multidimensional scaling research vocational psychology · 2016. 5. 18. · 491 multidimensional...

491

Multidimensional Scaling Researchin Vocational PsychologyJames B. Rounds, Jr.State University of New York at Buffalo

Michael A. ZevonRoswell Park Memorial institute

This review summarizes and evaluates the use ofmultidimensional scaling in vocational psychology.Multidimensional scaling applications are found in twoprimary areas: vocational interests and occupationalperceptions. These areas correspond to the two majoruses of multidimensional scaling: configural verifica-tion and dimensional identification. Two issues—the

relationship between multidimensional scaling and al-ternative data analytic methods, and the selection ofoccupational stimuli—are discussed. A number of de-veloping areas for the application of multidimensionalscaling are identified.

Multidimensional scaling (MDS) refers to a classof techniques that are used to investigate the struc-ture underlying data. More specifically, the tech-nique organizes proximity data by portraying thesimilarities among a set of objects as spatial rela-tionships. Although MDS techniques have a widerange of potential applications, the purpose of thepresent discussion is to summarize and to evaluatethe use of these techniques in vocational research.

It is not unusual for a considerable period oftime to elapse between the introduction of a newstatistical method and its application to substantiveissues in a scientific discipline. Although the al-gorithms for metric (Torgerson, 1952, 1958) andnonmetric (~r~slc~l9 1964) MDS have been avail-able for approximately three decades, the extensive

application of MDS to research in the behavioralsciences is a much more recent phenomenon. Evenmore recent is the use of MDS in vocational psy-

chology. An examination of this literature showsthat systematic applications of MDS have occurredprimarily in two areas, vocational interests and occupational perceptions. These areas have been cho-sen as central to this review. The decision was

buttressed by several prior reviews (Gottfredson,1982; Holcomb & Anderson, 1977) that identifiedvocational interests as the dominant research areain vocational psychology. Vocational perceptions,the other area characterized by programmatic MDSapplications, is also treated at length. I

In organizing the discussion two major purposesfor using MDS will be considered: configural ver-ification and dimensional identification. In config-ural verification the investigator examines the fit

between proximity data and prior expectations aboutthe nature of the stimulus configuration. To ex-emplify this approach, the focus will be on the

1The specific studies reviewed herein were assembled by iden-tifying a number of major outlets for vocational research andapplications of MDS. These sources included, but were notlimited to, the Journal of Vocational Behavior, Applied Psy-chological Measurement, Journal of Applied Psychology, Jour-nal of Counseling Psychology, Multivariate Behavioral Re-

search, Journal of Occupational Psychology, PersonnelPsychology, and Organizational Behavior and Human Perfor-mance. These journals were reviewed for the period 1970 to1983 for reports of research which applied MDS to vocationaltopics.

Downloaded from the Digital Conservancy at the University of Minnesota, http://purl.umn.edu/93227. May be reproduced with no cost by students and faculty for academic use. Non-academic reproduction

requires payment of royalties through the Copyright Clearance Center, http://www.copyright.com/

492

structure-of-interest models and the use of factorialand hierarchical methods as alternative data ana-

lytic methods for configural verification.Next, dimensional applications of 1~~~ in the

occupational perceptions area will be discussed.The purpose of dimensional applications it to iden-tify the attributes that individuals attend to in re-sponding to a class of stimuli. The ~~d~r~~l~~~~a-ledged use of MDS in this area has importantimplications for the understanding of vocationalbehavior. Using the research on occupational per-ceptions as an example, the sampling of occupa-tional titles will be discussed.

Finally, a number of promising applications ofMDS will be discussed. Although only limited workhas been accomplished in these areas to date, thediscussion will identify directions for future re-

search and applications of MDS.

Configural iTerifieation

Vocational Interests

Past and current research on vocational interestshas focused almost exclusively on their measure-ment, classification, and structure. Furthermore,much of what is reliably known about the structureand classification of vocational interests is based

on factor analytic methods (Dawis, 1980). Roe’s(1956) and Holland’s (1973) structure-of-interestmodels, for example, can be traced directly to Guil-ford, Christensen, Bond, and Sutton’s (1954) fac-tor analysis of 33 interest scales. This initial andinfluential study of the interest domain identifiedseven vocational interest factors (Mechanical, Sci-entific, Social Welfare, Aesthetic Expression,Clerical, Business, and Outdoor Work) as equiv-alent across two air corps samples. As detailed byRounds and Dawis (1979), these Guilford factorswere the most explicit forerunners of Roe’s eightinterest fields and Holland’s six interest types.A representation of Roe’s circular model of in-

terest fields and Holland’s hexagonal model of in-terest types is shown in Figure 1. Although Roe(1956; Roe & ~hs9 1969) hypothesized an overallcircular ordering, the internal relationships amongthe interest fields were never specified. Holland’s

hexagonal model, on the other hand, defines theinternal relationships among the interest types suchthat the distances between the types are &dquo;inverselyproportional to the theoretical relationships be-

tween them&dquo; (Holland, 1973, p.5), i.e., adjacenttypes on the hexagon are most related, whereasopposite types are least related, with alternatingtypes of an intermediate level of relationship.As a result of the apparent similarity of Holland’s s

and Roe’s models, Holland (1976) and Lunneborg(1975) have suggested a parallelism between com-ponents of the models. Specifically, the followingelements are hypothesized to represent similar do-mains : Holland’s Realistic (R) and Roe’s Tech-nology (Te) and Outdoor (0d); Holland’s Inves-tigative (1) and Roe’ Science (Se); Holland’s Artistic(A) and Roe’s ~~~e~°~1 Cultural (GC) and Arts aidEntertainment (AE); Holland’s Social (S) and RoesService (Sv); Holland’ Enterprising (E) and Roe’s sBusiness Contact (Bu); and Holland’s Gonven-tional (C) and Roe’s Organization (Or).Studies ®~ ~~~’s and Holland’s structural l

models. Strong support for both the circular andhexagonal structure-of-interest models has been

provided by the application of principal compo-nents analyses to intercorrelation matrices based oninterest scale scores (e.g., Hanson & Cole, 1973;Prediger, 1982). The use of MDS analysis withsimilar sets of data has, on the other hand, providedonly equivocal support for these models. Meir

(Feldman & Meir, 1976; I~~i~9 1973; Meir, Bar,Lahav, & Shalhevet, 1975; Meir & Ben-Yehuda,has conducted a series of IN4DS studies ~~~-amining the St of the internal relationships amongscores on a Hebrew interest inventory (Ramak; Heir& Barak, 1974) and the Hebrew version of theSelf-Directed Search (SDS; Holland, 1972) to Roe’scircular and Holland’s hexagonal models, respec-tively. In each study the structure was tested withIsraeli subjects by means of the Guttman-LingoesSmallest Space Analysis (SSA-1; Guttman., 19&~9Lingoes, 1965).

In one of the first applications of MDS to interestdata, Meir (1973) examined the fit of Roe’s field-by-level model to the intercorrelations amongRamak scale scores. This model classifies occu-

pations according to three levels of function and



493

Figure 1Schematic Representations of Roe’s Circular Model

and Holland’s Hexagonal Model -

eight fields defined by their primary focus of ac-tivity. The Ramak interest inventory was scoredby three levels and eight fields for a total of 24scale scores. As shown in Table 19 the arrangementof the fields in two-dimensional space for the 12th-

grade female and male students did not conformto Roe’s circular ordering of interest fields. A vis-ual inspection of the configurations shows that thesescale points formed a horseshoe shape with the Teand Sc scale points and the Or and Bu scale pointsclustering together. Although the three levels foreach tended to cluster within fg~lds9 the order ofthe levels was not in the hypothesized sequence.As can be seen from Table 1, two subsequent stud-ies by Meir and his associates produced slightlydifferent Ramak configurations, none of which matchthe Roe circular ordering of interest fields.

Meir has also examined the fit of Holland’s

RIASEC hexagonal model to the internal relation-ships among two measures of Holland’s interesttypes, i.e., the Interest Inventory for Females (IIF)and the Hebrew translation of the SDS. For the

IIF, Feldman and Meir (1976) reported circularoff IRAESC for a female 1 lth-grade sam-ple and an IRASEC circular ordering for an em-female There was considerableoverlap of the scale points representing the S andE interest fields for both For the Hebrewversion of the SDS, Meir and Ben-Yehuda ( 1976)

reported a two-dimensional horseshoe-shaped con-figuration of RISACE with the l~-I9 S-A, and C-Escale points clustering together for a combinedsample of male and female ninth-grade students.

