover all possible personality dimensions. The Osgood, Suci, and Tan- nenbaum(1957) three dimensional semantic differential (eval- uation, activity, potency) guided the selection of scales, because numerous studies in social psychology and nonverbal commu- nication have found some variant of these dimensions to be useful (e.g., Brown et al., 1974; Exline, 1972; Hayes & Meltzer, 1972; Scherer, 1974; Wish, 1978). This was supplemented with the Buss and Plomin (1975) temperament scales. However, there may well be other attributions that judges could make reliably based on these kinds of information that were not represented in the selection of scales used here. Third, the information types could have been defined differently, as in studies that distinguish be- tween transcripts of verbal content and content-filtered vocal characteristics. The intent in designing this study was to use relatively naturalistic expressive behaviors; rather than asking actors to "act out" specific emotions or attempt to convey a "warm" or "cold" impression, the stimuli were obtained by re- cording the actors' spontaneous reactions to the request to speak, write, and pose for a photograph. Control over the specific in- formation content of the stimuli was sacrificed to obtain some improvement in the "ecological validity" (Brunswik, 1956) of the social stimuli used as the basis for the judges' attributions. The results of this study corroborate earlier findings of differ- ential information about personality contained in facial appear- ance, handwriting, and speech (cf. Burns & Beier, 1973; Ekman etal., 1980; and Zuckerman et al., 1982). The results imply that, rather than attempting to show how much weight is given to each information channel in general, it may be necessary to ask how much weight is given to each information channel for each spe- cific attribute that is being judged. References Adams, G. R. (1977). Physical attractiveness research: Toward a devel- opmental social psychology of beauty. Human Development, 20, 217- 238. 798 REBECCA M. WARNER AND DAVIDB. SUGARMAN Aiken, L. R., & Zweigenhaft, R. L. (1978). Signature size, sex, and status in Iran. Journal of Social Psychology, 106, 273-274. Allport, G. W. (1961). Pattern and growth in personality. New York: Holt, Rinehart & Winston. Allport, G. W, &Cantril, H. (1934). Judging personality from voice. Journal of Social Psychology, 5, 37-55. Allport, G. W, & Vernon, P. F. (1933). Studies in expressive movement. NewYork: MacMillan. Baron, R. M. (1981). Social knowing from an ecological-event perspective: A consideration of the relative domains of power for cognitive and perceptual modes of knowing. In J. H. Harvey (Ed.), Cognition, social behavior, and the environment. Hillsdale, NJ: Erlbaum. Herman, H. J., Shulman, A. D., & Marwit, S. J. (1976). Comparison of multidimensional decoding of affect from audio, video and audiovideo recordings. Sociometry, 39, 83-89. Berscheid, E., & Walster, E. (1974). Physical attractiveness. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 7, pp. 157- 215). NewYork: Academic Press. Brown, B. L., Strong, W. L., & Rencher, A. C. (1974). Fifty-four voices from two: The effects of simultaneous manipulations of rate, mean fundamental frequency, and variance of fundamental frequency on ratings of personality from speech. Journal of the Acoustical Society of America, 55, 313-318. Brunswik, E. (1956). Perception and the representative design of psycho- logical experiments. Berkeley: University of California Press. Bull, R., & Stevens, J. (1979). The effects of attractiveness of writer and penmanship on essay grades. Journal of Occupational Psychology, 52, 53-59. Burns, K. L., & Beier, E. G. (1973). Significance of vocal and visual channels in the decoding of emotional meaning. Journal of Commu- nication, 23, 118-130. Buss, A., & Plomin, R. (1975). A temperament theory of personality development. New York: Wiley. Cronbach, L. J. (1955). Processes affecting scores on "understanding of others" and "assumed similarity". Psychological Bulletin, 52, 177- 193. DePaulo, B. M., Rosenthal, R., Eisenstat, R. A., Rogers, P. L., & Fm- kelstein, S. (1978). Decoding discrepant nonverbal cues. Journal of Personality and Social Psychology, 36, 313-323. Dion, K., Berscheid, E., & Walster, E. (1972). What is beautiful is good. Journal of Personality and Social Psychology, 24, 285-290. Edinger, J. A., &Patterson, M. L. (1983). Nonverbal involvement and social control. Psychological Bulletin. 