UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Structural validity in the measurement of Holland's vocational typology : a measure of Holland's types… Trapnell, Paul David 1989

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-UBC_1989_A8 T72.pdf [ 5.51MB ]
JSON: 831-1.0054543.json
JSON-LD: 831-1.0054543-ld.json
RDF/XML (Pretty): 831-1.0054543-rdf.xml
RDF/JSON: 831-1.0054543-rdf.json
Turtle: 831-1.0054543-turtle.txt
N-Triples: 831-1.0054543-rdf-ntriples.txt
Original Record: 831-1.0054543-source.json
Full Text

Full Text

S T R U C T U R A L V A L I D I T Y IN T H E M E A S U R E M E N T O F H O L L A N D ' S V O C A T I O N A L T Y P O L O G Y : A M E A S U R E O F H O L L A N D ' S T Y P E S S C A L E D T O A N E X P L I C I T C I R C U M P L E X M O D E L by P A U L D A V I D T R A P N E L L B.A., University of British Columbia, 1982 A T H E S I S S U B M I T T E D IN P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F A R T S in T H E F A C U L T Y O F G R A D U A T E S T U D I E S (Department of Educational Psychology) We accept this thesis as conforming to the required standard T H E U N I V E R S I T Y O F BR IT I SH C O L U M B I A October 1989 © Paul Dav id Trapnell In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. The University of British Columbia Vancouver, Canada Department DE-6 (2/88) ii Abstract A simple principal components procedure for constructing a multi-scale test to fit a circumplex structural model was illustrated. Loevinger's concept of structural validity, which specifies that test structure should faithfully reflect postulated structure within a domain, provided a psychometric rationale for applying circumplex scale construction procedures to the domain of vocational interests: although Holland's influential theory of career choice specifies an explicit two-dimensional circumplex model to describe the relations among the pr imary categories of vocational interest, and despite the extensive degree of validity established for the model, no measures of those interest variables have apparently been constructed explicitly to fit a two-dimensional circumplex model. In doing so, this study also tested the hypothesis that the two dimensions accounting for the well-known hexagonal arrangement of Holland's six vocational categories are continuous and that, within the two dimensional interest space, Holland's six categories represent but six points along what is a circular continuum. This hypothesis was tested by attempting to construct evenly spaced "octant" scales to mark the two-dimensional vocational interest space defined by Holland's six Vocational Preference Inventory (VPI) scales. Circumplex analysis procedures were applied to a large pool of occupational titles which included a large subset of VP I items. Items with the best circumplex properties in this pool were assembled into a set of scales to mark ideal octant locations around the two interest dimensions. Results provided strong support for a circumplex view of the dimensions accounting for Holland's hexagon, and evidence, in a validation sample, of a remarkably good fit of the new eight scales to an ideal circumplex structure. Implications of the presents results for Holland's theory and for circumplex approaches to measurement are discussed, as well as suggested applications of this efficient 64-item circumplex measure of Holland's "big two" dimensions of interests. iii T A B L E O F C O N T E N T S Abstract L ist of Tables v List of Figures vi Acknowledgments v i i Introduction 1 Literature Review .- 1 Holland's Theory 7 The typology 8 Hexagonal model 10 Importance of hexagon: Construct of congruence 11 Evidence supporting hexagonal model 12 Dimensionality 13 Bipolarity 14 Circularity 16 Introduction to circumplex models 19 Formal definition of. 19 vs factor models 20 Measurement implications of. 21 A circumplex model for measuring Holland's interest space 25 Methods & results 26 Overview of the studies 26 Phillips' Circumplex Analysis Program 2 7 Controlling for a general interest factor 28 Study 1 30 Method 30 Results 32 Discussion 37 Study 2 39 Method 39 Results 47 Derivation of circumplex scales 47 Structural evaluation of circumplex scales 49 Internal consistency 49 Structural stability 50 Dimensionality 51 Circularity 52 Discussion 56 General Discussion 57 Holland's theory 59 Circumplex models and assessment 60 Research applications 61 References 65 Appendices ^8 V L I ST O F T A B L E S Table 1. Definition of Holland's types 73 Table 2. Formal definition of a circulant matrix 75 Table 3. Intercorrelations among VP I scales 76 Table 4. Angles and communalities of VP I items 77 Table 5. Internal consistency of octant scales 78 Table 6. Intercorrelations among octant scales 79 Table 7. Comparative evaluation of fit to a circulant model 80 Table 8. Item content of final eight circumplex scales 81 vi L I ST O F F I G U R E S Figure 1. Holland's hexagonal structure of interests model 82 Figure 2. Circumplex structure of original VP I scales 83 Figure 3. Circumplex structure of VP I marker scales 84 Figure 4. Circumplex structure of octant scales, sample A 85 Figure 5. Circumplex structure of octant scales, sample B 86 Figure 6. Eigenvalue (scree) plot for final octant scales 87 Acknowledgments This study was'made possible by the encouragement and supportof many people. For putting up with me for the past year I wish to thank first of all my mother and father, who, of course, can hardly wait to read this tome. Thanks'to my sisters Jenny for her encouragement, Anne for her love, and to my brother Andy for his boundless sense of f un-his nightly dial-a-jokekept me human through more than a few of the uglier moments of blocage d'ecrivain. I wish also to thank A r t More and Dave Whittaker for agreeing to supervise the latter stages of this project and for providing critical insight, gentle encouragement and, in the closing moments, adminstrative wizardry. A special thanks-for-your-support also to Todd Rogers without whose intervention I wouldn't have been granted the opportunity to begin work in this area. A warm thank-you also to Del Paulhus whose passion for ideas, frank criticality, and unceasingly nurturant manner has done much to keep the intellectual fire in my eyes the past few years. Lastly, and foremostly, I wish to thank Jerry Wiggins, who in his luckless role as my life teacher, mentor, patron saint, and friend, has allowed me the useful illusion that at least some of my more clever ideas are not his own. A Circumplex Measure of Interests 1 I. Introduction Since the concept of "construct validity" was introduced to psychology by Cronbach & Meehl in 1955, the importance of theoretical as opposed to pragmatic types of test validity has been increasingly stressed in psychological measurement (Loevinger, 1957; Jackson, 1971; Anastasi , 1986). Most psychological test construction procedures today incorporate at least some of the main features of the theory-based approach to test construction and validation outlined by Loevinger (1957) known as "construct-oriented" measurement. This point of view defines test validity as the isomorphism between test and theory in three areas: (1) the substantive aspect, the correspondence between the theoretical definition of a construct and the denotive content of the test; (2) the structural aspect, the correspondence between the internal correlational structure of a test and that demanded by the structure of non-test manifestations of the construct; and (3) the external aspect, the correspondence between the actual pattern of relations with other measures and those expected by the theory of the construct (Loevinger, 1957; Wiggins, 1973). By stressing the unity of test and theory, the construct point of view also implies that the validation process begins at the first stages of test development: to the extent that theory explicitly guides the sequential stages of construct definition, and item construction, selection, and statistical refinement, validity, from a construct-oriented perspective, is something built into a test at the outset rather than something to be established after test development (Anastasi, 1986; Jackson, 1971) as in the traditional "criterion-oriented" view of test validation. One reason why explicitly theory-based approaches to measurement have made a "comeback" in psychology, albeit in a more sophisticated form, after their temporary fall from favor during the radical empiricist days of the 1940s, may be acceptance of the view that for tests to be useful for scientific uses, as opposed to specific practical applications, it is less important that they somehow "work" (predict, discriminate, etc.) than that their A Circumplex Measure of Interests 2 content and structure facilitate understanding about why they work (Loevinger, 1957). Thus, the principal advantage of construct-oriented measurement, i.e., designing a test explicitly to conform to theory, over criterion-oriented approaches, is its greater scientific usefulness. By attempting to make all assumptions and hypotheses in the selection of test content and methods of scoring explicit, construct-oriented approaches minimize ambiguity in the interpretation of test scores and provide an explicit network of logical inferences between test scores and theory that enhances the utility of such tests for examining hypotheses derived from the theory, i.e., enhances their usefulness as instruments for advancing theories. This was the principal rationale offered in favor of construct-oriented and against criterion-oriented tests by Loevinger (1957) in her classic monograph on construct validity. A similar rationale underlies two quite different approaches to scale construction, advocated by Jackson (1970) and Wiggins (1979), each of which amounts to a programmatic application of Loevinger's (1957) ideas to multivariate personality assessment. Jackson's sequential approach highlights the importance of careful, exhaustive definition of constructs, and their possible behavioral exemplars, and empirical refinement of scale content to simultaneously maximize scale homogeneity, discriminant validity, and freedom from the effects of response biases. Although Jackson's interpretation and measurement approach to Loevinger's concept of structural validity is not without its critics (e.g., Loevinger, 1972; Wiggins & Broughton, 1985), the critical acclaim awarded to Jackson's Personality Research Form (Wiggins, 1972; in press) attests to the utility of his general approach. Wiggins approach, although similar to Jackson's with regard to substantive concerns in test construction, differs radically from Jackson's in its approach to structural validity. Whereas Jackson adopts the position of seeking minimal correlation among resulting scales, i.e., he assumes, on pragmatic grounds, an orthogonal model of personality structure, Wiggins (1979, 1980) argues for the adoption of explicit f/ieory-based structural A Circumplex Measure of Interests 3 models in test construction. Since considerations of structural validity require that test structure conform to theoretical expectations (e.g., Loevinger, 1957), Wiggins argues that where explicit structural models have been articulated for a particular variable domain, and when sufficient evidence exists in support of that model, then structural criteria for combining items into scales should be explicitly derived from that structural model. In other words, postulated structural models, when sufficiently validated, should be adopted as explicit measurement models in those domains. In a significant departure, therefore, from traditional factor analytic approaches to scale construction, Wiggins (1979; Wiggins, Trapnel l , & Phillips, 1988) has attempted to maximize the structural fidelity between theory and measurement of interpersonal dispositions by adopting a specific, well-validated structural model of that domain (a two-dimensional circumplex model) as a criterion for item selection. Beginning with a rationally constructed pool of interpersonal trait adjectives, Wiggins developed circular structural criteria to combine items into scales such that the resulting scales intercorrelated in an optimally circular manner. The resulting Interpersonal Adjective Scales (Wiggins, 1979) were later found to possess the best circumplex properties, and hence, Wiggins argued, structural validity, of any interpersonal measures recorded in the research literature to date (Wiggins, Steiger, & Gaelick, 1981), a not unexpected outcome since this was the explicit objective. In light of this success, and with Loevinger's concept of structural validity as a rationale, this study extended these innovative procedures to another psychological domain, vocational interests, for which a well-validated circular structural model exists but in which no measures had yet been constructed with reference to that model. O f the six or so prominant theories of vocational choice (Brown & Brooks, 1984), two postulate circular structural models describing relations among the pr imary categories of vocational interests. After reviewing factor analytic studies of interests, Roe (1956) and Holland (1966) independently proposed an octagonal and an hexagonal model, respectively, A Circumplex Measure of Interests 4 to describe the degree of relation among the pr imary clusters of vocational interests. Although Roe's and Holland's theories differ considerably, there is general agreement in the literature that six of Roe's categories correspond closely to Holland's, and that the structural models are virtually identical (Holland, 1973, p.81; Lunneborg, & Lunneborg, 1975; Zytowski, 1987). Because Holland's theory is considered superior to Roe's and has generated, by far, the greater research interest and evidence of validity (Brown & Brooks, 1984), attention will be restricted to Holland's model. Because Holland's first published references to the hexagonal model (1969) appeared after he had reported constructing a measure of the types (1958) one can assume that the notion of a circular ordering of the types did not influence the selection of items in his original measures. This assumption is supported by Holland's statement that the circular ordering was "accidently discovered" upon inspection of factor plots from principal components analyses of the the original six Vocational Preference Inventory (VPI) scales (1973, p.72). Examination of the test manuals, however, for each of the later revisions of the VP I and Holland's other measure of the types, the Self Directed Search (SDS), gives no indication that Holland has subsequently incorporated structural criteria for item selection derived explicitly from the structural aspect of the theory. The likelihood that no published measures of the types have incorporated theory-based (i.e. hexagonal) structural criteria for combining items into scales is suggested by the structure of measures reported in the literature to date (e.g., Cole & Cole, 1970; Cole & Hanson, 1971; Edwards & Whitney, 1972; Lunneborg & Lunneborg, 1975; Rounds, Davison, & Dawis, 1979) Review of the factor plots and M D S configurations of the various Holland type measures published in these studies led Holland to conclude that the "actual " structure of the typology " in real world data" more closely resembles a "mishapen polygon" than an even-spaced hexagon (1979; p.43). i A Circumplex Measure of Interests 5 Holland implies here that the discrepancy between theory and measures reflects limitations of the theory. A n equally justifiable conclusion, however, is that the measures are imperfect. If one defines the structural validity of a test, as do Loevinger (1957) and Wiggins (1973), as the correspondance between test structure and the theory's structural model, then, strictly speaking, the degree of "mishapeness" of current measures might also be interpreted as reflecting their degree of structural invalidity. Since no existent measures of the typology appear to have employed circular structural criteria for selecting items, the possibility that one might improve substantially the structural validity of these measures by employing such criteria remains an open question. Recently, Phillips (1983) has developed an elegant principal components technique for scaling test items to a circumplex model. Originally developed to refine further the structural properties of the Wiggins' Interpersonal Adjective Scales (Wiggins, Trapnel l , & Phillips, 1988), the technique is a general one which may be used to construct optimally circular, two-dimensional sets of measures for any domain of variables hypothesized to conform to a circumplex model. With a strongly validated circular model, and apparently -no measures yet developed according to a circular measurement model (i.e., circular criteria for item selection), Holland's vocational typology would seem to be an ideal candidate for Phillips' circumplex scaling procedures. With a few exceptions (Hogan, 1983; Cavanaugh & McCormick, 1981), Holland's hexagonal model of interests is rarely cited in the personality literature as an example of a circumplex model of personality, despite the fact that it is explicitly a personality theory (see below), and, with over 400 published studies or dissertations focusing on Holland's model or its related hypotheses, it is, by far, the most widely researched example of a circumplex model in psychology (Hogan, 1986). The present study proposes that Holland's model be viewed explicitly as a circumplex model, under which the same structural assumptions of other circumplex models in personality (e.g., interpersonal circumplexes) would be held to apply. F rom this point of view, Holland's six category hexagonal model A Circumplex Measure of Interests 6 may be viewed as a circular array on a latent continuum of two psychological dimensions (Prediger, 1982). Because the central and unique goal of circumplex measurement procedures is to reliably measure as many points along the circular continuum as possible, adopting a circumplex measurement model for Holland's typology holds open the possibility of representing locations along the circular continuum of the Holland "plane" other than the six locations (variables) currently represented in the typology, a possibility that also has direct theoretical implications. Because identifying the nature of the dimensions underlying Holland's typology has been an important research question (Prediger, 1982), a circumplex approach to measuring the types, by attempting to "fill i n " the gaps around the hexagon, affords the possibility of clarifying the nature of the continuum through the "mini-hypotheses" involved in writing items to fill these locations. I believe there is sufficient evidence to support the validity of Holland's hexagonal model, and therefore to justify bringing theory and measurement of the typology into alignment by adopting a circumplex measurement model for the domain. Before reviewing the structural evidence, a brief overview of Holland's theory will be presented in the next chapter. The chapter concludes with a description of the measurement model being proposed for Holland's typology~a circumplex. A Circumplex Measure of Interests 7 II. Literature Review 1. Holland's Theory Both a theory of vocational behaviour and of personality, Holland's theory of vocational choice is doubly distinguished as the most influential career development theory and occupational taxonomy in vocational psychology (Brown & Brooks, 1984), and as the most widely researched and validated personality typology (Hogan, 1986). Since its publication in book form in 1973, the theory has generated the greatest number of new assessment instruments in vocational counseling (Zytowski & Borgan, 1984), and has prompted the development of scoring keys for the six Holland types on virtually all widely used interest inventories, including the Strong-Campbell and Kuder Inventories. More recently, all of the leading on-line career guidance systems (DISCOVER, SIGI, CHOICES, etc.) have incorporated Holland's deceptively simple 6-category framework for organizing occupational information and career exploration. The influence of the theory on research is equally notable. Holland cites, for example, in the revised 1985 edition of his book, over 400 published studies or dissertations in the fields of education, psychology or counseling involving the theory. In a recent citation count of authors and contributions to journals in the field of vocational behaviour (Watkins et. al, 1986), Holland's 1973 book, Making Vocational Choices was the most cited book or article by a wide margin (three times as many citations as the runner-up, Osipow's (1967; 1973) Theories of Career Development. Clearly, Holland's typology is a topic of central importance to practioners and researchers in psychology and education. Holland's theory consists of four parts: (1) a set of assumptions forming the basis of the theory, which includes a postulated six-category personality typology; (2) a classification system for occupations based on the 6 types; (3) a structural model specifying the degree of psychological similarity among the types and (4) a set of propositions relating A Circumplex Measure of Interests 8 the typology and structural model to aspects of vocational attitudes and behaviour such as career choice, change, stability, and satisfaction. Summary of the Theory Four assumptions form the basis of the theory: (1) Persons can be meaningfully categorized in terms of six personality types: realistic, conventional, investigative, artistic, social, and enterprising; (2) corresponding to the personality types are six types of environment, realistic, conventional, investigative, artistic, social, and enterprising, which are defined by the dominant personality type populating that environment (e.g, the predominance of realistic types in an automotive repair shop define that environment as realistic); (3) Persons of a type tend to seek out environments corresponding to that type in order to exercise their particular skills, abilities and prefered roles, and allow expression of their particular attitudes and values; (4) the degree and kind of "f it" between personality type and environmental type determines important aspects of a persons behaviour (i.e., vocational choice, job satisfaction, job changes, etc.). These four assumptions characterize Holland's theory as a "structural-interactive" approach to explaining vocational behavior (Weinrach, 1984): the theory provides a method of categorizing person and environment charcteristics, of structuring the characteristics, and of deriving predictions concerning important vocational and social outcomes via the interactions between these person and environment characteristics. The Typology Believing that a psychological schema offered the most parsimonious approach for an occupational taxonomy, and that the structure of human interests, values and traits could be adequately represented by just a small number of broad classes, Holland (1958) proposed that persons and interpersonal environments can be meaningfully classified in A Circumplex Measure of Interests 9 terms of six ideal " types": Realistic, Investigative, Artistic, Social, Enterprising, and Conventional. Table 1 presents Holland's psychological definitions of each of the six types. Insert Table 1 about here The content of these definitions reveals what is most notable about Holland's typology and its main contribution to vocational psychology. As simply a list of pr imary interest dimensions the typology is not unique. Guilford, Christensen, Bond, & Sutton (1954), in a large-scale factor analytic study of human interests, recovered among their 28 interpretable factors what they described as "the six well-known [vocational interest] factors" (p.28), each of which corresponds very closely to one of Holland's types. This comment implies a consensus in the literature on the basic vocational interest factors prior to Holland's writings (1958, 1966), although Holland's (1985, p.6) comment that this study "reassured" h im implies he conceived of the types prior to 1954. What distinguishes Holland's vocational taxonomy is that it is explicitly a personality taxonomy. Holland postulates that interests, including vocational interests, are an aspect of personality, and, that vocational interest inventories are, therefore, essentially personality inventories: The choice of an occupation is an expressive act which reflects the person's motivation, knowledge, personality, and ability... the choice of an occupational title represents several kinds of information: the S's motivation, his knowledge of the occupation in question, his insight and understanding of himself, and his abilities. In short, item responses may be thought of as limited but useful expressive or projective protocols. (Holland, 1973, p.5). Holland, therefore, interprets the six "well-known" vocational interest factors as representing six personality syndromes, or types, i.e., six distinct constellations of personal traits (i.e., abilities, dispositions, values, etc.) that coalesce by early adulthood into a set of six core interest patterns (Hogan, 1988). A Circumplex Measure of Interests 10 To operationalize the typology for research and counseling use, Holland developed the Vocational Preference Inventory (1958; 1966; 1978; 1985), a checklist consisting solely of occupational titles. Holland later developed the Self-Directed Search (1971; 1979; 1985), a more clinically-focused instrument which retains the VP I vocational scales as a subsection and provides additional scales tapping activity preferences, aspirations, and self-rated competencies and abilities. It is important to note here that Holland's notion of a type is as an "ideal type", a theoretical type, not a statistical one in the sense of a cluster of correlated traits, or a continuous trait dimension (e.g. Eysenck, 1985). Holland's use of cumulative, continuous, scales to index the types, however, implies a continuous, dimensional interpretation of the construct measured by each scale (Wiggins, 1973; p.404-405). While a detailed discussion of the differences between typological and dimensional views of personality is beyond the scope of this review, the relevant point here is that no explicit link exists in Holland's theory between his concept of a type, and his use of continuous, homogeneous scales to operationalize it. Hogan (1983) suggests that there exists an alternative measurement model that does "make explicit the concept of a type"—a circumplex model. This point will be returned to in a later section. The Structural Model: Holland's Hexagon Holland proposes that "the relationships within and between types or environments can be ordered according to an hexagonal model in which the distances between the types or environments are inversely proportional to the theoretical relationships between them." (1973, p.5). Thus, the relative psychological similarity among the six types can be represented graphically, in two dimensions, in the circular order R, I, A , S, E, C , as depicted in Figure 1. Holland describes this structural model as the "calculus" of the theory, from which most (although not all) of the important theoretical propositions of the theory are derived (see A Circumplex Measure of Interests 11 below). Weinrach (1984) describes its use as "essential to understanding the theory, instruments, and classification system." (p. 66). For purposes of the present study, two points concerning the hexagon deserve emphasis. Holland's formulation makes explicit that the basis of the hexagonal ordering is psychological, i.e., the distances in the model refer to psychological similarity, rather than work tasks or non-psychological job aspects. Not mentioned by Holland but implicit in the hexagonal formulation is the inference that the psychological relations among the types is two-dimensional (Prediger, 1982), i.e., that underlying the relations among the types are two higher order (psychological) dimensions, and that the relation between the types and these latent dimensions varies continuously around the circle. This interpretation of the hexagon makes plain the correspondence between Holland's hexagonal model and that of other psychological circumplexes (emotions, interpersonal traits, etc.). Holland's typology is explicitly a personality circumplex model, a fact noted only rarely in the personality and psychometric literatures (Hogan, 1983; Cavanaugh & McCormick, 1981). Importance of the Hexagon to the Theory: Person-Environment Fit, or Congruence Osipow (1987) has described the concept of "person-environment" interaction as the "heart of vocational psychology and career development" (p.333). Holland's theory is an explicit attempt to provide one kind of model of person-environment interaction, one essentially based on personality characteristics (i.e., the typological definitions), and for this reason he describes the notion of person-environment interaction as "the guts of the theory" (1985, p. 12). To tie the typology and structural model to predictions concerning person-environment interaction Holland developed four constructs-consistency, differentiation, identity, and congruence. Consistency, differentiation, and identity are of lesser importance and essentially concern the degree of focus vs diffusion of a person's or A Circumplex Measure of Interests 12 an environment's psychological profile. These variables serve primari ly in the role of moderators of person-environment interactions. The pr imary interaction construct, and in a sense the "heart" of Holland's theory is the concept of congruence: "Vocational satisfaction, stability, and achievement depend on the congruence between one's personality and the environment in which one works... In the present theory, the congruence of person and environment is defined in terms of the structure of personality types and environmental models." (1985, p. l u -l l . Emphasis from the original). Thus, degree of person-environment congruence, or fit, is defined as the distances between typological categories in the hexagonal model: identical types are predicted to be most congruent (a social person in a social environment), adjacent types to be second most congruent (a social person in an enterprising environment), and types opposite on the hexagon to be least congruent (a social person in a realistic environment). Since the principal predictions of Holland's theory (concerning vocational choice, change, satisfaction, and achievement) derive from the notion of person-environment congruence or fit, and since this concept is operationalized in the theory specifically by the structural model, it is clear that this model--a two dimensional circle, i.e., circumplex model-is centrally important to the theory. 2. Review of Structural Evidence on Holland's Hexagon Although the hexagonal model is central to Holland's theory, and although modern notions of construct validity would seem to require that measures of the types be constructed so as to conform to this model, there appear to be no measures of the typology, including those constructed by Holland (the VPI , and SDS), that have been constructed explicitly to do so. Adopting such a measurement strategy, as this study proposes to do, first requires, of course, sufficient evidence of the hexagonal model's validity. Some of this A Circumplex Measure of Interests 13 evidence is now reviewed. Because, however, the large number of these studies (in the hundreds) precludes a detailed review, only a few studies will be briefly reviewed, namely those representing three types of evidence which are particularly relevant to the question of circumplexity in the interest domain-evidence pertaining to the dimensionality, bipolarity and circularity of vocational interests. Dimensionality Holland's choice of six rather than a greater or lesser number of vocational types was based partly on the factor analytic literature of the time (i.e., the 1954 Guildford et. al. study which identified six vocational interest factors) and partly on a faith in the heuristic value of parsimony (Holland, 1985, p.6-7). There are most certainly a greater number of identifiable vocational interest dimensions than six. Jackson (1977), for example, found ten higher order factors from the 34 basic interest scales of the Jackson Vocational Interest Survey. Holland's decision to focus on six specific dimensions is primari ly a theoretical one, based on the belief that these dimensions were most important and that they summarized vocational interests at a level of complexity most useful for developing theories to explain vocational behavior. Consequently, the issue of dimensionality should refer not to the universe of all possible vocational interests, but to the universe defined by Holland's theory. For the hexagonal model to be valid, the factor structure of the set of six scales measuring the types should be two-dimensional. A re they? Although several studies have found more than two factors underlying scale intercorrelations in Holland's VP I (Lunneborg & Lunneborg, 1975) and S D S (Edwards & Whitney, 1972), a re-analysis by Prediger (1982) of 24 published correlation matrices from seven different kinds of inventories measuring the types provides convincing support for a two factor interpretation. In each data set Prediger extracted three factors which accounted for an average of 7 7 % of the total variance among the six scale scores. The first factor in each data set, a general factor on which all six scales loaded highly and A Circumplex Measure of Interests 14 positively, Prediger interpreted as a response set factor, reflecting "overall checking-rate", i.e., the tendency to endorse many versus few interest items across all scales. Prediger argued that such a factor has no theoretical import as a dimension of vocational interests and that rotation of three factors is therefore unjustifiable in that it would confound content and method factors. Us ing two methods of extracting factors, principal components and a theory-based method (Cooley and Lohnes, 1971, pp. 137-143), which extracts factors to fit a pre-specified target matrix, Prediger found strong support for two substantive dimensions underlying each data set, which he interpreted as a "people vs things" dimension, and a "data vs ideas" dimension. Thus the six interest categories comprising Holland's typology, and identified by Guildford et. al. in their pioneering 1954 study, may be thought of as six first order factors of interests, which can be further summarized in terms of two higher order dimensions. As to the representativeness of Holland's six pr imary factors, it is worth noting that after four years of test development, beginning with an enormous item pool with representativeness as a principle criterion, Jackson & McCar thy (1986) found seven factors underlying their final 15-scale Career Directions Inventory (designed for non-college bound youth), six of which are essentially identical to Holland's types (the seventh factor consisted of service industry occupations). Whether a higher order (circumplex?) structure might be discernable among the CDI 's seven factors is not known, but an interesting possibility. Bipolarity Both the definitions of the types in Table 1 and the hexagonal model clearly imply bipolar relations among the types. Artistic types are not simply uninterested in conventional occupations (e.g., accountant) they find them aversive; conventional types likewise find artistic occupations aversive. Similar contrasts are implied for social vs realistic types, and enterprising vs investigative types. Despite the bipolaritiy of the A Circumplex Measure of Interests 15 theory, however, the correlations among the six scales measuring the types are usually all positive (Holland et. al., 1969). This implies, of course, that those, for example, with an above average interest in conventional occupations also tend to have above average interests in artistic interests, a notion which clearly contradicts the theory. The simple explanation for this incongruity is that ratings of interests result in a general response set factor, as suggested by Prediger (1982) and Hanson, Prediger and Schussel (1977), reflecting a disposition to like many versus a few interests. Bipolar interest factors are clearly evident when this tendency is controlled for, either through the use of an ipsative rating technique (e.g., ranking of interests) or by statistically separating response set variation from substantive variation (partialing out an unrotated general factor from subject's scores, or subtraction of subject means from scores, etc.). In unrotated factor pattern matrices from VPI , SDS, and other measures of Holland's types, the response set factor is usually the first factor. Factors two and three are clearly bipolar, and correspond to the two coordinate dimensions of Holland's hexagon. Additional support for the bipolarity of interests lies in the pattern of external correlations for scales measuring the types—especially when response set is controlled for. Inspection of the many tables of personality correlates in the 1977 VP I manual shows that where a large positive correlation exists for a VP I scale on one side of the hexagon, a corresponding negative correlation exists for the scale on the other side of the hexagon. The most convincing evidence of this sort may be the study by Costa, McCrae & Holland (1984), who identified two broad factors of personality, extraversion and openness to experience, as the probable coordinate dimensions underlying Holland's vocational circumplex. When response set was controlled for, SDS scores demonstrated a substantial, almost circular, bipolar relation to these two personality factors. A Circumplex Measure of Interests 16 Circularity Most central to the viability of a circumplex model is the demonstration of circularity in the pattern of relations among a set of variables. The circularity requirement has two aspects, the ordering of the variables, and the degree to which they are evenly-spaced along the circumference of the two latent dimensions on which they are ordered. In the case of Holland's theory, the postulated ordering for the types is R,I,A,S,E,C,. Because there are six variables, the even-spacing requirement is defined with reference to a hexagon, although, as this study will demonstrate, the relations between adjacent catergories are in fact continuous, rather than discrete, and therefore Holland's hexagon is more properly designated a (six-pointed) circular model rather than an hexagonal one. Evidence for the invariance of the R,I,A,S,E,C ordering is substantial, having been demonstrated across a variety of vocational interest instruments assessing the types (Cole & Hanson, 1971; Prediger, 1982), across sex (Cole, 1973), race (Walsh, Hildebrand, Ward & Matthews, 1983), age (Edwards, Nafziger, & Holland, 1974), and using methods other than interests ratings to assess the types, such as trait adjectives (Bobele, Alston & Wakefield, 1976), and biographical data (Eberhardt & Muchinsky, 1984). The replicability and generalizability of the circular ordering is an important finding since " invariance" of structure is a crucial aspect of any structural model's validity and has proved quite difficult to achieve in factor analytic models, the other most common type of structural model in personality psychology (Plutchik, 1980). Invariance of circular ordering is not, by itself, however, sufficient evidence of circular structure. If the locations of the six Holland variables along two latent dimensions are stable but tend to fall close to dimensions'axes, in simple structure fashion, rather than fanning out in a quasi-circular fashion around the dimensions as expected by a circumplex model, arguments for a circumplex interpretation of the dimensions would be decidedly weakened. A Circumplex Measure of Interests 17 Assessing the equal spacing evidence is, however, somewhat problematic. One problem relates to the fact that the structure of existing tests is determined, to a certain degree, by the implicit or explicit structural assumptions of the test developer (Loevinger, 1957). This point is especially crucial to the equal spacing issue because use of traditional factor analytic methods of test construction, which adopt the principle of simple structure as structural criteria (as in the most recent revision of Holland's VP I and SDS) will tend to create scales that fall close to factor axes rather than scales that locate between factor axes. Simple structure criteria are in one sense the opposite of the criteria for constructing a circularly structured test (this point is elaborated below, p. ). Consequently, use of traditional factor analytic test construction methods may have biased the structure of these measures against the circular model. Although some empirical suggestion of circumplexity in existing measures seems necessary to justify postulating a circular model, evaluating the viability of the hexagonal model by examining the structure of existing measures must make allowance for the structural assumptions underlying the manner in which items were selected for these measures. A second problem relates to the decision criteria for evaluating the equal spacing requirement. Exact ly how hexagonal does the structural evidence need to be to be considered supportive of Holland's theory? Formal methods do exist for evaluating the equal spacing requirement of circular models, which include statistical "goodness-of-fit" indices, (Stern, 1970; Wiggins, Steiger & Gaelick, 1981), although there appear to be no studies that have applied such procedures to the vocational interest domain. However, as Wiggins et. al. (1981) note, because perfectly even spacing would not be expected in empirical data due to measurement error, and because formal circular hypotheses would therefore virtually always be false, given sufficiently large samples, formal goodness of fit values are informative only in a comparative rather than a statistical sense. Consequently, use of statistical tests is not more appreciably objective a method for evaluating fit to the model than much simpler methods such as visual inspection of factor plots and other A Circumplex Measure of Interests 18 graphic representations (MDS configurations) of empirical structure. These latter two methods have in fact predominated in studies of the hexagonal model's validity. Despite these two problems-use of measures potentially biased against Holland's model, and rather subjective methods of deciding whether obtained structure confirms or disconfirms a circular hypothesis~the vast majority of studies examining the validity of the hexagonal model have concluded in favor of the model. These studies are primari ly factor analyses of various measures of the typology: (e.g., Cole, 1973; Cole & Hanson, 1971; Cole, Whitney & Holland, 1971; Wakefield & Doughtie, 1973; Meir & Barak, 1974; Meir & Ben-Yehuda, 1976; Meir , Sohlberg, & Barak, 1973; Prediger, 1982), and multidimensional scaling studies (Rounds, Davison, & Dawis, 1979; Gati , 1979). In summary, evidence of two-dimensionality, bipolarity and circularity among the six pr imary categories of interests defined by Holland's types strongly supports the hexagonal model. Holland concludes in the latest edition of his book that overall "the hexagonal model has received relatively clear and positive support..." (1985, p.94). Evidence supporting Holland's hexagonal model may also be interpreted as support for a circumplex model of vocational interests since, as Hogan (1983) and Cavanaugh & McCormick (1981) have noted the hexagonal model may be viewed as a six category variant of the more general circumplex model. Such models have been postulated for a wide variety of psychological phenomena including mental abilities, perceived similarity among visual and auditory stimuli (cited in Wiggins et. al., 1981), as well as emotions (Plutchik, 1980; Russell, 1980), interpersonal behaviour (Kiesler, 1983; Wiggins, 1980), and interpersonal problems (Alden, Wiggins, & Pincus, 1989). Viewing Holland's typology as a circumplex model, with properties similar to other psychological circumplex models, immediately raises the possibility of applying both Wiggins' (1979) structural validity rationale and Phillips (1983) methodology to the development of circumplex measures of the "big two" dimensions of vocational interests defined by Holland's typology. The unique A Circumplex Measure of Interests 19 properties of circumplex models and their implications for test construction are now introduced. 