m ■IIP ii\/i.ii.i ASo n Figure 1. Relations Yielding Certain AS Scores The first investigations in our series used a forced-choice instrument to obtain the descriptions (3,4). That instrument yielded low reliabilities for RS scores, and not very high reliabilities for AS (7). "While that study was in progress, factors which limit the reliability and validity of forced- choice and Q-sort techniques, which had not been previously recognized, were uncovered by mathematical analysis (1,8). Hence we shifted to a free- response instrument in the present surveyor study. (It may be renaarked parenthetically that the relative merits of forced-choice and free -response instruments in personality study are somewhat uncertain at present. Neither .^vidence nor theory now available is wholly consistent in support of either position.) Statennents were prepared to be judged in one of seven categories, in the naanner of a Likert-type attitude scale. Responses fell on a continuunrx from "definitely true" to "definitely untrue." Items were written to tap several areas of personality which were judged a priori to have possible relevance to group effectiveness. The pool of items was pretested on 200 radio technicians at Chanute Air Force Base. Items were selected from the pool on the basis of various criteria designed to enhance discrimination and to yield fairly homogeneous clusters of items. Cf the 60 itenr^s selected fronm the original pool, only 43 appeared to belong to one or another cluster. The names given to the clusters are somewhat arbitrary, since the psychological unity underlying the inte re or relations is sometimes obscure. At this point in our investigations, the value of cluster scoring was not certain, and efforts to produce clusters were not carried as far as they would be in instrument developnaent where the clusters were a miajor concern. Data for the present analysis come from the population of student surveyors described in (2). 39 subjects tested at Urbana and 33 others tested in Minnesota were used in general. In sonne instances, only a part of the cases was used where the analysis was not inaportant enough to warrant treat- ing all cases. There is no reason to think that, where only part of the cases was used,, this selection biased results. The Clusters: Homogeneity and Intercorrelation Internal consistency of each cluster was determined by the split-half technique, each cluster being split into two parts on a priori grounds to assure better -than-average equivalence of the parts. It will be noted that clusters contained only eight or nine items. Under these circumstances we could not expect the split-half reliability to be very stable from split to split. The resulting split-half coefficients, corrected by the Spearnrian-Brovvn formula, are given in Table 1. We discuss them cluster by cluster. C luster I. Self-Confidence . The eight items in this cluster included the following: 1 am likely to try out my ideas and not worry about the opinion of others. I often find that I have made up my mind too late. (-) The direction of scoring was such that the total score represented confidence. The symbol (-) indicates an item reversed in scoring. The cluster appears homogeneous under directions s^, p, andj^. For some reason, the cluster loses this homogeneity under the n directions- This leads to interesting reflections about the perception of personality. Qualities perceived together, in thinking about the ideal or preferred person, are separated in thinking about the not-preferred person. Such an analysis is irrelevant to our present purpose, and the data at hand are too limited to warrant present pursuit of this lead. Cluster II. Gregariousness. Naming this trait proved difficult- It was originally intended to be a conventional nneasure of sociability. The items that survived, however, had a distinctive quality of insulation, or rejection of others. Itenns include: I would not want to take another person fully into my confidence. (-) I am. very discriminating in my choice of friends- (-) This cluster has fairly satisfactory homogeneity, considering its length. The lower reliability for the _i directions need not be regarded as significant. Cluster III. Reflectiveness. This trait is perhaps similar to the Thinking Introversion and Theoretical traits of other tests. Items include: I try to take an active part in civic and government affairs. I ana more interested in what a person does than in why he does it. (-) Consistency is relatively low for this scale, probably because items having only peripherally related content were used to lengthen the scale. Cluster IV. (Doubtful meaning). The low internal consistency for this trait indicates that it was probably assembled on the basis of chance intercorrelations in the Chanute sample on which we had pretested items. It will not be treated further in this report. Cluster V. Aesthetic Values. Items include: I can get very affected, pcrhapr. feel like crying, in a sad movie- Foul language disgusts me. I ain sensitive to color and color schemes. -4- Table 1 Split -Half Reliabilities, Means and Standard Deviations for Cluster Scores (N = 72) Cluster I Cluster II Cluster III Self-Confidence Gregariousness Reflectiveness Mean S.D. Mean S.D. Mean S.D. s P .58 36.0 6.9 .46 36.1 5.3 n (-.02)* 32.2 4.6 i .61 42.6 6.1 51 34.2 6.6 .21 30.5 5.1 61 32.6 5.6 .29 31.0 5.8 74 30.2 7.2 .47 30.0 5.7 21 35.3 6.1 .50 35.0 5.2 Cluster IV r Mean S.D. Cluster V Aesthetic Values r Mean S.D. s (-.07)-;^ 32.0 UoU .31 32.5 7.0 P .15 31»6 " 5oO c61i 31.0 6.7 n •Uo 29o7 Uo9 e70 31.U 8o6 1 (-.06)-;^ 30o3 k.