ililiiti L I B R.ARY OF THL U N 1 VLRSITY or ILLl NOIS 370 The person charging this material is re- sponsible for its return on or before the Latest Date stamped below. Theft, mutilation and underlining of books are reasons for disciplinary action and may result in dismissal from the University. UNIVERSITY OF ILLINOIS LIBRARY AT URBANA-CHAMPAIGN Di:C :\r. 13' L161— O-1096 Digitized by the Internet Archive in 2012 with funding from University of Illinois Urbana-Champaign http://www.archive.org/details/processesaffecti10cron no. lO GROUP EFFECTIVENESS RESEARCH LABORATORY Department of Psychology University of Illinois Urbana, Illinois PROCESSES AFFECTING "UNDERSTANDING OF OTHERS" AND "ASSUMED SIMILARITY" LEE J. CRONBACH College of Education, University of Illinois TECHNICAL REPORT NO. 10 Study performed under Contract N6-ori-07135 with the Office of Naval Research PROJECT ON SOCIAL PERCEPTION AND GROUP EFFECTIVENESS AUGUST, 1954 THE LIBRARY OF thf SEP 27 1954 UNIVERSITY OF lUINOIS PROCESSES AFFECTING SCORI]S OH "Ul€)ERSTi.l\IDING OF OTHFRS" AND "ASSUIED SBCLi^-RITY"-'' Lee J» Cronbach College of Education, University of Illinois How one person judges another is important both for its theo- retical implications and for its practical significance in leadership, clinical assessment, and teaching skill. Recent studies of "social per- ception", as this area is termed, have been chiefly concerned with dif- ferences among perceivers, either in terms of their accuracy, or in terms of their tendency to view others as similar to themselves. These studies have usually been built aroimd a particiilar operation in which a judge (J) "predicts" hoi^ another person (o) will respond. Often, for example, both persons describe themselves on a personality inventory, and J is then asked to fill out the inventory as he thinks did. The extent to which the prediction agrees with 0*s actual response is talcen as a measure of J's accuracy of social perception (or "empathy", "social sensitivity", or "diagnostic competence", etc). Measurements obtained in this manner are difficult to interpret and several investi- gators have obtained evidence of distressingly low reliability or con- sistency for such scores (11,17,19,28). This paper seeks to disentangle some of the many effects xjhich contribute to social perception scores, and to identify separately meas- urable coniponents. This analysis has several results: "* Appreciation is expressed to Mary E. Ehart, who assisted in all stages of this paper from initial conception to final interpretation; and to Urie Bronfenbrenner and associates, for helpfully providing data and for their courtesy in exchanging ideas throughout our rather similar inve s tig ations , This study was conducted under 01© Contract N6-ori-07135s Fred E, Fiedler, Principal Investigator, 1# It shows that investigators run much risk of giving psychological interpretation to mathematical artifacts, when measures are used which combine the components, 2, It sheds light on the extent to which adapta- tion to individual differences is advisable, when the differences are not judged accurately, 3. It directs attention to some especially inter- esting aspects of social perception left untouched by the usual approach. Our analysis of social perception scores may also be instructive regarding research strategy generally. This area of research has developed in an ultra-operationalist manner^ of late, workers have seemed content to regard "empati^" as "what empathy tests measure". The principal research activity has been corre- lating "empathy", so defined, with other variables, ;/e shall show, however, that the operation involves many unsuspected sources of variation, so that scores are impure and results un- interpretable. Studies based on myopic operationism are largely waste effort, x^rhen the operation does not correspond to potentially meaningful constructs. Defining a measure operationally is only a first stage, preliminary to analytic studies which can refine the measure and bring it closer to the intended construct. It ms^ be of interest to note that this paper relies alir.ost entirely on al- gebraic analysis as its research method. Even though analysis is based on a model rather than actual behavior, it generates severrl psychological hypotheses. Although our report deals with a specialized area of per- ceptual research, it shares much of the perspective of Postman* s important general revievj of perception (21;), His remarks are peculiarly pertinent to studies of social perception, even thougji he was referring especially to the "New Look" studies of perception of words and objects: "At this juncture of debate, we shall do xiell to pull up short a moment and reconsider the fundamental operations of our perceptual experiments, particularly as they bear on the validity of the theoretical constructs linking perception to motivation and personality*.. Experiments have shared a common tendency vdiich may be called the projective bias a selective emphasis on central motivational determinants at the expense of adequate attention to the verbal and motor response dispositions of the subject and the relation of these dispositions to the dimensions of the stimulus.,.. We must, then reaffirm the critical importance of a full and precise analysis of the responses as well as the stimuli which furnish the basic data of perceptual experiments," A Mathematical Resolution of Social Perception Scores into Components Model and notation When data are gathered by means of a test consisting of items (i = a, b, c, ••• k), these items define a k-dimensional space, and the responses of any person define a point in that space (10), One point is defined by O's actual responses, and another ty J*s prediction for 0, x^^ denotes