4. Circumplex Models The notion that relations among traits can be ordered along a circular continuum is an old one in psychology, attributed first to Wundt in 1903 (Eysenck, 1985). Guttman (1954) first articulated the formal properties of such structures within his set of procedures known as "order analysis", which he offered as a complement to factor analysis, useful, he suggested, for identifying important patterns in a correlation matrix (such as circular patterns) which might be the "most striking feature of the data" (Guttman, 1966; p.453), but which traditional factor analytic procedures (i.e., simple structure criteria) are insensitive to. He defined one such pattern as representing a circular structure and labelled the matrix model for such a structure a circumplex. Formal Definition of a Circumplex Table 2 illustrates the formal properties of a circulant matrix (after Wiggins et. al., 1981, Table 1), which are essentially two: correlations within diagonals are equal to one another, and the magnitude of correlations within each diagonal is highest along the main diagonal, gradually decreasing then increasing again within successive minor diagonals (such that p i > p2 > p3 >; see Table 2). Although Guttman developed the concept of a circumplex with reference to abilities data, where correlations tend to be all positive, Wiggins (1981 et. al., p. 263) note that Guttman intended the circumplex to be a general model applicable to bipolar data (i.e., personality data) as well. Wiggins et. al. (1981) have also illustrated how principal components analysis (without rotation) may be used as an alternative to Guttman's order analysis for identifying circumplex patterns in a correlation matrix. A Circumplex Measure of Interests 20 Guttman's original formulation of a circumplex model was non-metric, concerning only the ordering of correlations and not their absolute magnitudes. However, because a circular ordering implies latent continua underlying the structure, a circumplex can be represented as a (two) dimensional circular array, and may therefore be given a metric interpretation, with variables interpreted as points along two continuous dimensions, or as vectors in conventional Euclidean space. This use of a circumplex as a formal geometric model has a long history in interpersonal behaviour theory (LaForge, Leary , Naboisek, Coffey, & Freedman, 1954; Lorr & McNai r , 1965) and has been used as an explicit measurement model to construct measures of interpersonal behaviour (Wiggins, Trapnel l , & Phillips, 1988). Circumplex Models vs Factor Models Circumplex models differ in fundamental ways from the traditional model used to structure personality and vocational interest variables, the factor analytic model. Factor models, which take their name from the statistical method they employ (factor analysis) to define structure, impose order on a set of variables by attempting to create maximally distinct categories such that variables will have membership across the least number of categories. In methodological terms this goal corresponds to the factor rotational criteria known as "simple structure" (Thurstone, 1947; Tucker, 1955), which assumes that a particular pattern of factor loadings (i.e., simple rather than complex) is most interpretable and therefore most useful to define the categories for ordering the phenomena. By defining category membership in terms of discrete "all-or-nothing" criteria, factor representations have been described as a form of digital coding of relations (Plutchik, 1980, p. 196). (Note: The term "discrete" refers to the manner of defining category boundaries and not to quantitative attributes of the factors themselves which are obviously continuous, i.e., dimensional). Plutchik contrasts this with analogue types of representations such as spatial A Circumplex Measure of Interests 21 models, and suggests that for some psychological domains analogue techniques are more useful than digital ones for properly representing domain structure. Circumplex models are a type of analogue model. Rather than assuming discrete boundaries between categories as in factor models, circumplex models assume overlapping boundaries, with the content of one category shading gradually into the next. This feature of circumplex models makes them ideal for representing categories of phenomena that persons structure naturally as "fuzzy sets" (Zadeh, 1965), categories lacking "necessary and sufficient" boundary criteria, such as colours, natural language categories (Rosch, 1978), and, most importantly, categories of person perception—e.g., traits and stereotypes (Cantor & Mischel, 1979). The facility with which circumplex models can accommodate the blending property of trait categories makes them ideal, Hogan (1983) argues, for representing type concepts in personality psychology: The advantage of circumplex models over factor lists is that they may represent a more "ecologically va l id" method of trait classification because, in reality, only certain traits tend only to occur together, and they form types, (p.66-67) Because Holland's entire theory rests on both typological and person perception concepts (the notion that interest inventories work because they essentially tap powerful psychological stereotypes-see Holland, 1985, p.9-10), a circumplex would appear an appropriate structural model for Holland's vocational interest categories on conceptual grounds alone, that is, independently of the fact that they also appear to be empirically structured this way. Measurement Implications of Circumplex Models As a construct-based approach to measurement, circumplex methods share with other construct-oriented methods the goals of high within-scale item homogeneity, i.e., high inter-item correlations (so as to maximize reliability and interpretability of scale scores), and lower correlations of items between scales (so as to justify scoring each scale separately A Circumplex Measure of Interests 22 and interpreting each differently). A circumplex measurement model thus resembles the traditional factor model for multi-scale test construction in the use of both internal consistency and discriminant validity criteria for item selection. Circumplex models differ radically from traditional factor models, however, in between-scale structural criteria for item selection. Factor models generally seek to minimize between-scale correlations. Both in orthogonal factor models, which seek uncorrected resulting scales, and in oblique models, which assume some correlation among scales, item selection proceeds by identifying those items most clearly marking one and only one factor. Items correlating highly with more than one factor are discarded. Circumplex models, on the other hand, seek adequate representation of a continuum of phenomena, which, in a dimensional interpretation of a circumplex, means adequate representation of all significant blends of the two underlying dimensions. Item selection criteria under a circumplex model is, in a sense, the opposite of the factor model's simple structure criteria. In a circumplex approach, items correlating highly with both dimensions are not discarded but deliberately sought on the assumption that intermediate regions between dimensions are substantively meaningful, can be empirically discriminated by subjects, and therefore deserve to be represented in measurement separately from those scales marking the dimensional axes themselves. Because a circumplex model assumes that all locations along the circular continuum are of potential theoretical interest, the concept of representativeness or content validity has a unique aspect under such a model: the degree to which the entire circular continuum spanned by the two dimensions of interest are represented in measurement, which may be termed the angular representativeness of the item pool. Because this can easily be determined by inspecting a plot of the item projections onto the circumplex dimensions, one advantage of a circumplex model is that it immediately reveals deficiencies in content coverage of a set of measures, or in an item pool, by drawing attention to "gaps" in the circular structure of the variables. A Circumplex Measure of Interests 23 Under a circumplex measurement model the traditional concept of discriminant validity also has a unique aspect. Empir ical overlap or correlation among a set of circumplex scales is expected by definition. The closer together any pair of variables are along the perimeter of the two circumplex factors, however, the higher their correlation with one another. Thus there is trade off between the goal of representing in measurement as many points along the continuum as possible (content validity) and maintaining sufficient distance between scales so that they are not too highly correlated (discriminant validity). This trade-off is maximized, however, by ensuring equal spacing among the variables: the highest absolute correlation among any two in a set of variables defining a circumplex is at a min imum when the variables are equally spaced. This structural criterion, unique to a circumplex model, may be termed circular validity. A final unique aspect of structural validity under a circumplex measurement model is the requirement that the covariance among the resulting set of scales be accounted for by no more than two factors, and that these factors account for approximately equal proportions of variance. The first part of this requirement is analogous to the traditional notion of "factorial validity" (Messick, 1980). In the case of a circumplex model, the expected dimensionality is always two, and circumplex measures would be expected to demonstrate two dimensionality as a requirement of construct validity. The second part of the dimensional requirement, that the two dimensions account for about equal proportions of total variance, ensures a circular rather than elliptical structure among the set of scales. It is this aspect of factorial validity that distinguishes circumplex from factor models. Together these two dimensional criteria may be termed the dimensional validity of circumplex measures. Two possible caveats to this third type of structural validity should be briefly noted. The dimensionality requirement refers to substantive dimensionality among the circumplex scales, and would not be considered invalidated by the presence of additional components of variation among the scales such as method and response set factors, if those factors are A Circumplex Measure of Interests 24 considered substantively unimportant on theoretical grounds (e.g., a checking factor in adjective checklists). Secondly, different circumplex domains would be expected to exhibit different levels of cohesiveness or degree of intercorrelation among their scales, with the result that the magnitude of circumplex components may vary considerably from one domain to the next. For example, two dimensions may typically account for 7 0 % of total scale variance in circumplex measures of interpersonal dispositions, whereas a value of only 6 0 % may be typical for circumplex measures of interpersonal problems. The "weaker" degree of dimensional validity in the latter case may stem not from poor item sampling but from the nature of the phenomena being measured. Thus the requirement of dimensional validity must take into account the "characteristic intercorrelation" expected within particular psychological domains (Loevinger, 1957, p. 662). To summarize, circumplex models differ qualitatively from factor models in assuming continuous rather than discrete relations between categories of variables, a feature which makes circumplex models particularly well-suited to represent the fuzzy category boundaries and semantic overlap inherent in trait and type concepts. As an approach to test construction, circumplex measurement models are novel in providing an explicit structural definition of the universe of content implied for each circumplex domain. As a construct-oriented technique they employ psychometric criteria similar to traditional construct approaches, and at the same time possess a unique and distinctive set of structural criteria for item selection. These criteria include angular representativeness (coverage of all points on the interest circle), circular validity (even spacing), and dimensional validity (maximum percent of variance accounted for, and in approximately equal proportions, by the two circumplex components). A Circumplex Measure of Interests 25 5. Toward a Circumplex Measurement Model for Holland's Typology In view of the conceptual and empirical viability of a circumplex model for Holland's typology, and in view of the explicit role of that model in Holland's theory, consideration of the structural aspect of validity, specifically, the requirement that test structure be isomorphic to the theory's structural model, argues for a circumplex measurement model for Holland's typology. Consequently, this study sought to to construct for the two interest dimensions underlying Holland's typology a set of structurally pure circumplex marker scales via application of Phillips (1983) circumplex scaling procedures. The specific hypotheses were: (1) that adoption of an explicit circumplex measurement model for Holland's typology would permit construction of a set of eight vocational interest scales possessing excellent circumplex properties; (2) that the factor structure of the resulting scales in the validation sample would show exceptionally clear two-dimensionality, providing additional support for a circumplex view of Holland's structural model which implies (but does not theoretically elaborate) two higher order dimensions ordering the types. A Circumplex Measure of Interests 26 III. Method & Results Overview of the studies The present study involved two related studies. In Study 1, Phillips' Circumplex program was used to calculate the circumplex coordinates of all 160 items from Holland's Vocational Preference Inventory. A subset of items possessing the best circumplex properties were scored as a set of six marker scales for Study 2, each representing one of the 6 original VP I vocational scales, but each somewhat shorter in length. These "purif ied" markers of the VP I served to initialize the factor space in Phillips' program in Study 2. Calculation of the circumplex properties of the VPI 's item pool in Study 1 also served to inform the process of selecting new items for inclusion in Study 2. Comparison of the content of VP I items in relation to their circumplex properties (their locations in the factor plane, and degree of correlation with the two factors) provided evidence suggesting what types of items might be written to fill in the "gaps" around the vocational interest circle. In Study 2, Phillips' procedures were used to assemble from a large item pool a set of eight vocational interest scales possessing optimal structural validity according to a circumplex model of scale structure. Structural validitj' of the resulting scales were evaluated in a second, independent sample. The within-scale component of structural validity was defined in terms of traditional construct-based criteria, i.e., homogeneity and internal consistency, and evaluated using traditional indices of these criteria. The between-scale component of structural validity was defined, under a circumplex model, with reference to three criteria (1) minimal scale variance beyond the two circumplex dimensions, (2) equal magnitudes of the two latent roots corresponding to the circumplex components; and (3) equal spacing of the scales around these two components. The structural validity of the set of circumplex octant scales derived on the first sample was evaluated according to these criteria on a second independent sample. A Circumplex Measure of Interests 27 Phillips'(1983) Circumplex Analysis Program Phillips' program for constructing circumplex scales involves a principal components analysis with a targeted rotation of two specified components, the calculation of polar coordinates (angle and vector length) of item score projections on the plane defined by the desired components, and a routine to select items within pre-specified angular sectors on this plane, which are then combined into scales to represent the sectors. For both Study 1 and Study 2, Phillips' program was utilized in the following manner: (1) a set of initializing scales (the six vocational scales from Holland's VPI) was intercorrelated and factored by the method of principal components. (2) The components were rotated to a criterion that minimized the least squares differences between pre-specified angular locations of the scales and their empirical locations (e.g, for a set of six scales, under a circumplex model the specified ideal angular locations were 0 ° , 60 ° , 120° , 180 ° , 240 ° , and 300 ° ). (3) The correlation between each item in the pool and the two component scores was calculated and interpreted as an item's location between each of the components. These values were converted to polar coordinates for each item: item angular locations were calculated by taking the arctan of the ration of the two correlations (e.g., theta = arctan (.4/.6) = 33.7 degrees); item vector length, i.e., communality, was computed as the sum of the squares of the item or scale correlations with the component 9 9 9 scores (e.g., h =( .4)^ +(.6) = .52). The program was then instructed to divide the factor plane into a number of equally spaced sectors corresponding to the desired number of resulting scales and to select those items falling within each sector having the highest communality values. A set of scales corresponding to each sector was then scored from these best within-sector items. These resulting scales were then used as initializing scales for a second iteration, the resulting scales from that iteration initialized the next iteration, and so on. This procedure was continued until no further improvements in structure of the resulting scales were observable. In previous applications of the program using personality adjectives or questionnaire statements, circumplex values generally stabilized after only 1-A Circumplex Measure of Interests 28 2 iterations, so an equally small number of iterations was anticipated for the vocational interest data employed here. •Controlling for the General Factor in Vocational Interest Measurement A critical problem undermining validity in self assessments of personality is score variation due not to real differences in level of the trait being measured, but to differences in the use of the rating format, or due to item properties other than their intended content (e.g., social desirability, ambiguity, or direction of wording). These individual differences, labelled "response set" (Cronbach, 1950) or "response styles" (Jackson & Messick, 1962) can potentially confound interpretations of trait relations, particularly when using multivariate data reduction techniques such as factor analysis. When a response style is prominant (accounting for a large proportion of score variance) and general (operating across all traits being measured) it is possible that all traits being assessed will be positively correlated, even those considered semantically opposite (e.g., Bentler, 1969; Russell, 1980). Factor analysis of such a correlation matrix will usually yield a substantial general factor on which all variables load highly and positively, a factor corresponding to this response style. If rotation is not performed on the solution, emergence of such a factor is of little consequence to the interpretation of the other factors. Rotation when such a factor is present, however, can lead to interpretive problems such as the confounding of trait and response style variance within each rotated factor. More importantly for circumplex analysis, however, the presence of a general response set factor in circumplex data may lead to the decision to rotate more than the two circumplex factors when in fact only two substantive factors, the circumplex ones, are present. This practice would eradicate any circumplexity present in the data as the circumplex variance is re-distributed in favor of simpler structure across the 3 factors. Prediger (1982) has suggested this practice is one reason some confusion exists in the literature as to the number of interpretable factors in the VP I and SDS. A Circumplex Measure of Interests 29 Ratings of vocational or avocational interests are plagued by a particular type of response set known as "checking rate", a general tendency to endorse many or few interest items. In factor analyses of interest scales this response style appears as a general factor, usually the first and largest factor, on which all the interest factors have large and positive loadings. In six different measures of Holland's typology reviewed by Prediger (1982), including Holland's VP I and SDS , this general "checking-rate" factor accounted for an average of 4 2 % of total scale variance in each test. Arguing that the tendency to like or dislike many interests is irrelevant to the question to identifying basic interest dimensions, i.e., directions of interests, Prediger used two methods of partitioning this response set variance from the circumplex variance: (1) examining unrotated principal components, and (2) utilizing targeted rotational methods (Cooley & Lohnes' 1971, pp. 137-143) to extract pre-specified substantive factors first. He concluded that each method produced identical results (the two substantive factors, which showed the expected circular pattern of loadings, were identical across methods). A simple, alternative procedure for eliminating the effects of a general "checking-rate" factor, which is equivalent to residualizing subject's scores on such a factor, is to standardize each subject's scores with respect to their individual means and standard standard deviations computed across all variables, i.e., to ipsatize scores, before factoring. Although there appears to be no previous mention of this technique within the vocational interest literature, there are several precedents for its use in the personality literature (Alden, Wiggins, & Pincus, 1989; Russell, 1980), and Phillips (1983) has incorporated such a procedure as an option within his circumplex analysis program. The decision to use ipsatization rather than partial correlation techniques is somewhat arbitrary. However, if a general factor is considered to be "nuisance" variance on a priori grounds (as in vocational interest data) it seems logical to remove its effects prior to computing correlations rather than after. The availability of the ipsatization option in the circumplex A Circumplex Measure of Interests 30 program made it an expedient method to eliminate the effects of "checking-rate" in the present study. Study 1 Method Subjects The subjects for Study 1 were 113 Vancouver Community students (68 women and 45 men) enrolled in introductory psychology courses. Although this sample was somewhat smaller than desirable for calculating stable estimates of circumplex properties of items (sample sizes of 175 subjects or greater are recommended; Wiggins, Steiger & Gaelick, 1981), it was sufficient for the purpose of providing the seed items for the item pool of Study 2. Measures The only measure used in Study 1 was the 1978 edition of Holland's Vocational Preference Inventory (VPI). The VP I consists of 160 occupational titles to which a respondant indicates preference or interest by checking " Y e s " (scored as "1") or " N o " (scored as "0") . Eleven scales are scored from these 160 items by summing responses to items keyed for each scale. Six of these scales measure degree of resemblance to the 6 vocational personality types postulated by Holland's theory (see pp. 8-11 above, and Table 1). The remaining 5 scales measure non-vocational constructs (acquiescence, masculinity, etc.). Because all 160 items are occupational titles, all 160 items were included in the item pool for Study 1, though only the 84 scored by Holland (1978) for the vocational type scales were used to initialize Phillips' Circumplex program for the present Study. Although Holland's Self-Directed Search (SDS) has to some extent superceded the Vocational Preference Inventory as the preferred measure of Holland's typology in both A Circumplex Measure of Interests 31 counseling and research uses, several factors make the VP I