93, 30-56. Ekman, P., Friesen, W. V., O'Sullivan, M., & Scherer, K. (1980). Relative importance of face, body, and speech in judgments of personality and affect. Journal of Personality and Social Psychology, 38, 270-277. Exline, R. V. (1972). Visual interaction: The glances of power and pref- erence. In J. Cole (Ed.), Nebraska Symposium on Motivation, 1971. Lincoln: University of Nebraska Press. Feldt, L. S. (1961). The use of extreme groups to test for the presence of a relationship. Psychometrika, 26, 307-316. Guise, B. J., Pollans, C. H., & Turkat, I. D. (1982). Effects of physical attractiveness on perception of social skills. Perceptual &Motor Skills, 54, 1039-1042. Hall, C. S., & Lindzey, G. (1970). Theories of personality (2nd ed.). New York: Wiley. Hayes, D. P., & Meltzer, L. (1972). Interpersonal judgments based on talkativeness. 1: Fact or artifact? Sociometry, 35, 538-561. Jones, W. H., Hannson, R. C., & Phillips, A. L. (1978). Physical attrac- tiveness and judgments of psychopathology. Journal of Social Psy- chology, 105, 79-84. Jorgenson, D. O. (1977). Signature size and dominance: A brief note. Journal of Psychology, 97, 269-270. Krauss, R. M., Apple, W., Morency, N., Wenzel, C, & Winton, W. (1981). Verbal, vocal, and visible factors in judgments of another's affect. Jour- nal of Personality and Social Psychology, 40, 312-320. Mehrabian, A., & Ferris, S. R. (1967). Inference of attitudes from non- verbal communication in two channels. Journal of Consulting Psy- chology, 31, 248-252. Nie, C. H., & Hull, N. H. (1981). SPSS Update 7-9. NewYork: McGraw- Hill. Osgood, C. E., Suci, G. J., & Tannenbaum, P. H. (1957). The measurement of meaning. Urbana: University of Illinois Press. Ostwald, P. F. (1965). Acoustic methods in psychiatry. Scientific American, 212. 82-91. Rosenthal, R., Hall, J. A., DiMatteo, M. R., Rogers, P. L., & Archer, D. (1979). Sensitivity to nonverbal communication: The PONS test. Bal- timore: Johns Hopkins University Press. Scherer, K. R. (1974). Acoustic concomitants of emotional dimensions: Judging affect from synthesized tone sequences. In S. Weitz (Ed.), Non- verbal communication: Readings with commentary. NewYork: Oxford University Press. Scherer, K. R., & Giles, H. (1979). Social markers in speech. Cambridge: Cambridge University Press. Unger, R. K., Hilderbrand, M., & Madar, T. (1982). Physical attractiveness and assumptions about social deviance: Some sex-by-sex comparisons. Personality and Social Psychology Bulletin, 8, 293-301. Williams, C. E., & Stevens, K. N. (1972). Emotions and speech: Some acoustical correlates. Journal of the Acoustical Society of America, 52, 1238-1250. Wish, M. (1978). Dimensions of dyadic communication. In S. Weitz (Ed.), Nonverbal communication: Readings with commentary (2nd ed.). NewYork: Oxford University Press. Zuckerman, M., Amidon, M., Bishop, S., & Pomerantz, S. (1982). Face and tone of voice in the communication of deception. Journal of Per- sonality and Social Personality, 43, 347-357. Zweigenhaft, R. L. (1977). The empirical study of signature size. Social Behavior and Personality, 5, 177-185. (Appendix follows on next page) ATTRIBUTIONS OF PERSONALITY 799 Appendix Rating scales were developed in a preliminary instrument development study, using a separate group of 273 judges to provide data on factor structure and internal consistency. For the initial item pool, 32 pairs of bipolar adjectives were generated. Eight adjective pairs were chosen by the principal investigators and their research assistants for each of the Mowing four domains: activity, potency, social evaluation, and intellectual evaluation. Face validity was the criterion for inclusion in this initial phase. A questionnaire was set up incorporating these 32 bipolar adjective pairs with a 5-point rating scale. Ratings were to be made relative to the "average college student." Thus, if the bipolar adjective pair quiet/noisy were rated on a 5-point Likert scale: about as noisy as an average college student (3); a little bit quieter than average (2); much noisier than average (5). The position of the more socially desirable adjective was varied ran- domly so that sometimes the left side was the more socially desirable and sometimes the right side was the more socially desirable; however, to facilitate scoring and