Feldman and Meir’s (1976) finding that the Hol-land model for females was IRAESC or IRASECwas not confirmed in a study with American sub-jects (Rounds, Davison, & Dawis, 1979). UsingTORSCA-9 nonmetric scaling (Young & Torger-son, 1967), Rounds et al. (1979) examined the fitof the Strong-Campbell Interest Inventory (SCII)and the Vocational Preference Inventory (VPI) scalesto Holland’s hexagonal model. As shown in Table1, five scaling solutions were obtained with anidentical RIASEC scale arrangement. Neverthe-

less, a visual examination and a statistical test

(Wakeneld & Doughtie, 1973) of how well thescaling representation fit the model showed that thefemale data met the expectations from Holland’smodel less often than the male data. The fit of theSCII MDS results to the hexagonal model for fe-males was not good, with a near reversal of the Sand E scales. For males the fit of the SCII and VPIMDS results to the hexagonal model was satisfac-tory.When reviewing the above studies, it is apparent

that the MDS research by Meir and his colleagues(Feldman & Meir, 1976; Meir, 1973; Meir et al.,1975; Meir & Ben-Yehuda, 1976) provides no sup-



494

Table 1

Multidimensional Scaling Studiesof Vocational Interests



495

port for Roe and Holland’s hypothesized interestconfigurations. On the contrary, each study hasfound a different circular ordering of the interestfields and different scale point clusters. Some cau-tion is required, however, when interpreting I~leir’s sinability to replicate the hypothesized interest con-figurations. On the one hand, it is important torealize that these studies were conducted amongdifferent age groups, with the youngest subjectsbeing 14 to 15 years old. On the other hand, andperhaps more importantly, cultural differences mayaccount for the failure to find the hypothesizedinterest configuration.A study that addresses the issue of cultural dif-

ferences and the structure of vocational interests

was reported by Meir, Sohlberg, and Barak (1973).The investigation examined the fit of the correla-

tions among the Ramak scales and the CoursesInterest Inventory (CII) to Roe’s circular configu-ration with an Arab (nonwestem culture) 12th-gradesample and an Israeli (western culture) universityapplicant sample. For intercultural comparisons the12th-grade Ramak data for both sexes, previouslyreported by Meir (1973), were reanalyzed. As shownin Table 1, the arrangement of the Ramak and CMscales does not conform to Roe’ hypothesized or-der, and the Ramak scale order differs from theCII scale order for the Arab and the Israeli samples.Because of the scale order variability within theArab and Israeli samples, it is almost impossibleto accurately summarize the scale order betweenthese samples.The four interest configurations for the Israeli

samples can best be described as &dquo;rnisshap~r~ poly-gons,&dquo; a term Holland (1979) applied to the hex-agons resulting from real-world data. The two in-terest configurations for the Arab samples are bestrepresented by a triangle with the Sc and Te scalesat the apex of the triangle, and the AE, Sv, GC,Or, and Bu scales defining the base of the triangle,a configuration that suggests a degenerate solution.At least for these samples the Arab students dif-ferentiated the Sc and Te occupations from otheroccupations more so than did the Israeli students.The question of the adequacy of the Ramak as

a measure of Roe’s eight interest fields is one pos-sible explanation for the lack of fit between theRamak data and Roe’s circular model. Since Roe’s s

hypothesized circular model is culturally deter-

mined, another plausible explanation concerns thediversity and breadth of American occupations incomparison to Israeli occupations. None of theseexplanations, however, received support from

Gati’s (1979) SSA-I of the Vocational Interest In-

ventory (VII; Lunneborg, 1975) scale correlationsbased on male and female American university stu-dents. Aside from the Ramak, the VII appears tobe the only instrument developed based on Roe’ssystem. Gati has presented an arrangement of Roe’sfields that is identical to Meir’s (1973) circulararrangement of Te-Sc-Od-AE-Sv-GC-Or-Bu, afinding that leads to questions concerning the ad-equacy of Roe’s model.

Studies comparing Roe’s and Holland’s

models. Two studies (Gati, 1979; Meir & Ben-

Yehuda, 1976) have investigated the parallelismbetween Holland’s and Roe’s models. Meir and

Ben-Yehuda (1976) displayed a two-dimensionalsolution with four scale clusters defining two or-thogonal dimensions: Sc and Od opposing E, C,Bu, and Or; and Te, R, and I opposing AE, A,Gc, Sv, and S. In spite of the apparent correspon-dence between fields, the parallel fields are notalways nearest to each other: Bu lies nearer to Cthan to E, and Od lies nearer to S than to R.As previously noted, Gati (1979) presented a

two-dimensional SSA-I scaling solution of corre-lational matrices from Lunneborg and Lunneborg’s s(1975) factor analytic study of the interest modelsof Roe and Holland. The configuration was a con-centric circle with Holland’s RIASEC forming theinner circle and Roe’s Te-Od-Sc-AE-GC-Sv-Or-

Bu forming the outer circle. This concentric patternof the interrelationships between Holland and Roescales is not unexpected; intercorrelations amongthe Holland scales have previously been found tobe quite high, in some instances almost as high ascorrelations between measures for the same Hol-land interest type (Rounds et al . , 1979). Gati (1979)reported that expectations based on parallel fieldsare not met in several cases: Or is between Bu and

E, C is nearer to E than to Or, and S is nearer toGC than to Sv.

Summary. MDS studies of the structure-of-

interests have highlighted several problems of thehexagonal and circular models. With respect to



496

Roe’s circular model, there is a disagreement be-tween the theoretical and empirical ordering of theinterest fields. The inconsistencies in the orderingare usually interchanges of adjacent fields and oc-cur across samples in an unpredictable manner.With respect to ~c~ll~r~d9s model, the empirical datagenerally conform to the RIASEC ordering; theshape of the configurations, however, rarely ap-proximates a hexagon. In addition, female datameets the Holland model expectations less oftenthan does the male data. Finally, the expected par-allelism between Holland and Roe’s models has

yet to be demonstrated.The aforementioned less than perfect relation-

ship between interest theory and data led Gati (1979, 91982) to propose that a hierarchical model fits theempirical interrelationships between interest fieldsmore adequately than the circular or hexagonalmodels. It is important to note that Gati’s (1979)initial claims for a hierarchical model were basedon the comparison of MDS or factor analytic so-lutions to hierarchical cluster solutions. C~~.ti’s (1982)most recent claims for a hierarchical model as abetter representation of the Holland and Roe dataare based on a direct (raw proximity data) com-parison of the distances between interest fields.Gati has buttressed his conclusion with argumentsthat the hierarchical approach offers a more ap-propriate framework for studying the process ofoccupational choice and career development. Tver-sky and Gati (1978) have also questioned the vi-ability of distance measures for psychological sim-ilarity (see Krumhansi, 1978, for a response).The following discussion will examine only Gatfs s

initial claim concerning dimensional, factorial, andhierarchical representation. Although the issue ofthe appropriateness of specific analytic proceduresis pertinent to a number of vocational research areas, 9the comparison of 1~DS to other data analytic meth-ods seems best exemplified in the context of vo-cational interest research.

Representations and I-lierarchicalRepresentations

Two issues concerning the relationship betweendimensional, factorial, and hierarchical methods

will be addressed. The first, more global, issue

involves understanding the similarities and differ-ences among these methods. The second and more

specific issue is which data analytic technique-nonmetric scaling, principal components, or hier-archical clustering-best accounts for the interre-lationships among interest scales.

Comparisons among methods. Comparisons ofMDS to factor analysis (Davison, 1983; Mac-G~lla~~9 1974; Shepard, 1972) and cluster analysis(Kruskal, 1977; Sattath & Tversky, 1977; Shepard& Arabie, 1979) have generally focused on themost appropriate type of proximity data for eachprocedure, the assumptions underlying the analyticprocedures, and the manner in which stimuli arerepresented. Although all three methods essentiallyanalyze similarity measures, factor analysis, withonly a few notable exceptions (e.g., 9 ~l~r~~r~, 1954),has rarely been used with similarity judgments.