preferable to the SDS for use in this study. First, the SDS is not strictly a vocational "interest" measure in that it employs self-ratings of abilities and competencies as well as activity and occupational preference ratings to derive Holland type scores. The multi-method nature of the SDS is a desirable property for practical assessment purposes in that it casts a broader net around the constructs thereby improving generalizability of the measures. For this study, however, the different sections would have necessitated four separate applications of the procedures (for each subsection) which would have unnecessarily complicated this demonstration of a new scale construction technique. Secondly, the SDS takes approximately four times as long to complete as the VP I , which precluded its use for the subject samples available (it was too long to complete during the regularly scheduled class periods that were available for testing). F inal ly, the VP I is the original measure with which Holland discovered the hexagonal model (Holland, 1985), and is considered by Holland to be as valid a measure of the types as the SDS (1985, p.42). Recent evidence, in fact, casts some doubt on the construct validity of several of the subsections of the SDS (Lowman & Wil l iams, 1987). For all of these reasons, the VP I will be used here rather than the SDS to operationalize Holland's typology. Procedure Subjects completed the 160 item Vocational Preference Inventory (Holland, 1978) during regularly scheduled class periods in the fall of 1984. Subjects were told that participation was entirely voluntary and that anyone who wished not to participate could be excused from the class prior to or at any time during the 20 minute testing session. No subjects declined to participate. Following testing, subjects were given a brief lecture on personality and vocational interests and the nature of the study was explained in detail to them. A Circumplex Measure of Interests 32 Analyses VP I scale scores were computed by summing item responses keyed for each scale as per Holland (1978). VP I scale and item scores from the present sample were then analysed by Phillips' Circumplex program in the manner described above. A subset of 51 VP I items judged to possess the best circumplex properties (angular location and communality) were selected to provide a set of VP I marker scales to initialize Phillips' program in Study 2. Results Circumplex Analysis of Holland's VPI Table 3 presents the matrix of intercorrelations among VP I scales scores from the Langara Community College sample of 113 subjects. Insert Table 3 and Figure 2 About Here Figure 2 presents the plot of loadings of VP I scales on the first and second principal components extracted from their intercorrelations (following ipsatization to remove the general response set factor). Eigenvalues for the first three VP I components extracted from this ipsatized data were 2.08, 1.63, and 1.05, and the proportions of total scale variance accounted for by each component were 34.7, 27.2, and 16.9 % , respectively. Comparison of Figure 2 with similar VP I factor plots reported in previously studies (Holland et. al., 1969; Lunneborg & Lunneborg, 1975) suggests a close degree of correspondance in VP I scale structure across these samples. To quantitatively evaluate this degree of correspondance, factor congruence coefficients (Wrigley & Neuhouse, 1955) were computed between VP I factors extracted from the present sample and those extracted from a similar matrix obtained from a published report (Holland, et. al., 1969). A Circumplex Measure of Interests 33 The latter VP I matrix was based on a large (n= 1234), representative sample of male two-year college students. Since this comparison matrix is based on unipsatized VP I data, the Langara sample VP I scale intercorrelation matrix for this factor congruence analysis was likewise based on unipsatized scores. For each matr ix three components were first extracted and rotated to a target matrix suggested by Prediger's (1982) review of the VPI 's structure: Factor I was designated a general factor (the expected VP I response set factor) and Factors II and III target values were geometric values (sines and cosines) representing an ideal 6-variable circumplex. Factor congruence coefficients (Wrigley & Newhouse, 1955) calculated between the three pairs of corresponding factors were .99, .98, .98, respectively. Given the twenty-year difference in time of administration, differing population characteristics, and different editions of the VP I between these two samples, these coefficients may be considered a conservative estimate of VP I factor replicability in the Langara sample. They suggest that the VP I structure in this sample is trustworthy as a basis for estimating item coordinates in VP I factor space. Table 4 presents the results from Step 4 of the circumplex analysis of the VP I , the location of each VP I item, expressed in polar coordinate form, with the theoretically rotated VP I circumplex components. Items in Table 4 are those 84 of the 160 VP I items that are scored for the six vocational type scales, and are grouped in Table 4 according to this original scoring. Also included in Table 4 are three additional items not originally scored for the vocational scales but which were retained for the final marker scales due to their possessing good circumplex properties (e.g., "Department Store Manager" ) or content judged prototypical of a content region in the VP I space (e.g., "Social Worker" at 0 ° ) ) . Within each type, items in Table 4 are sorted, in ascending order, by their angular location within the VP I circumplex space. Insert Table 4 About Here A Circumplex Measure of Interests 34 The circumplex analysis reveals several notable structural characteristics of the original 84 VP I vocational items. F irst , the location of items in relation to their content provides rather striking support for the hypothesis that Holland's hexagonal model may be viewed as a circumplex model: as one reads down through the list of VP I items in Table 4, changes in content between the items appears to form a circular continuum, with no beginning and no endpoint (e.g., note the similarity in content between the Realistic items listed first in Table 4 and those Conventional items at the bottom of Table 4). This patterning is all the more remarkable when one considers that these items were not in any way selected with reference to their circular properties. A t the item level as well as the scale level, the VP I shows evidence of a circular continuum underlying its structure. Secondly, inspection of the angular location of VP I items in relation to their scale assignment (the VP I scale in which they are grouped in Table 4) reveals several apparently "miskeyed" items. The Realistic items "F i sh & Wildlife Specialist" and "Tree Surgeon", for example, have angular locations closer to the theoretical midpoint of the Investigative region (at 240° ) than half of the Investigative items. Other examples of item mislocations, which occur for all but one (Investigative) of the six scales, are "Commercial Art ist" (338 ° ) , "Cl inical Psychologist" (327 ° ) , "School Principal" (50 ° ) , "Business Executive" (97 ° ) , and "Bank Tel ler" (76 ° ) . These mislocations imply that some items are scored for the wrong scale in terms of the psychological meaning of the items (and scales) relevant to Holland's theory, which defines this meaning with reference to the hexagonal model (see above, p. 11). Empirically, this meaning corresponds to the circumplex components of scale scores rather than other common (e.g. the " g " factor) or scale-specific dimensions of score meaning. Consequently, items which measure regions of the circumplex other than those of the scale they purportedly represent may be said to be measuring psychological characteristics other than that assumed for that scale, and are, in this sense, "miskeyed". A Circumplex Measure of Interests 35 It is possible, however, that a VP I item found scored for the wrong scale (in terms of its circumplex location) may, nevertheless, correlate more highly with the total score of the keyed scale than the total score of the scale it is actually closer to in circumplex space (although this should be a rare occurance since such a large proportion of scale variance is accounted for by the circumplex components). This being so, use of standard item-total criteria to select items for scales within such domains in potentially problematic in that items may be selected that are good markers of something unique to a particular scale but that are poor markers of the psychological properties of the scale that are of central interest. The "mislocations" evident in Table 4 illustrate, therefore, an advantage of the circumplex measurement model employed here, compared to a traditional psychometric model, when constructing measures to represent circumplex domains: circumplex approaches define and display "item-total" relations only with reference to "total-score" variance relevant to the theory, and provide, therefore, a better psychometric basis for item selection in such domains. Column 2 of Table 4 presents the communality of each item with the two circumplex components, expressed as the squared multiple correlation of each item score with the two component scores. Communality indicates the degree to which an item's variance is accounted for by the circumplex components. Higher values indicate that an item is a good marker of that region of the plane. Lower values indicate that an item is measuring something other than the two dimensions underlying the VP I hexagonal structure and that such an item is, therefore, a relatively poor marker of that region. Item communalities for the 84 VP I vocational scale items ranged from .007 ("Speculator", at 53.9 ° ) to .527 ("Certified Public Accountant", at 116.9 ° ) . Communalities tended to be vary systematically in relation to angular location. Items falling between the Conventional region midpoint and the the Enterprising region bordering Conventional (roughly 120° to 9 0 ° ) tended to have higher communalities. Items falling within the Social region and between the Social and Enterprising regions (roughly 327 ° to 58 ° ) tended to have smaller A Circumplex Measure of Interests 36 communalities. This pattern of communality variation at the item level reflects the differences in scale communalities. In this sample, Conventional scale scores had the highest squared, multiple correlation with the VP I circumplex components (.74) and Social had the lowest (.61). These differences suggests that individual differences in occupational interests are determined by the psychological dimensions underlying the hexagon (the circumplex factors) to a greater degree for some Holland categories than for others. Selection of VPI Initialization Items for Study 2 The angular location and communality values in Table 4 provided a basis for item selection for a set of intialization scales to be used in Study 2. Ideally, all 84 original VP I items would have been included in this next study to initialize the component space in Phillips' program, since the original VP I scales provide a well-validated initial definition of the vocational circumplex components. Due to the limited testing time available for data collection in Study 2, however, it was decided that a more efficient approach toward the final goal of Study 2, the development of structurally-optimal circumplex scales, was to capitalize on the present findings and omit from consideration those VP I items found in Study 1 to possess poor circumplex properties. This permitted more new items, which were written with the benefit of knowledge obtained in Study 1, to be included in Study 2. Several exceptions to the empirical item-selection criteria were made on the basis of theoretical considerations. In the case of several retained Zoiocommunality items, the decision was based on the conceptual significance of their angular locations in relation to their content. For example, the item "Anthropologist", although having a communality value of only .098, fell at exactly the midpoint of one of the larger "gaps" on the VP I plane, that between the Artistic and Investigative regions. Since one of the goals of Study 2 was construction of thematically meaningful new scales to fill in the larger of the "gaps" apparent between scales in VP I circumplex space, the location of "Anthropologist" was of theoretical interest in that it suggested a new scale might be constructed to mark the A Circumplex Measure of Interests 37 region between Investigative and Artistic by splitting the Investigative scale into " H a r d " and "Soft" science clusters, the former falling nearer in circumplex space to the Realistic occupations, and the latter closer to the Artistic occupations. The decision not to include certain /ijjj/i-communalities items was based on a variety of rational considerations. These included: content redundancy with items having higher communality values (e.g., "Auto Mechanic" [.129] vs "Airplane Mechanic" [.302]) and angular locations midway between scales already located somewhat too "close" on the plane ("Business Executive", which projected at 97 ° , midway between Enterprising and Conventional). The final items selected for the initialization scales of Study 2 are underlined in Table 4. Discussion In Study 1 the circumplex structure of the 1978 VP I scales and items were investigated using Phillips (1983) circumplex analysis program. Following ipsatization of item scores, circumplex properties of the VP I were investigated by projecting VP I scales and items onto the two rotated,, circumplex-structured components underlying the VP I scale intercorrelations. The VP I scale structure in the present sample was determined to be highly similar to that reported previously for a much larger, representative sample. The distribution of the 84 VP I vocational scale items around the two largest (ipsatized) VP I components generally supported Holland's hexagonal model of scale relations, and strongly suggested a circumplex view of this model: items scored for a particular scale which located near a second scale were more similar in content to that scale than items which located farther from the second scale. This pattern of item similarity was evident for all six VP I scales and appeared to form a circular continuum of meaning. A Circumplex Measure of Interests 38 Examination of item angles and communalities in comparison with their original scale assignments highlighted psychometric deficiencies within the VPI scales and illustrated the advantages of circumplex methods in the measurement of phenomena for which such a measurement model is appropriate. Item angles and communalities in the present sample also provided a basis for reduction of the VPI item pool to a smaller set of marker scales, representing each of the six original VPI vocational scales, to be used as initialization scales in Study 2. Lastly, examination of each item's content in relation to it's circumplex properties indicated content "gaps" in the original VPI item pool, and suggested what types of occuptions might locate in those poorly marked regions of the VPI circumplex. Two regions of the plane seemed to offer the best possibilities for two new scales that might be thematically distinct from the six original VPI scales. These were the regions between Investigative and Artistic, and between Social and Enterprising. The gap between Realistic and Conventional scales was, as in previous studies of VPI structure, even larger than the that between the former two pairs of scales. The circumplex analysis, however, revealed very few items locating between these regions, and none that suggested a meaningful major occupational category that could be located to mark that region. Results from Study 1 suggested that a set of eight, evenly-spaced scales to mark the circumplex dimensions underlying Holland's typology might be constructed by splitting the Investigative scale and Enterprising Scales into two themes each, and by identifying a number of items that might "pull" the Conventional and Realistic scales somewhat closer together on the circumplex. These were the major objectives in the construction of the item pool for Study 2. A Circumplex Measure of Interests 39 Study 2 Method Subjects The subjects for Study 2 were 515 University of Brit ish Columbia students (313 women and 202 men) enrolled in introductory psychology courses. Students in these classes represent a fair cross-section of university faculties (e.g., Commerce, Science, Ar ts , Engineering, and Nursing) and so constitute a fairly heterogeneous sample with respect to Holland's six vocational types. Measures Preliminary item pool. Based on the results of Study 1, 101 new occupational titles were generated for the present item pool using the following procedure. The relation between item content and angular location in Study 1 was examined in order to deduce what kinds of occupational titles might locate in regions poorly sampled by the VP I item pool. The content of items locating midway between Enterprising and Social scales (e.g. "Public Relations Manager" , "Personnel Manager" , and "Hotel Manager") suggested an occupational theme midway between the commercial orientation of the Enterprising theme and the human welfare orientation of the Social theme. Several new items having a commercial/social flavor (e.g., "Trave l Agency Director", "Tour ism Director", "Social Events Organizer") were written, therefore, in an attempt to create a separate scale to mark this theme. A similar approach was followed to create a scale intermediate in content between the original Artistic and Investigative scales of the VPI . The principal objective in creating two new scales was to enable a conventional "octant" representation of the vocational circumplex dimensions. Such a set provides better coverage of these dimensions A Circumplex Measure of Interests 40 than the original six scales of the VP I in that it provides for assessment of all four poles of the underlying dimensions as well as all four intermediate regions between the two axes. Other items were created or identified that might "re-align" the position of some of the original themes. For example, several Realistic theme items were created that, it was hoped, would project in the Realistic region bordering Conventional (e.g., "Business Equipment Technician", "Da ta Systems Technician") thereby serving to "pu l l " these two scales somewhat closer together in factor space. This process of identifying which items best mark the points of interest in factor space and subsequently building up the item pool with similar items is analogous to traditional factor-based methods of scale construction methods, where the content of items loading highest on the factors or subfactors of interest guides the writing of items to better measure those factors. The 101 occupational titles thus generated were combined with the 49 VP I items retained from Study 1 for a final pool of 152 items. These items were assembled into a 5-page booklet entitled the Job Preference Rating Form (JPRF; see appendix I). Items were arranged in two different random orders, labelled JPRF-I, and JPRF-II, to minimize any possible effects of item order on item correlations. Rating instructions for the JPRF were adapted from the Strong-Campbell Interest Inventory (Campbell, et. al., 1985). These instructions are very similar to those of the VP I but were felt to be a bit more appropriate for the JPRF due to a change in response format (see below). Response Format. The traditional response format used to scale ratings of vocational interest is a three-option, " L i ke " "?" "Dis l ike" scale. This format originated from T . L . Kelley's doctoral dissertation on interest measurement, supervised by E.L. Thorndike, in the early 1920's, and is retained today in the most widely used interest inventory, the Strong-Campbell (Hansen & Campbell , 1985). The response format chosen for the most recent edition of Holland's Vocational Preference Inventory (1985), and Self-Directed A Circumplex Measure of Interests 41 Search (1985) differs slightly from this traditional format, and from earlier editions, by omitting the middle option (the subject is instructed to make "no marks " if uncertain or undecided), leaving a dichotomous " Y " (like or interested), " N " (dislike or indifferent) format. Following arguments by Dawis (1980) that an operational definition of "interest" requires use of a "Like-Dislike" rating dimension (rather than other possible dimensions such as "famil iar ity" , "preference", "self-description" or "interest"), "Like-Dislike" was retained as the rating dimension in the instructions and as anchor labels on the rating scale. For a number of reasons, however, Holland's dichotomous " Y e s ' V N o " format was replaced with an 8-place, bipolar, L ikert format. This study involves selection of items with reference to principal components derived from scale intercorrelations. There are two reasons why a L ikert format would be preferable to a dichotomous format for such a study. There is consistent evidence that item and scale scores are more reliable (Nunnally, 1967), valid (Flamer, 1983), and their covariance structures more differentiated and interpretable (Comrey & Montag, 1982; K ing , K ing , & Klockars, 1983; Velicer & Stevenson, 1978) when using a larger number of response options in the rating format. Secondly, and related to the first point, is the fact that dichotomously-scored items are more susceptible than multi-category-scored items to skewness in their score distributions, which can severely limit the magnitude of their correlations (Comrey, 1978; MacLennan, 1988) and therefore potentially bias item selection. A second reason to opt for a greater number of response options than in traditional formats is that subjects apparently prefer a more differentiated scale, and judge them to be more accurate, reliable, interesting and less ambiguous that a dichotomous format (Jones, 1968; cited in Jackson, 1971). A final point concerns Holland's theory. Although Holland's theory defines interests as being bipolar (i.e., attraction to some activities and avoidance of others), and although A Circumplex Measure of Interests 42 Tyler (1960) has presented evidence that dislikes might even be more important than likes for the development of interest patterns, Holland's scoring method is unipolar, making no discrimination between "indifferent" and "dislike". Both the present two-category and earlier three-category response formats of the VP I and S D S are scored in the manner of a checklist: " Y e s " responses are scored as a " 1 " and all other responses ("No", "Indifferent" or "Uncertain") as " 0 " . The insensitivity of a checklist format to interest intensity and to interest negativity would seem to render it too crude a method for assessing interests and a method not