the response given by Other (O) when describing himself on item i, VJhen a judge (j) predicts what will say, x-ie have a prediction, Jq^a . x is alw^s used to indicate self -descriptions, y to indicate pre- dictions. Error in prediction is represented hj the discrepancy between Xqj|_ and yoiT* I^ some studies J*s accuracy is measiored by summing ly .. - X .1. \/e shall instead measure accuracy by the distance from the predicted to the actual location in k-space, as determined by the sum over items of squared differences. This formula is easier to treat mathematically than the sum of absolute differences^ and will ordinarily give about the same results, tJhen all items are of a Yes - No form, so that the error on any prediction can only be 1 or 0, the two formulas give identical results. Our measure has the important property of being invariant under orthogonal rotation of axes (10), We define the Accuracy with vjhich J perceives ty ACC^ . = 2 (y . . - X . )^ (1) oo 1 033 01 ' Either ACC or ACC msy be used as a measure. Because it is much easier to analyze ACC^ than ACC, we treat ACC^ throughout this paper. This should be reiaembered in applying results. Now we define Grand mean of aH self -description re- sponses; average ele- vation) • Mean of all persons on any item Mean of argr person on all items; his elevation We define y 9 7 - and y correspondingly. Then x . - x ••3 '^J o,j .1 would be the mean score on an item, figured as a deviation from the overall mean, x - x is a score on the item, figured as a devia- 01 .1 tion from the item mean. We shall let x' . stand for x - x 01 oi o, X . + X , and define y' . similarly. These are scores figured ,3. • • 03.2.'' as deviations from both nt^m and pp^rson mean, i »<=;., "ceiitRi-orl on tests and persons," X • • - - Z Z X N k i oi ^i' N ""oi X = 0, izx k i oi The Accuracy Score and Its Components j^cciiracy in predicting one on one item may be measured simply by the absolute value of the error. We have the folloTO.ng identity: ACC ..»i y..-x.l» j (y .-x ) * iij^ . -y i ) - ( ^ - X )) O.J ..J 0» *• + (( y .. - y . ) - ( X -. X )) •ij ..J •j-j •• + ( y» .. - x» . )!! (2) ^ ^ oij 01 ^' ^ ' In the right member of (2), the first term compares the elevation of J*s (the Judge's) predictions to the average elevation of the criterion responses. Each person using a scale forms his own frame of reference, and some people tend to use different portions of the scale than others (?)• This first term -will be small if J uses about the same part of the scale as the average 0, The second term represents J's ability to predict the relative elevation of the particular 0, It measures J • liJ .. 6 Host research has been concerned with the accuracy of J as a judge of all Others, This coiiLd be represented by an average of his accuracy scores (ACC . ) with particular Others, ACC^ = i 2 ACC^ = k ( f - 2r )^ J N o oj ..j •• * r2 (( y . -y . ) - ( X -x ))^ N O.J ..J o» •• ■*-2 (( y - J ) - (X -x ))^ 1 • ij • • 3 • 1 • • + i 2 2 (( yt -x« )f (3) N o i oij oi These four terms are attributable respectively to the dif- ference of J's elevation from the average, his errors in predict- ing individual deviations in elevation, errors in predicting item means, and errors in predicting individual deviations from the item mean (after correction for elevation). We shall refer to these as the Elevation (E) component , the Differential Elevation (DE) component , the Stereotype Accuracy (SA) component , and the Differ- ential Accuracy (DA) component . The last three of these require separate discussion. Differential Elevation QJE) The Differential Elevation component measiires J's errors in judging the "elevation" of O's responses. In some tests ele- vation reflects insignificant response sets, and we should ignore this component (cf. 10, p, k^3)» In other tests this component reflects J»s judgment of the overall "desirability" of each 0, and if so, it may be very important. The Differential Elevation component may be broken down by- using the formula for sioms of squares of correlated differences: 2 2 2 DE = k ( a- + a- - 2o- a- r- • ) {k) •^O.J O. O.J o« o. -^cj 2 The variance o- measures J*s tendency to predict that Others differ in elevation. It represents Assumed Dispersion in Ele- vation, later seen to be a component of "Assumed Similarity". 2 o- is the true dispersion in elevation. The correlation r- ^o. ^o* ^o.j (to be symbolized DEr) represents J»s ability to judge which 0*s rate highest on the elevation scale. If every item measures morale, for instance, the correlation shows how well J can judge which O's say they have the highest morale. Stereotype Accuracy (SA) As used here "stereotype accuracy" refers to the person^s ability to predict the norm for Others, It might be called "accuracy in predicting the generalized other" (3). This score depends on J*s knovjledge of the relative frequency or popularity of the possible responses. In contrast to our stereotype inferred from responses on many items. Gage measures an explicit stereotype. He asks J»s to predict the model response among Others of a specified type (l6, p, 8-11), Similar stereotype predictions are obtained in studies of ability to estimate group opinion (e.g., 6, 18, 29). Evidence comparing these two types of stereotype, would be valuable. We may write; SA^ = k ( a^ + oS - 2 a- a- r- - ) {$) a y.. y. ^ n^ ± ^ ±-\^ ± Here each variance is computed over items. The variance a« 8 2 expresses hcu much J expects item means to vary, o- is the X 2 'O scatter of the actual means, SA represents ability to judge the shape and scatter of the profile of item means, r- - (Stereotype y.ij ^.i Correlation, SAr) represents accuracy in judging mean profile shape without regard to errors in judging profile scatter (i,e,, spread in difficulty). Differential Accuracy (DA) Differential Accuracy measures ability to predict differences between persons on any item. This component is a sum over items. The component for ar^ item brealcs down: (6) 2 2 2 DA = a + a • -2a a r ij y' x» y' xi y» X oij oi oij oi oij oi 2 2 DA , summed over items, yields DA • Each variance in the formula ij J is taken over Others, a , is the Assumed Dispersion on the item (see below). It resembles closely Gage^s concept of "rigidity" or "adherence to stereotype" in prediction (l5, p. l6j 17 )• The correlation (DAr) in (6) is a measure of ability to judge which Others have highest scores on the item, when the score is taken as a deviation from the Others* mean. There is one such correlation for each item. Implications Seven aspects of J*s performance have been separated: 1, Elevation component: difference of predicted average response from actual average 2, Assumed Dispersion in Elevation "\ } Differential 3, Elevation Correlation DEr j Elevation U* Predicted variation in item means ~; I Stereo bypf» Ar.p.nrany 5o stereotype Correlation SAr 6« Assumed Dispersion on ar^r item ' (elevation held constant) [ Differential \ Accuracy 7» Differential Correlation DAr j The fact that the components are mathematically distinct does not imply that they are necessarily uncorrelated. Change in argr of these msy alter the Accuracy score. Surely these aspects of social perception do not all reflect the same trait, A person who uses the same region of the response scale as other persons (Elevation is low) need not have superior insight. ^-^And while judging which items have the highest mean seems to require acquaintance with the norms of the group, a person might possess such knowledge to a very high degree and yet lack diagnos- tic skill which would permit him to differentiate accurately betxieen individuals. At best, f ailiire to recognize the presence of distinct components makes interpretation ambiguous, Chowdiy and Newcomb (6) requested group members to predict what percen- tage of their group would agree with each of many attitude state- ments. Ability to make this prediction was judged by a difference score, and this score correlated significantly with le adership status. This score, however, combines our Elevation and Stereo- type Jlccuracy components. We cannot conclude that their leaders are better able to judge the specific attitudes in the group. Until the components are separately measured we cannot rule out the possibility that leaders simply used the correct range of the scale more often than non-leaders. This, in turn, might reflect willingness (or unwillingness) to use extreme percentages rather than arQT more subtle perceptiveness. That such effects do occur is shoTrin in a stucfy \sj Lorge and Diamond, who required juiiges t^ y^ artj* r. 10 estimate what propoition of 0*s would pass ability test items. They found that poor judges were greatly helped simply by being told the difficulty of a few items, "apparently the difference between 'poorest ' , 'mediocre', and 'best' judges is that the 'best* judges have some experi- ential reference for the per cent of the population that can pass an item. Giving such referents to the 'poorest' and 'mediocre* judges,., leads to a significant reorientation of such judgments," (20,p,33) When judges re- sponded only to the items, the best judges had a mean Stereotype Correla- tion of ,73, and the poorest one of ,56, /.fter information indicating an appropriate reference level was given, the same groups had mean correla- tions of ,77 and ,73« The difficulty encoiintered in inteipneling the Chow- dry-Newcomb study does not arise in two recent treatments of the saniO problem (18,29) where subjects are asked to predict what ranks will be assigned to certain stimuli. The ranking method eliminates elevation and dispersion differences from the responses, and therefore confines scores to the Stereotype Correlation, An Alternative and more informative method might be to analyze data of the Chowdry-Newcomb type in terms of the sepa- rate components so as to determine how leaders behave on each. At worst, failure to identify the ccmpcnents of the Accuracy score leads to arti- f actual correlations. Only a few of the mar^ examples in the literature need be cited, Norman and Ainsworth (CO) report a large number of corre- latioiis between Accuracy ("Empathy") and Assumed Similarity ("Projection"), Since +he accuracy score contains assumed similarity components, there would necessarily be an overlap between Those two scores even in a situa- tion where both sets of responses are determined strictly by chance. The correlations have no psychological meaning. In Esrmond's study (12) it was reported that persons with high Accuracy are also most easily judged. But a person who uses the scale in a typical manner will have a lo;>r Elevation Component; and thus will have lower Elevation erros in judging him simply because of this typicality. This would happen even if the other predicted his responses mth^ut ever meeting himi Perhaps social psychologists should take what comfort they can from Bertrand Russell's remark that physicists "have not yet reached the point where they can distinguish between facts about relativity and mathematical operations which may have nothing to do thereTiith", n Analysis of Assumed Similarity Score into Components Assumed Similarity (AS) may be determined (see lU, for example) for a single Other by the formiila: AS^ = I ( y - X f (7) JO ^ oij ji y is the perception of by J, and x is J*s statement about oij ji himself, (Sometimes "ASo", Assumed Similarity between two Others selected in a certain manner, is computed). Some investigators have measured AS over many Others, to get a general