Cluster, factor, and MDS procedures also reston very different assumptions regarding the rela-tionship between the proximity data and its spatialrepresentation. Nonmetric MDS assumes that theproximity data are a monotonic function of dis-tances between objects in Euclidean space. Factoranalysis assumes that the proximity data are lin-early related, not to distances between objects, butto cosines of angles between vectors. Hierarchicalcluster analysis assumes that the proximity data aremonotonically related to distances in ultrametricspace (see Davison, 1983, pp. 208-211, for a dis-cussion of the concept of ultrametric distances).Although factor analysis and MDS both providespatial representations of the stimulus structure,MDS has the advantage of ease of interpretability,i. e. , it is easier to interpret distances between pointsor scale values than to interpret the angles betweenvectors as required by factor analysis. Hierarchicalcluster analysis, on the other hand, represents stim-uli in terms of stimulus groups portrayed in a treediagram. It is important to realize that the tree

diagram representation and the spatial representa-tions usually capture different aspects of the samedata.

Although these generalizations concerning thecomparison of dimensional, factorial, and hierar-chical methods provide overall guidelines, Davison



497

(1981, 19831 has more systematically examined therelationship of factor analysis and ~1~~. Davisoninvestigated the relationship of nonmetric scalingand principal components solutions based on thesame intercor-relation matrices for ability and in-terest measures. His results suggest that the prin-cipal components solution will usually contain ageneral factor that has Do dimensional counterpartin the scaling solution. Scale values, when appro-priately transformed, provided a close approxi-mation to the factor loadings. This type of system-atic comparison of MDS and factor analysis is anarea of increased activity in several psychologicaldomains (Luxenberg & ~~~r~~9 1982; Zevon, Lux-

enberg, & Rounds, 1983).~/f!’~fy of the methods. Two specific data an-

alytic methods used in the interest area employprincipal component analyses to study the rela-tionships among interest scales. The principal com-ponents method used to provide initial support forHolland’s circular arrangement of interest fields

(e.g., Cole, 1973; Cole, ‘~Ihit~ey9 & Holland, 1971;Lunneborg & Lunneborg, 1975) was a configuralanalysis or ’’analysis of spatial configuration&dquo; (Cole& Cole, 1970). Using a two-stage principal com-ponents analysis, this method treats variables as

vectors and fits a &dquo;smaller dimensioned&dquo; space tothe vectors.

Gati (1979) compared solutions based on con-6gural analysis, Guttman-Lingoes SSA-1, andADDTREE hierarchical cluster analysis (Sattath &Tversky, 1977) in a reanalysis of Lunneborg andLunneborg’s (1975) VPI and VII correlation mat-rices. The ADDTREE solution was compared tothe SSA-1 solution via the coefficient of alienation

(COA), an ordinal measure of fit. The COA forthe VPI data analyzed with the dimensional meth-odology was .012, whereas the hierarchically basedCOA was .075. The dimensional COA for the VII

data was .014; the hierarchically based COA was.095. Variance accounted for was used to comparethe configural solution to the ADDTREE solution.The values reported for the VPI data were 77.5%and 68.4% for the factorial and hierarchical solu-tions, whereas the corresponding VII values were75.9% and 77.1%. These comparisons of the dif-fering representational methods indicate that di-

mensional and factorial methods a slightiybettir fit than factorial methods provide a slightlybetter fit than the hierarchical method for the VP’~~Holland data. For the VII Roe data the hierarchicalmethod represented the distances in the data as wellas the dimensional and factorial methods.

Investigations of i-~~ii~.~d9s hexagonal model nolonger employ Coif and Cole’s (1970) analysis ofspatial configurations. Cooley and ~~h~~s’ ~ (1971, 9pp. 137-143) FACTOR program, a principal com-ponent technique for the extraction of arbitrary fac-tors, has become the dominant method for exam-

ining the fit of the interrelationships among interestscales to Holland’s RIASEC hexagonal model. TheCooley and Lohnes algorithm allows the researcherto specify a target matrix of loadings. Each factoris extracted in succession with the restriction thateach successive factor must be orthogonal to allpreviously extracted factors. Prediger (1982), us-ing the FACTOR program, examined the extent towhich two theory-based dimensions--clata/ideasand things/people&horbar;6t 24 sets of Holland scale in-tercorrelations. He reported the RIASEC scale ar-rangement for 23 of the 24 data sets.No direct comparisons exist between solutions

resulting from the Cooley and ~~h~~s ~ i. ~~ k ) tech-nique and nonmetric scaling procedures. However,Rounds et ~i. ~ ~ ~7~) and Rounds and Dawis (1980)have examined, with the TORSCA-9 scaling pro-gram, six of the same data sets found in Prediger(1982). Visual inspection of these two sets of so-lutions shows (1) an identical RIASEC circular or-dering, ~~) ~ a dissimilar configuration for four ofthe six 2-dimensional solutions, and (3) a differentdimensionality for two of the six solutions. Al-

though no firm conclusions are possible based onthis ad hoc comparison, MDS seems to more faith-fully reproduce the observed proximity data.

Dimensional ~~pp~~~~t~~~~

Occupationai ~~g ~~~~:~~~.~

One of the best kept secrets in vocational psy-chology is the extensive MDS research on occu-pational perceptions. Holland (i~ ~39 1976), for ex-ample, reviewed the occupational perceptionsliterature, yet did not cite any MDS studies. Like-



498

wise, in a review of the empirical evidence sup-porting career development theories, Osipow (1983)cited studies on occupational stereotyping but failedto cite any of the MDS studies of occupationalperceptions. It should be noted, however, that muchof this literature is reported in European journalsand/or is conducted within disciplines other thanpsychology.

of Occupational Perceptions

Perceptions of occupations have been studiedunder a variety of labels, e.g., vocational or oc-cupational images, stereotypes, and preferences.With the exception of several factor analytic studies(e.g., Gonyea & Lunneborg, 1963) and the MDSliterature, research on how individuals perceive oc-cupations has primarily focused on perceptions ofjob incumbents (e.g., Dipboye & Anderson, 1961)or has equated occupational perceptions with re-sponses to interest inventories (e.g., Edwards,Nafziger, & Holland, 1974). Where direct judgedsimilarities have been used, several additional cri-teria become salient in investigations of occupa-tional perceptions. First, a large and representativesample of occupations must be studied. Second,the basis for comparison must be chosen by thesubject rather than imposed by the investigator.Finally, the method of data analysis must be ableto both accommodate large numbers of occupationsand to recover the dimensions on which the oc-

cupations have been compared.Studies of occupational stereotypes that purport

to investigate occupational perceptions tend to meetnone of the above requirements. The typical ex-perimental design involves the a priori manipula-tion of, at most, two dimensions for only a fewoccupations and the subsequent assessment of themanipulation on a third dimension (e.g., prefer-ence). Studies that have directly investigated thejudged similarity of large numbers of occupationsrely on either content analysis (e.g., 9 ~r~nes~ 1957)or principal components analysis (e.g., Stone &

Bassett, 1972). In the former case, the descriptiveanalyses relied on rather arbitrary and intuitiveclusters. In the latter case the principal componentsanalyses yielded an unmanageable number of di-

mensions., whereas the higher order factors provedtoo complex for clear and interpretable results. Fur-th~;rrrr~®re9 principal components solutions usuallyemploy derived indices of similarity rather thandirect similarity judgments, a characteristic that maycontribute to a loss of information. Because factor

analysis is an unsatisfactory method for analyzingsimilarity judgments, work on the structure of oc-cupational perceptions was largely abandoned untilthe development of MDS. MDS, employed on arepresentative sample, does satisfy the aforemen-tioned criteria.

As is illustrated in Table 2, the bulk of the re-search in the area of occupational perceptions hasbeen conducted by only a small number of inves-tigators. The most systematic work has been con-ducted by Reeb (1959, 1971, 1974, 1979) usingnonmetric scaling. Reeb has been concerned pri-marly with identifying the characteristics by whichindividuals organize occupational perceptions andwith examining the generalizability of these di-mensions.

Guided by census data, Reeb (1959) selected the15 most frequently entered occupations for 15- to19-year-old males in London. Using a category T2 2

rating technique, he obtained direct similarity judg-ments from British Youth Employment Officertrainees. Two dimensions were obtained with Tor-

~~rs®r~9s (1952, 1958) algorithm and were visuallyinterpreted as a craft versus clerical dimension andan occupational level dimension. Using the sameMDS technique and set of occupations, Reeb (1971)attempted to replicate his prior findings with ex-perienced British Employment Officers and twogroups of 14- to 15-year-old school dropouts ofdifferent socioeconomic levels. Instead of direct

similarity judgments, however, occupations werejudged according to suitability (viz., &dquo;For a boyyou are advising, if one job of the pair is the mostsuitable of all possible jobs, then how suitable isthe other? ’ ’ ) .