$ o35 36,U 6»8 Median r** ^uncorrected **Cluster IV omitted .31 .17 .35 -5- Reliabilities for Cluster V ran moderately high. It is strange that in this trait, where people are describing their preferences rather than their actions, the means should be what they are. In this trait the mean ideal score is high, relative to _s,_p, and n; but the preferred person is said to have less of this trait than either self or the rejected person. The difference between _g_ and _i is large enough th^t it might have arisen on a non-chance basis. Reading down Table 1, we find that the reliabilities for the four sets of directions are connparable. By this criterion, all directions appear equally usable. It is noteworthy that individual differences in ideal are as distinguishable as differences in self description. It might have been anticipated that, on the contrary, people would so agree in their ideals that no reliable differentiation was possible. Table 2A presents the intercorrelations of clusters under £ directions. The clusters are moderately independent., although I overlaps III, and II overlaps V, in the_s and other responses. These intercorrelations will later be compared with intercorrelations of AS measures for the clusters. Independence of Separate Descriptions and AS Measures There would be little value in obtaining several descriptions from each man, if the responses of one trial could be predicted from responses under other directions. Table 3 presents the correlations within clusters, and the median over clusters, omitting IV. The correlations are generally positive, except, as expected, that n correlates negatively with the other sorts. The correlation between £ and p and that between s and i are so high relative to the reliabilities as to suggest that these three procedures do not provide appreciably different information. Assumed Similarity in a cluster can be measured by the difference between two cluster scores, ordinarily without regard to sign. Thus ASPt would be Is, - pj ^ ,. , , , 1 I II • According to the usual theory regarding the reliability of. difference scores, these cluster scores are so highly intercorrelated that we could not expect the cluster-scored assumed similarity to be reliable. The AS measures involving the n sort provide an exception. V/ith low correlations between n and s or p, we can expect cluster ASn and cluster ASo to be as reliable as the cluster scores themselves. When we turn to an examination of AS scored in the usual manner, item by item, we have a quite different picture. To get an item ASp score, -6- Table 2A Intercorrelations of Clusters under £ Directions (N = 72) I II UI V Self- Gregar- Reflect- Aesthetic Confidence iousness iveness Values I. Self-Confidence II. Gregariousness III. Reflectiveness V. Aesthetic Values (.^8) .Ok (•51) ^26 -•18 (.21) • 20 -•31 .25 (.31) Table 2B Median Intercorrelations between Clusters * (N^72) Median I r I _s directions £ directions n directions i directions .22 .21 .09 .12 * Cluster IV omitted. •7- Cluster 1 Self -Confidence Table 3 Intercorrelations of Descriptions within Clusters (N = 72) Cluster II Gregariousness n s (.58) o36 c32 -.21 P i.kS) -32 -.05 i (e6l) -.05 ** (.00) 3 (.51) .53 -.08 -.27 P (.61) .07 -^36 i (..21) -,05 n (.7i|) Cluster III Reflectiveness s p i n s (.21) ,k2 .U7 -o20 P (.29) i •2U -c09 (c50) -.21; n (.U7) Cluster V Aesthetic Values n s («3i) ,U8 .6U -.12 P (.61i) .55 -.01 1 n (.35) -.06 (.70) Median Intercorrelations n s P i n (.Ul) .U5 .39 -.20 (,53) .28 -.07 (cii2) -.06 (.58) -8- we take the_s score on the item, the p score on the item for the same person, and square the difference. These Values are summed over items to get an AS score. Taking the square root is required to transform this to the distance scale which is most interpretable, but use of rank correlations makes this transfornnation unnecessary in our calculations. The split-half reliabilities for^ASscores obtained item by item are presented in the diagonals of Table 4, and are extremely high. Obviously, the correlation between descriptions, cluster scoredi does not prevent reliable measurement of AS item-by-item. This will be explained below. The intercorrelations between AS scores are also given in Table 4. AS n and ASo are so highly correlated as to suggest that they will yield little distinct infornnation. One or the other should be dropped fronn further work, although they might be combined into a single measure. ASp is more independent of the other scores. AS p and AS Ip have so much overlap that retention of both in future work seems not likely to be profitable. Now let us look at the correlation pf AS measures from cluster to cluster. ASo was computed for each person on each cluster by two methods. For cluster scoring, the absolute difference, |p - n j , on each cluster was used. For itenn scoring, differences on items were squared and summed, for the items in each cluster. The correlations are given in Tables 5A and 5B. V/e imnnediately see that for cluster scoring (Table 5B) the correlations are low. There is a slight trend to positive correlations, suggesting that some overall general quality has very small loadings in all these ASo scores. When we look at ASo item scored, however, we find marked positive intercorrelations even with Cluster IV which is not reliable as a trait measure. For Cluster II a split -half reliability of ASo item scored was found to be .46; the intercorrelations are also of this magnitude. These higl^ correlations are to be explained, we think, in only one way. This conclusion was originally urged as probabfe by