wholely consistent with Holland's bipolar theory of interests. Use of a middle category, as in past editions of the VP I and SDS , would retain bipolarity in the scale-provided the middle and dislike categories are scored differently (e.g., " Yes " , "No , and "B lank" options scored as 1, 0, -1). Better still would be a scale providing for degrees of dislike as well as like, i.e., a bipolar, L ikert format. For all of the reasons cited above, this study employed an 8-point, bipolar Likert scale ranging from "strongly dislike" to "strongly like". Eight scale intervals were chosen on the basis of studies showing that psychometric improvements due to increasing the number of scale points tend to level off at about 7-11 scale points (Nunnally, 1967). Anchor words were used only at the ends of the scale, as suggested by Dixon, Bobo, & Stevick (1984), who found the end-defined format superior, in terms of marginally higher individual difference variance, to all-defined, or end and middle-defined formats. Procedure Subjects completed the 152 item JPRF, as part of a larger personality assessment study, in groups of 1 to 40 subjects per session, over a 6 month period from October 1987 to March 1988. The assessment was conducted in the Individual Differences Laboratory of the U B C Psychology Department. A l l subjects completed the personality measures first, followed by the JPRF (having one of two random item-orders). After completing all of the measures subjects were given a written debriefing form, which described in detail the t A Circumplex Measure of Interests 43 purpose of the study, and a credit slip granting two points toward their Psychology 100, or 200 course grade. Analyses Subjects' responses to the 152 occupational items of the JPRF provided the data for Study 2. Item order effects were checked by evaluating the stability of items' circumplex coordinates across the two item orders. Items were then re-ordered permitting the combining of samples across different item orders, and the resulting total sample of 515 subjects was divided into two subsamples of 257 (sample A) and 258 subjects (sample B) by random assignment of subjects. Scale derivation was based upon JPRF item responses in sample A. The structural validity of the new scales was evaluated in sample B. Scale Derivation. The 51 items retained from the VPI in Study 1 were scored from the JPRF to represent the original six VPI scales. The correlations among these scales provided the initial definition of the vocational circumplex space in the first run of Phillips' program. All 152 JPRF items were projected onto these two rotated components and initial estimates of angular location and communality provided the basis for re-combining items into a set of preliminary octant scales, which then were used to initialize a second circumplex analysis. Several iterations of Phillips' program followed in which item locations and communalities were re-calculated, and composition of the eight scales was adjusted. Scale assignments of items were determined on each iteration initially by Phillips' program but were adjusted afterward, if necessary, on substantive grounds in those instances where the program placed items falling near the border of two scales into the more thematically-dissimilar scale relative to that item (e.g., "Rehabilitation Counselor" assigned to the Artistic scale instead of the more thematically appropriate Social scale). In this manner, a set of eight, thematically and psychometrically homogeneous scales, consisting of 8 items each, and possessing optimal circumplex structural characteristics in that sample was constructed. A Circumplex Measure of Interests 44 Structural Evaluation. The structural validity of the new scales was evaluated in a second, independent sample (sample B) using the following methods. Within-scale structural validity was assessed via conventional internal consistency analyses (coefficient alpha estimates of reliability, and mean inter-item correlation estimates of homogeneity). The between-scale aspect of structural validity was evaluated with reference to a circumplex model. The dimensionality component of circumplexity was evaluated by plotting the magnitude of eigenvalues of the resulting scale intercorrelation matrix against the number of components, and inspecting the resulting curve for that point at which the values level out. This non-statistical procedure, known as a "scree" test (Cattell, 1966), is recommended as one of the most accurate, reliable, and useful number-of-factor tests (Zwick & Velicer, 1982). In the present study, it was hypothesized that such a screeplot would show a sharp drop and rapid leveling-off in the curve after the second factor, clearly indicating a two-dimensional factor structure. The second criteria of circumplexity, circularity, or "even-spacing", was evaluated via a procedure recommended by Wiggins, Steiger, & Gaelick (1981) for evaluating circumplexity in personality data. In the present study this procedure was employed as follows: A chi-square estimate of goodness-of-fit between the intercorrelation matrix of new scales obtained in the validation sample (sample B) and an ideal 8x8 circulant matrix (shown in Table 2) was computed via Steiger's MULTICORR program. Secondly, MULTICORR chi-squares of circumplex fit were also generated for a total of 16 comparison matrices. Wiggins et. al (1981) suggest that the degree of departure of the sample matrix from the model should not be interpreted in a strict Neyman-Pearson manner (e.g., when statistical failure to reject the null hypothesis means failure of the circulant model) because statistical rejection of complex structural hypotheses is virtually inevitable in sufficiently large samples. Rather, they argue, the degree of circumplexity must be evaluated relativistically. Consequently, a sample of correlation matrices from the vocational and personality literatures which have been interpreted as having a circumplex A Circumplex Measure of Interests 45 structure, were obtained to provide empirical "benchmarks" of circumplexity for this study. Eight of these comparison matrices were 6x6 scale intercorrelation matrices for measures of Holland's types (e.g, Holland's VPI , SDS, American College Test Program's UN IACT- IV , and the Strong-Campbell General Occupational Theme Scales). A n additional 3 were 8x8 matrices of scale intercorrelations for measures of Roe's (1956) eight occupational themes, whose circular structure corresponds closely to Holland's model. These 1 ^ matrices were selected either because they have featured prominantly in vocational interest structural research or because they appeared to be among the best examples of circular structure in this literature. Two additional comparison matrices were 8x8 scale intercorrelation matrices for circumplex measures of personality that had been constructed via the same circumplex methodology employed here (Phillips' program) and have been found to exhibit a very close empirical fit to the circumplex model. Although representing a domain of variables quite different from vocational interests, these latter matrices provide somewhat of an "upper-bound" value for empirical examples of circumplexity due to their manner of construction, and so have been included here for that reason. The remaining two comparison matrices were those obtained in this study for the VP I and the VP I marker scales (in the Langara and U B C samples, respectively). Because under a circumplex structural hypothesis, matrices of different order have different degrees of freedom, resulting chi-squares for each matrix were transformed to standard normal deviates by means of the formula suggested by Wilson & Hilferty (1931), so that the value of their chi-squares could be compared directly. Chi-squares estimates of model fit, such as that obtained from MULTICORR and other techniques for analyzing covariance structures (e.g., LISREL, COSAN) are notoriously sensitive to sample size, a fact that hinders comparisons across samples differing greatly in size. Wiggins et. al. (1981) suggest that the mean squared residual (the mean squared deviation between corresponding elements in the sample correlation matrix and the circulant population correlation matrix estimated from it by MULTICORR) may be A Circumplex Measure of Interests 4 6 used as a secondary quantitative estimate of fit since it's value is unweighted with respect to sample size. Because sample sizes of the comparison matrices employed in this study ranged very widely (from 113 to 1,851 subjects) the mean squared residual (MSR) was employed here as the primary comparative index of circularity. A Circumplex Measure of Interests 47 Results 1. Derivation of Vocational Circumplex Scales Figure 3 presents the results of Step 1 of the circumplex analysis of the VP I marker scales (VPI-m), the plot of each scale's loadings on the First two rotated principal components extracted from their intercorrelations. Comparison of this figure with the factor plot of original (ipsatized) VP I scales in the Langara Community College sample (Figure 2), shows a close correspondance between the two structures. Factor congruence coefficients (Wrigley & Newhouse, 1955) computed between corresponding circumplex factors across these two samples confirms this visual impression of very similar structure. These values were .990 for the factor aligned through Social and Realistic and .994 for the factor aligned between Conventional and Enterprising at the top of the circumplex and between Investigative and Artistic at the bottom. Eigenvalues corresponding to the first three VPI-m components (ipsatized data) were 2.21, 1.97, and .69, and the proportions of total scale variance accounted for were 36.9, 32.8, and 11.4, respectively. Comparison of these results with corresponding values for the (ipsatized) Langara VP I data in Study 1 (p. 32, above) reveals a somewhat cleaner circumplex structure for the VPI-m than for the original VP I scales in the Langara sample. Note that the third eigenvalue in the Langara VP I data was approximately twice as large as that for the U B C VPI-m data (1.016 vs .565), and that the circumplex components accounted for more of the total variance in the latter sample (61.9 vs 69.7 %). Because these two samples differed with respect to item composition of the scales (total VP I vs. subset of VP I items), subject charcteristics (community college students vs. university undergraduates), and response format (checklist vs. Likert) it is not possible to know which among these factors (or others) might be responsible for the structural differences just mentioned. This comparison does offer evidence, however, that the VPI-m A Circumplex Measure of Interests 48 provided a reasonably good definition of the VPI circumplex dimensions for the purposes of initializing Phillips's program in this Study. Insert Figure 3 About Here Step 2 of Phillips' program projected all variables (items and scale scores) onto the space defined by the VPI-m scales and outputed the circumplex values of these variables in the form of polar coordinates (angle and communality) with reference to that space. Following re-division of the VPI-m component space from 6 categories (corresponding to the six initialization scale locations) into 8 categories (corresponding to ideal circumplex octant locations), the initial item circumplex values provided the basis for re-grouping and selecting items so as to construct a set of eight marker scales for that space. (Communalities and angular locations for all 152 JPRF items with reference to the VPI-m components are located in Appendix C). Using eight scales for each subsequent iteration of the program, the item selection procedure detailed in the Analyses section above (p. 43) was repeated until a final set of eight, 8-item scales were scored from the JPRF pool that were judged to possess optimal structural and substantive characteristics in this sample. A scale length of eight items was chosen, rather than a greater or lesser number, because for circumplex measures that have been constructed for other domains such a length appears to provide a good balance between individual scale reliabilities on the one-hand, and adminstrative economy on the other (Wiggins, et. al., 1988). Insert Figure 4 About Here Figure 4 presents the plot of these final resulting scales on the two rotated circumplex components. Eigenvalues for the first three components extracted were 3.07, 2.86, and A Circumplex Measure of Interests 49 .675, with the circumplex components accounting for an impressive 74.2 % of the total variance. The circumplex structure and the strong two-dimensionality of these scales compares favorably with circumplex octant measures constructed for other domains, (e.g., Wiggins, et. al., 1988; Alden, et. al., 1989), and suggests an excellent fit of the new scales to a circumplex model. Because, however, both the item composition of the scales and their component structure shown in Figure 4 were derived on the same sample, the circumplex structure may be partly artifactual, depending on sample-specific properties of the items. The circumplexity apparent in the Figure 4 may, in other words, depend in part on selection of items whose circumplex properties are a result of chance variation in this sample, properties which, by virtue of being random would not replicate. Structural validity of the final scales was therefore evaluated in a second independent sample of subjects from the same population (sample B). 2. Structural Evaluation of the Circumplex Octant Scales Internal Consistency Because component scores are linear composites of the original scale scores, the criteria for assigning items to scales in the present study, namely correlations with component scores, virtually guaranteed for the resulting scales a high degree of internal consistency, and of discriminant validity with respect to item-scale correlations. This follows from the fact that each item's two component-score correlations are a summary of the eight scale correlations for that item. Items selected on the basis of this summary must of necessity correlate more highly with its own resulting scale than with the other scales. It is useful, never-the-less, to empirically verify the assumption of scale homogeneity underlying the practice of summing items into a single scale score (or, in this case, 8 scale scores): summing items assumes that each item within a composite measures a similar construct. This assumption is traditionally evaluated for a scale v ia estimates of internal A Circumplex Measure of Interests 50 consistency. Table 5 provides these estimates in the form of coefficient alphas and mean inter-item correlations for each of the new octant scales in sample B. Since the general factor in interest ratings serves to inflate these estimates, and because that variance did not feature in the theoretical or empirical interpretation of scale scores in this study, estimates of internal consistency based on ipsatized data are also provided in Table 5. Overal l the internal consistency of of the scales was quite high, with coefficient alphas ranging from .85 (IA scale) to .94 {AR scale) for unipsatized data, and .77 (IA scale) to .92 (AR scale) for ipsatized data. Insert Table 5 About Here Structural stability To evaluate the structural stability of the new scales, principal components were extracted from the above matrix and scale and item locations on the rotated circumplex components were compared across samples A and B. Scale locations in circumplex space were compared visually. Figure 5 presents the plot of scale scores in sample B on their circumplex components, rotated, as before, to maximize fit to the theoretical octant positions. Comparison of Figure 5 with Figure 4 indicates a highly similar structure across the two samples: the eight variables in each figure show an identical ordering around the circle and show very similar angular locations and distances from the origins. With the exception of a minor shift in the location of the Social scale, the structure of the octant scales across the two samples appears remarkably similar. Insert Figure 5 About Here A Circumplex Measure of Interests 51 A more fine-grained evaluation of structural stability was provided by correlating circumplex coordinates of the 64 scale items across the two samples. These correlations were .99 and .78 for angle and communality, respectively. The high value of the former correlation suggests that the difference, or relative size, between the two component correlations for each item was essentially invariant across the two samples. The latter correlation indicates a moderate degree of variation in item communalities across the two samples, but this result may be partly an artifact of range restriction-only high communality items were retained for the scales. When this correlation was re-computed using the entire set of 152 J P R F items it increased to .88. Overall , the structure of the new octant scales appears to be very stable in this student population. Dimensional validity. Under a circumplex model the meaning of scale scores is defined in relation to the two circumplex components of score variance. For scales constructed to fit such a model it is expected that the lion's share of each scales' variance should be accounted for by the circumplex components and that remaining variance be due to measurement error rather than systematic variance attributable to non-circumplex factors. The degree of two-dimensionality of the new J P R F octant scales was evaluated in sample B by means of a plot of eigenvalues against the number of components. This screeplot is presented in Figure 6, along with a tabulation of the percent of total variance accounted for by each component. A s expected the curve of the scree line shows a pronounced drop and a rapid flattening out after the first two components, indicating a strongly too-dimensional accounting of total scale variance. Systematic components of variation beyond the two circumplex components would appear to be relatively trivial for this set of eight scales. Insert Figure 6 About Here A Circumplex Measure of Interests 52 A second aspect of dimensional validity under a circumplex model is that the two circumplex components account for roughly equal proportions of total scale variance. The difference in magnitude of the latent roots corresponding to the circumplex components was .39, or a difference of 6.2% in proportion of total variance accounted for by these components. This difference compares favorably with a mean value across nine samples of 6 .1% for another circumplex measure reported by Wiggins et. al (1988). This measure was also constructed using Phillips (1983) program, and is considered to possess an excellent circumplex structure (Wiggins, et. al., 1988). Together, results from this comparison and the screeplot suggest the new scales possess a relatively good degree of dimensional validity for a circumplex instrument. Circularity Although each of the three aspects of structural validity assessed so far-stability of structure, scale homogeneity, and dimensional validity-is important in evaluating how well the new scales conform to the structural assumptions of a circumplex measurement model, none of these criteria are strictly unique to that model. Traditional factor methods of constructing scales also make use of similar criteria-stability of item or scale factors, assumptions of scale homogeneity, and assumptions concerning the relative magnitude or importance of different factors within a given test. Quite unique to a circumplex model, however, and of central importance to the present study, is the validity requirement that the inter-scale relations conform to a circular pattern. Table 6 presents the matrix of intercorrelations among the new scales in sample B (based on ipsatized data). Comparison of the relative differences in magnitude of correlations along the minor diagonals of this matrix with the pattern expected under a circumplex model (Table 2) suggests a close fit to a circular model. Insert Table 6 About Here A Circumplex Measure of Interests 53 Quantitative estimates of the degree to which an observed matrix fits the circulant patterning displayed in Table 2 may be obtained, as demonstrated by Wiggins et. al (1981), by estimating what the ideal circumplex population correlation might be for each diagonal of the observed correlation matrix and by then obtaining a chi-square estimate of fit between the original observed matrix and this estimated matrix. The diagonal entries for the target population matrix may be estimated from the sample matrix by assuming that the correlations within each diagonal of the sample matrix represent-under a circumplex hypothesis- samples of a single population value. By means of this assumption, circularity may be evaluated independent of the absolute magnitude of correlations for a given sample matrix, which may vary between populations and between different domains of variables. Table 7 presents the results of a MULTICORR (Steiger, 1979) chi-square test of circularity for the intercorrelation matrix of scales developed in this study (the matrix shown in Table 5) and for each of 16 comparison matrices. Three goodness-of-fit indices, shown on the far right of Table 7, are presented for each sample matrix: multivariate chi square (x ), Wilson-Hilferty standardized chi square (W-H Z), and mean squared residual (MSR). For the reasons discussed in detail above (p.45-46) the 17 samples are rank-ordered with reference to the right-most index, the mean squared residual. F r om Table 7 it can be seen that the three measures constructed by Phill ips' program (IIP, IASR, and J P R F octant scales from this study), which appear at the top of the table, have the lowest mean square residual values of the 17 different samples. Next below them is the U N I A C T IV, a scale which measures Holland's six types by means of activity items rather than occupational titles. Although the present author was unable to discern from published literature on the U N I A C T IV whether circumplex structural criteria were used to construct this measure, factor plots of the U N I A C T scales reported in Hanson, Prediger, & Schussel (1977) show almost perfect circuplexity. The U N I A C T correlation matrix analyzed here was in fact based on the same data. It is interesting to note that Cole & A Circumplex Measure of Interests 54 Cole (1970) in a brief footnote (p. 13) attribute to G. Hanson the idea of employing item projections onto component scores as a method of item analysis during scale development. Since the same Hanson is also a co-author of the U N I A C T scales (Hanson, Prediger & Schussel, 1977), it seems a possibility that the low M S R found for the U N I A C T matrix may be due in part to Hanson's adoption of a circumplex measurement model for this domain, as advocated in this paper. If so, the excellent circumplex fit of the activity-based U N I A C T