analysis, the data were receded with 5 always cor- responding to the more positive end of the scale. The questionnaire also asked for the sex, age and year in college of the rater, and asked about previous familiarity with the actor. A group of 273 judges each rated one actor/information combination using the 32 bipolar adjectives. The data were pooled across the 18 con- ditions and factor analyzed. This procedure involved the implicit as- sumption that there was homogenous factor structure across the 18 con- ditions, an assumption that was not tested directly because the number of subjects within each cell was too small to obtain reliable estimates of within-cell correlations and structure. The initial principal factors analysis was conducted with varimax rotation. All 32 bipolar adjectives were in- cluded, and there were seven factors with eigenvalues greater than one. Examination of these factors indicated that the last four factors consisted of single items or pairs of items. The set of eight adjectives that were supposed to detect an "activity" dimension did not appear together on a factor; instead, these items tended to load moderately highly on the factors which seemed primarily interpretable as intellectual evaluation, social evaluation, and potency. Based on this initial analysis it was decided to retain five items as measures of intellectual evaluation; five items as measures of social evaluation; and three items as measures of potency. A second factor analysis was performed using this reduced set of 13 items; the rotated factor loadings are shown in Table 1. Factor 1 was labeled an Intellectual Evaluation factor; Factor 2 as a Social Evaluation factor; and Factor 3 was named a Potency factor. Before rotation, the first three factors accounted for 60%of the variance, and only these three factors had eigenvalues greater than one. Scale scores were created using unweighted linear composites for each of these three groups of variables. Factor score coefficients could have been used to create factor scores; however, simple unweighted linear composites were chosen for several reasons including simplicity; consis- tency of scoring of the Buss scale and the new scales created for this study; and the availability of simple statistics such as Cronbach's alpha to describe the internal homogeneity for scales that are simple unweighted linear composites. Cronbach alphas were calculated to assess the internal homogeneity of the scales created by averaging the scores for the groups of items. The social evaluation scale alpha was .773; the intellectual eval- uation alpha was .833; and the potency scale alpha was .747. Correlations among these three scales tended to be small; the largest correlation was r = .354 for the intellectual and social evaluation scales. These three scales seemed to summarize the major dimensions of person perception that were implicit in our original selection of 32 adjectives reasonably well. After the results of the main study were obtained, the ratings of the 382 judges were factor analyzed to evaluate the stability of the factor structure. The rotated factor loadings are reported in Table 1. The results were essentially identical to those in the instrument development phase except that the first three factors only accounted for 55%of the variance prior to rotation, and the order of the factors was different. Alpha coef- ficients were calculated to assess the internal homogeneity of the three scales, which were composed of the same items that were used previously. There was some shrinkage of the alphas, but the scales were still sufficiently reliable to be used: the social evaluation scale alpha was .776; the intel- lectual evaluation scale alpha was .759; and the potency score alpha was .706. Analysis was also done to check the internal consistency of the items in the four Buss EASI scales within our sample. Alphas were .65 for the emotionality scale, .79 for the sociability scale, .82 for the activity scale, and .45 for the impulsivity scale. Due to the low alpha value, the im- pulsivity scale was not included in subsequent analyses. Also, the item "is independent of others" was omitted from the Buss sociability scale, because it had a negative correlation with other items on this scale for men and a positive correlation for women (Buss & Plomin, 1975, p. 26). Received April 4, 1985 Revision received August 23, 1985