score called "projection" or "identification" (22), IJe msy brealc AS into components as we did ACC, If, as before, we measure AS by a distance formula based on sums of squares, AS^ « i Z ? AS^ = k ( y . - X . )2 J N o 1 oij ..J .0 2 + k , a- :-^ . . . a t-2 (( y -y ) - ( X -X )) 1 -IJ ••J IJ "J + 2 o2 i y (8) oio Assumed Similarity, therefore, contains four components. Equation (8) is simpler than the corresponding formula for ACC because some terms vanish, Assui-ncd Elevation (AE) The first term we may call the Assumed Elevation (AE) component. It measures J«s tendency to assvime that Others have the same average response as he does. This component is important if items are polarized so that a high score on each represents good adjustment or some other interpretable qualityj the score then shows whether J regards the average as similar to himself in this central dimension, tiBR^ 12 Assumed Dispersions (APE, API), The second component is the Assumed Dispersion among Others in elevation. The fourth is the Assumed Dispersion on specific items after differences in elevation are removed. These dispersions have already been encountered in equations (U) and (6) as components of ACC« ¥e shall refer to them as Assumed Dispersion in Elevation (ADE) and Assumed Dispersion within Items (API), respectively. Assumed Self-Tsrpicality (AST) > The third component measures the discrepancy between J*s percep- tion of the average and his self-description. This component tells whether J regards his own profile as typical in shape. Or, we might say, this component shows the similarity of J's self -perception to his implicit stereotype of Others (elevation held constant). We follow Gage in calling this Assumed SeUT-Typicality (AST) (l6, p,17). Of the four components, only AST divides into separate variance and correlational terms, AST = k ( a? / a^ - 2 a a- r - ) ^^^ y .^ X. . X. . y . . X .y . . .ij 10 10 .13 iO .10 The variance among the y's represents the tendency of the predictor to predict different means for different items. The correlation represents the similarity between his self -description and the average profile, after removing differences in elevation and scatter from consideration. We may call it the Self -Typicality Correlation (STr ), To summarize: the components of AS are of tvio types, ADE and ADI involve Assumed Similarity between Others ; i.e., a tendency to differ- entiate, iJ?S and /ST represent Assumed Similarity of self to average Other , These types seem logically distinct, but a subsequent section will indicate the probable desirability of combining AE with ADE, and AST with ADI. ■4= 13 Optimizing Predictive Decisions Insofar as our mathematical model is an acceptable approxi- mation to real conditions, we can reason mathematically to de- termine how a person may improve his judgments. We have assumed that the goodness of predictions can be evaluated b^ the mean 2 square error. Taking the derivative of each component of ACC , and setting that derivative equal to zero, we find that ACC be- comes smaller, and therefore prediction improves, xirhen (a) J has a typical response set, (b) c- approaches r- - a- , Here the variance is y.io \±y.±3 \± over items. This means that a- should not exceed y ii a- y and should be near zero if the Stereotype Correlation is low. If this correlation is low, .he the more/differentiates among items, the poorer is his accuracy. (c) o approaches ^^, , ^x« ' * ^^® variance being over Others, This means that o ^ should not exceed a , , and y X shoiild be near zero if the Differential Correlation is low. This principle is true for accuracy of prediction on any single item, and for the elevation score. It has not been possible to determine the conditions which maxi- mize accuracy as measured by other formulas (such as the mean of ACC .)but a result of the same general character would be expected. Effec ts of differentiation on prac tical decisions These formal principles indicate that there is an optimal degree of differentiation in maldng judgments, jf a Jiidge cnn 11; make accurate judgments as to the relative location of Others on a continuum, then he is vdse to make a as large as a — never larger. But if he is forced to base his judgment on inadequate cues or if the available personality theoiy and situ- ational knowledge do not permit trustt\rorthy inference, then he should treat people as if they were very nearly alike. The person who attempts to differentiate individuals on inadequate data introduces error even when the inferences have validity greater than chance . This is consistent with Gage»s evidence that judges predict a stranger more correctly when they describe the typical person of his group than when they try to describe him as an individual (l6, p.lO), Tlie variation of J's predictions indicates how much he differentiates. For example, a teacher estimating IQ's in a class might spread them from 90 to 110, or from 70 to 130. We would expect the judge who perceives greater differences to spply viore sharply differentiated treatments to the persons. His a is essentially a weighting, or an expression of his con- fidence in his own discriminations (cf, 19, p,201), A person who knows that the expected s,d, for IQ»s is l6 might try to predict so that his estimates would have this s,d, but unless he is a perfect judge, this is unwise. He will have less error if his predicted s.d, is less than l6— how much less depending on the correlational accuracy of his predictions. If two diagnosticians can each judge some trait with corre- lational validity ,i;0, the one xfho differentiates strongly (i,e*, makes extreme statements) will mslce far more serious absolute errors than -on© who diffeientj.ates moderately. Indeed, the as'rtoi, e- r.njjp- 15 person "who makes extreme differentiations on the basis of a validity of ♦1^0 msgr make worse predictions, judged by absolute magnitude of errors, than a judge who has zero correlational validity but makes no false differentiations. "Every pupil has his oX'jn pattern of readiness, and the teacher must fit methods to that pattern, not treat the pupil in terns of the statistical average" (9, p«73)« Statements such as these, commonly made in teacher-training, now appear to require qualification. From our evidence the degree of adaptation desirable depends on the adequacy of the diagnostic information. If the teacher is not well informed regarding the unique patterns of his pupils, he should probably treat them by a standard pattern of instruction xjhich has been carefully fitted to the typical pupil. Modifying his plans drastically on the basis of limited diagnostic information may do harm, A similar argument applies to clinical diagnosis and industrial leadership. Differentiation is harmfiil if the extent of adaption or differentiation exceeds the amount justified by our accuracy in differentiating. This is a distinct reversal of the view that judgment is always improved by taldjig into account additional information which has validity greater than zero. Investigators have noted a "central tendency of judgment", vrhich leads to lower dispersion among estimates than among objects, Ii/hereas formerly "the central tendency of judgment" was regarded as a source of inaccuracy (1, p,521) our analysis shows that this tendency may have beneficial consequences* Teachers may properly modify treatments considerably to fit individual differences provided they are well able to ju^dge. those 16 differences* They might be expected to judge differences in past achievement in arithmetic qiiite accurately j if so, they could profitably provide quite different treatments (e.g., dif- ferent assignments) for different individuals. But if it is hard to judge some other quality (e,g», creative potential in art), then it is a great mist alee to differentiate treatment. Treatment from individuals should depart substantially from that suited to the average of the group only when dependable information is available to guide the adaptation. Illustrative Analysis of Cornell Data To illustrate our system of analysis, we use data kindly provided by Bronf enbrenner and Dempsey. The data were gathered at Cornell University primarily for the purpose of pilot analyses such as ours. Only eight subjects and nineteen items are involved; and we actually employ only eight of the items. In the Cornell experiment (U)^ the eight subjects vrere can- didates for employment as interviewers. Each person interviewed each of the seven others. In each interview, each man was to obtain information about his partner. Following the interview, each person filled out a form stating his o^^m reaction and predicting what his partner would say. There are eight items, each to be judged on a four-point scale. One item is: To what extent did you feel at ease during the interview? a, very much b. a good bit c, only slightly d, not at all The respective responses are scored 1-2-3-U, Completion of the design provides seven self -descilpti ons and. Gcvon prerli rit.i ons by each man (also, seven for each man). 17 We have taken two simplifying steps which might be illegitimate for purposes other than demonstration. In every instance, we have used the average of 0*s responses over all seven interviews as his true response, x . • This discards in- formation on 0*s variation from interview to interview* Secondly, we treat J*s self-description as a "prediction of himself". This "prediction" is taken as perfectly accurate. By this device, we deal at all times with eight Judges and eight Others, and the criterion is made the same for every person. Accuracy Scores for Eiglit Persons Table 1 presents the ACC score for each person, and his score on each component. Table 2 organizes the same data to shox/ the person* s relative position in the group. Based as they are on only eight cases and eight items, these data and subsequent numerical results are illustrative, and not a proper basis for generalization. They m^ be useful to guide future studies. Magnitude of components The Differential i\ccuracy Component has substantially larger variance than the others, and therefore has much greater influence on ACC, Although Elevation has a smaller mean than Differential Elevation and Stereotype Accuracy, the variancesfor these three scores are nearly equal; they pl^ an equal part in determining individual differences in Accuracy. The correlations DEr and DAr are generally low but positive. The Stereotype Accuracy correla^ tions, however, averaged ,7l4. Relatior. of differentiation to accuracy The data illustrate our mathematical principle that any accuracy component is made smaller as the predicted standard Ji^Xii: «=! »— 4 g o CO ro O vn. O CO H oo O O -J 4=- o CO oo VjJ VJT. 1-3 ON Vn •fr- UJ ro H Q • o On • -0 \J-L • * OS O vn. • r\3 (N3 • V^ ro o ro ro t-' o PO ro O ro ro oo ro OS O a o o V*J ~0 OS vn ro VjJ On M ro CD ro ON vn ro oo NO vn O l\5 ■P- ro ro H H r.) • • • • • • vn ro o vn f \jj On o On On vn \jj ro CD H ro VjO ~o ro XT' o • • • o ro ro H vn NO • • • • • • H ro Va) H Vjl> ro CO -J vn vn oo P- ON •P- ro \-> O ON M ro P- O oo o VjO ■Cr- • O • ->3 H vn P- ro 4=- O ON vn •p- vn H CD —J NO ro ro H —0 oo ON NO ON ON —J ro vn CO vn V;l) VO Va> Vu ro Vx> vn V;J oo ro OS O Xr- ro VjJ O oo ro fV) oo ro ro P- P- ro ^ 4 W CD 18 > o o cj. ro m H y"*"^ CD w ^ ° 1^ Ci- (0 " CO t-1 s ^ > 5 «) 5. ct t+ p. o^ 4 H I o C_l. *<{ tN3 I ft g o Co o o CO O o I CO o o CO a £3* I (W CD CO p. CD p. O O o CD CO >!e ro u> vjT. OS 03 H CD -0 ON vn VjJ ro O VjJ oo jv> vn oo On vn fO Ui