provides evidence of generalizability in the utility of this approach within the vocational interest domain. Comparisons across the chi-square values, their Wilson-Hilferty transformed values, and the MSRs show little apparent relation. Since all three indexes purportedly estimate the same parameter, circularity, this observation seemed puzzling. To examine these relations, correlations were computed among them across the 17 entries in Table 7. Chi square values correlated .98 with their transformed values. This correlation is less than perfect due to the small degree of variation across the 17 matrices in their order (6x6 vs 8x8). M S R was, however, uncorrelated with the other indices of circularity (.08, and .07 with Chi square and W-Hz values, respectively). The correlations between these circularity indices and sample size of the 17 matrices may, however, explain this lack of relation: While Ch i square and W-Hz values correlated .64 and .67, respectively, with sample size, the corresponding value for MSRs was -.46! Other things being equal, one would expect circularity to improve rather than deteriorate as sample size increases, since sampling error in the estimation of population correlations decreases with larger samples. The negative correlation between M S R and sample size in Table 7 is consistent with this reasoning, where-as the chi square relation appears not to be, at least in the context of its use here as a comparative index of cirularity. This difference, of course, between the chi square and M S R estimates of circularity is a consequence of the computational sensitivity of the chi square (and its W-H z transform) to sample size, and the MSR 's insensitivity to it, as discussed previously. The empirical A Circumplex Measure of Interests 55 relation of these indices to sample size reported above supports the use here of the M S R rather than the chi square index to comparatively evaluate circularity across the samples in Table 7. Chi square values and their transforms are reported for the benefit of other investigators who might wish to use them for other purposes. Insert Table 7 About Here Although some trends are discernible in the degree of circularity evident for different types of measures in Table 7 (e.g., measures of Roe's typology-e.g., R A M A K A N D VII-tended to do worse than Holland measures), there is not sufficient representation of sample matrices for the various kinds of measures to permit inferences to drawn in this regard. The rather high degree of circularity of the two larger sample VP I matrices deserves some comment, however. The VP I matrix obtained from Holland, Whitney, Cole & Richards (1969) is that cited by Holland (1973) as the basis for "discovery" of the hexagonal model. This same matrix was analyzed by Rounds et. al. (1979) to assist in evaluating the fit of the SCII Holland theme scales to that model. The equally low M S R value for the Wakefield & Doughtie (1973) VP I sample is of interest for a similar reason: their study is among those most widely cited as evidence supporting Holland's hexagonal model. It is notable that these two data sets demonstrate such an excellent fit to a perfect circumplex, better than all but the matrices for measures constructed explicitly to fit this model (excepting the U N I A C T , which might have been so constructed). Comparsion of Table 7 of this study with Table 5 from Wiggins, et. al. (1981), which displays MSRs for purported circumplex matrices from the interpersonal behavior literature, indicates that the VP I appears to possess better circumplex structure than all but two of the 19 matrices of interpersonal measures reported in that table. This finding certainly lends confidence to a "circumplex" A Circumplex Measure of Interests 56 view of the vocational interest domain, especially in view of the fact that the VP I was constructed without reference to circumplex structural criteria. A corollary to this positive assessment of the VP I is the conclusion that the scales developed in this study, which demonstrate in Table 7 an even closer fit to a circulant model than the VP I , give evidence of being one of the best examples of a circumplex structure in the vocational and personality literatures. Item content of the final set of eight vocational circumplex scales are presented in Table 8. Included in this table are preliminary normative circumplex values (angular location and communality), based on the combined sample of 515 subjects (UBC sample A + U B C sample B). Insert Table 8 About Here Discussion In the first part of Study 2, a large pool of occupational titles, which included those selected from Study 1 to mark the VP I scales plus 101 new titles generated on the basis of findings in Study 1, were administered to a new sample of university undergraduates and submitted to a circumplex analysis. Structure of the initializing scales comprised of the VP I marker items was found to be highly similar to the full VP I administered in Study 1. Most of the new items written to mark circumplex locations undersampled in the original VP I were found to project exactly where intended on the VP I circumplex dimensions. Selecting items from the entire Study 2 item pool, Phillips' program was used to construct a set of eight vocational interest scales, consisting of eight occupational titles each, whose intercorrelational structure approximated an ideal, two-dimensional circumplex. A Circumplex Measure of Interests 57 In the second part of Study 2, the various aspects of structural validity of the new octant scales were systematically evaluated in a second, independent sample. The new scales were found to each possess a high degree of internal consistency, and the circumplexity of their component structure was found to be quite stable across the derivation and validation samples. Evaluation of the degree to which the structure of the new scales conformed to the structural model assumed for the domain~a circumplex model--revealed an excellent fit of these scales to a circumplex model: dimensional validity was demonstrated by a strongly two-dimensional component solution for the scale intercorrelations, and by an acceptable degree of equivalence in the size of the two components. Circularity was demonstated by a quantitative degree of fit to an ideal circulant matrix that exceeded that of 15 comparison matrices of scale intercorrelations for several different measures of Holland's and Roe's vocational typologies. Two comparison matrices demonstrated a closer fit to a circumplex than the new scales devloped in Study 2. These matrices were, however, based upon measures also constructed, v ia Phillips' procedures, to fit a circumplex model. General Discussion The purpose of the present study was to construct a set of measurement scales so as to maximize their degree of validity in relation to a theoretical structure that had been postulated for the variable domain represented by the scales. The rationale for this goal was a sub-aspect of Loevinger's (1957) notion of construct validity which she denoted "structural" validity. In the context of multivariate personality assessment, the concept of structural validity has been elaborated by Jackson (1971) and Wiggins (1973) to mean the empirical fit between the observed pattern of relations among test variables and that expected on theoretical grounds for the constructs represented by the test variables. Wiggins (1973) called this the structural fidelity of a test. A Circumplex Measure of Interests 58 With the concept of structural fidelity as a point of departure, an explicit set of structural criteria was adopted both as a measurement model guiding item selection and as a set of criteria for evaluating the outcome of the test construction effort. The measurement model adopted was a circumplex model, for which a precise set of structural criteria have been specified, and for which an elegant set of procedures have recently been developed to construct measurement scales according to this model. The domain of variables chosen was the vocational interest domain as defined by J.L. Holland's theory of vocational choice, and operationally defined in this study by the scale structure and item content of Holland's Vocational Preference Inventory. As a specific application of Wiggins' structural fidelity rationale for test construction, this domain was selected to demonstrate the circumplex approach to test construction because a well-validated circumplex model of structure exists for this domain (Holland's hexagonal model), and because no existing measures of Holland's constructs have been constructed explicitly to fit this model. A secondary goal of the present study was the demonstration that Holland's hexagon may be viewed as a continuous, two-dimensional, circumplex model. Specifically, it was hypothesized that by means of judicious selection of items, properly scaled with reference to a circumplex structure, it would be possible to construct a set of eight separate scales to represent the same dimensions as mapped by Holland's original six scales, that would each be thematically meaningful as a distinct occupational interest cluster, would demonstrate excellent circumplex structure, and, would-as a consequence of the latter fact-exhibit very clear two-dimensionality. In Study one, application of circumplex item analysis procedures to item data from Holland's Vocational Preference Inventory offered preliminary support for the hypothesis that the dimensions accounting for the well-known "hexagonal" structure of VPI variables may be viewed as circumplex dimensions: item locations on the two dimensions gave evidence of a circular continuum ordering relations both at the scale and item level. Item locations also revealed some discrepancies between item keying and location on the A Circumplex Measure of Interests 59 circumplex in relation to scale locations. Some items located in regions corresponding to scales in which they were not scored. In Study 2, results of the item analysis from Study 1 guided the generation of new items in an attempt to better represent the circumplex dimensions underlying the VPI. The overall objective was construction of an octant representation of these dimensions, a set of eight, thematically and psychometrically homogeneous scales that would be maximally distinct from one-another in relation to the circumplex dimensions, i.e., evenly spaced around these dimensions. The hypotheses for Study 2 were confirmed to an impressive degree. Items newly written to locate in specific regions of the circumplex located, in virtually every case, precisely where expected to. Although not all items had strong projections on the circumplex, a surprisingly high percentage of them did so, enough that the task of constructing a set of evenly-spaced octant scales from this item pool was simple and straight-forward. It should be stressed here that for an item to have projected strongly onto the circumplex it must have possessed a specific correlation pattern across six (and subsequently eight) scales, i.e., one that approximated a circumplex pattern. The ease in identifying so many items that succeeding in fitting such a constrained structural hypothesis provides strong support for the circumplex assumptions guiding these efforts, i.e., supports a circumplex interpretation of the VPI's well-known hexagonal structure. Results from the structural validation efforts in the second part of Study 2 were equally positive. The excellent circumplex properties of the new scales replicated well in a second independent sample. In fact, the circumplex structure obtained in that sample was found to be among the best yet reported in either the vocational or personality literatures. Implications for Holland's Theory Because the item pool for this study was derived from Holland's VPI, because item selection criteria was based on (slightly purified) VPI principal components, and because A Circumplex Measure of Interests 60 almost one-half (29) of the final 64 items selected in Study 2 appear in the VP I , one may conclude from these structural findings that the "B ig Two" interest factors corresponding to the dimensions of Holland's hexagon form a circular continuum, in fact one of the best examples of such a continuum in the psychological literature. The clear and unambiguous two-dimensionality of the new scales-which are two more scales in number than Holland's original markers of these dimensions-strongly supports Holland's decision to "keep-it-simple", and model the psychological similarity among his six vocational types in two rather than a greater number of dimensions, and does not support the view that the hexagonal model is an unnecessarily gross simplification of the structure of interests (e.g., Lunneborg & Lunneborg, 1975). Adopting such a simple model is not to claim, of course, that two dimensions provide a complete picture of vocational interest variation. So much of the total variation in occupational interests, however, may be represented by factors corresponding to Holland's six types (e.g., Guilford, Sutton, Bond, & Christiansen, 1955; Jackson & McCar thy , 1986), which in turn may be represented by a two-dimensional circumplex, that the psychological significance of those two dimensions in determining career interest patterns would seem to dwarf whatever additional interest factors might be lying off the circle. Implications for Circumplex Approaches to Personality Structure and Measurement For domains of variables where such a model is appropriate, the principal components technique introduced by Phillips (1983) offers a simple but effective method for scaling a pool of items to a two-dimensional circumplex model. The structural success of the present effort supports, at least, the utility of the technique. The psychological significance of the result of such efforts has been questioned, however. In a critique of the circumplex both as a meaningful psychological structure, and as an approach to measurement in personality, Jackson & Helmes (1979) have argued that many if not all psychological circumplexes reported in the literature are artifacts of two A Circumplex Measure of Interests 61 general "response style" factors common to personality assessment by self-report, namely "social desirability" and "acquiescence". The close to perfect circumplex structure of the resulting scales in this study is, however, difficult to reconcile with this view. A n acquiescence interpretation of the general factor in vocational interest ratings might be plausible. However, that factor is usually represented as orthogonal to the circumplex components, and, was, in anycase, controlled in this study by ipsatization procedures. It is not clear to what extent the strongly bipolar relation between science and business occupations or between engineering and social welfare occupations could be attributable to either socially desirable responding or to acquiescence. A substantive interpretation, namely psychological differences in values (economic vs theoretical) or personality traits (nurturance) seems more plausible. The excellent circumplex structure reported here would seem to contradict the view that such structures appear only as a consequence of response artifacts. Research Applications of the Vocational Circumplex Scales The clear two-dimensional structure and precise geometric properties of the scales developed in this study make them ideal for a number of applications related to Holland's theory. One useful application would be in research seeking to determine the psychological meaning of the dimensions underlying Holland's hexagon. The new scales provide factorially unambiguous, and "dimensionally-representative" assessment of those dimensions (i.e., all-possible blends of the two dimensions). Furthermore, the mapping of the intermediate regions between the dimensions that a circumplex instrument provides permits an internal replication of relations between the circumplex dimensions and outside variables: external correlates of the circumplex dimensions should show a circular pattern of correlations across the set of octant scales. A s structurally precise markers of Holland's typological dimensions, the circumplex scales developed in this study are ideally suited to a second application-typological A Circumplex Measure of Interests 62 diagnosis. One of the great advantages of circumplex representations is their ability to represent relations within the space with reference to the geometric properties of a circle. (LaForge, Leary , Naboisek, Coffey, & Freedman, 1954; Wiggins, 1980). In the vocational literature this fact has not gone unnoticed. Exploitation of the geometric potentialities of the hexagonal model was first suggested by Cole, Whitney & Holland (1971), and Prediger (1976, 1981) has expanded on this idea by developing the A C T "World-of-Work-Map", and by advocating the use of theoretical weights based on the geometry of the hexagon to assign persons and occupational groups to locations within the hexagonal space. Because the use of geometric procedures in this manner, however, is justifiable only if the structure of the scales on which subjects (or groups) are scored reflects the geometric properties assumed by the weighting scheme (Wiggins, et. al., 1989), the precise geometric properties of the scales constructed in this study make them ideal for the kinds of diagnostic classification advocated by Prediger. The even-spacing of the scales implies each provides maximal ly discriminant information relating to the classification dimensions, and their high communality implies relative purity in the assessment those dimensions. In addition, the empirical octant locations of the scales are such as to permit confidence in the use of geometric weights to calculate dimension coordinates from the set of scale scores for the kinds of uses suggested by Prediger. Finally, the use of eight rather than six scale scores to estimate such coordinates has the advantage of greater reliability and content validity. One further area of application for the new circumplex scales is in the testing of predictions derived from Holland's theory that are based on the hexagonal model, such as those concerning differentiation, consistency, and congruence. O f course, the minor differences in fidelity of measurement between other measures of Holland's dimensions and this circumplex version would not be expected to alter the course of research history on these constructs. However, in certain kinds of studies the better discrimination on the hexagon dimensions provided by the octant method of assessment may possess certain advantages. With congruence defined as distance between projections on the Holland plane, A Circumplex Measure of Interests 6 3 for example, the superior geometric diagnostic capabilities of the new scales in comparison to other measures of Holland's types may permit more sensitive testing of differences between diagnostic groups since the classification of subjects is likely to be more reliable, and the groups more homogeneous. Although the new circumplex scales were found to pass the grade in terms of the structural criteria adopted here, a number of limitations of the present study should be noted. The first is the relatively small size and homogeneity of the samples used to derive and to validate the new measure. Sample sizes were large enough in this study to conduct component analysis of scale matrices, however, they must be considered of only marginally adequate size for the purpose of inspecting over 152 item correlations with these component scores, and basing final item selections on these correlations. Although, cross-sample stability of these correlations was never-the-less found to be high in the small samples employed here, replication of item coordinates on a much larger sample would be desirable before closing the door on final scoring of the scales. Use of larger samples would also permit analysis of sex differences which was not possible with only 258 subjects in the validation sample used here. Secondly, the good structure of the resulting scales in the validation sample permits generalizations about the scales' structure to be made only to the relatively homogenous population of college undergraduates of which both sample A and B of this study were drawn. Although the structure of Holland's Vocational Preference Inventory appears to be impressively robust across a multitude of population parameters, including age, educational level, and ethnic group, this fact does not, however, guarantee the generalizability of the structure of those 58% non-VPI items contained in the circumplex scales, whose structural properties have been demonstrated only for the undergraduate population sampled for this study. Some caution is in order, therefore, in evaluating the present findings until such time as these results are replicated on a larger and more heterogeneous population. A Circumplex Measure of Interests 6 4 Development of a set of circumplex scales for measuring Holland's typological dimensions may provide a useful research tool for continued mining of his enormously rich theory of careers, and may provide an efficient (brief) assessment instrument as a complement to other interest inventories in a clinical setting. It is hoped, in addition, that construction of these scales provides a useful demonstration of a novel approach to multivariate scale construction and an illustration, in the spirit implied in the title to Loevinger's famous 1957 monograph, of how explicitly theory-based approaches to scale development might enhance the utility of the resulting scales for advancing psychological theory. A Circumplex Measure of Interests 65 Bibliography Alden, L, Wiggins, J .S., & Pincus, A . L . (1989). Construction of circumplex scales for the  Inventory of Interpersonal Problems. Manuscript submitted for publication. Anastasi , A . (1986). Evolving concepts of test validation. Annua l Review of Psychology, 37, 1-15. Bentler, P.M. (1969). Semantic space is (approximately) bipolar. Journal of Personality, 71, 33-40. Bobele, R.M., Alston, H.L. , Wakefield, J .A . , Jr., & Doughtie, E.B. (1975). A comparison of Holland's model using constructs measured by two different methods. Journal of  Vocational Behavior, 9, 245-50. Brown, D. & Brooks, L. (1984). Career Choice & Development. N.Y.: Wiley. Buros, O.K. (Ed.) (1972). The seventh mental measurements yearbook. Highland Park, N.J.: Gryphon Press. Cantor, N., & Mischel, W. (1979). Prototypes in person perception. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology, Vol . 