Multimethod Replication of the AB5C Model John A. Johnson Penn state DuBois Campus

. ~.~ \:... ~ \. \. l, (!()P~I \' >' ~\ "., Multimethod Replication of AB5C ~. 1 Multimethod Replication of the AB5C Model John A. Johnson Penn state DuBois Campus Paper presented at the sixth Conference, European Association for Personality psychology, Groningen, The Netherlands, June 18, 1992. correspondence to: Dr. John A. Johnson Penn State DuBois Campus DuBois, PA 15801 USA Phone: (814) 375-4774 FAX: (814) 375-4784 BITNET: J5J@PSUVM.BITNET Internet: j5j@psuvm.psu.edu

---------~-- --- Multimethod Replication of AB5C 2 Abstract Personality psychologists who use inventories to assess personality have sometimes criticized the lexical tradition, which includes the AB5C version of the Five-Factor Model (FFM) as being overconcerned with the structure of language and underconcerned with the structure of personality in real people. This paper argues, first, that purely lexical AB5C circumplexes are very similar to inventory-generated circumplexes, and second, that AB5C analyses might be useful in interpreting inventory scale scores. To support these claims, AB5C circumplexes were built by mapping trait words gathered via acquaintance ratings onto scales from two widely used self-report inventories--the California Psychological Inventory (both American and German versions) and the Hogan Personality Inventory. The resulting inventory by trait rating circumplexes, when properly rotated, correspond well to circumplexes generated entirely from trait ratings alone. Practical implications for configural interpretation of personality inventory scores via AB5C are discussed.

--~-- ~~----- ~-~-~----- Multimethod Replication of AB5C 4 traits. The structure of trait adjectives is not necessarily equivalent to the structure of traits." (R. R. McCrae, personal communication, May 23, 1991). Rooted squarely in the lexical approach, the Abridged Five Dimensional Circumplex (AB5C: Hofstee, de Raad, & Goldberg, in press) is formally no different from the FFM with regard to the internal structure of personality. The AB5C simply represents an integration of the traditional FFM with the Wiggins (1979) two-dimensional circumplex model of personality. Hofstee et ale note that Wiggins's circumplex locates traits within a plane defined by the Extraversion and Agreeableness factors of the FFM, but that a complete integration of the FFM and circumplex model would plot traits in the 10 planes generated by all possible combinations of the five factors. The circumplex whose two factors show the largest multiple B with a trait determines to which of the 10 circumplexes an adjective is assigned. For example, "respectful" is a blend of Agreeableness and conscientiousness (11+111+ or A+C+ in AB5C terms), while "unconventional" is a blend of (low) Conscientiousness and Openness (111-V+ or C-O+). Despite its lexical basis, AB5C perhaps comes a little closer to what personality inventory users would like from a model of the trait descriptive universe. Because the AB5C model considers traits as factor blends, it contains a sort of dynamism lacking in the FFM. Consider an observation by McCrae and John

Multimethod Replication of ABSC (in press) along these lines: "Open people are inquisitive. If they are also conscientious, their curiosity may take the form of sustained and systematic study of a topic, if they are low in C, theirs will be an idle curiosity, absorbed by the passing interest of the moment. Theoretical elaboration of such interactions of factors can bring a more dynamic flavor to trait psychology." To further persuade personality inventories users of the merits of ABSC, this paper attempts to demonstrate that AB5C circumplexes generated by plotting acquaintance ratings on selfreport inventory scores are virtually identical to AB5C circumplexes generated from the "purely cartographical" or lexical approach of Hofstee, et al. The lexical approach factoranalyzes ratings (self- and/or acquaintance) on individual trait words, and then regresses all trait ratings on all possible pairs of 5 (varimax-rotated) factors, generating ten circumplexes. The present study regresses acquaintance trait ratings on all possible pairs of 5 (standardized) scale scores from three different self-report inventories (two English and one German). The inventory-generated circumplexes are compared to the lexical circumplexes and to each other. It was hypothesized that the circumplexes would be simply rotated versions of each other. Method Subjects ~ Measures Three subject groups were used: 154 (68 male, 86 female) 5

