The Spherical Model of Vocational Interests in Germany

Abstract

The aim of the present investigation is to examine key aspects regarding the validity of the spherical model of vocational interests in Germany, namely, its structural validity, its convergent validity—with an instrument assessing vocational interests according to the RIASEC (R—Realistic, I—Investigative, A—Artistic, S—Social, E—Enterprising, and C—Conventional) model—and its construct validity regarding the pattern of gender differences. To this end, the Personal Globe Inventory (PGI) was translated and completed by a sample of German university students. Results of randomization tests of hypothesized order relations provided support for the structural validity of the instrument in female and male university students. Principal components analyses with target rotations identified two particular scales as outliers. RIASEC scores derived from the PGI correlated strongly with corresponding scales of another instrument, and gender differences were in line with previous findings reported in the literature. Overall, our results provide evidence for the validity of the spherical model in German university students.

Keywords

interest structure personal globe inventory spherical model RIASEC gender differences

Vocational interests have proven to be important characteristics of human behavior and perceptions in organizational and educational settings (e.g., Rounds & Su, 2014; Warwas, Nagy, Watermann, & Hasselhorn, 2009). Findings in both fields consistently document the importance of vocational interests in predicting educational and occupational choices and demonstrate the relevance of interest congruence for educational and professional success (e.g., Etzel & Nagy, 2016; Lee, Lawson, & McHale, 2015; Tracey & Robbins, 2006).

Research on this topic has created a large body of theories that attempt to group the extensive range of vocational interests into meaningful interest domains. Besides Holland’s famous RIASEC model (1997), postulating six vocational personality types (R—Realistic, I—Investigative, A—Artistic, S—Social, E—Enterprising, and C—Conventional), the works of Knapp and Knapp (1974), Meir (1973), and Roe (1956) provided well-known systems for grouping vocational interests. However, recent works on the structure of vocational interests emphasize that the key aspect of their similarity structure is best described by a continuous model, while the specification of groups of vocational interest is, to a certain degree, arbitrary (Prediger, 1982; Tracey & Rounds, 1995). Today, there is ample evidence that the similarity structure of vocational interests can be described by their locations on the perimeter of a circle (i.e., a circumplex; Guttman, 1954). Interests that are located closer to each other are more similar than those that are further apart from each other (Holland, 1997). There is extensive empirical support for the structural validity of the circumplex model of vocational interests across different countries and social groups (e.g., Day & Rounds, 1998; Nagy, Trautwein, & Lüdtke, 2010; Rounds & Tracey, 1996).

Circumplex models are often described by two orthogonal axes, representing the qualities underlying the circularly ordered constructs. Regarding vocational interests, the most influential model was introduced by Prediger (1982), who specified one axis to indicate the preference for interpersonal versus impersonal tasks (People–Things) and another axis to describe the preference for internal versus external tasks (Ideas–Data). This framework allows for the integration of vocational interests with other variables (Armstrong, Day, McVay, & Rounds, 2008) and has proven to be useful for analyzing gender differences (Lippa, 1998).

Despite its long tradition and strong empirical support, the circumplex model of vocational interests has the limitation that it does not account for a person’s preferred level of occupational prestige, an aspect that has repeatedly been identified as an important facet in subjective perceptions of occupations (Gottfredson, 1996). In order to integrate this aspect into the realm of vocational interests, Tracey and Rounds (1996) introduced a spherical model of vocational interests. In this model, the interest circumplex, defined by Prediger’s People–Things and Ideas–Data axes, is expanded by an orthogonal axis representing a person’s preference for vocational activities or occupations at different levels of prestige. Occupational prestige is thereby understood as a broad construct, summarizing—inter alia—concepts of socioeconomic status, level of difficulty and responsibility, as well as required education and training. Our study addresses the question of whether there is empirical support for the validity of this model in Germany.