Reeb investigated the generality of the MDS rep-resentations-and hence the difference between

similarity and suitability representations-througha comparison of the solutions based on the 1959counselor trainee sample and the 1971 1 experiencedcounselor sample. The resulting dimensions for the



499

combined counselor group were easily interpreted,one being occupational level and the other blue-collar versus while-collar. For the combined ado-lescent sample, the dimensions were lab~l~d 6 ‘pr~s-tige&dquo; and &dquo;desirability.&dquo; The very high correla-tions between the stimulus coordinates for the two

groups of counselors (r== 92 for the level dimen-sion and r=.94 for the blue/white collar dimen-

sion), and between the ratings given on two oc-casions for the experienced counselors (r=.98),indicate a well-founded perceptual structure unaf-fected by type of judgment task (similarity versussuitability) or experience. The test-retest reliabili-ties and subgroup analyses for the male adolescentsalso indicated a stable and generalizable occupa-tional structure. More importantly, as Reeb (1971)notes, &dquo;The most striking result, however, is theextreme simplicity of the dimensional structure ...with advantages in intuitive understanding and ineconomy and clarity of exposition as to how thesegroups saw these occupations&dquo; (p. 242).

Using M-D-SCAL (Kruskal & Carmone, 1969)to scale 12 occupational titles, Reeb (1974) par-tially replicated his 1971 findings with 125 Israeliadolescent boys. The group suitability judgmentswere found to be highly reliable, and MDS yieldedtwo dimensions unaffected by the subjects’ socio-economic level, intelligence, or job preference. Thefirst dimension, prestige, is similar to that foundin Great Britain with adolescent male samples. Thesecond dimension, however-blue collar versuswhite collar-does not seem to have a counterpartin the dimensional representation of the British ad-olescents.

Reeb was able to study up to 15 occupations, alevel that required a subject to make 105 pairedcomparison judgments. In contrast, the experi-mental task in which subjects sort occupations intocategories can be used to scale large numbers ofoccupations without overwhelming subjects. Bur-ton (1972) was able to study 60 occupations withsuch a sorting task, a maximum determined by thelimitations of Kruskal ’ I~-~-~CAl~pr®~rar~. Sub-jects were instructed to sort a deck of cards withoccupational terms on them into any number ofpiles ’ ’ so that occupations which seemed the samewere in the same pile&dquo; (Burton, 1972, p. 59).

Scaling of the joint probability measure of occu-pational similarities indicated three dimensions-dependency, prestige, and skill.

Burton’s study, pertaining as it does to only 60occupations and 54 respondents, is limited in scope.Kraus, Schild9 and Hodge (1978), on the otherhand, reported the first comprehensive investiga-tion of a large sample of occupations (~V == 220) andrespondents (N=463) representative of, respec-

tively, the occupational domain and the generalpopulation. The subject sample included individ-uals aged 20 and over randomly sampled from thethree largest urban areas in Israel. Pretests of theoccupational sorting task showed that respondentshad difficulty coping with more than approximately90 occupations. Consequently, Kraus et al. (1978)randomly assigned the respondents to three sub-samples. Respondents in each subsample were pre-sented with 90 occupations to sort, 25 of whichwere common to all subsamples and 65 of whichwere unique to the three subsamples. The sym-metric similarity matrix, the entries of which arethe proportion of respondents sorting the two oc-cupations in the same category, was analyzed withthe SSA-1 program. Correlations between the firstand second unrotated stimulus coordinates com-

mon to the three random subsamples and foursubgroups indicated a unidimensional occupationalstructure. This single dimension was highly cor-related with both the respondents’ ratings of theoccupations &dquo;social standing&dquo; (r = .98) and Hart-mann’s (1975) prior assignment of occupationalprestige scores (r = .92).

Individual clifferercce.s. In the previous studiesthe MDS solutions were obtained by analyzing dataaveraged across subjects. Although most of thesestudies included subgroup analyses (e.g., socio-

economic level), the nonmetric scaling techniqueswere not designed for investigating individual dif-ferences and may, therefore, misrepresent the per-ceptual structure for some subjects (Wish, l7eutsch9& Biener, 1972).Coxon and Jones (1974a, 1978) and Shubsachs

and Davison (1979) have studied individual dif-ferences in occupational perceptions. The use ofindividual difference scaling in such studies pro-vides important information concerning (1) the



500



501



502

amount of variance in a subject’s data accountedfor by the configuration, (2) the pattern and salienceof the dimensions, and (3) the importance off thehypothesized attributes. In Coxon and Jones (1974a)the relative weights of occupational groups for theirtwo-dimensional solutions were examined. On the

dimension labeled educational requirement;, highweights tended to occur among theology studentsand nurses, whereas low weights characterized

building and engineering apprentices and collegeof education students. On the second dimension,people orientation, the above weights for the oc-cupational group relationships were reversed.

Shubsachs and Davison ( 1979) compared the rel-ative weights attached by vocational experts, lib-eral arts students, and engineering students to

INDSCAL dimensions. Differences among the

subject groups were found only along the compen-sation dimension; the vocational experts, a groupcomposed of vocational counselors and research-ers, attached less importance to this dimension rel-ative to the other groups. As shown in Table 2

Shubsachs and Davison reported a four-dimen-sional group solution based on the judged similarityof all possible pairs of 18 occupations. For in-terpretive purposes they regressed 25 hypothesizedoccupational attributes on the four INDSCAL di-mensions. Unfortunately, the number of significantmultiple and zero-order correlations confounded thestraightforward interpretation of these dimensions.Nevertheless, the regression analysis did demon-strate that occupational reinforcers (Dawis, Lof-quist, & ~I~iss9 1968; Lofquist & Dawis, 1969)were an important component of occupational per-ceptions.

Surprisingly, neither the Coxon and Jones (1974a, 91978) nor Shubsachs and Davison (1979) studiesreported finding a prestige dimension. Shubsachsand Davison did, however, report finding a sig-nificant multiple correlation (R=.80) betweenprestige ratings and their four dimensions, a findingwhich indicated that occupational prestige is rep-resented in the spatial solution. Coxon and Jonesrelied on a visual examination of the stimulus co-ordinates and, ostensibly, did not test for a prestigedimension.

Possibly the most extensive applications ®f ~DS

to vocational data are the investigations of occu-pational cognitions conducted by Coxon and Jones(1978, 1979a, 1979b). The results of their exten-sive analyses, and sometimes obtuse reporting oftheir findings, are published in three separate vol-umes and are too extensive to report in detail. Theirconclusions concerning the application of individ-ual differences scaling are, however, particularlygermane to the present discussion. These conclu-sions were ( 1 ) configurations obtained from scalingpairwise and triadic formats are very similar; (2)while the INDSCAL group space is similar to non-metric solutions, the INDSCAL solutions requirea large number of dimensions; (3) a horseshoe-likestructure was observed indicating that one dimen-sion is sufficient to describe the occupational space(see Kruskal & Wish, 1978, for a more extensivediscussion of the horseshoe phenomenon); and (4)subjects’ verbalizations of the predicates they usedin making occupational judgments indicated intra-and inter-individual variations in the level of gen-crality of the judgments.

This latter conclusion resulted from asking sub-jects to state the way (or ways) in which occupa-tional pairs were alike or different. This methodsystematically elicits occupational attributes with-out the imposition of the investigator’s preconcep-tions. Nevertheless, Coxon and Jones ( 1979a) ob-served, ’Perhaps one of the more striking findingswas that the context of a sorting often tells a com-pletely different story from the verbal descriptions.... Some people, for example, claimed that socialclass had no place in their thinking and then pro-ceeded to use it liberally&dquo; (p. 189).

Perceptions vs. clc~s,~~f’acc~ta®~a systems. Siess and

Rogers (1974) and Reeb (1979) investigated thedegree to which the perceived similarity of occu-pations corresponded to occupational classificationsystems. These studies are examples of investi-gations that are concerned with both the identifi-cation of the dimensions underlying occupationalperceptions (dimensional approach) and the ar-

rangement of occupations in N-dimensional space(configural approach).