6 introductory psychology students I taught at Penn state in 1984 85: 75 (30 male, 45 female) similar students I taught in 1987; and 89 (31 male, 58 female) University of Bielefeld students who volunteered to participate in research I conducted in 1991. Subjects tested in 1984 were rated by three acquaintances, in 1985 by one acquaintance, and in 1987 and 1991 by two acquaintances on 49 bipolar adjective ratings scales (BARS; Johnson, 1987, 1991a; Johnson, Germer, Efran, & overton, 1988). Ratings were averaged within the 1984, 1987, and 1991 samples. The BARS was designed specifically to assess personality according to the FFM, and the individual Likert scales can be aggregated to produce five scale scores. However, for the present study, individual adjective scales were used (reliability/validity data for some of these individual scales is presented in Johnson, Cheek, & Smither, 1983). All American subjects had taken the 480-item version of the California Psychological Inventory (CPI; Gough, 1975); 1987 subjects also took the Hogan Personality Inventory (HPI; Hogan, 1986). German subjects took a prepublished, but officially authorized, 462-item version of the California Psychological Inventory, kindly provided by Ansfried Weinert. Analyses The purely lexical circumplexes were generated by regressing acquaintance trait ratings on the factor scores from a 5-factor, varimax rotated factor analysis of all of the American trait

7 ratings. Both the 480-item and 462-item versions of the CPI were scored by a method previously described (Johnson, 1991b) to yield E,A,C,S, & 0 scores from the FFM. HPI scale scores for Sociability, Likeability, Prudence, Adjustment, and Intellectance were used because they represent the best HPI markers for E,A,C,S, & 0 (Goldberg, 1992). All scale scores were standardized to ~-scores within sample norms for each gender. Peer ratings for the 98 BARS adjectives were plotted on all possible pairings of the standardized CPI or HPI FFM scales, where the horizontal distance from the origin is the Pearson product-moment correlation between the adjective and the x-axis inventory score, and the vertical distance by the adjective's correlation with the y-axis inventory score. The angular displacement of each adjective from the x-axis is given by e = tan-'(y/x), and the length of the vector, which is the multiple B of the two inventory scores on the adjective score, is given by B=(x2+y2)~. The above procedure is essentially the converse of the procedure employed by Wiggins and Broughton (1991), who plotted scale scores on axes defined by their adjectives. Following the convention of Hofstee, et ale (in press), only traits with vector lengths (Bs) of at least.20 were retained for further analyses of the circumplexes. The following comparisons were made: (1) the ten lexical circumplexes with their corresponding American CPI, HPI, and German CPI circumplexes;

(2) American CPI with HPI circumplexes; and (3) American with German CPI circumplexes. Angular location of all traits common to the circumplexes being compared were correlated. These correlations were calculated for the 10 circumplexes as a group and for each of the 10 circumplexes separately. (In cases where the two angular locations differed by more than 180, 360 was subtracted from the larger angle). If corresponding circumplexes generated from different methods are essentially rotated versions of each other, the sets of angular locations should correlate highly (>.9) with each other. Furthermore, examination of the regression equation, y=bx+~ should yield a slope (Q) close to unity and also indicate with the intercept constant (~) how many degrees the second circumplex must be rotated to correspond to the first. Circumplexes were plotted and superimposed upon each other, and then the second circumplex was manually rotated by the amount ~. This provided a visual representation of goodness-of-fit between corresponding circumplexes. Results Table 1 summarizes the number of adjectives demonstrating significant (>.20) vector lengths (Bs) within each method of assessing the FFM. Far more significant vector lengths were found in the internally generated circumplexes than in the circumplexes generated with the CPI and HPI. This is not surprising, given that CPI and HPI circumplexes involve both 8