vn. CD ro \^i OS -<] \J\ CO CN fo VjJ VA On I\5 ro 03 ON CD vn vuo ro vn On oo ro vo TO o o 3 f '-*' CO CD C+- o (D o 4 ^ O 0) o':^ ^ ^ CD tt tt> k; o H} a CD 4 CD O c+ 2 t-' h) 2 CD H, ^ ON 2aa % % O 3 W P' M H, CD H, •< CD p hi c+ cn p. q^ 3 P- ^ CD O CD §^ CD w O O SJ o o 4 o H p- o o O O P O ■^ O I o d- cn h3 CD 21 over all items, others consistently poor« But when we exairdne the coinponents of DA, we find that (a) c is reliable over items ( cC « .79) (b) DAr, the measure of accuracy in locating others, is not ( jjC = ,18) In this sample Differential Accurajcy shows reliability because some persons have consistently low assumed similarity* This makes them consistently inaccurate predictors because Differ- ential Accuracy Correlation is generally low). No adequate estimate could be obtained for the reliability of Elevgtion, of Stereotype Accuracy, or of Accuracy over Others. We examined wl^ Accuracy has reliability much lower than Differential Accuracy, one of its components. Apparently this occurs be- cause the sign of the stereotype error has a substantial effect on accuracy on any one item and therefore lowers the correlation from item to item. Our limited data suggest the accuracy components tend to be unreliable except where reliable differences in assumed similarity affect the component. Stone and Leavitt (28) report very low consistency (-.0? to 30) of accxiracy scores in pre- dicting different children on a fairly long test, but a median consistent of .63 between two halves of the test for the same child. They then trace the latter consistency to consistent favorable sets toward a given child, and to assumed similarity. Further work is needed to establish which independent components of Accuracy can be reliably measured. In Table 2 we note that Number 1 is consistently superior 22 on various components of accuracy and ^k is consistently inferior. But #7, the best predictor as judged 1:^ Differential Accuracy Corfelxticnls the poorest on Differential Elevation CbrrcsLstion and next-to-poorest on Stereotype Aceuracy- Ccmpcnent* Table 3 presents the intercorrelations of the eight measures of accuracy. M asterisk indicates pairs of variates which are experimentally linked; these correlations are higher than would be expected from independent measurements. Being based on only eight cases, the correlations cannot be interpreted confidently. The correlations are low but many of them are as high as the accorrpanying reliabilitfes. Only one firm recommendation can presently be made. Future studies of predictive accuracy should measure the com- ponents separately, preferably using two independent sets of items and Others, Such measurement mXL permit accurate determina- tion of reliabilities, of the relation between the components, and of their relation, if ar^r, to external criteria. Only after such research can we decide how many important components are present i^thin the overall Accuracy score presently used in most research on social perception and which unwanted components must be suppcreeBedby appropriate design of tests and scoring keys. Illustrative Analysis of Assumed Similarity Scores In Table I;, the Assumed Similarity scores of the eight Judges are divided into components. Table $ presents the same information in rank form, and Table 6 presents the intercorrela?- tion. The relatively large variance of ADI, Assumed Dispersion on each item, indicates that it bs© great influence on individual .'o'f'i t- 22a CD CO O 'OT. UJ ro H o H Uj t- ro H rsD v^ • • • • w • • H o H P vn ro f=- ^=- H VJT. ■p- H tK ■p- U) H • • • ON -o OD w CO VO ro VjJ H ro ro O O H g On O On VjJ vn H H ro vn On ^ ro VJ-l. NO H V\ ON vrv ON NO CX) CO Vn. On H H On O H On CO CO CO -0 VjJ o Va) O NO vn VjJ o ro p OD VO -0 ro NO GO NO ON H NO O On O NO 4=r ro ro VjJ H ^ ON •P- o VjJ —3 CD ro vn P o NO Va) On NO -0 w CD cn H« CD P ^ hi p W W- O e £3 CO ^=• O 03 CD ct- H- S> D^ CO &3 H'TJ en J:s o c; HMO) c^- H- p. CD O P 3 M CO CD P^ H ^. w o o »i CD CO 1-3 H CD 23 o s- CD 2 O 3 & * I ON ON NO * VJT. VA * I H NO On CX) ^U O Hj O Hj • ro ^ ^ {=- p CD O 2 '<^ d- H- Jr.' o o o ^ 0) H !xi P? ^ b«j H- X H c+ (D « CD H- (D o CD p. 3 o 3 O cl- ^ ^ o i 03 ^ {r^ d- o O CD 13 O 4 CD fi CD 3 4 O d- P .<^ co ^^ CD 2li differences in overall Assumed Similarity, The correlations show great overlap of Assumed Elevation Xidth Assumed Dispersion in Elevation, and Assumed Self -Typicality with Assumed Dispersion \d.thin Items, The tendency to differen- tiate emcng Others is accompanied by a tendency to differentiate the average Other from oneself. This result is partly an arti- fact, resulting from using each person's self-description as one of his "predictions". Even allowing for this, our correlation suggests treating only two components of AS: Assumed Similarity in Elevation (AE + ADE) and Assumed Similarity in Pattern (AST + ADI), The correlation between these two variables is negligible ( X> = •21), Further evidence is required to determine how to divide Assumed Similarity snd which components merit serious investigation. Correlation of Assumed Similarity with Accuracy Table 7 gives the correlations of Assumed Similarity components with Accuracy components. The Judge «s "Implicit Personality Theory" We turn now to an aspect of social perception data which msy prove to be particularly significant, VJhen a Judge makes predictions for a large number of Others, these predictions define a corresponding distribution of points in the variate space. This distribution mary be regarded as a description of the generalized Other, representing the Judge's view of both central tendency and individual differences. The Judge's generalized perception may be an insert ant indicator of his expectations regarding others, lie shall discuss the general VjJ OD ro 4=- VjT. H O oo On VJT. VuO ro H \jj 4=- CO -0 [^O o H ro ON CO vn v>^ OO -o On ro \JT. •Fr- \x> VjJ va OO ON !