Siess and Rogers (1974) compared judged sim-ilarities made by college freshmen to Roe’ (1956)classification system with the expectation that Roe’s



503

classification would be reflected in the dimensional

configuration. Four illustrative occupational titleswere selected from five of Roe’s eight fields withlevel held constant. Application of Torgerson’s(19529 1958) algorithm to direct similarity judg-ments provided support for the relative homogene-ity of judgments with fields and no support for theirpostulated circular order.Reeb (1979) compared direct suitability judg-

ments on 15 occupations made by 101 male andfemale Israeli students to Roe’s (1956), Holland’s(1973), and Flanagan, Shaycoff, Richards, &

Claudy’s (1971) occupational classifications and tofour academic areas. Since the occupations wereselected to correspond with the occupations of thestudents in the four academic areas, it is not sur-

prising that the normetric scaling procedure yieldeda two-dimensional configuration with occupationsgrouping into four clusters corresponding to thefour academic areas. The comparisons of male andfemale occupational configurations were accom-plished by rotation to best fit of the female stimuluscoordinates to the male target matrix. Visual in-

spection revealed minor variations in location ofthe occupations for the female and male students.Reeb (1979) also reported a nonmetric internal

analysis of the student’ rank-ordered occupationalpreferences. The K4DPREF procedure (~~r~°®11~ 1972;Chang & Carroll, 1974) yielded common stimulusspace configurations for each se~c9 which were ro-tated for comparison with the occupational simi-larity solutions. The fields previously found fromthe similarity judgments remain largely intact onthe preference maps; however, the relative loca-tions of the fields and the location of the occupa-tions with each field changed for the male andfemale students .

Summary. Judgments of the perceived simi-larity between occupations are efficiently repre-sented by MDS, usually in a two-dimensional space.The dimensions are reproducible within subgroupsof samples and are easily interpretable. The studiesreviewed above have demonstrated the existence

of occupational perceptual structures across a rangeof samples. In comparison to the configural veri-fication approach, the dimensional approach hasbeen characterized by careful attention to sampling

of both occupations and subjects and to tests ofprior attribute hypotheses. This greater attention tomethodology distinguishes this application of MDSfrom many other multivariate procedures.

of Occupational Titles

A consistent finding across studies of occupa-tional perceptions is that subjects organize theirperceptions according to prestige. I~ev~rtheless9 theconsistency of these findings is overshadowed bythe variability in the number and types of otherdimensions reported in this literature. One possibleexplanation for this dimensional variability is thatthe selection of occupational titles affects the num-ber and types of dimensions that result from scalinganalysis.Coxon and Jones (1974b) have discussed three

concerns related to the selection of occupationaltitles: (1) comparability with prior investigations,(2) limitations in the number of stimuli that can be

concurrently judged, and (3) adequacy of the stim-ulus domain sample. The subjects’ familiarity withthe occupations under investigation is also an im-portant consideration.

C’®~p~aa~~bila~. Comparability with prior in-

vestigations refers to a concern about the selectionof occupations that are consistent with previousstudies. Reeb (1959, 1971), for example, retainedhis entire set of occupational titles across both in-vestigations, thereby allowing him to directly com-pare the judged similarities between the 1959 vo-cational counselors trainee sample and the 1971

experienced vocational counselor sample. Coxonand Jones (1978) selected a set of eight occupa-tional titles from the Hall-Jones ( 1951 ) classifica-tion system and were able, therefore, to comparetheir findings with the results of other investiga-tions. This type of cumulative study of a commonset of occupational titles encourages programmatic,as opposed to fragmented, research efforts, Un-fortunately, the acceptance of a uniform set of titleshas yet to occur, although it seems to be a desirablegoal.Number of stimuli. The number of occupa-

tional titles used in any investigation varies withthe type of experimental task. In general, occu-



504

pational perception research has used a categoryrating task in conjunction with a complete pairedcomparison design. As the number of occupationaltitles increases, however, the number of pairs, 9I(I - 1)/2, increases rapidly, thereby placing limitson the number of occupations that subjects can beexpected to judge within the paired comparisonformat. With the use of incomplete designs (Tor-gerson, 1958), however, the number of stimuli canbe increased substantially. Recent work byh4acCallum ( ~ 9~ ~) and Spence and Domoney (1974)discusses the advantages of different types of in-complete designs for nonmetric MDS. With largenumbers of occupational stimuli (that is, greaterthan 35), sorting strategies such as those used byBurton (1972) are necessary (see Rosenberg & Kim,1975).Adequacy ®~~ ~,~~ ~~a~~tli. It is important to re-

call that the set of occupational titles in a particularstudy is a sample from a large domain. Existingoccupational classifications have often served asdefinitions of the domain in studies of occupationalperceptions. The adequacy of the set of occupa-tional titles used in a study then becomes a sam-pling issue. None of the studies so far reviewedhave reported drawing a random sample of occu-pational titles. The most frequently encounteredsampling strategy involves the experimenter se-lecting occupations that are judged to be represen-tative of the domain in question. The major ad-vantages of this nonprobability sampling techniqueare convenience and economy. On the other hand,this procedure lacks an objective verification of therepresentativeness of the sample. Given an unre-presentative sample, the generalizability of the

findings becomes an issue. These considerationsbecome particularly salient when the domain underinvestigation is extremely large, e.g., The Dic-

tionary of Occupational Titles (U.S. Departmentof Labor, 1977).

Several studies have attempted to define the oc-cupational domain by simply asking subjects to listoccupations. Burton (1972), using a free recall task,reported that subjects listed only high prestige andcreative occupations. Reeb (1974) found no rela-tionship between census data and subjects’ listingsof occupations in response to the question, ‘6~~

what jobs do people most commonly work?&dquo; (p.These two studies indicate little correspon-dence between an individdaal’s naturalistic occu-

pational classifications and those imposed by in-vestigators . The effect of this discrepency has notbeen explored, and although it is perhaps only aninteresting observation at this point, it seems wor-thy of future investigation.with the stimuli,. Very little is known

about the effects of subjects’ occupational famil-iarity on scaling solutions. Of the nine occupationalperception studies reviewed herein, only one di-rectly assessed the subject’s familiarity with theinvestigated occupations (Shubsachs ~z Davison,1979).Research on vocational maturity (~’~esY~~°~c~~~9

1983) and cognitive complexity (~~a~s~9 Reed, &Winer, 1979) indicates that occupational knowl-edge (accuracy of information) is related to appro-priateness of vocational choice and, further, thatoccupational information decreases rather than in-creases cognitive complexity in the occupationalrealm. Thus, the question of familiarity with oc-cupations is an issue considerably more complexthan many investigators had initially anticipated.At the minimum, therefore, it is suggested thatresearchers incorporate familiarity checks into MDSdesigns and regress the familiarity ratings over thecoordinates of the configuration. This procedurewould indicate the manner in which familiarity ef-fects the scaling solution.

Developing ~~.pp~n~~¢~~~3~~

Several topics within, vocational psychology rep-resent areas in which MDS analysis has, to date,been only minimally applied. A number of thesetopics appear to represent promising new directionsfor vocational research and areas where MDS tech-

niques can be fruitfully applied. One such directionis exemplified by a study investigating stage the-ories of vocational development (Jepsen lt Grove,1981). Hitherto, vocational models have largelybeen untested. Tests that have been conducted con-sisted of demonstrating mean differences on vo-cational maturity indices among age groups. Jepsenand Grove applied MDS procedures to measures



505

of Tiedeman and O’Hara’s (1963) vocational de-cision-making paradigm. The authors reported tworelated purposes for their study: first, to test thehypothesis that vocational decision-making stagesfollow the exploration, crystallization, choice, andclarification stages postulated by Tiedeman andO’Hara; and second, to extend earlier work on Dav-ison’ (1977) metric unfolding model. The resultsof the nonmetric scaling generally supported Tiede-man and ~’l~~.ra’s stage order, with the exceptionof a reversal of the choice and clarification stages.The authors’ hypothesis regarding the domi-

nance of the vocational decision-making stages andthe association of the student’s responses with a

single stage was not supported. A major finding ofthe study was the usefulness of the metric unfoldingmodel for testing stage theories with cross-sectionaldata. Although the authors noted that the unfoldingprocedure is not a substitute for longitudinal data,its use in this and related studies testing stage se-quence models is amply supported (see Davison,King, l~itche~~r, ~ Parker, 1980; Davison, Rob-bins, & Swanson, 1978).A second study (Krau, 1982), examining vo-

cational stages, assessed a career model for im-

migrants with two samples-one recent immigrantgroup and one group assessed 5 years after im-

migration. The analysis, using Guttman-Lingoesnonmetric SSA, is a good example of a data re-duction application where the intent is to representthe similarity data in a simpler form. The hypoth-esized model was supported by both the Guttman-Lingoes analysis and the multiple regression of cri-terion variables on a number of success predictors.Even though the investigation is of interest, sincelittle empirical work examines applications of ca-reer stage models to an immigrant sample, the useof nonmetric scaling procedures provides minimalinformation compared to that which would resultfrom other data analytic techniques. Application ofDavison’s unfolding model, for example, wouldprovide a better test of this career stage model.