The Personal Globe Inventory (PGI)

The PGI (Tracey, 2002) is an instrument used to assess vocational interests as conceptualized by the spherical model. It consists of eight basic scales at an intermediate prestige level (Social Facilitating, Managing, Business Detail, Data Processing, Mechanical, Nature/Outdoors, Artistic, and Helping), five higher prestige scales (Social Sciences, Science, Business Systems, Financial Analysis, and Influence), and five lower prestige scales (Quality Control, Basic Service, Personal Service, Construction/Repair, and Manual Work). Prediger’s People–Things and Ideas–Data axes constitute the equator of the sphere with the eight basic interest scales distributed on its perimeter. While the Influence and Manual Work scales define the upper and lower poles of the sphere, the remaining higher and lower prestige scales are assumed to exhibit a circumplex structure at different levels of prestige. Figure 1 displays the theoretical hemispheres as viewed from above (higher prestige scales) and below (lower prestige scales).

Figure 1.
Spherical model of vocational interests. Concentric circles (from outer to inner) are at 0°, 45°, and 67.5° latitudes.

Three different item types can be assessed with the PGI: activity preferences, activity competence beliefs, and occupational preferences. Additionally, a composite of the three item types can be derived (i.e., their average). The hypothesized spherical structure is said to hold for all item types. Furthermore, it is possible to convert the PGI scale scores into RIASEC scores. This allows for the comparison of the PGI with other RIASEC-based interest inventories. Empirical support for the cross-cultural structural and convergent validity of the instrument has been found in English-speaking countries (Darcy, 2005; Tracey, 2002), the Balkan States (Hedrih, 2008; Šverko, 2008), China (Long, Adams, & Tracey, 2005), and the Netherlands (Holtrop, Born, & de Vries, 2015).

However, a number of studies indicate cross-cultural and group-specific differences for the circumplex model (e.g., Hansen, Collins, Swanson, & Fouad, 1993; Leong, Austin, Sekaran, & Komarraju, 1998; Tracey, Watanabe, & Schneider, 1997) and the spherical model (Wilkins, Ramkissoon, & Tracey, 2013). Researchers must therefore challenge the validity of the structural invariance assumption before applying an instrument in a new context, especially if it has been translated into another language. The validity of cross-cultural comparisons of interest-outcome relationships is only ensured if the invariance assumption holds to a reasonable extent. While there is support for the validity of the two-dimensional circumplex structure of vocational interests in Germany (Nagy et al., 2010), there is not yet any such evidence for the validity of the spherical model.

The Present Investigation

The present research addresses the validity of the spherical model of vocational interests—as assessed by the PGI—in Germany, by analyzing a sample of N = 526 university students. Following the abovementioned argumentation concerning the theoretical configuration of the 18 PGI scales, we expected to find evidence for the structural validity of the circumplex structure of the six RIASEC scales, the eight basic interest scales, the spherical structure of the 18 spherical scales, and, thus, the underlying dimensions: People–Things, Ideas–Data, and Prestige. In line with the findings of other authors, the structure was expected to be invariant across gender groups and item types (e.g., Darcy & Tracey, 2007). We employed confirmatory methods to assess global model fit as well as exploratory graphical methods to identify the PGI scales responsible for the misfit between the theoretical and the empirical interest structure.

In order to examine the convergent validity of the PGI, we analyzed its relationship to another well-established interest inventory: the General Interest Structure Test (GIST; Allgemeiner Interessen-Struktur test; Bergmann & Eder, 2005). The GIST is the state-of-the-art interest inventory in Germany for assessing vocational interests according to Holland’s RIASEC model. Extensive support (Nagy et al., 2010) has been found for its structural validity. We expected to find strong associations between RIASEC scores derived from the PGI and the GIST.

Finally, we investigated gender differences in the PGI scales, thereby examining whether the pattern of gender differences corresponds to the results known from the literature. Building upon the findings of Lippa (1998), we expected the majority of gender differences to concentrate on scales located near to the poles of the People–Things axis. On average, female participants were expected to score higher on interest domains that are located close to the People pole, whereas male participants were expected to score higher on those domains close to the Things pole. For scales located near the poles of the Data–Ideas axis, we expected to find nonsignificant or at least smaller gender differences. To the best of our knowledge, the relationship between gender and the Prestige axis has not yet been examined systematically. We therefore approached this issue as an open research question.