different sources of information (acquaintance versus self) and different instrumental methods (rating scales versus inventory scales). 9 Insert Table 1 about here Table 1 also indicates the number of traits with significant vector lengths in pairs circumplexes, because comparisons of circumplexes are limited necessarily to traits shared in common. This table shows that roughly half of a circumplex's significant traits are also significant in other circumplexes. An informal perusal of traits that did not "replicate" across circumplexes found that often their vector lengths barely failed to reach the.20 cutting point. Also, their meaning was not radically different from neighboring traits that did replicate. For example, "conservative" and "conventional" might lie within a few degrees of each other in a circumplex, but only one of the traits had a vector length >.20 in a second circumplex. This indicates that the information "lost" from a circumplex in the comparisons is not especially unique or important. Table 2 summarizes the comparisons of the lexical circumplex (generated by American BARS traits on BARS factors) with circumplexes generated by plotting traits on American CPI, HPI, and German CPI scale scores. This table indicates a high degree of correspondence of the ordering of adjectives on the lexical

Multimethod Replication of ABSC 10 and inventory circumplexes, with Pearson correlations ranging from.94 to.99 for the American CPI,.88 to.99 for the HPI, and.83 to.98 for the German CPl. In the regression equations, the b weights ranged from.77 to 1.05, and the A values ranged from 23 to 59.5. Insert Table 2 about here Table 3 shows comparisons between the American CPI and HPI circumplexes, and Table 4 shows comparisons between the American CPI and German CPI circumplexes. In both cases, ordering of traits was again highly similar across circumplexes. Insert Tables 3 and 4 about here Figures 1-3 graphically illustrate examples of the circumplex comparisons. Figure 1 compares the lexical and American HPI conscientiousness by Openness circumplexes. Figure 2 compares the American CPI and HPI Extraversion by Agreeableness circumplexes. Figure 3 compares the American CPI and German CPI Agreeableness by Emotional Stability circumplexes. Insert Figures 1-3 about here

Multimethod Replication of ABSC Discussion ABSC circumplexes generated by different methods (ratings on internally constructed factor scores, ratings on American CPI and HPI scores, ratings on German CPI scores) corresponded very closely to each other. Angular locations correlated near unity across circumplexes, which is consistent with the idea that the circumplexes are simply rotational variants of each other. The amount of rotation is indicated by the value of the constant in the regression equation. The degree of correspondence is quite astonishing, given the number of factors that should destabilize the results: relatively small sample sizes, differences in methods for assessing the five factors, and language differences. The correspondence between lexical and inventory results indicate that the findings of "purely lexical" research may indeed be of some interest to practitioners who use personality inventories such as the CPI and HPI. One of the great strengths of the lexical tradition is the vast, practically exhaustive cataloging it has provided on trait words. To tap into this lexical information, inventory users need only determine the rotational differences between lexical circumplexes and inventory-generated circumplexes. The present study, which uses only 98 trait words, is only suggestive of the way in which a much larger lexical data base could inform circumplex models of inventory scales. Inventory users might yet fail to see the usefulness of 11

12 mapping trait words onto scales in circumplex fashion. Let me describe briefly one possible application which I discuss in much more detail elsewhere: Configural interpretation (Johnson, 1992). Configural interpretation of personality scale scores is determining what someone is like who scores high/high, low/low, or high/low on a pair of scales (Gough, 1991; McAllister, 1986). It is difficult to tell from the published literature the degree to which hard data versus clinical intuition forms the basis of configural interpretation. The AB5C model provides an empirically-based, systematic way of interpreting all combinations of two scales at a time from a set of FFM scales. Perhaps the most appropriate way to build some validity into the configural interpretations would be to use the following procedure. A set of standardized scores (or factor scores, if one prefers) from an FFM-based personality inventory could be used as axes for the ten circumplexes, and peer ratings from a relatively large set of trait adjectives could be mapped onto the circumplexes. Adjectives that possess both a sufficiently large distance from the origin and a projection closer to the bisectrix of the octant blend than to the primary factor dimensions would form the basis of a configural interpretation that combination. For example, if the peer adjective ratings cheerful, warm, confident, and relaxed fall close to the bisectrix of Extraversion and Agreeableness and the multiple regression of E and A on these ratings is statistically significant, then one has