12. New York: Academic Press. Cattell, R.B. (1966). The scree test for the number of factors. Multivariate Behavioral  Research, 1, 140-161. Cavanaugh, J . A . & McCormick, C .C. , & (1981). Scaling Interpersonal Checklist items to a circumplex model. Applied Psuchological Measurement, 5, 421-447. Cole, N.S. & Hanson, G.R. (1971). A n analysis of the structure of vocational interests. Journal of Counseling Psychology, 18, 478-486 Cole, N.S. (1973). On measuring the vocational interests of women. Journal of Counseling  Psychology, 20, 105-112. Cole, N.S., & Cole, J .W .L . (1970). A n analysis of spatial configuration and its application  to research in higher education. ( A C T Research Report No. 35.). Iowa City IA: The American College Testing Program. Cole, N.S., Whitney, D.R., & Holland, J . L . (1971). A spatial configuration of occupations. Journal of Vocational Behavior, 1, 1-9. Comrey, A . (1978). Common methodological problems in factor analysis. Journal of  Consulting and Clinical Psychology, 46, 648-659. A Circumplex Measure of Interests 66 Comrey, A . & Montag, L. (1982). Comparison of factor analytic results with two-choice and seven-choice personality item formats. Applied Psychological Measurement, 6, 285-289. Cooley, W.W., & Lohnes, P.R. (1971). Multivariate Data Analysis. New York: Wiley. Costa, P.T., Jr., McCrae , R.R., & Hol land, J . L . (1984). Personality and vocational interests in adulthood. Journal of Applied Psychology, 69, 390-400. Cronbach, L.J. (1950). Further evidence on response sets and test design. Educational and  psychological measurement, 10, 3-31. Cronbach, L.J. & Meehl, P.E. (1955). Construct validity in psychological tests. Psychological Bulletin, 52, 281-302. Dawis, R.V. (1980). Measuring interests. In D.A. Payne (Ed.), New directions for testing  and measurement: Recent developments in affective measurement. San Francisco: Jossey-Bass. Dixon,P.N. , Bobo, M , Stevick, R.A. (1984). Response differences and preferences for all-category-defined and end-defined Likert formats. Educational and Psychological  Measurement, 44, 61-66. Eberhardt, B.J., & Muchinsky, P.M. (1984). Structural validation of Holland's hexagonal model: Vocational classification through the use of biodata. Journal of Applied  Psychology, 69,174-181. Edwards, K.J., Nafziger, D.H. , & Holland, J . L . (1974). Differentiation of occupational perceptions among different age groups. Journal of Vocational Behavior, 4, 311-318. Edwards, K.J., & Whitney, D.R. (1972). Structural analysis of Holland's personality types using factor and configural analysis. Journal of Counseling Psychology, 19, 136-145 Eysenck, H.J. & Eysenck, M. (1985). Personality and individual differences. New York: P lenum Press. F lamer , S. (1983). Assessment of the Multi-trait Multi-method Matr ix Val idity of Likert Scales V i a Confirmatory Factor Analysis. Multivariate Behavioral Research, 18, 275-308. Gat i , I. (1979). A hierarchical model for the structure of interests. Journal of Vocational Behavior, 21, 90-106. Gottfredson, G.D. , Holland, J . L . , & Ogawa, D.K. (1982). Dictionary of Holland  Occupational Codes. Palo Alto, C A : Consulting Psychologists Press. A Circumplex Measure of Interests 67 Guilford, J .P., Christensen P.R., Bond, N.A. & Sutton, M.A. (1954). A factor analysis study of human interests. Psychological Monographs, 68, (4, Whole No. 375). Guttman, L. (1954). A new approach to factor analysis: the radex. In P.R. Lazarsfeld (Ed.), Mathematical thinking in the social sciences. Glencoe, 111: Free Press. Guttman, L. (1966). Order analysis of correlation matrices. In R.B. Cattell (Ed.), Handbook of multivariate experimental psychology. Chicago: Rand McNa l l y & Company. Hansen, J . C . , & Campbell , D.P. (1985). Manua l for the SVIB-SCII (4th ed.). Stanford, C A : Stanford University Press. Hanson, G.R., Prediger, D.J., & Schussel, P. (1977). Development of the U N I A C T : A Unisex Edition of the A C T IV Interest Inventory. (ACT Research Report No, 67). Iowa City: American College Testing Program, 1977. Hogan, R. (1983). A socioanalytic theory of personality. In M . M . Page (Ed.), Nebraska  Symposium on Motivation: Personality: Current theory and research (pp 55-89). Lincoln: University of Nebraska Press. Hogan, R. (1986). Hogan Personality Inventory manual . Minneapolis: National Computer Systems, Inc. Hogan, R. & Hogan, J . (1988). Personality and status. In D.G. Gilbert and J . J . Conley (Eds.), Personality, social skills, and psychopathology: A n individual differences  approach. New York: Plenum. Holland, J . L . (1958). A personality inventory employing occupational titles. Journal of  Applied Psychology, 42, 336-342. Holland, J . L . (1966). The psychology of vocational choice. Waltham, Mass. : Blaisdell. Holland, J . L . (1973). Mak ing vocational choices: A theory of careers. Englewood Cliffs, N J : Prentice-Hall. Holland, J . L . (1978). Manua l for the Vocational Preference Inventory. Palo Alto, C A : Consulting Psychologists Press. Holland, J . L . (1979. Professional manual for the Self-Directed Search. Palo Alto, C A : Consulting Psychologists Press. Holland, J . L . (1985a). Making vocational choices: A theory of vocational personalities and  work environments (2nd ed.). Englewood Cliffs, N J : Prentice-Hall. Holland, J . L . (1985b). Professional manual for the Self-Directed Search. Odessa, F l : Psychological Assessment Resources. A Circumplex Measure of Interests 68 Holland, J . L . (1985c). Professional Manua l for the Vocational Preference Inventory. Odessa, F L : Psychological Assessment Resources. Hol land, J .L . , Whitney, D.R., Cole, N.S., & Richards, J . M . (1969). A n empirical occupational classification derived from a theory of personality and intended for  practice and research. ( A C T Research Rep. No. 29). Iowa City, IA: Amer ican College Testing Program. Jackson, D .N . & Helmes, E. (1979). Personality structure and the circumplex. Journal of  Personality and Social Psychology, 37, 2278-2285. Jackson D.N. , & Messick, S. (1958). Content and style in personality assessment. Psychological Bulletin, 55, 243-252. Jackson, D .N . (1970). A sequential system for personality development. In C D . Spielberger (ed.), Current topics in clinical & community psychology, Vol . 2, Academic Press, N.Y. Jackson, D .N . (1971). The dynamics of structured personality tests: 1971. Psychological  Review, 78, 229-248. Jackson, D .N . (1977). Jackson Vocational Interest Inventory manual. Port Huron, Mich: Research Psychologists Press. Jackson, D.N. , & McCar thy , J . M . (1986). A n overview of the Career Directions Inventory. N A T C O N : Twelfth National Consultation on Vocational Counseling. Employment & Immigration, Canada. Keisler, D. (1983). The 1982 Interpersonal Circle: A taxonomy for complementarity in human transacions. Psychological Review, 90, 185-214. K ing , A . L . , K ing, D.W., & Klockars, A . J . (1983). Dichotomous and multipoint scales using bipolar adjectives. Applied Psychological Measurement, 7, 173-180. Laforge, R., Leary , T . F . , Naboisek, H. , Coffey, H . S., & Freedman, M.B. (1954). The interpersonal dimension of personality: II A n objective study of repression. Journal  of Personality, 23, 129-153. Loevinger, J . (1957). Objective tests as instruments of psychological theory. Psychological  Reports, 3, 635-694. Loevinger, J . (1972). Personality Research Form. In O.K. Euros (Ed.), The seventh mental  measurements yearbook. Highland Park, N.J.: Gryphon Press. Lor r , M , & McNai r , D . M (1965). Expansion of the interpersonal behavior circle. Journal  of Personality & Social Psychology, 2, 823-830. A Circumplex Measure of Interests 69 Lunneborg, C .E . , & Lunneborg, P.W. (1975). Factor structure of the vocational interest models of Roe and Holland. Journal of Vocational Behavior, 7, 313-326. MacLennan, R.N. (1988). Correlation, base-rates, and the predictability of behaviour. Personality & Individual Differences, 9, 675-684. Meir , E.I. (1973). The structure of occupations by interests-A Smallest Space analysis. Journal of Vocational Behavior, 3, 21-31. Meir , E.I. & Barak, A . (1974). A simple instrument for measuring vocational interests based on Roe's classification of occupations. Journal of Vocational Behavior, 4, 33-42. Meir , E.I., & Ben-Yehuda, A . (1976). Inventories based on Roe and Holland yield similar results. Journal of Vocational Behavior, 8, 269-274. Meir , E.I., Sohlberg, S., & Barak, A . (1973). A cross-cultural comparison of the structure of interests. Journal of Cross-Cultural Psychology, 4, 501-508. Messick, S. (1980). Test validity and the ethics of assessment. American Psychologist, 35, 1012-1027. Mount, M.K., & Muchinsky, P.M. (1978). Person-environment congruence and employment job satisfaction: A test of Holland's theory. Journal of Vocational  Behaviour, 13, 84-100. Nunnal ly, J . C . (1967). Psychometric Theory. New York: McGraw-Hil l . Nunnal ly, J . C . (1978). Psychometric Theory. (2nd ed.). New York: McGraw-Hil l . Osipow, S.H. (1987). Apply ing person-environment theory to vocational behaviour. Journal of Vocational Behavior, 31, 333-336.0sipow, S .H. (1983). Theories of  career development (3rd ed.). Englewood Cliffs, N J : Prentice-Hall Osipow, S .H. (1967). Theories of career development (1st ed.). Englewood Cliffs, N J : Prentice-Hall Osipow, S .H. (1973). Theories of career development (2nd ed.). Englewood Cliffs, N J : Prentice-Hall Phillips, N. (1983). A circumplex generator/analyser program. Unpublished manuscript, University of British Columbia. Plutchik, R. (1980). Emotion: A Psychoevolutionary Synthesis. New York: Harper & Row. Prediger, D.J. (1976). A world of work map for career exploration. Vocational Guidance  Quarterly, 24, 198-208. A Circumplex Measure of Interests 70 Prediger, D.J., (1981). Mapping occupations and interests: A graphic aid for vocational guidance and research. Vocational Guidance Quarterly, 30, 21-36. Prediger, D.J., (1982). Dimensions underlying Holland's hexagon: Missing link between interests and occupations?. Journal of Vocational Behaviour, 21, 259-287. Roe, A . (1956). The psychology of occupations. New York: Wiley. Rosch, E. (1978). Principles of categorization. In E. Rosch and D.B. L loyd (Eds.), Cognition  and categorization. Hillsdale, N.J.: Er lbaum. Rounds, J .B. , Jr., Davison, M.L. , & Dawis, R.V. (1979). The fit between the Strong-Campbell Interest Inventory General Occupational Themes and Holland's hexagonal model. Journal of Vocational Behavior, 15, 303-315. Russell, J . A . (1980). A circumplex model of affect. Journal of Personality & Social  Psychology, 39, 1161-1178. Steiger, J . H . (1979). A computer program for fast, accurate, small-sample testing of correlational pattern hypotheses. Educational and Psychological Measurement, 39, 677-680. Stern, G .G . (1970). People in context: Measuring person-environment congruence in  education and industry. New York: Wiley. Thurstone, L .L . (1947). Multiple factor analysis. Chicago: University of Chicago Press. Tucker, L. (1955). The objective definition of simple structure in linear factor analysis. Psychometrika, 20, 209-211. Ty ler L. (1955). The development of "vocational interests": I. The organization of likes and dislikes in ten-year-old children. Journal of Genetic Psychology, 86, 33-44. Velicer, W.F . & Stephensen, J . F . (1978). The relation between item format and the structure of the Eysenck Personality Inventory. Applied Psychological  Measurement, 2, 293-304. Wakefield, J .A . , Jr., & Doughtie, E.B. (1973). The geometric relationship between Holland's personality typology and the Vocational Preference Inventory. Journal of  Counseling Psychology, 20, 513-518. Walsh, W.B., Hildebrand, J .O . , Ward, C M . , & Mathews, D.F. (1983). Holland's theory and non-college-degreed working black and white women. Journal of Vocational  Behavior, 22, 182-190 Watkins, Jr. C .E . , Bradford, B.D., Lew, D.E, & Himmel l , C D . (1986). Major contributors and major contributions to the vocational behaviour literature. Journal of  Vocational Behaviour, 28, 42-47. A Circumplex Measure of Interests 71 Weinrach, S. (1984). Holland's theory. In D. Brown & L. Brooks (Eds.) Career Choice and  Development. N.Y.: Wiley. Wiggins, J .S . & Broughton, R. (1985). The interpersonal circle: A structural model for the integration of personality research. In R. Hogan & W. Jones (Eds.), Perspectives in  Personality, Vol . 1., pp. 1-47. Greenwich, C N : JA I Press. Wiggins, J .S . (1972). Personality Research Form. In O.K. Buros (Ed.), The, seventh mental  measurements yearbook. Highland Park, N.J.: Gryphon Press. Wiggins, J .S . (1989). Personality Research Form. Buros, O.K. (Ed.), The tenth mental  measurements yearbook. Highland Park, N.J.: Gryphon Press. Wiggins, J .S . (1973). Personality & prediction: Principles of personality assessment. Reading, M A : Addison-Wesley. Wiggins, J .S . (1979). A psychological taxonomy of trait descriptive terms: The interpersonal domain. Journal of Personality & Social Psychology, 37, 395-412. Wiggins, J .S . (1980). Circumplex models of interpersonal behaviour. In L. Wheeler (Ed.), Review of Personality & Social Psychology, Vol 1. Beverly Hil ls: Sage Publications. Wiggins, J .S . (1982). Circumplex models of interpersonal behavior in clinical psychology. In P.S. Kendal l & J . N . Butcher (Eds.), Handbook of Research Methods in Clinical  Psychology (pp. 183-221). New York: Wiley Wiggins, J .S. , Steiger, J . , & Gaelick, L. (1981). Evaluat ing circumplexity in personality data. Multivariate Behavioral Research, 16, 263-289. Wiggins, J .S., Trapnel l , P., & Phillips, N. (1988). Psychometric and geometric properties of the revised Interpersonal Adjective Scales (IAS-R). Multivariate Behavioral  Research, 23, 517-530. Wilson, C.B. & Hilferty, M .M . (1931). The distribution of Chi-square. The Proceedings of  the National Academy of Sciences, 17, 684-688. Wrigley, C. & Neuhaus, J .O . (1955). The matching of two sets of factors. American  Psychologist, 10, 418-419. Zadeh, L .A. (1965). Fuzzy sets. Informational and Control, 8, 338-353. Zytowski, D .G & Borgan, F .H . (1984). Annua l review of research in vocational behaviour, 1983. Journal of Vocational Behaviour, 26. Zytowski, D. G. (1987). Comparison of Roe's and Holland's occupational classifications: Diverse ways of knowing. Journal of Counseling Psychology, 33, 479-481. A Circumplex Measure of Interests 72 Zytowski, D.G. , & Borgen, F .H . (1983). Assessment. In W.B. Walsh & S.H. Osipow (Eds.), Handbook of vocational psychology. Hillsdale, N.J: Er lbaum. Zwick, W.R., & Velicer, W.F . (1982). Factors influencing four rules for determining the number of components to retain. Multivariate Behavioral Research, 17, 253-271. Table 1 Definitions of Holland's Six Vocational Personality Types 1. Realistic preference for activities that entail the explicit, ordered, or systematic manipulation of objects, tools, machines, and animals aversion to educational and therapeutic activities traits: asocial, materialistic, self-effacing, conforming, natural,inflexible, frank, normal, thrifty, genuine, persistent, uninsightful, hard-headed, practical, uninvolved occupations: machinist, radio operator, auto mechanic, locomotive engineer, surveyor, electrician 2. Investigative preference for activities that entail the observational, symbolic, systematic, and creative investigation of physical, biological, and cultural phenomena in order to understand and control such phenomena aversion to persuasive, social, and repetitive activities traits: analytical, independent, rational, cautious, intellectual, reserved, critical, introspective, retiring, complex, pessimistic, unassuming, curious, precise, unpopular occupations scientific research worker, chemist, biologist, anthropologist, medical lab tech 3. Artistic preference for ambiguous, free, unsystematized activities that entail the manipulation of physical, verbal, or human materials to create art forms or products aversion to explicit, systematic, and ordered activites traits: complicated, imaginative, intuitive, disorderly, impractical. nonconforming, emotional, impulsive, original, expressive, independent, sensitive, idealistic, introspective, open occupations: poet, journalist, musician, composer, sculptor, portrait artist, concert singer 74 4. Social preference for activites that entail the manipulation of others to inform, train, develop, cure, or enlighten aversion to explicit, ordered, systematic activities involving materials, tools, or machines traits: ascendant, helpful, responsible, cooperative, idealistic, sociable, patient, empathic, tactful, friendly, kind, understanding, generous, persuasive, warm occupations: personal counselor, playground director, speech therapist, marriage counselor 5. Enterprising preference for activities that entail the manipulation of others to attain organizational goals or economic gain aversion to observational, symbolic, and systematic activities traits: acquisitive, adventurous, flirtatious, adventurous, exhibitionistic, optimistic, agreeable, excitement-seeking, self-confident, ambitious, extroverted, sociable, domineering, talkative occupations: sales manager, business executive, speculator, buyer, publicity director, hotel manager preference aversion traits: occupations: 6. Conventional for activities that entail the explicit, ordered, systematic manipulation of data, such as keeping records, filing materials, reproducing materials, organizing written and numerical data according to a prescribed plan, operating business machines and data processing machines to attain organizational or economic goals to ambiguous, free, exploratory, or unsystematized activities careful, inflexible, persistent, conforming, inhibited, practical, conscientious, methodical, prudish, defensive, obedient, thrifty, efficient, orderly, unimaginative bookkeeper, budget reviewer, bank examiner, payroll clerk, tax expert, financial analyst 75 Table 2 Representation of a C i r c u l a n t Matrix Var iab le where p i > p2 > p3 > p4 1 1 2 Pi 1 3 P2 Pi 1 4 P3 p2 Pi 1 5 P4 P3 p2 Pi 1 6 p3 P4 P3 P2 Pi 1 7 P2 P3 P4 P3 p2 p i 1 8 Pi p2 P3 P4 P3 p2 p i 76 Table 3 Intercorrelations Among the Six VPI Scales, Langara Community  College Sample, n=113. RE IN AR SO ES REALISTIC — INVESTIGATIVE 44 — ARTISTIC 02 43 — SOCIAL 18 23 39 — ENTERPRISING 16 12 14 29 — CONVENTIONAL 16 -02 -31 05 57 77 Table 4 Angular Location and Communality of VPI liens on Circumplex Components Defined by the Original VPI  Vocational Scales, Langara Community College Sample. n=113. Angle Commun. Angle Commun. 70.1 0.054 P i l l i n g Station Worker (SO) 326.9 0.092 139.8 0.207 Construction Inspector 327.2 0.080 158.6 0.275 Electrician 332.8 0.143 165.8 0.116 Power Shovel Operator 338.2 0.143 170.6 0.305 Auto Mechanic 340.7 0.129 180.8 0.311 Airplane Mechanic 346.4 0.196 182.0 0.191 Machinist 350.8 0.135 184.1 0.108 Radio Operator 353.6 0.213 187.6 0.241 Carpenter 358.5 0.040 188.5 0.144 Surveyor 359.4 0.118 192.3 0.240 Locomotive Engineer 2.9 0.190 196.1 0.040 Long Distance Bus Driver 5.4 0.189 231.4 0.140 Fish & Wildlife Specialist 7.2 0.238 235.9 0.109 Tree Surgeon 18.4 50.0 0.061 0.095 213.3 0.110 Physicist 219.1 0.205 Geologist (EN) 28.0 0.013 220.7 0.192 Chemist 42.2 0.120 220.7 0.210 Meteorologist 49.9 0.144 220.8 0.048 Writer of Scienti f i c Articles 53.9 0.007 228.6 0.116 Medical Laboratory Technician 57.9 0.053 229.7 0.185 Scientif i c Research Worker 67.6 0.249 233.3 0.252 Biologist 74.9 0.109 235.8 0.061 Editor of a Scientific Journal 75.1 0.175 236.6 0.180 Botanist 83.4 0.213 238.0 0.197 Zoologist 83.5 0.086 239.8 0.247 Independent Research Scientist 83.7 0.267 262.1 0.136 Astronomer 88.2 0.197 264.6 0.098 Anthropologist 90.1 90.9 0.241 0.437 283.8 0.124 Cartoonist 96.7 0.480 285.1 0.241 Musician 300.9 0.362 Sculptor/SculDtress (CO) 76.0 0.141 303.5 0.305 Poet 77.2 0.029 305.6 0.231 Symphony Conductor 92.6 0.164 306.2 0.332 Author 94.1 0.227 306.8 0.408 Novelist* 96.6 0.418 307.4 0.343 Free-lance Writer 98.1 0.259 308.8 0.302 Composer 104.6 0.418 309.7 0.295 Portrait Artist 105.2 0.392 312.2 0.151 Musical Arranger 105.8 0.324 312.9 0.232 Journalist 110.3 0.261 318.3 0.223 Playwright 111.6 0.391 338.7 0.103 Commercial Artist 116.9 0.336 346.1 0.156 Concert Singer 116.9 135.0 0.527 0.249 Cl i n i c a l Psychologist Social Science Teacher Speech Therapist Psychiatric Case Worker Social Worker* Playground Director Sociologist Youth Camp Director Juvenile Delinquency Expert Vocational Counselor  Marriage Counselor  Director of a Welfare Agency  Personal Counselor Highschool Teacher School Principal Television Producer Publicity Director Restaurant Manager Speculator Master of Ceremonies Salesperson Buyer Real Estate Salesperson  Advertising Executive Sports Promoter Department Store Manager  Hotel Manager Manufacturer's Representative  Sales Manager Business Executive Bank Teller Court Stenographer Payroll Clerk Business Teacher Bank Examiner Credit Investigator Financial Analyst Cost Estimator Budget Reviewer Tax Expert Bookkeeper Inventory Controller Certified Public Accountant IBM Equipment Operator M e : Labels are Realistic Type (RE), Investigative Type (IH), A r t i s t i c Type (AR), Social Type (SO), Enterprising Type (EH), Conventional Type (CO). Items are ordered by angular location, and grouped according to original keying of scales. * Items not originally scored but retained as marker items. Table 5 Coe f f i c i e n t Alpha and Mean Inter-item Correlation f o r F i n a l  JPRF Octant Scales, Sample B, n=258. alpha* r i : * (SO) SOCIAL .92 ( .90) .60 ( .53) (E-S) ENTERPRISING-SOCIAL .90 ( .84) .53 ( .40) (EN) ENTERPRISING .91 ( .85) .55 ( .43) (CO) CONVENTIONAL .91 ( .88) .55 ( .46) (RE) REALISTIC .89 ( .82) .50 ( .34) (I-R) INVESTIGATIVE-REALISTIC .92 ( .87) .57 ( .46) (I-A) INVESTIGATIVE-ARTISTIC .85 ( .77) .41 ( .29) (AR) ARTISTIC .94 ( .92) .65 ( .57) * Values i n brackets are for ipsatized data. 79 Table 6 Intercorrelations Among the F i n a l JPRF Octant Scales.  Sample B. n=258 SO ES EN CO RE IR IA SOCIAL — ENTERPRISING-SOCIAL 53 — ENTERPRISING -03 50 — CONVENTIONAL -37 -06 54 — REALISTIC -56 -48 -03 37 — INVESTIGATIVE-REALISTIC -54 -75 -51 -09 54 — INVESTIGATIVE-ARTISTIC -14 -49 -70 -51 -05 54 — ARTISTIC 14 -04 -45 -59 -44 -12 29 Table 7 Estimates of Goodness-of-Fit to a Circulant Hatrix for Intercorrelations Anong Final JPRF Octant  Scales and for 16 Comparison Matrices Sample N Sex Heasure X2 W-H Z HSR: Alden, Wiggins & Pincus (1989) ipsat. 974 HF IIP 165.5 9.25 507 Wiggins & Pincus (1989) 581 MF IASR 280.4 11.91 602 Trapnell (1989) OBC sample B ipsat. 258 HF JPRF 141.4 8.32 718 Hanson, Prediger & Schussel (1977) 1851 HF CNIACT 308.6 14.20 766 Wakefield & Doughtie (1973) 373 HF VPI 78.3 6.24 782 Holland, Whitney, Cole & Richards (1969) 1234 H VPI 297.7 13.95 866 Wigington (1981) 993 HF s e n 177.3 10.56 917 Holland (1985) 85 M SDS 30.5 2.47 1038 Campbell et. a l (1977) 200 H s e n 49.2 4.28 1066 Campbell & Holland (1972) 150 HF SVIB 25.5 1.92 1078 Trapnell (1989) OBC sample A ipsat. 257 HF VPI-m 75.9 6.11 1391 Heir (1973) 220 H RAHAK 76.9 4.90 1786 Trapnell (1989) Langara Community College 113 HF VPI 42.2 3.68 1845 Trapnell (1989) Langara Community College ipsat. 113 HF VPI 45.3 4.27 1870 Heir & Ben-Yehuda (1976) 217 HF RAHAK 227.6 11.54 2169 Holland (1985) 200 H SDS 90.2 6.92 2437 Lunneborg & Lunneborg (1975) ipsat. 