Another promising area for the application of1~~~ is the investigation of the structure of jobsatisfaction. ee studies were located (Ben-Porat,1978, 1981 Katz & l~~ar~e~9 1977). In an ex-

ploratory study, Ben-Porat (1978) showed that job

factors are arranged in a circular order divided byextrinsic and intrinsic dimensions. In an extensionof this study, Ben-Porat (1981) evaluated Schniederand Locke’s (1971) two-factor theory of job sat-isfaction by embedding event and agent as twodomain facets of a job satisfaction content universe(see Guttman, 1954, and Shapira and Zevulun, 1979,for a discussion of facet analysis). Based on anIsraeli sample of 104 blue-collar workers from eightindustrial organizations, an intercorrelation matrixof 11 I items of job satisfaction and one item meas-uring overall satisfaction were submitted to the SSA-1 program. The results confirmed an a priori hy-pothesis of a radex structure when job satisfactionis defined by two domain facets.

Katz and Maanen (1977) examined the relation-

ship between components of work satisfaction anda number of work environment design variables(e.g., assigned tasks, work assistance, and com-munications). Work satisfaction was measured bya modified version of the Minnesota Satisfaction

Questionnaire (Dawis & Weitzel, 1974) adminis-tered to 3,080 subjects from four government or-ganizations. The authors subsequently applied anonmetric scaling procedure (TORSCA; Torger-son, 1965; Young & Torgerson, 1967) to only thejob satisfaction intercorrelation matrix. Three dis-tinct clusters (job properties, interaction context,and organizational policies) were identified with avisual inspection of the two-dimensional solutionand the application of a neighborhood interpreta-tion. The two dimensions were interpreted as ashort- versus long-term element in job satisfactionand the traditional intrinsic-extrinsic dimension.

The primary advantage of MI7S as applied instudies of job satisfaction is that nonmetric pro-cedures typically yield fewer dimensions than fac-tor analysis, thereby providing greater simplicityand facilitating interpretation. A potential disad-vantage is the influence of investigator bias in vis-ual interpretation. A possible alternative to ex-

ploratory uses of nonmetric scaling is the use ofconfirmatory scaling procedures (Carroll & Arabie,1980; Davison, 1983). At this point, however, theadvantages associated with confirmatory proce-dures are limited by the lack of an agreed uponmeasure of fit.



506

A research area related to job satisfaction is thetopic of work outcomes, also known as needs andvalues in the counseling literature. A good exampleof a dimensional application of MDS in this areais a study by Billings and Cornelius (1980). Thepurpose of this exploratory study was to examinethe perceptual structure of work outcomes and todemonstrate the appropriateness of MDS in thiseffort. The authors developed two questionnaires,one based on similarity judgments and the otherrequiring likelihood judgments. Both question-naires consisted of paired comparisons of the 21work outcomes used in the Dyer and Parker (1975)survey and asked the subjects to rate eight hypoth-esized attributes for each of the 21 work outcomes.

Using the individual differences weighted Eu-clidian model incorporated in the ALSCAL pro-gram (Takane, Young, & de Leeuw, 1977), theauthors determined the dimensionality by correlat-ing the stimulus coordinates of the similarity spaceswith the likelihood spaces. The cross-correlationsthat best met the requirements for convergent anddivergent validity indicated a three-dimensional so-lution for both the likelihood and similarity data.The fit of the eight attribute vectors indicated thatthe three dimensions were best interpreted as &dquo;so-

cietal values,&dquo; &dquo;underlying needs,&dquo; and &dquo;extentinherent in work&dquo;; no support was found for theintemal-extemal categorization of work outcomes.

This later finding is particularly interesting inlight of the study by Ronen, Kraut, Lingoes, andAranya (1979). This investigation examined theintercorrelations among importance ratings of 14work outcomes made by 800 salesmen and 1,800repairmen using the SSA-1 program. A two-

dimensional solution was derived separately for thesalesmen and repairmen occupational groups; theCOAs were .14 and .15, respectively. Neighbor-hood interpretations supported several different apriori ~°®upin~s-intrir~sic/extri~sic, Maslow’s needhierarchy, and Alderfer’s existence and relatednessneeds-an eloquent demonstration of the pitfallsof visual interpretation.

These developing applications are not meant tobe an exhaustive representation of multidimen-sional scaling applications in vocational research.

Indeed, a number of additional applications areevident in the literature, including the areas of oc-cupational reinforcers (Rounds, Shubsachs, Dawis,& Lofquist, 1978), perceptions of vocational coun-seling roles (Brook, 1979), potential work mobility(Aranya, Jacobson, 81 Shye, 1976), design of workenvironments (Kenny & Canter, 1981), job anal-ysis and classification (Brown, 1967; Sackett, Cor-nelius & Carron, 1981; Smith & Siegel, 1967), andcareer preferences (Soutar & Clarke, 1983). Thediversity of these applications bodes well for thefuture of multidimensional scaling in vocationalpsychology research.

References

American College Testing Program. Handbook for theACT career planning program. Iowa City IA: Author,1968.

Aranya, N., Jacobson, D., & Shye, S. A Smallest SpaceAnalysis (SSA) of potential work mobility. Multivar-iate Behavioral Research, 1976, 11, 165-173,

Ben-Porat, A. The correlation matrix of job satisfactiontreated by Guttman Smallest Space Analysis. Psycho-logical Reports, 1978, 42, 784-786.

Ben-Porat, A. Event and agent: Towards a structuraltheory of job satisfaction. Personnel Psychology, 1981,34, 523-534.

Billings, R. S., & Cornelius, E. T. Dimensions of workoutcomes: A multidimensional scaling approach. Per-sonnel Psychology, 1980, 83, 151-161.

Brook, J. A repertory grid analysis of perceptions ofvocational counseling roles. Journal of Vocational Be-havior, 1979, 15, 25-35.

Brown, K. R. Job analysis by multidimensional scaling.Journal of Applied Psychology, 1967, 51, 469-475.

Burton, M. Semantic dimensions of occupational names.In A. K. Romney, R. N. Shepard, & S. B. Nerlove

(Eds.), Multidimensional scaling: Theory and appli-cations in the behavioral sciences (Vol. 2). New York:Seminar Press, 1972.

Campbell, D. P. Manual for the Strong- Campbell In-terest Inventory (2nd ed.). Stanford CA: Stanford Uni-versity Press, 1977.

Carroll, J. D. Individual differences and multidimen-sional scaling. In R. N. Shepard, A. K. Romney, &S. B. Nerlove (Eds.), Multidimensional scaling: The-ory and applications in the behavioral sciences (Vol.1). New York: Seminar Press, 1972.

Carroll, J. D., & Arabie, P. Multidimensional scaling.Annual Review of Psychology, 1980, 31, 607-649.



507

Chang, J. J., & Carroll, J. D. How to use MDPREF, a

computer program for multidimensional analysis ofpreference data. Murray-Hill NJ: Bell Laboratories,1974.

Cole, N. S. On measuring the vocational interests ofwomen. Journal of Counseling Psychology, 1973, 20,105-112.

Cole, N. S., & Cole, J. W. L. An analysis of spatialconfiguration and its application to research in highereducation (ACT Research Report No. 35.) Iowa CityIA: The American College Testing Program, 1970.

Cole, N. S., Whitney, D. R., & Holland, J. L. A spatialconfiguration of occupations. Journal of VocationalBehavior, 1971, 1, 1-9.