=- M rv:) OO ON vrv VjJ J-J o 05- CD M H« CD go. <+ H- t- IS' M cn 3 CD d •-J 3 H M CD <+ H- p. CD O CO CO 3 Cfl CO n 4 1-^ CD vn. ^f o c+ O CD % o p- p 4 B- O o O •t:^ W Sj (!> CO 1 Jj- H- CD 5-§ o- O M H' c+ to O 1 t' o H- ;::) y-j Ho ^ G CD c a poor predictor, we determined the mean, variance, and covariance of his predictions. The matrix of co- variances was factored by a pivotal method akin to square-root factor analysis (31)> intended to yield interpretable factors. Table 8 shoxjs the loadings on three factors with item means and ■'.'ij. ■".:?■- 29 variances. The means for Judge 3 show no striking features, especially xrhen considered in relation to the true means presented below. The variances indicate that 7^3 regards others as f airily uniform in their awareness of him (item 3)^ and as varying especially in their openness, ease, and feeling of dominance (items 1,6,8), The first two dimensions of #3*s perceptual space crc plotted as Figure 1, Little confidence can be placed in factors based on eight cases, but we would otherwise interpret Factor I as repre- senting a feeling of being under pressure. It is notable that #3 regards those persons who are most open (item 1) as being least at ease (item 6). Factor II shows a link between items U and 5^ getting and giving information. Factor III is indis- tinct. It is notable that items 6 and 7 are correlated; a "good interviewing relation" is perceived by #3 as one wheire the inter- viewer is at ease J Such a finding regarding #3's perception, if better substantiated, might have much diagnostic importance. The literature contains many studies of correlation betx-jeen ratings which bear on the perceiver*s frame of reference. The studies of halo effect in rating suggest the existence of a strong general good-bad factor. These studies have not examined raters separately, Frenkel-Brunswik reports that ethnocentric individuals see others in black-and-white terms, the "good", "strong" traits going together (l5> See also 25), She does not present correla- tional data, but she is essentislly stating that halo effects are stronger in such raters. In our language, their covariance matrix is loaded with one factor, vrliile non-authoritarians use 'li'.'i-i.uj 30 maDy factors and do jiot emphasize the general evaluative dimension, Steiner (2?) has substantiated this conclusion, and discusses his results in terms of the perceiver's "trait contingencies", that is, in terms of the perceiver's frame of reference. "The indi- vidual's assumption that certain attributes belong together is expected to influence his percept of the person with xirhom he is interacting" (p.3l49). Steiner 's data are restricted to group differences, but his theory is not. Our position differs slightly from steiner 's in that we emphasize the implicit contingencies of which the perceiver may be quite unaware, Steiner 's method, in its present form, requires the psrcefcier to say explicitly what contingencies he expects. Two other studies show differences in the perceptual ref- erence frame of groups, Wickman's well known stucfy (30) showed that teachers expected different traits to correlate with mental health than did mental hygienists, Moore (21) performed a factor analysis of ratings given non-commissioned officers by their subordinates, and also of ratings given by their superiors. The factor patterns differed. For instance, superiors coupled "leadership" with eagenness and responsibility, but the sub- ordinates viewed "leadership" as closely linked with intelligence and skill. None of these studies of groups examines the perceptual space by which an individual describes personality, but the evidence supports the belief that important individual differ- ences exist. In view of our interpretation of the perceptual distribution as an implicit personali+y theory, special interest at-, V 31 would attach to studies or ratings given by clinical psychologists or psychiatrists of different schools, or having different amounts of training. One objective of instruction in the field of person- ality is to modify oversimple views students may hold. If our procedure does reveal covert and unconscious conceptions, it may be a usef-ul device for evaluation. Effect on accuracy scores . The Judge's distribution of Others has been interpreted here as a standing system of meanings which delimts the space within which he locates Others, It is obvious that any such delimitation would affect social perception scores. Discrepancies between perceived mean and actual mean lower Stereotype Accuracy, 1/e have shown earlier that Accuracy declines if Assumed Dispersion departs from an optimal value. The corre- • lational effects are a bit less easy to perceive. Correlations describe the shape of the distribution of Others, If traits 1 and 2 are uncorrelated, then Others will have a roughly circular bivariate distribution. If a Judge re- gards 1 and 2 as correlated, attributing both to the same persons, his perceived distribution will be elliptical. His perceived dis- persion along the dimension 1+2 will be greater than in the true responses, and his accuracy will suffer, ife can view the example in another w^. Suppose the judge predicts variate 1 perfectly but believes that variates 1 and 2 correlate 1,00 when they have a true correlation of zero — - then he must have substantial error in predicting variate 2, He can predict 2 accurately only if he perceives the covariance of 1 xd-th 2 accurately. ,;♦^.^^? 