13 some grounds for describing a high scorer on the HPI scales Sociability and Likeability with a statement such as "People perceive you as a cheerful, warm, confident, relaxed person." Elsewhere (Johnson, 1992) I suggest that these kinds of analyses might form a useful framework for computer-generated personality narrative reports.

14 References Goldberg, L. R. (1992). The development of markers for the Big Five factor structure. Psychological Assessment, 1, 26-42. Gough, H. G. (1975). Manual ~ tb California Psychological Inventory. Palo Alto, CA: Consulting Psychologists Press. Gough, H. G. (1991, August). Scales ~ combinations of scales: What do they tell us, what do they mean? Invited address, Division of Evaluation and Measurement, 99th Annual Meeting of the American Psychological Association, San Francisco, CA. Hofstee, W.K.B., de Raad, B., & Goldberg, L. R. (in press). Integration of the Big Five and circumplex approaches to trait structure. Journal of personality and Social Psychology. Hogan, R. (1986). Hogan Personality Inventory manual. Minneapolis, MN: National Computer Systems. John, o. P. (1990a). The "Big-Five" factor taxonomy: Dimensions of personality in the natural language and in questionnaires. In L. A. Pervin (Ed.), Handbook 2t personality theory and research (pp. 66-100). New York: Guilford Press. John, O. P. (1990b). The search for basic dimensions of personality: A review and critique. In P. McReynolds, J. C. Rosen, & G. L. Chelune (Eds.), Advances in psychological assessment (Vol. 7, pp. 1-37). New York: Plenum Press.