Method

Participants and Procedure

We distributed an online questionnaire via social media platforms, specifically posting the link in groups of various majors at universities all over Germany. The instruments in question were embedded in a larger study aiming to assess the psychometric quality of additional scales that were not relevant for the present research (e.g., trait self-control). To account for the length of the instruments, we decided to implement a multiple group design with planned missing data. Participants were randomly assigned to one of the three groups consisting of (1) the complete PGI and the GIST; (2) the PGI activity items, the GIST, and additional measures, and (3) the PGI occupation items and a larger set of additional measures. Thus, all relevant instruments were assessed in at least two of the three groups.

During the survey period, 939 people opened the link to the questionnaire. In a first step, we excluded those participants who did not reach the psychometric part of the questionnaire (n = 182) and those who could not be identified as students (n = 69). In addition, we excluded participants who dropped out of the survey very early (n = 162). We defined an early dropout as someone who did not answer at least half of the first PGI instrument. The rationale behind this cutoff is coherent with the research question. Due to the order of the items, participants who did not reach the second half of the first instrument did not provide any information for the spherical scales. The final sample consisted of N = 526 university students (66.54% female) from over 50 different majors. The average age was 22.44 with a standard deviation of 3.18. Coverage across the three groups was balanced (Group 1: n = 158, Group 2: n = 169, and Group 3: n = 199). Note that all analyses were conducted on multiply imputed data sets with N = 526 per set.

There were no significant differences between the final sample and the early dropouts with respect to gender, χ²(1, 685) = 1.25, p = .26], number of semesters enrolled, t(644) = −1.20, p = 23), and average school grade, t(675) = .866, p = .39. The age difference between our final sample and the dropouts was significant, [t(683) = −2.57, p < .01, but of negligible magnitude (|Δ_M| = .72).

Measures

PGI

The instrument consists of 108 activity preference items, 108 activity competence belief items, and 108 occupational preferences items. Participants were asked to indicate how much they like the activities or occupations on a 7-point rating scale ranging from 1 = very strongly dislike to 7 = very strongly like. For the activity items, participants were also asked to rate their perceived competence on a 7-point rating scale ranging from 1 = unable to do to 7 = very competent. The items are organized in the same 18 scales for all item types and their composite (see Figure 1).

We translated the instrument using a translation-back translation approach: First, we translated the original items from English into German. These translations were then given to a native English-speaker working at our institute and translated back into English. The back translations were then compared with the original items and discrepancies were discussed. On this basis, we revised the instrument and created the final version. In our sample, all Cronbach’s α coefficients were acceptable to very good (α = .70 to α = .95) with two minor exceptions: the activity item types for the Personal Service subscale (α = .61 and α = .65).

GIST

The GIST (Bergmann & Eder, 2005) is an instrument used to assess vocational interests according to Holland’s RIASEC types. The instrument has been extensively validated in Germany. It consists of 60 items describing different work activities. Participants are asked to rate how interested they are in certain activities on a 5-point scale. In our sample, internal consistencies were high throughout the six scales (α = .84 to α = .91). The authors provide strong evidence for the internal and convergent validity of the instrument. In addition, the circumplex structure of the GIST is well established in Germany (Nagy et al., 2010).

Statistical Analyses

Missing data

Missing data were handled via multiple imputations with the mice package (van Buuren & Groothuis-Oudshoorn, 2011) implemented in the R statistical software (Version 3.2.1). Due to the large number of items, we decided to impute on the scale level. We computed the scale means prior to imputation, using all available information. This method yields results with a similar precision to that of item-level imputation, while reducing computational effort (Enders, 2010). The background variables used in the imputation model were sex, age, and several school grades. The analyses of the convergence of the algorithm were inconspicuous. Using this method, we created 100 complete data sets.