32 Data reported by Crow (11, p.86) show this phenomenon clearly. As part of a larger stuc^, he asked Judges to predict what would be the first word missed by a patient on a vocabulary test and what was the highest level attained (called tasks Dl and D2.) The correlation of Judges' accuracy on Dl t.jith accuracy on 12 was positive and significant for five of ten patients, but negative and significant on two patients. Judges tended to expect a correlation between the two scores, and when there was a true correlation they did well; where it was negative, the Judges could not be accurate on both predictions. There was a rank correlation of ,97 (over patients) between consistency of accuracy scores, and consistency of the patient's performance. The Cornell data were examined to determine the covariance between items in self-descriptions. The resulting "criterion" matrix was factored, with the results shown in Table 9 and plotted in Figure 2 (first two factors). This pattern is different from that of #3 (Table 8) in several respects. Notably #3 overdifferen- tiates on all items. The first factor for #3 lumps openness and lack of receptivenessj these variables are divided among two factors in the criterion. In the criterion, being at ease (item 6) is positively related to openness. It is especially interesting that "feeling like the person being interviewed" is, for the group as a whole, positively correlated with being at ease; but for #3 these items are negatively correlated. With a view of people so discrepant from the facts, it is not surprising that #3 has a low ACC score. hat.. 33 o o CD CD o CO p CD P p 3 H> ci- c+ H- 03 <^ O ►9 CD H> p: O CD ro O • ro H VJT. —3 M • O ON ro -&^ ■&?^ 4r- VjJ (DO • H CX) c» ro CD • -v3 f>0 On -J CD c:o t o VJT, |:r Vo ro 1 i^^ 1 M iro « !• • • ■ • • CO ;vo ON 4=- -J KjJ \y^ OD iro CD CJN On ro ivn 1 1 I H M • • • • • • o U) -^ H VjJ t-J CO o OS -0 vn. ON NO ON CO \f ^ ro CO CJN o iro On ON ro On M VJ-l H ro H ro -J • « • • • • • • 4r- ro NO OD o ro On VjO to On O ON 4=- CO vn. VJT. ON Vjj ro NO ro On • • (r- ro ro ro M Vjl) -J • • • • • H VoJ OO -J \n ON ■pr M -J CD e; —3 vn NO V>J H ro • • CO \jj O NO M O 3 •-d o C/3 w CD Pj p M O 5 M 3 OP 3 CD C+ O p. H, o t3 ►^ M tJ M O M ^ < CD H- CD H- O 02 CO c+ o c+ o c_ 1-3 p; pj Cb o' CD CD VjO O o -9, in press (Abstract), k» Bronfenbrenner, U. and others. The analysis of social sensitivity (empathy), Amer, Psychologjist, 1952, 7 (Abstract), 5, Cass, Loretta K, Parent-child relationships and delinq- uency, J, abnorm, soc, Psychol, , 1952, U7, 101-10!|, 6, Chowdry, Kalma, and Newcomb, T, M, Ihe relative abilities of leaders and ncn-leaders to estimate opinions of their own groups, J, abnorm, soc. fsychol. , 1952, U7, 51-57. ""■ 7, Cronbach, L, J, Further evidence on response sets and test design, Educ. psychol, Ileasmt , 1950, 10, 3-31, 8, Cronbach, L, J, Coefficient alpha and the internal structure of tests, Psychometrika, 1951, l6, 297-33U. 9* Cronbach, L, J, Educational psychology . New York: Harcourt, Brace, 195 U, 10. Cronbach, L, J., and Gleser, Goldine C, Assessing similarity between profiles, Psychol, Bull ., 1953, 50, U56-U73. 11« Crow, W, J, A methodological study of social perceptiveness. Ph. D, Thesis, U. Colorado, 195U. 12, Dymond, Rosalind F, Personality and empathy. J, consulting psychol. 1950, lU, 3U3-350. 13* Fiedler, F. E. A method of objective quantification of certain counter-transference attitudes. J. clin. Psychol., 1951, 7, 101-107. lU, Fiedler, F, E. Assumed similarity measures as predictors of team effectiveness, J, abnorm. soc. Psychol. , 195U, h9, 381-388. 15« Frenkel-Brunswik, Else. A study of prejudice in children. Human Relations, 19U8, 1, 295-306. l6» Gage, N. L» Judging intei'ests from expressive behavior. Psychol, Ifonogr ., 19^2, 66, No, l8 (VJhole No. 3^0). 17. Gage, N, L, Accuracy of social perception and effectiveness in interpersonal relationships, J. Person ., 1953 ^ 22, 128-ltil. "~ 18, Greer, L, F,, Galanter, E. H., and Nordlie, P. G. Inter- personal knowledge and individual and group effectivoriess, J. abnorm. soc, Psychol ., 19$kf h9, i4ll-l4lU. !?• Kelly, E, L, and Fiske, D. W, The prediction of performance in clinical psychology . Ann Arbor: Univer, of Michigan Press, 1951. 20, Lorge, I, and Diamond, Lorraine K, The value of information to good and poor judges of item difficulty, Ed. Psychol, Measmt., 195U, Ih^ 29-33. 21, Moore, J, V, Factor analytic comparisons of superior and subordinate ratings of the same WCO supervis'-rs. HRRC Teclinical Report 53-2li, July, 19^3. 22, Norman, R, D. and Ainsworth, Patricia. The relationships among projection, empathy, reality, and adjustment, operationally defined, J. consult. Psychol , 195U, 18, 53-58. "~ 23, Osgood, C. E, The nature and measurement of meaning, Psychol. Bull ,, 1952, U9, 197-237. 2I|.. Postman, L, Perception, motivation and behavior, J« Person ., 1953, 22, 17-31. 25, Scodel, A, and Mussen, P. Social perceptions of authori- tarians and nonauthoritarians. J. abncrm* soc. Psychol., 1953, U8, 181-18U. 26, Sopchak, A, L, Parential "identification" and "tendency toward disorders" as measured by the Minnesota Multi- phasic Personality Inventory. J. abnomi. soc. Psychol. , 1952, U7, 159-165. 27, Steiner, I, D, Ethnocentrism and tolerance of trait "incon- sistency", J. abnorm. soc, Psychol. , 195U, U9, 3h9-35U» 28, Stone, G, C, and Leavitt, G. S, Generality of accuracy in perceiving standard persons. Paper presented to Mid- western Psychological Association, M^, 195U. 29, Talland, G. A, The assessment of group opinion by leaders, and their influence on its formation. J, abnorm. soc , Psychol., 195U, i±2j i431-U3U. Il2 30, Wickman, E, K» Children's behavior and teachers' attitudes . New York J Commonwealth Fund, 1925. 31» Urigley, C. and McQuitty L. L, The square root method of factor analysis: a re -examination and a shortened procedure, Ilimeogrsphed, Univer, of Illinois, 19^3,