Cooley, W. W., & Lohnes, P. R. Multivariate data

analysis. New York: Wiley, 1971.Coxon, A. P. M. Occupational attributes: Constructs and

structure. Sociology, 1971, 5, 335-354.Coxon, A. P. M., & Jones, C. L. Occupational simi-

larities : Subjective aspects of social stratification.

Quality and Quantity, 1974, 8, 139-157. (a)Coxon, A. P. M., & Jones, C. L. Problems in the se-

lection of occupational titles. Sociological Review,1974, 22, 369-384. (b)

Coxon, A. P. M., & Jones, C. L. The images of oc-cupational prestige. New York: St. Martin, 1978.

Coxon, A. P. M., & Jones, C. L. Class and hierarchy:The social meaning of occupations. New York: St.

Martin, 1979. (a)Coxon, A. P. M., & Jones, C. L. Measurement and

meaning: Techniques and methods of studying occu-pational cognitions. New York: St. Martin, 1979. (b)

Davison, M. L. On a unidimensional, metric unfoldingmodel for attitudinal and developmental data. Psy-chometrika, 1977, 42, 523-548.

Davison, M. L. Multidimensional scaling versus factoranalysis of tests and items. Paper presented at theannual meeting of the American Psychological As-sociation, Los Angeles, August 1981.

Davison, M. L. Multidimensional scaling. New York:Wiley, 1983.

Davison, M. L., King, P. M., Kitchener, K. S., & Par-

ker, C. A. The stage sequence concept in cognitiveand social development. Developmental Psychology,1980, 16, 121-131.

Davison, M. L. Robbins, S., & Swanson, D. Stagestructure in objective moral judgments. Develop-mental psychology, 1978, 14, 137-146.

Dawis, R. V. Measuring interests. In D. A. Payne (Ed.),New directions for testing and measurement: Recentdevelopments in affective measurement. San Fran-cisco : Jossey-Bass, 1980.

Dawis, R. V., Lofquist, L. H., & Weiss, D. J. A theoryof work adjustment: A revision (Minnesota Studies inVocational Rehabilitation, No. 23). Minneapolis : Uni-

versity of Minnesota, Vocational Psychology Re-search, 1968.

Dawis, R. V., & Weitzel, W. Worker attitudes and ex-pectations. In O. Yoder & H. G. Heneman, Jr. (Eds.),Handbook of personnel management and industrialrelations. Washington DC: Bureau of National Af-fairs, 1974.

Dipboye, W. J., & Anderson, W. F. Occupational ste-reotypes and manifest needs in high school students.Journal of Counseling Psychology, 1961, 8, 296-304.

Dyer, L., & Parker, D. F. Classifying outcomes in workmotivation research: An examination of the intrinsic-extrinsic dichotomy. Journal of Applied Psychology,1975, 60, 455-458.

Edwards, K. J., Nafziger, D. H., & Holland, J. L. Dif-ferentiation of occupational perceptions among dif-

ferent age groups. Journcal of Vocational Behavior,1974, 4, 311-318.

Ekman, G. Dimensions of color vision. Journal of Psy-chology, 1954, 38, 467-474.

Feldman, S., & Meir, E. I. Measuring women’s interestsusing Holland’s vocational classification. Journal ofVocational Behavior, 1976, 9, 345-353.

Flanagan, J. C., Jr., Shaycoff, M. F., Richards, J. M.,Jr., & Claudy, J. G. Five years after high school.Palo Alto CA: American Institutes for Research, 1971.

Gati, I. A hierarchical model for the structure of voca-tional interests. Journal of Vocational Behavior, 1979,15, 90-106.

Gati, I. Testing models for the structure of vocationalinterests. Journal of Vocational Behavior, 1982, 21,164-182.

Gonyea, G. G., & Lunneborg, C. E. A factor analyticstudy of perceived occupational similarity. Journal ofApplied Psychology, 1963, 47, 166-172.

Gottfredson, L. S. Vocational research priorities. TheCounseling Psychologist, 1982, 10, 69-84.

Grunes, W. F. Looking at occupations. Journal of Ab-normal and Social Psychology, 1957, 54, 86-92.

Guilford, J. P., Christensen, P. R., Bond, N. A., &Sutton, M. A. A factor analytic study of human in-terests. Psychological Monographs, 1954, 68 (4, WholeNo. 375).

Guttman, L. An outline of some new methodology insocial research. Public Opinion Quarterly, 1954, 18,395-404.

Guttman, L. A generalized nonmetric technique for find-ing the smallest coordinate space for a configurationof points. Psychometrika, 1968, 33, 469-506.

Haase, R. F., Reed, C. F., & Winer, J. L. Effect of

positive, negative, and mixed occupational informa-tion on cognitive and affective complexity. Journalof Vocational Behavior, 1979, 15, 294-302.

Hall, J., & Jones, D. C. Social grading of occupations.British Journal of Sociology, 1951, 1, 31-35.



508

Hanson, G. R., & Cole, N. S. (Eds.), The vocationalinterests of young adults (Monograph No. 11). IowaCity IA: American College Testing Program, 1973.

Hartmann, M. The grading of occupations by sociolo-gists. Paper presented at the annual meeting of theIsraeli Sociological Association, 1975. (in Hebrew)

Holcomb, W. R., & Anderson, W. P. Vocational guid-ance research: A five-year overview. Journal of Vo-cational Behavior, 1977, 10, 341-346.

Holland, J. L. Professional manualfor the Self-DirectedSearch. Palo Alto CA: Consulting psychologists Press,1972.

Holland, J. L. Making vocational choices: A theory ofcareers. Englewood Cliffs NJ: Prentice Hall, 1973.

Holland, J. L. Vocational preferences. In M. D. Dun-nette (Ed.), Handbook of industrial and organiza-tional psychology. Chicago: Rand McNally, 1976.

Holland, J. L. Professional manualfor the Self-DirectedSearch. Palo Alto CA: Consulting Psychologists Press,1979.

Jepsen, D. A., & Grove, W. M. Stage order and dom-inance in adolescent vocational decision-making pro-cesses : An empirical test of the Tiedeman-O’Hara par-adigm. Journal of Vocational Behavior. 1981, 18,237-251.

Katz, R., & Maanen, J. V. The loci of work satisfaction:Job, interaction, and policy. Human Relations, 1977,30, 469-486.

Kenny, C., & Canter, D. A facet structure for nurses’evaluation of ward designs. Journal of OccupationalPsychology, 1981, 54, 93-108.

Krau, E. The vocational side of a new start in life: Acareer model of immigrants. Journal of VocationalBehavior, 1982, 20, 313-330,

Kraus, V., Schild, E. O., & Hodge, R. W. Occupa-tional prestige in the collective conscience. Social

Forces, 1978 , 56, 900-918.Krumhansi, C. L. Concerning the applicability of geo-

metric models to similarity data: The interrelationshipbetween similarity and spatial density. PsychologicalReview, 1978, 85, 445-463.

Kruskal, J. B. Multidimensional scaling by optimizinggoodness of fit to a nonmetric hypothesis. Psycho-metrika, 1964, 29, 1-27; 115-129.

Kruskal, J. B. The relationship between multidimen-sional scaling and clustering. In J. Van Ryzin (Ed.),Classification and clustering. New York: AcademicPress, 1977.

Kruskal, J. B., & Carmone, F. How to use M-D-SCAL(Version 5M) and other useful information. MurrayHill NJ: Bell Laboratories, 1969.

Kruskal, J. B., & Wish, M. Multidimensional scaling.Beverly Hills CA: Sage, 1978.

Lingoes, J. C. An IBM 7090 program for Guttman-Lin-goes Smallest Space Analysis-I. Behavioral Sciences,

1965, 10, 183-184.Lofquist, L. H., & Dawis, R. V. Adjustment to work.New York: Appleton-Century-Crofts, 1969.

Lunneborg, C. E., & Lunneborg, p. W. Factor structureof the vocational interest models of Roe and Holland.Journal of Vocational Behavior, 1975, 7, 313-326.

Lunneborg, P. W. Manual for the Vocational InterestInventory. Seattle: University of Washington, Edu-cational Assessment Center, 1975.

Luxenberg, M., & Zevon, M. A. The structure of moodcovariation patterns over time via an individual dif-ference multidimensional scaling analysis. Paper pre-sented at the annual meeting of the Psychometric So-ciety, Montreal, Canada, 1982.