235 HF VII 334.1 14.55 3495 Note: Heasure labels are Inventory of Interpersonal Problems (IIP), Revised Interpersonal Adjective Scales (IASR), Final JPRF Octant Scales (JPRF), American College Testing Program DHIACT-IV (UNIACT), Vocational Preference Inventory (VPI), Strong-Campbell Interest Inventory, General Occupational Theme Scales (SCII), Self-Directed Search (SDS), Strong Vocational Interest Blank, Holland Theme Scales (SVIB), VPI Harker Scales, selected in Study 1 (VPI-m), RAHAK Interest Inventory (RAHAK), Vocational Interest Inventory (VII). Ipsat. indicates those correlation matrices based upon ipsative or ipsatized data. 1 Each value has been multiplied by 105 for ease of comparison. Table 8 Anoular Locations and Communalities of Final 64 Vocational Circumplex Items. n=515* 81 (EM) ENTERPRISING 83.7 Sales Representative (.357) 85.3 Marketing & Promotions Manager (.437) 86.8 Marketing Representative (.512) 87.5 Department Store Manager (.380) 89.3 Sales Manager (.420) 97.8 Retail Industry Buyer (.317) 105.2 Chamber of Commerce Manager (.544) 106.0 Business Management Consultant (.490) (I-A) INVESTIGATIVE-ARTISTIC 248.4 Biologist (.384) 259.2 Ecologist (.451) 260.7 Science Reporter/Writer (.344) 268.0 Wildlife Researcher (.397) 273.7 Archeologist (.324) 279.4 Park Naturalist (.287) 288.7 Anthropologist (.337) 303.5 Social Sciences Researcher (.156) (CO) CONVENTIONAL 120.1 Bank Inspector (.472) 121.9 Financial Auditor (.564) 122.7 Credit Investigator (.408) 124.2 Accounting/Payroll Expert (.569) 127.1 Cost Control Analyst (.555) 129.7 Inventory Controller (.331) 154.3 Data Processing Expert (.277) 156.0 IBM Equipment Operator (.434) (AR) ARTISTIC 303.5 Sculptor (.338) 309.6 Classical Musician (.368) 311.8 Painter (Artist) (.362) 316.0 Novelist (.360) 319.7 Music Professor (.400) 320.3 Jazz Artist (.344) 320.4 Art Expert (.367) 321.5 Poet (.431) (RE) REALISTIC 152.0 Production Plant Manager (.266) 165.8 Data Systems Technician (.433) 170.6 Business Equipment Technician (.410) 174.0 Building Contractor (.217) 181.5 Power Station Operator (.327) 191.5 Electronics Technician (.464) 192.7 Industrial Machinist (.298) 196.5 Electrical Engineer (.485) . (SO) SOCIAL 340.3 Rehabilitation Counselor (.293) 353.8 Crisis Center Counselor (.412) 358.9 Children's Aid Director (.374) 0.0 Social Worker (.386) 0.6 Guidance Counselor (.447) 11.2 Director of Social Services (.501) 12.2 Neighbourhood Center Director (.323) 22.5 Social Coordinator (.535) (I-R) INVESTIGATIVE-REALISTIC 201.3 Industrial/Mechanical Engineer (.520) 202.3 Computer Science Researcher (.364) 222.1 Physicist (.380) 222.5 Geophysicist (.539) 227.8 Research Chemist (.486) 231.0 Scientific Research Worker (.499) 231.6 Fisheries Scientist (.356) 235.9 Geologist (.491) (E-S) ENTERPRISING SOCIAL 34.8 Social Events Organizer (.413) 37.9 Travel Agency Director (.397) 38.7 Tourism Director (.334) 44.6 Publicity Director (.393) 45.4 Personnel Director (.379) 47.2 Public Relations Manager (. 525) 48.3 Public Agency Spokesperson (.309) 63.6 Hotel Manager (.391) Combined derivation (n=257) and validation (n=258) samples. 82 CONVENTIONAL / / / / / / REALISTIC / / / / < \ \ \ \ \ \ \ \ \ \ INVESTIGATIVE ENTERPRISING \ \ \ \ \ \ \ \ \ \ > SOCIAL / / / / / / / / / ARTISTIC Figure 1. Holland's Hexagonal Model Describing the Degree of Psychological S i m i l a r i t y Among Six Vocational Personality Types. 83 CO. EN .8-1 •4H RE -.4 .8 SO • 4 H IN AR .8-1 Figure 2. Circumplex Structure of Original VPI Scales, Langara Community College Sample, n=113. 84 CO EN .4H •.4 RE •4H IN -.8-so AR Figure 3. Circumplex Structure of VPI Marker Scales, UBC Sample A, n=257. EN CO ES -.4 IR - . 4 H -.8-1 SO AR IA Figure 4. Circumplex Structure of JPRF Octant Scales, UBC Sample A (derivation sample), n=257. EN CO .8-1 •4H ES iS -.8 -.4 IR •4H .8-J IA AR Figure 5. Circumplex Structure of JPRF Octant Scales, UBC Sample B (validation sample), n=258. 87 3.192 + E I G E N V A L U E S 2.696 + 698 + ,413 + 230 + 000 +-1 2 3 4 5 -+-6 7 -* 8 FACTOR EIGENVALUE % TOTAL VAR CUMULATIVE % 1 3.19175 39.9 39.9 2 2.69630 33.7 73.6 3 .69763 8.7 82.3 4 .47689 6.0 88.3 5 .41289 5.2 93.5 6 .22964 2.9 96.4 7 .19684 2.5 98.9 8 .09804 1.2 99.9 Figure 6. Eigenvalue (Scree) Plot and Percent of Total Variance Accounted f o r by Components of F i n a l JPRF Scale Intercorrelations, Sample B, n=258. 88 Appendix A 89 THE VOCATIONAL PREFERENCE INVENTORY This is an inventory about your feelings and attitudes about many kinds of work. 1. Show the occupations which interest or appeal to you by blackening Y for "Yes". 2. Show the occupations which you dislike or find uninteresting by blackening N for "No". 3. Make no marks when you are undecided about an occupation. (Y) (N) 1. Criminologist (Y) (N) 2. Private Investigator (Y) (N) 3. Restaurant Worker (Y) (N) 4. Detective (Y) (N) 5. Photoengraver (Y) (N) 6. Truck Gardener (Y) (N) 7. Physical Education Teacher (Y) (N) 8. Humorist (Y) (N) 9. Photographer (Y) (N) 10. Diplomat (Y) (N) 11. Airplane Mechanic (Y) (N) 12. Meteorologist (Y) (N) 13. Sociologist (Y) (N) 14. Bookkeeper (Y) (N) 15. Speculator (Y) (N) 16. Poet (Y) (N) 17. Deep Sea Diver (Y) (N) 18. Stock Clerk (Y) (N) 19. Dramatic Coach (Y) (N) 20. Lawyer (Y) (N) 21. Fish and Wildlife Specialist (Y) (N) 22. Biologist (Y) (N) 23. High School Teacher (Y) (N) 24. Business Teacher (Y)(N) 25. Buyer (Y) (N) 26. Symphony Conductor (Y) (N) 27. Wrecker (Building) (Y) (N) 28. Veterinarian (Y) (N) 29. Elementary School Teacher (Y) (N) 30. Physician (Y)(N) 31. Auto Mechanic (Y) (N) 32. Astronomer (Y) (N) 33. Juvenile Delinquency Expert (Y) (N) 34. Budget Reviewer (Y) (N) 35. Advertising Executive (Y)(N) 36. Musician (Y) (N) 37. Prizefighter (Y) (N) 38. Post Office Clerk (Y) (N) 39. Experimental Laboratory Engineer (Y) (N) 40. Bartender (Y) (N) 41. Carpenter (Y) (N) 42. Medical Laboratory Technician (Y) (N) 43. Speech Therapist (Y) (N) 44. Certified Public Accountant (Y) (N) 45. Manufacturer's Representative (Y) (N) 46. Author (Y)(N) 47. Firefighter (Y) (N) 48. Airline Ticket Agent (Y) (N) 49. Entertainer (Y) (N) 50. Novelist (Y) (N) 51. Power Shovel Operator (Y) (N) 52. Anthropologist (Y) (N) 53. Marriage Counselor (Y) (N) 54. Credit Investigator (Y) (N) 55. Television Producer (Y) (N) 56. Commercial Artist (Y) (N) 57. Wild Animal Trainer (Y) (N) 58. Administrative Assistent (Y) (N) 59. Physical Therapist (Y) (N) 60. Cashier (Y) (N) 61. Surveyor (Y) (N) 62. Zoologist (Y) (N) 63. School Principal (Y) (N) 64. Court Stenographer (Y) (N) 65. Hotel Manager (Y) (N) 66. Free-Lance Writer (Y) (N) 67. Stunt Man/Stunt Woman (Movies) (Y) (N) 68. Route Salesperson (Y) (N) 69. Professional Athlete (Y) (N) 70. Flight Attendant (Y) (N) 71. Construction Inspector (Y) (N) 72. Chemist (Y) (N) 73. Playground Director (Y) (N) 74. Bank Teller (Y) (N) 75. Business Executive (Y) (N) 76. Musical Arranger (Y) (N) 77. Jockey (Y) (N) 78. Interior Decorator (Y) (N) 79. Airplane Pilot (Y) (N) 80. Banker (Y) (N) 81. Radio Operator (Y) (N) 82. Independent Research Scientist 90 (Y) (N) 83. Clinical Psychologist (Y) (N) 84. Tax Expert (Y) (N) 85. Restaurant Manager (Y) (N) 86. Journalist (Y) (N) 87. Motorcycle Driver (Y) (N) 88. Department Store Manager (Y) (N) 89. Referee (Sporting Events) (Y) (N) 90. Mail Carrier (Y) (N) 91. Filling station Worker (Y) (N) 92. Writer of Scientific Articles (Y) (N) 93. Social Science Teacher (Y) (N) 94. Inventory Controller (Y) (N) 95. Master of Ceremonies (Y) (N) 96. Portrait Artist (Y) (N) 97. Blaster (Dynamiter) (Y) (N) 98. Police Officer (Y) (N) 99. English Teacher (Y) (N) 100. U.N. Official (Y) (N) 101. Tree Surgeon (Y) (N) 102. Editor of a Scientific Journal (Y) (N) 103. Director of a Welfare Agency (Y) (N) 104. IBM Equipment Operator (Y) (N) 105. Salesperson (Y) (N) 106. Concert Singer (Y) (N) 107. F.B.I. Agent (Y) (N) 108. Probation Agent (Y) (N) 109. Astronaut (Y)(N) 110. College Professor (Y) (N) 111. Long Distance Bus Driver (Y) (N) 112. Geologist (Y) (N) 113. Youth Camp Director (Y) (N) 114. Financial Analyst (Y) (N) 115. Real Estate Salesperson (Y) (N) 116. Composer (Y)(N) 117. Mountain Climber (Y)(N) 118. Cook/Chef (Y) (N) 119. Stage Director (Y) (N) 120. Ticket Agent (Y) (N) 121. Locomotive Engineer (Y) (N) 122. Botanist (Y) (N) 123. Personal Counselor (Y) (N) 124. Cost Estimator (Y) (N) 125. Publicity Director (Y) (N) 126. Sculptor/Sculptress (Y) (N) 127. Explorer (Y) (N) 128. Nursery School Teacher (Y) (N) 129. Quality Control Expert (Y) (N) 130. Judge (Y) (N) 131. Machinist (Y) (N) 132. Scientific Research Worker (Y) (N) 133. Psychiatric Case Worker (Y)(N) 134. Payroll Clerk (Y) (N) 135. Sports Promoter (Y) (N) 136. Playwright (Y) (N) 137. Test Pilot (Y) (N) 138. Computer Programmer (Y) (N) 139. Clothing Designer (Y) (N) 140. Truck Driver (Y) (N) 141. Electrician (Y) (N) 142. Physicist (Y) (N) 143. Vocational Counselor (Y) (N) 144. Bank Examiner (Y) (N) 145. Sales Manager (Y) (N) 146. Cartoonist (Y) (N) 147. Racing Car Driver (Y) (N) 148. Forester (Y) (N) 149. Social Worker (Y) (N) 150. Sales Clerk 91 (Y) (N) 151. Funeral Director (Y) (N) 152. Mind Reader (Y) (N) 153. Architect (Y) (N) 154. Shipping & Receiver Clerk (Y) (N) 155. Criminal Psychologist (Y) (N) 156. Insurance Clerk (Y) (N) 157. Barber (Y) (N) 158. Bill Collector (Y) (N) 159. Ward Attendant (Y) (N) 160. Masseur/Masseuse 92 Appendix B I .0. # Male Female Age JOB INTERESTS RATING FORM I LISTED ON THE NEXT FOUR PAGES ARE MANY DIFFERENT OCCUPATIONS. FOR EACH ONE SHOW HOW YOU WOULD FEEL ABOUT DOING THAT KIND OF WORK. NEXT TO EACH OCCUPATION CIRCLE A NUMBER FROM 1 TO 8 THAT BEST DESCRIBES YOUR FEELINGS FOR THAT KIND OF WORK. USE THE SCALE AS SHOWN BELOW. DON'T WORRY ABOUT WHETHER YOU WOULD BE GOOD AT THE JOB, OR ABOUT NOT BEING TRAINED FOR IT. FORGET ABOUT HOW MUCH MONEY YOU WOULD MAKE OR WHETHER YOU COULD GET AHEAD. THINK ONLY ABOUT HOW MUCH YOU LIKE (OR DISLIKE) THE KIND OF WORK DONE IN THAT JOB. WORK QUICKLY. PLEASE ANSWER EVERY ONE. USE THE FOLLOWING SCALE: STRONGLY DISLIKE 1 2 3 4 3 6 7 8 STRONGLY LIKE FOR EXAMPLE. CONSIDER THE OCCUPATION "COMPUTER TECHNICIAN". IF YOU MODERATELY LIKED THAT KINO OF WORK. YOU WOULD CIRCLE A " 6 " NEXT TO IT: 2 3 4 5 @ 7 8 Computer T e c h n i c i a n I F , ON THE OTHER HAND. YOU MODERATELY DISLIKED THAT KIND OF WORK. YOU WOULD CIRCLE * 3 " NEXT TO IT. . . . e t c . 1 2 (5) 4 5 6 7 8 Computer T e c h n i c i a n STRONGLY 1 DISLIKE 1 2 3 4 5 6 7 8 Data P r o c e s s i n g Expert 1 2 3 4 5 6 7 8 Rec r e a t i o n D i r e c t o r 1 2 3 4 5 6 7 8 P a i n t e r ( A r t i s t ) 1 2 3 4 5 6 7 8 P u b l i c i t y D i r e c t o r 1 2 3 4 5 6 7 8 Business Management Consultant 1 2 3 4 5 6 7 8 Sales Manager 1 2 3 4 5 6 7 8 Astronomer 1 2 3 4 5 6 7 8 Travel Coordinator 1 2 3 4 5 6 7 8 C o n s t r u c t i o n Inspector 1 2 3 4 5 6 7 8 A r c h e o l o g i s t 1 2 3 4 5 6 7 8 Cost Estimator 1 2 3 4 5 6 7 8 S o c i a l Worker 1 2 3 4 5 6 7 8 Machine-Too) Oestgner/Bultder 1 2 3 4 5 6 7 8 Automotive Engineer/Technician 1 2 3 4 5 6 7 8 School P r i n c i p a l 1 2 3 4 5 6 7 8 Investment Promoter 1 2 3 4 5 6 7 8 Commodity Trader 1 2 3 4 5 6 7 8 E l e c t r i c a l Engineer 1 2 3 4 5 6 7 8 Volunteer Agency D i r e c t o r 8 STRONGLY LIKE 1 2 3 4 5 6 7 8 R e t a i l Industry Buyer 1 2 3 4 5 6 7 8 Community S e r v i c e s D i r e c t o r 1 2 3 4 5 6 7 8 Marketing Representative 1 2 3 4 8 6 7 8 Debts C o l l e c t i o n Manager 1 2 3 4 5 6 7 8 A i r p l a n e Mechanic 1 2 3 4 5 6 7 8 Youth Counselor 1 2 3 4 5 6 7 8 Research Chemist 1 2 3 4 5 6 7 8 P o l i t i c a l Campaign Organizer 1 2 3 4 5 6 7 8 Independent Research S c i e n t i s t 1 2 3 4 5 6 7 8 H i s t o r y P r o f e s s o r 1 2 3 4 5 6 7 8 B a c t e r i o l o g i s t 1 2 3 4 5 6 7 8 F i n a n c i a l Analyst 1 2 3 4 5 6 7 8 C r e a t i v e w r i t e r 1 2 3 4 5 6 7 8 T r a v e l l i n g Sales R e p r e s e n t a t i v e 1 2 3 4 5 6 7 8 D i r e c t o r of S o c i a l S e r v i c e s 1 2 3 4 5 6 7 8 F i n a n c i a l A u d i t o r 1 2 3 4 5 6 7 8 Composer 1 2 3 4 5 6 7 8 P u b l i c R e l a t i o n s Manager 1 2 3 4 5 6 7 8 Locomotive Engineer STRONGLY 1 2 3 DISLIKE 1 2 3 4 5 6 7 8 S c i e n t i f i c Research Worker 1 2 3 4 5 6 7 8 P r o f e s s i o n a l Musician 1 2 3 4 5 6 7 8 Hotel Manager 1 2 3 4 5 6 7 8 Guidance Counselor 1 2 3 4 5 6 7 8 R e h a b i l i t a t i o n Counselor 1 2 3 4 5 6 7 8 Science F i c t i o n W r i t e r 1 2 3 4 5 6 7 8 L i t e r a r y Expert 1 2 3 4 5 6 7 8 P o r t r a i t A r t i s t 1 2 3 4 5 6 7 8 Gym/Fitness Center Manager 1 2 3 4 5 6 7 8 B i o l o g i s t 1 2 3 4 5 6 7 8 Biochemist 1 2 3 4 5 6 7 8 W i l d l i f e Researcher 1 2 3 4 5 6 7 8 Personnel R e c r u i t e r / I n t e r v i e w e r 1 2 3 4 5 6 7 8 Taxation Expert 1 2 3 4 5 6 7 8 Medical Research S c i e n t i s t 1 2 3 4 5 6 7 8 Scu l p t o r 1 2 3 4 5 6 7 8 Chamber of Commerce Manager 1 2 3 4 5 6 7 8 C h i l d r e n ' s A i d D i r e c t o r 1 2 3 4 5 6 7 8 Department Store Manager 8 STRONGLY LIKE 1 2 3 4 5 6 7 8 Tourism D i r e c t o r 1 2 3 4 5 6 7 8 E n t e r t a i n e r ' s Promoter/Agent 1 2 3 4 5 6 7 8 Science Magazine E d i t o r 1 2 3 4 5 6 7 8 A d v e r t i s i n g E x e c u t i v e 1 2 3 4 5 6 7 8 Documentary Film-matter 1 2 3 4 5 6 7 8 C e r t i f i e d P u b l i c Accountant 1 2 3 4 5 6 7 8 Cost C o n t r o l A n a l y s t 1 2 3 4 5 6 7 8 H o s p i t a l i t y Consultant 1 2 3 4 5 6 7 8 M i l i t a r y Radar Operator 1 2 3 4 5 6 7 8 Genetics Researcher 1 2 3 4 5 6 7 8 N o v e l i s t 1 2 3 4 5 6 7 8 Jazz A r t i s t 1 2 3 4 5 6 7 8 S o c i a l Events Organizer 1 2 3 4 5 6 7 8 E l e c t r o n i c s T e c h n i c i a n 1 2 3 4 5 6 7 8 E c o l o g l s t 1 2 3 4 5 6 7 8 Personnel D i r e c t o r 1 2 3 4 5 6 7 8 S o c i a l C o o r d i n a t o r 1 2 3 4 5 6 7 8 Business Conventions C o o r d i n a t o r 1 2 3 4 5 6 7 8 I n t e r n a t i o n a l T rading Consultant VO STRONGLY t 2 DISLIKE 1 2 3 4 S 6 7 8 Geophystcist 1 2 3 4 3 6 7 8 Budget Planner 1 2 3 4 5 6 7 8 I n d u s t r i a l Machinist 1 2 3 4 5 6 7 8 Science R e p o r t e r / w r i t e r 1 2 3 4 5 6 7 8 Marine S c i e n t i s t 1 2 3 4 5 6 7 8 C r i s i s Center Counselor 1 2 3 4 5 6 7 8 Business Equipment Technician 1 2 3 4 5 6 7 8 A r t s Reporter 1 2 3 4 5 6 7 8 Travel Agency D i r e c t o r 1 2 3 4 5 6 7 8 Astronomy Research A s s i s t a n t 1 2 3 4 5 6 7 8 Sales Representative 1 2 3 4 5 6 7 8 B u i l d i n g Contractor 1 2 3 4 5 6 7 8 Marriage Counselor 1 2 3 4 5 6 7 8 S c i e n t i f i c Journal E d i t o r 1 2 3 4 5 6 7 8 Real E s t a t e Salesperson 1 2 3 4 5 6 7 8 C u l t u r a l H i s t o r i a n 1 2 3 4 5 6 7 8 P h y s i o l o g i s t 1 2 3 4 5 6 7 8 Dramatist/Playwright 1 2 3 4 5 6 7 8 Music P r o f e s s o r 8 STRONGLY LIKE 1 2 3 4 5 6 7 8 B o t a n i s t 1 2 3 4 5 6 7 8 Business Teacher 1 2 3 4 5 6 7 8 Pub I Ic Agency Spokesperson 1 2 3 4 5 6 7 8 Inventory C o n t r o l l e r 1 2 3 4 5 6 7 8 Heal t h Sciences Researcher 1 2 3 4 5 6 7 8 Entertainment D i r e c t o r 1 2 3 4 5 6 7 8 Computer Science Researcher 1 2 3 4 5 6 7 8 Park N a t u r a l i s t 1 2 3 4 5 6 7 8 A c c o u n t i n g / P a y r o l l Expert 1 2 3 4 5 6 7 8 Oceanographer 1 2 3 4 5 6 7 8 A n t h r o p o l o g i s t 1 2 3 4 5 6 7 8 Experimental P s y c h o l o g i s t 1 2 3 4 5 6 7 8 High School Science Teacher 1 2 3 4 5 6 7 8 Data F i l e s Manager 1 2 3 4 5 6 7 8 Manufacturer's R e p r e s e n t a t i v e 1 2 3 4 5 6 7 8 Stocks & Bonds S e l l e r 1 2 3 4 5 6 7 8 A d v e r t i s i n g Salesperson 1 2 3 4 5 6 7 8 S o c i o l o g i s t 1 2 3 4 5 6 7 8 P r o d u c t i o n Plant Manager CT. STRONGLY 1 OISLIKE 1 2 3 4 5 6 7 8 Philosophy wr1ter/teacher 1 2 3 4 5 6 7 8 J o u r n a l i s t 1 2 3 4 S 6 7 8 C l a s s i c a l Musician 1 2 3 4 5 6 7 8 Aquarium Research A s s i s t a n t 1 2 3 4 5 6 7 8 G e o l o g i s t 1 2 3 4 5 6 7 8 Welfare Agency D i r e c t o r 1 2 3 4 5 6 7 8 Labour R e l a t i o n s Counselor 1 2 3 4 5 6 7 8 Neurosclences Researcher 1 2 3 4 5 6 7 8 Information Bureau D i r e c t o r 1 2 3 4 5 6 7 8 C r e d i t I n v e s t i g a t o r 1 2 3 4 5 6 7 8 Employment Counselor 1 2 3 4 5 6 7 8 Neighbourhood Center D i r e c t o r 1 2 3 4 5 6 7 8 Poet 1 2 3 4 5 6 7 8 Bank Inspector 1 2 3 4 5 6 7 8 Z o o l o g i s t 1 2 3 4 5 6 7 8 A i r p o r t S e c u r i t y D e t e c t i v e 1 2 3 4 5 6 7 8 Charity/Agency Fundraiser 1 2 3 4 5 6 7 8 Business Developer/Entrepreneur 1 2 3 4 5 6 7 8 P s y c h i a t r i c Researcher STRONGLY LIKE 1 2 3 4 5 6 7 8 Business Data Manager 1 2 3 4 5 6 7 8 Land Surveyor 1 2 3 4 5 6 7 8 S o c i a l Sciences Researcher 1 2 3 4 5 6 7 8 P r o f e s s i o n a l Bookkeeper 1 2 3 4 5 6 7 8 F i s h e r i e s S c i e n t i s t 1 2 3 4 5 6 7 8 Power S t a t i o n Operator 1 2 3 4 5 6 7 8 O f f i c e Systems Manager 1 2 3 4 5 6 7 8 Art Expert 1 2 3 4 5 6 7 8 Marketing & Promotions Manager 1 2 3 4 5 6 7 8 IBM Equipment Operator 1 2 3 4 5 6 7 8 C h i l d Abuse Counselor 1 2 3 4 5 6 7 8 P h y s i c i s t 1 2 3 4 5 6 7 8 Data Systems T e c h n i c i a n 1 2 3 4 5 6 7 8 M e t e o r o l o g i s t 1 2 3 4 5 6 7 8 E l e c t r i c i a n 1 2 3 4 5 6 7 8 I n d u s t r i a l / M e c h a n i c a l Engineer 1 2 3 4 5 6 7 8 C r u i s e Ship Steward/Hostess 1 2 3 4 5 6 7 8 F i n a n c i a l Records O f f i c e r 1 2 3 4 5 6 7 8 E l e c t r i c a l C o n t r a c t o r VO Appendix C Angular Locations and Communalities for A l l 152 JPRF Items With Respect to the VPI-m Circumplex  Components, Sample A, and the Final Octant Scales' Circumplex Components. Sample B VPI-m Final Octant Scales Angle Comm* Angle Comm* 322.9 252 344.2 265 43 336.9 373 0.0 393 82 339.5 304 . 2.1 387 25 357.4 157 2.7 225 19 342.9 389 4.9 370 12 341.8 354 5.8 320 56 345.1 383 7.2 292 89 335.7 371 9.0 374 144 339.2 443 9.6 439 42 358.4 139 12.1 165 120 355.6 080 13.8 037 15 357.8 024 15.6 043 47 357.1 391 16.5 505 34 355.9 241 16.5 342 126 355.5 054 16.8 125 2 7.8 197 18.6 301 101 12.5 163 19.2 129 131 25.2 255 21.4 209 65 10.5 358 24.1 563 74 9.7 286 28.2 281 125 27.2 162 29.3 155 150 12.9 271 30.8 338 21 26.1 243 31.8 446 70 36.6 269 32.7 389 85 25.4 140 34.9 206 8 34.8 230 37.6 316 58 44.8 281 38.5 435 4 33.0 144 40.5 162 121 29.0 201 41.9 278 59 38.6 359 43.4 364 73 29.9 159 44.7 169 51 38.5 380 47.0 554 37 38.5 244 48.5 295 98 65.7 153 58.5 213 123 73.2 191 61.8 128 33 55.5 385 64.1 355 41 35.9 083 66.4 141 27 66.5 334 73.3 290 112 70.1 332 75.8 379 61 74.6 430 80.4 421 142 79.6 347 82.1 233 91 74.9 421 84.4 354 87 80.8 340 85.2 377 75 78.5 415 87.6 360 57 79.7 436 88.3 439 6 74.9 480 88.7 549 89.2 365 97.4 238 85.4 240 101.6 288 93.5 457 104.4 588 94.0 486 105.4 472 92.5 251 106.9 306 103.5 316 109.9 358 94.9 284 110.5 323 110.0 302 112.1 302 105.1 417 113.0 470 117.7 235 113.2 312 113.0 397 114.5 374 110.2 451 114.9 469 106.0 420 115.1 421 109.0 365 115.6 405 108.3 370 118.4 499 111.0 292 119.6 438 109.6 503 120.1 583 108.9 481 120.1 520 107.7 465 120.2 631 111.4 455 121.9 520 108.8 423 122.3 585 114.4 354 125.3 289 110.6 572 125.5 547 111.5 483 127.7 623 117.3 374 129.5 286 119.0 303 132.4 358 147.9 110 133.5 041 118.2 382 136.2 291 144.5 195 140.7 268 142.4 318 157.3 472 131.3 165 160.5 327 149.5 308 170.0 473 161.8 344 170.1 396 160.7 260 175.3 209 175.3 199 183.1 195 170.0 290 185.1 332 171.4 384 186.3 383 184.8 393 193.4 269 181.6 376 193.9 492 200.2 209 197.6 424 179.4 221 199.5 124 185.7 427 200.1 481 192.3 391 200.2 356 188.9 449 200.3 533 191.4 475 201.9 506 176.5 297 202.5 245 187.6 364 204.3 346 195.0 442 208.6 314 191.9 235 209.3 290 216.9 356 222.4 367 209.3 444 224.1 586 216.9 480 228.5 537 222.6 358 228.9 341 224.0 432 230.2 584 216.0 358 231.5 376 226.2 372 232.9 531 22 Marketing Representative 20 Retail Industry Buyer 97 Business Teacher . 55 Chamber of Commerce Manager 5 Business Management Consultant 76 International Trading Consultant 16 Investment Promoter 110 Manufacturer's Representative 132 Business Developer/Entrepreneur 151 Financial Records Officer 23 Debts Collection Manager 124 Credit Investigator 128 Bank Inspector 78 Budget Planner 111 Stocks & Bonds Seller 140 Office Systems Manager 17 Commodity Trader 35 Financial Auditor 63 Certified Public Accountant 134 Business Data Manager 31 Financial Analyst 52 Taxation Expert 137 Professional Bookkeeper 104 Accounting/Payroll Expert 64 Cost Control Analyst 99 Inventory Controller 109 Data Files Manager 130 Airport Security Detective 11 Cost Estimator 114 Production Plant Manager 143 IBM Equipment Operator 1 Data Processing Expert 146 Data Systems Technician 83 Business Equipment Technician 88 Building Contractor 66 Military Radar Operator 139 Power Station Operator 152 Electrical Contractor 79 Industrial Machinist 71 Electronics Technician 102 Computer Science Researcher 9 Construction Inspector 18 Electrical Engineer 14 Automotive Engineer/Technician 148 Electrician 149 Industrial/Mechanical Engineer 135 Land Surveyor 13 Machine-Tool Designer/Builder 24 Airplane Mechanic 38 Locomotive Engineer 145 Physicist 77 Geophysicist 26 Research Chemist 53 Medical Research Scientist 39 Scientific Research Worker 138 Fisheries Scientist 49 Biochemist 100 227.0 410 234.0 464 28 Independent Research Scientist 212.0 324 234.5 353 147 Meteorologist 227.3 349 236.1 387 30 Bacteriologist 228.1 315 236.2 262 86 Astronomy Research Assistant 222.9 364 236.7 393 122 Keurosciences Researcher 223.5 499 237.5 474 119 Geologist 240.1 344 238.8 458 48 Biologist 233.2 273 240.1 340 67 Genetics Researcher 268.5 107 241.9 220 90 Scientific Journal Editor 241.1 126 243.0 196 60 Science Magazine Editor 252.9 243 246.3 261 100 Health Sciences Researcher 234.4 270 249.1 176 7 Astronomer 232.4 474 250.7 451 81 Marine Scientist 237.1 217 251.2 236 93 Physiologist 237.1 335 254.1 383 96 Botanist 252.4 311 254.6 367 118 Aquarium Research Assistant 247.2 349 255.1 442 72 Ecologist 237.7 373 256.4 288 105 Oceanographer 243.6 332 256.8 413 129 Zoologist 255.5 279 257.6 332 80 Science Reporter/Writer 254.0 330 261.4 384 50 Wildlife Researcher 239.2 082 265.3 099 108 High School Science Teacher 258.5 286 275.2 290 10 Archeologist 283.1 177 276.4 174 44 Science Fiction Writer 260.8 175 278.3 308 103 Park Naturalist 267.7 263 293.0 361 106 Anthropologist 303.1 198 294.9 077 133 Psychiatric Researcher 298.4 328 301.3 366 54 Sculptor 297.0 172 303.8 120 136 Social Sciences Researcher 306.0 344 308.3 425 117 Classical Musician 312.6 387 310.3 305 46 Portrait Artist 303.7 111 310.3 096 107 Experimental Psychologist 305.6 291 311.9 323 36 Composer 313.1 090 312.6 147 29 History Professor 283.4 175 313.5 210 62 Documentary Film-maker 320.9 336 314.7 287 45 Literary Expert 312.7 438 314.9 358 68 Novelist 302.6 352 315.2 402 3 Painter (Artist) 312.5 190 315.7 272 92 Cultural Historian 319.0 503 317.5 408 127 Poet 295.6 278 318.2 314 40 Professional Musician 311.6 298 319.6 296 115 Philosophy writer/teacher 314.8 369 320.2 306 32 Creative Writer 313.8 389 320.8 365 141 Art Expert 308.3 411 321.8 407 95 Music Professor 306.4 328 321.9 349 69 Jazz Artist 323.4 452 329.1 422 94 Dramatist/Playwright 328.4 349 334.8 240 113 Sociologist 328.6 399 335.8 388 84 Arts Reporter 327.0 299 341.8 178 116 Journalist *Decimal points have been omitted. 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items