15 John, o. P., Angleitner, A., & Ostendorf, F. (1988). The lexical approach to personality: A historical review of trait taxonomic research. European Journal of personality, A, 171-203. Johnson, J. A. (1987). Influence of adolescent social crowds on the development of vocational identity. Journal Q1 Vocational psychology, ~, 182-199. Johnson, J. A. (1991a). Interpreter's guide t2 thg Bipolar Adjective Rating Scales (BARS). unpublished manuscript, pennsylvania State university. Johnson, J. A. (1991b). Interpreting the California Psychological Inventory ~ the AB5C model. Written version of an invited talk presented at the University of Groningen, June 21, 1991. Johnson, J. A. (1992). Generating computer narrative reports for the CPI, HPI. ~ other omnibus personality inventories with thg ~ model. Paper to be submitted for a proposed symposium on computer-generated personality reports, 101st Annual Convention of the American Psychological Association, August, 1993, Toronto, canada. Johnson, J. A., Cheek, J. M., & Smither, R. (1983). The structure of empathy. Journal of personality and Social Psychology, 45, 1299-1312. Johnson, J. A., Germer, C. K., Efran, J. s., & Overton, W. F. (1988). personality as the basis for theoretical

~~~----- ------ Multimethod Replication of AB5C 16 predilections. Journal Q{ Personality gog Social Psychology, 55, 824-835. Lanning, K. (1991). Consistency. scalability. and personality measurement. New York: Springer-Verlag. McAllister, L. W. (1986). A practical guide to CPI interpretation. Palo Alto, CA: Consulting Psychologists Press. McCrae, R. R., Costa, P. T., Jr., & Piedmont, R. L. (in press). Folk concepts, natural language, and psychological constructs: The California Psychological Inventory and the Five-Factor Model. Journal of Personality. McCrae, R. R., & John, o. P. (in press). An introduction to the five-factor model and its applications. Journal of Personality. Wiggins, J. 5. (1979). A psychological taxonomy of traitdescriptive terms: The interpersonal domain. Journal of Personality and Social Psychology, 37, 395-412. Wiggins, J. 5., & Broughton, R. (1991). A geometric taxonomy of personality scales. European Journal of Personality, 2, 343-365.

17 Author Notes I am indebted to the Alexander von Humboldt-stiftung for supporting my work as a visiting research fellow at the University of Bielefeld, to Alois Angleitner, Wolf Nowack, and Fritz Ostendorf for their assistance in the development of the German version of the BARS, and to Ansfried Weinert for providing the German 462-item version of the CPl.

Multimethod Replication of ABSC 18 Footnotes 'For convenience--without endorsing any researcher's specific views--i use the following labels for the five factors: Extroversion, Agreeableness, Conscientiousness, Emotional stability, and Openness, and abbreviate the factors either with Roman numerals, I through V, or with the letters E,A,C,S, and o.

19 Table 1 Count Qf Traits ~ Vectors Greater than ~ in Circumplexes Constructed ~ Different Methods Count of Common Traits Source of circumplex Source of circumplex Year Count 3 4 5 1 American BARS x CPI 1984,5 290 166 140 2 American BARS x CPI 1984,5,7 318 207 3 American BARS x HPI 1987 576 140 4 American BARS x BARS 1984,5,7 774 397 5 German BARS x CPI 1991 474

20 Table 2 Correspondence between Lexical Circumplexes gng Inventory-Bases Circumplexes circumplex Pearson Correlation Slope Intercept American CPI 1 Extraversion x Agreeableness.99 1.00-21.3 2 Extraversion x Conscientiousness.95 1.00-20.6 3 Extraversion x Stability.97.94 38.4 4 Extraversion x Openness.98 1.05 7.5 5 Agreeableness x Conscientiousness.99 1.02-9.7 6 Agreeableness x Stability.96.95 42.1 7 Agreeableness x Openness.97.98 19.2 8 Conscientiousness x Stability.94.95 21.8 9 conscientiousness x Openness.99.96 36.8 10 Stability x Openness.98.98 20.9 TOTAL, American CPI.96