Fiedler, but seemed inaprobable to Cronbach and therefore has been tested with considerable care. The data appear to have established beyond question that Assumed Similarity is a mental set or perceptual tendency which influences a person's behavior regardless of the content of the items he is marking. The original aim of cluster scoring was to reduce error. It was argued that error present in £ andji ratings would .accumulate to a. very substantia^ error in the final ASo score. In cluster scoring, errors in p and n were exjvoctcci to v'.ancei out to some degree. Yet, it is wlih 'item -9- Table 4 Rank Order Intercorrelations of AS Measures (N = 39) ASp ASn ASo ASIp ASp (,83) »53 «38 •62 ASn (.95) ,7U ASo («93) ASIp (.73) (Note: Diagonal entries are split-half reliabilities ( rho ).) -10- Table 5A Rank Order Correlations between Clusters, Item-Scored ASo Scores (N = 35) I II III IV V Self- Gregar- Reflect- Aesthetic Confidence iousness iveness Values I. Self-Confidence II. Gregariousness 1X1. Reflectiveness XV. V. Aesthetic Values >li8 .$9 •25 a32 .18 .31 .52 Table 5B Rank Order Correlations between Clusters, Cluster-Scored ASo Scores (N = 35) I II III Self- Gregar- Reflect- Confidence iousness iveness IV V . Aesthetic Values I. Self-Confidence II. Gregariousness III. Reflectiveness IV. V. Aesthetic Values (o3l) .20 (.56) .08 .10 (.33) .12 .Oil .10 (.28) (Note: Diagonal entries are split-half reliabilities (rho).) .18 -.08 .08 .22 (.53) -11- scoring" that appreciable cluster intercorrelations are found. This implies that homogeneity of content within a cluster is of less importance than the , reliability of the set to maintain differences between ratings, since a procedure which allows more error in itena responses produces a difference score more stable from cluster to cluster. Scores under one set of directions were correlated with those under another, often nearly to the extent of their reliabilities (Table 3). The reliabilities of AS scoires (Table 4) are substantial. An AS score, though, is the result of combining differences between responses under two different directions. The only way one can obtain reliable differences between variables which are themselves correlated to the extent of their reliabilities is to have "correlated error." In other words, subjects tend to have a pervading set to maintain differences between person* as they describe them, regardless of the statements on which this description is nnade or even regardless of whether the itenns have any consistency among themselves. Our analysis does not rule out the possibility that in addition to this general set there may be more specific tendencies. A person might be average in overall ASo, and have greater than average tendency to assunne sinnilarity of others in some one trait. If AS is a set which transcends item content, what does it mean? This we cannot say with any certainty. We have not yet established that a person who shows AS in a particular setting, with a pencil and paper test will show that same set in other relations with persons. The fact that AS scores have been significantly correlated with external criteria in some of Fiedler's studies is evidence that we are dealing with nnore than a transient verbal set. We are impressed by the possibility that AS represents a tendency to perceive others as alike, or conversely that the person who receives a low AS score tends to be alert to real or fancied differences among others. This would correspond to George Klein's evidence that people differ along a "leveling- sharpening" continuum, as judged by their tendency to differentiate when perceiving laboratory stimuli (6). Further work will be required to test this possible interpretation of AS. A person who seeks to differentiate may not be an accurate percciver of others. Gage provides evidence that a description of "people in general" is more likely to fit a particular other person than a description given under directions demanding a deliberate attempt to differentiate and predict him as an ii:idividuai (5). Hence the person who seeks to differentiate (low AS) may therefore be less accurate. -12- Conclusions The instrument in the surveyor study has excellent reliability for pleasuring Assumed Sinnilarity. ASo determined from differences on the items considered separately has a reliability of .93, ASp of .83, ASn of .95. This method of scoring is superior to a cluster-scoring procedure for determin- ing AS. The internal consistency of the separate clusters is low. One cluster proved worthless; the other clusters have reliabilities in the neighborhood of .40. The clusters are not intercorrelated to a substantial degree. Responses obtained under^, £, and ^ directions are so highly correlated that obtaining separate scores for each of thenn is not warranted. Similarly, it is found that ASo and ASn are so highly correlated that exanaining them separately in future studies is not advisable. However, in spite of the high correlation of s^ withjp, it is found that ASp can be nrxeasured with good reliability and is partly independent of J^Sn and ASo. ASo determined from differences on itenns within a cluster correlates with ASo in another cluster so highly as to indicate that the AS score does not depend on the content of the items. Assumed Similarity is therefore to be interpreted as a general attitude, or mental set, essentially independent of the content of the test items. The tendency to assume sinnilarity between persons may be an important aspect of personality. The relation of this trait to other aspects of personality should be studied further.