MacCallum, R. C. Relations between factor analysis andmultidimensional scaling. Psychological Bulletin, 1974,81, 505-516.

MacCallum, R. C. Recovery of structure in incompletedata by ALSCAL. Psychometrika, 1978, 44, 69-74.

Meir, E. I. The structure of occupations by interests—A Smallest Space Analysis. Journal of VocationalBehavior, 1973, 3, 21-31.

Meir, E. I., Bar, R., Lahav, G., & Shalhevet, R. Interestinventories based on Roe’s classification modified for

negative respondents. Journal of Vocational Behav-ior, 1975, 7, 127-133.

Meir, E. I., & Barak, A. A simple instrument for mea-suring vocational interests based on Roe’s classifica-tion of occupations. Journal of Vocational Behavior,1974, 4, 33-42.

Meir, E. I. , & Ben-Yehuda, A. Inventories based onRoe and Holland yield similar results. Journal of Vo-cational Behavior, 1976, 8, 269-274.

Meir, E. I., Sohlberg, S., & Barak, A. A cross-culturalcomparison of the structure of vocational interests.Journal of Cross-Cultural Psychology, 1973, 4, 501-508.

Osipow, S. H. Theories of career development (3rd ed.).Englewood Cliffs NJ: Prentice-Hall, 1983.

Prediger, D. J. Dimensions underlying Holland’s hex-agon : Missing link between interests and occupations?Journal of Vocational Behavior, 1982, 21, 259-287.

Reeb, M. How people see jobs: A multidimensionalanalysis. Occupational Psychology, 1959, 33, 1-17.

Reeb. M. Multidimensional perceptions by counselorsand 14-year-old boys of suitability for jobs, and theirprestige and desirability. Occupational Psychology,1971, 45, 233-242.

Reeb, M. The perceptions of occupational structure—

an intervening variable in vocational behavior. Jour-nal of Vocational Behavior, 1974, 4, 125-137.

Reeb, M. Occupational perception and school stream inrelation to job classifications, preferences, and pres-tige for 17-year-old girls and boys. Journal of Oc-cupational Psychology, 1979, 52, 113-127.



509

Roe, A. The psychology of occupations. New York:Wiley, 1956.

Roe, A., & Klos, D. Occupational classification. Coun-seling Psychologist, 1969, 1, 84-89.

Ronen, S., Kraut, A. I., Lingoes, J. C., & Aranya, N.A nonmetric scaling approach to taxonomies of em-ployee work motivation. Multivariate Behavioral Re-search, 1979, 14, 387-401.

Rosenberg, S., & Kim, M. P. The method of sorting asa data-gathering procedure in multivariate research.Multivariate Behavioral Research, 1975, 10, 489-502.

Rounds, J. B., Jr., Davison, M. L., & Dawis, R. V.The fit between Strong-Campbell Interest InventoryGeneral Occupational Themes and Holland’s hexa-gonal model. Journal of Vocational Behavior, 1979,15, 303-315.

Rounds, J. B., Jr., & Dawis, R. V. Factor analysis ofStrong Vocational Interest Blank items. Journal ofApplied Psychology, 1979, 64, 132-143.

Rounds, J. B., Jr., & Dawis, R. V. Multidimensionalscaling solutions for the Self-Directed Search (1979ed.). Unpublished manuscript, 1980.

Rounds, J. B., Jr., Shubsachs, A. P. W., Dawis, R. V.,& Lofquist, L. H. A test of Holland’s environmentalformulations. Journal of Applied Psychology, 1978,63, 609-616.

Sackett, P. R., Cornelius, E. T., & Carron, T. J. A

comparison of global judgment vs. task oriented ap-proaches to job classification. Personnel Psychology,1981, 34, 791-804.

Sattath, S., & Tversky, A. Additive similarity trees.Psychometrika, 1977, 42, 319-345.

Schneider, J., & Locke, E. A. A critique of Herzberg’sincident classification system and a suggested revi-sion. Organizational Behavior and Human Perfor-mance, 1971, 6, 441-457.

Shapira, Z., & Zevulun, E. On the use of facet analysisin organizational behavior research: Some conceptualconsiderations and an example. Organizational Be-havior and Human Performance, 1979, 23, 411-428.

Shepard, R. N. Introduction to volume 1. In R. N. Shep-ard, A. K. Romney, & S. B. Nerlove (Eds.) Multi-dimensional scaling: Theory and applications in thebehavioral sciences (Vol. 1). New York: Seminar Press,1972.

Shepard, R. N., & Arabie, P. Additive clustering: Rep-resentation of similarities as combinations of discrete

overlapping properties. Psychological Review, 1979,86, 87-123.

Shubsachs, A. P. W., & Davison, M. L. Individual dif-ferences in perceptions of occupations and occupa-tional reinforcers. Journal of Occupational Psychol-ogy, 1979, 52, 299-310.

Siess, T. F., & Rogers, T. B. Roe’s classification and

the multidimensional nature of occupational percep-tions. Journal of Vocational Behavior, 1974, 4, 403-415.

Smith, R. J., & Siegel, A. I. A multidimensional scalinganalysis of the job of civil defense direction. Journal

of Applied Psychology, 1967, 51, 476-480.Soutar, G. N., & Clarke, A. W. Examining business

students’ career preferences: A perceptual space ap-proach. Journal of Vocational Behavior, 1983, 23,11-21.

Spence, I., & Domoney, D. W. Single subject incom-plete designs for nonmetric multidimensional scaling.Psychometrika, 1974, 39, 469-490.

Stone, L. A., & Bassett, G. R. The MEDEX occupationconcept in subjective medical occupation-professionmultidimensional space. Psychological Reports, 1972,31, 167-174.

Takane, Y., Young, F. W., & de Leeuw, J. Nonmetricindividual differences multidimensional scaling: Analternating least squares method with optimum scalingfeatures. Psychometrika, 1977, 42, 7-67.

Tiedeman, D. V., & O’Hara, R. P. Career develop-ment : Choice and adjustment. New York: CollegeEntrance Examination Board, 1963.

Torgerson, W. S. Multidimensional scaling: I. Theoryand method. Psychometrika, 1952, 17, 401-419.

Torgerson, W. S. Theory and methods of scaling. NewYork: Wiley, 1958.

Torgerson, W. S. Multidimensional scaling of similar-ity. Psychometrika. 1965, 30, 379-393.

Tversky, A., & Gati, I. Studies of similarity. In E. Rosch& B. B. Lloyd (Eds.), Cognition and categorization.Hillsdale NJ: Erlbaum, 1978.

U.S. Department of Labor. Dictionary of OccupationalTitles (4th ed.). Washington DC: U.S. GovernmentPrinting Office, 1977.

Wakefield, J. A., & Doughtie, E. B. The geometric re-lationship between Holland’s personality typology andthe Vocational Preference Inventory. Journal of Coun-seling Psychology, 1973, 20, 513-518.

Westbrook, B. W. Career maturity: The concept, the

instrument, and the research. In W. B. Walsh & S. H.

Osipow (Eds.), Handbook of vocational psychology(Vol. 1). Hillsdale NJ: Erlbaum, 1983.

Wish, M., Deutsch, M., & Biener, L. Differences inperceived similarity of nations. In A. K. Romney,R. N. Shepard, & S. B. Nerlove (Eds.), Multidimen-sional scaling: Theory and application in the behav-ioral sciences (Vol. 2). New York: Seminar Press,1972.

Young, F. W., & Torgerson, W. S . TORSCA, a FortranIV program for Shepard-Kruskal multidimensional

scaling analysis. Behavioral Science, 1967, 12, 498.Zevon, M. A., Luxenberg, M., & Rounds, J. B., Jr.Mood covariation patterns over time: A comparison



510

of multidimensional scaling and factor analytic so-lutions. Manuscript submitted for publication, 1983.

Acknowledgments

Preparation of this paper was supported in part by theSpencer Foundation. The authors thank Elaine Shislerfor her assistance in the preparation of the manuscript.

Author’s Address

Send requests for reprints or further information to JamesB. Rounds, Jr., Department of Counseling and Educa-tional Psychology, 416 Christopher Baldy Hall,SUNYAB, Buffalo NY 14260, U.S.A.



multidimensional scaling research vocational psychology · 2016. 5. 18. · 491 multidimensional...

Documents