21 Table 2, continued Circumplex Pearson Correlation Slope Intercept HPI 1 Extraversion x Agreeableness 2 Extraversion x Conscientiousness 3 Extraversion x Stability 4 Extraversion x Openness 5 Agreeableness x Conscientiousness 6 Agreeableness x Stability 7 Agreeableness x Openness 8 Conscientiousness x Stability 9 Conscientiousness x Openness 10 Stability x Openness TOTAL, HPI.97.97 25.5.97 1.02-23.0.88.91 49.8.99 1.08-18.9.96.98 15.3.96.91 31.8.97.98 33.7.97 1. 00 35.3.99.98 59.5.95.97 19.8.94

22 Table 2, con't Circumplex Pearson Correlation Slope Intercept German CPI 1 Extraversion x Agreeableness.88.93-3.7 2 Extraversion x Conscientiousness.92.95 7.9 3 Extraversion x Stability.95.91 42.6 4 Extraversion x Openness.98.97 5.8 5 Agreeableness x Conscientiousness.88.88 17.3 6 Agreeableness x Stability.83.90 12.9 7 Agreeableness x Openness.88.90 16.9 8 Conscientiousness x Stability.93.90 43.9 9 Conscientiousness x Openness.97.94 38.2 10 Stability x Openness.87.77 31.1 TOTAL, German CPI.88

~---- Multimethod Replication of AB5C 23 Table 3 Correspondence between ~ and Hfl circumplexes Circumplex Pearson Correlation Slope Intercept 1 Extraversion x Agreeableness.96.95 41.1 2 Extraversion x Conscientiousness.99 1.00 2.3 3 Extraversion x Stability.94 1. 00 7.3 4 Extraversion x Openness.99 1.03-25.7 5 Agreeableness x Conscientiousness.99.96 27.8 6 Agreeableness x Stability.84.64 56.8 7 Agreeableness x Openness.95.95 23.8 8 Conscientiousness x Stability.97.85 38.3 9 Conscientiousness x Openness.93.93 16.7 10 Stability x Openness.96 1.01 2.5 TOTAL, CPI/HPI.94

24 Table 4 Correspondence between American ~ German CPI Circumplexes circumplex Pearson Correlation Slope Intercept 1 Extraversion x Agreeableness.94.86 27.0 2 Extraversion x Conscientiousness.95.96-3.2 3 Extraversion x Stability.99.98 5.2 4 Extraversion x Openness.99.94-2.9 5 Agreeableness x Conscientiousness.91.98 1.9 6 Agreeableness x Stability.91 1.06-26.8 7 Agreeableness x Openness.90 1. 06-62.7 8 Conscientiousness x Stability.96.89 19.0 9 Conscientiousness x Openness.94.96-6.6 10 Stability x Openness.96.95-11.5 TOTAL, American/German CPI.94

25 Imaglnatlv.e v+ careless 111+ talkative unreliable rule-avoldlng unlettered unreflect~conservative conventional down-tn-earth v- careful BARS (outside) and BPI (inside) circumplexes, unrotated Imaginative V+ experimenting IIberalx ~ ~~ ~~~ ~ "<-~ 'O{'t ~{'t ~ ~ {'t", "O! ~ careless-l\~~ ~... IS' ~ ~ 111 III ~'(.~ ~ ~.-z.. t",9i <' 4 -.~ ~~,,~o~ ~~. t".,,,, -~ ~....,~ f/ii' ~ <' l \ {'t,\ ('tit ~! V conservative conventional down-to-earth careful BARS (outside) and BPI (inside) circumplexes, rotated 60

26 11+ quiet shy Introverted quiet reserved solitary shy Introverted 1+ sociable outgoing extraverted self-assured talkative reserved solitary I: 11 CPI (outside) and BPI (inside) circumplexes, unrotated conservative.. 11+ agreeable I ~ goodnatured,<;...~ ~,.~ 4"~ goo <!'...6~ I;'J>. I'~ ~L... self-assured '~ 3 eo -c-' talkative '" ~~ ~... ~~ 1',.~ ('''I', ~eo. I~~ quiet <'0.., ~I'j!> t".~ shy.~~ iii"~" Introverted -:?*~ ~1Js_ " ~., "'l',~ reserved -4, ~o~ eo'l... ~ s outary ~.. ~<1.<!'~ "eo~ I:- -" ~J.. ~ it. 1+ ~ " (;o~_ sociable o o~ 'ri outgoing "61.; "C'q ~~l)\'-a extraverted 11 CPI (outside) and BPI (inside) circumplexes, rotated 41 0

27 ~.-- ~~- 11 11+ impatient critical moody nervous tense depressed unenergetic' '. d worm: quiet timid ~shy,introverted solnary reserved American CPI (outside) and German CPI (inside), unrotated outgo1in!l ~oc 81)1 COPlId:in Ch~o~ous relaxed extraverte \ poised self-assured bol ~I ~ l \ ~~ composed talkative ~.Jt.o.~~T"e.:O\fJ(f:t.~f, e.6, goodnatured '(}) ~G e~(f," o\f> 'ltf, ~... e"!l"!l" "'9~e.(. ~e." agreeable "!IiJ.e,,\'t'...\e. v ~ \! G~ e\." tolerant I g,o ~~\ ~oo 1J!i. 0" e..te I II. 11- \ e.!e~ ~' \ev. e. (fle. ~,...\~ r,'fo\ \~ r,t -'. impatient ~ /~',«\9.~ fj"l~t;, '(',,e.:.e.: " 00<"'!e. ~ e.~ ~f, critical '::.1 V\\'/~te."~~~~ti~\~~~~~~. 'I'table I \~"'.'!v. quiet In moody "_ r1mid nervous shy d tense introverte depressed. solitary unenerge~rried reserved American CPI (outside) and German CPI (inside), rotated 27 0