Spherical structure

In order to analyze the structure hypothesized by the spherical model of interests, we first conducted a series of randomization tests of hypothesized order relations (Hubert & Arabie, 1987) for the RIASEC scales retrieved from the PGI, the eight-type basic interest circumplex, and the 18-type spherical model. The logic of this test (hereafter only referred to as the randomization test) is to translate the hypothesized structure into pairwise order predictions of the correlations between the scales and to compare them with the order in the sample correlation matrix. The result of this test is summarized by a correspondence index (CI) that ranges from −1 to 1, indicating the proportion of predictions met (−1 = no predictions met, 1 = all predictions met).

Furthermore, it is possible to test whether or not the number of violated order relations in the sample correlation matrix is smaller than it would be if the rows and columns of the correlation matrix were randomly relabeled. We analyzed the male and female subsamples separately, as well as the full sample, using the RANDALL program (Tracey, 1997). All analyses were conducted on the pooled correlation matrices, derived by averaging the correlations across the 100 imputations.

Because tests of global model fit, such as the randomization test, typically do not yield perfect congruence between the data and the hypothesized model, we additionally conducted principal components analyses of the full sample correlation matrices for the three item types and their composite. The resulting component loadings allow for a more detailed examination of the relationships between the empirical data and the theoretical model. In line with Prediger (1998), the first component was expected to be a general response factor without substantial meaning. The second to fourth components were expected to constitute the three dimensions of the interest sphere. In order to maximize the comparability between the component loadings and the theoretical model, we standardized the vector lengths of each scale to 1. This was achieved by dividing the three component loadings λ_j of each scale j by the square root of the sum of squared component loadings (i.e., the vector length with respect to the three components: $\sqrt{λ_{j 1}^{2} + λ_{j 2}^{2} + λ_{j 3}^{2}}$ ). This procedure ensures that each scale is located on the surface of a unit sphere, just like the scales of the theoretical model. Finally, we performed orthogonal target rotations, using the R package “GPArotation” (Bernaards & Jennrich, 2005), thereby optimizing the correspondence between the theoretical and the empirical configuration.

The target matrix was constructed by identifying the spherical coordinates of the theoretical model (as displayed in Figure 1) and translating them into Cartesian coordinates. The Ideas pole served as the reference position for longitude (0° at three o’clock). We then went in steps of 45° counterclockwise around the circumplex, starting with the Artistic scale at 22.5°, to identify the eight basic scales. For latitude, we defined the Influence scale as the upper pole (0°) and the Manual Work scale as the lower pole (180°). Accordingly, the latitude for the eight basic scales was 90° because they are located at the equator of the sphere. For the lower and higher prestige scales, the first scale is located in between the Ideas and People poles (45° longitude). The other scales are separated by 90°, moving counterclockwise. To account for the difference in Prestige, the latitudes were set to 135° (lower prestige) and 45° (higher prestige). Given the 18 pairs of spherical coordinates and setting the radius to 1, we could determine the Cartesian coordinates according to the theoretical model.

The results of the target rotated component loadings were then compared with the theoretical scale positions in two ways: First, we graphically analyzed the congruence between the empirical and the theoretical scale positions. Second, we calculated the Euclidean distances between the empirical and the theoretical scale positions in order to quantify the deviation with respect to all three dimensions. In order to define what is considered a large deviation from theory, we specified a Euclidean distance of Δ_E = .765 as the cutoff criterion, which is the distance between two adjacent scales separated by the smallest possible angle in the theoretical model (i.e., 45°).

Gender differences and convergent validity

The convergent validity of the RIASEC scales of the PGI with the GIST RIASEC scales was examined by means of a monotrait–heteromethod correlation matrix, using the composite scores. For the analysis of gender differences, we used linear regressions of the PGI scales on a dummy coded indicator variable (1 = male, 2 = female). The regression weights were standardized in the dependent variable (i.e., y standardized), thus representing gender differences in the units of the scales’ standard deviations. In order to address whether or not gender was related to the Prestige axis, gender effects on PGI scales located on the same level of prestige were averaged (Figure 1) and the corresponding average effects were compared with each other. A systematic relationship would be manifested in a monotonous trend in average effects along the prestige axis (e.g., the lower the level of prestige, the larger the differences in favor of males would become). All of the analyses were conducted using the imputation module implemented in the Mplus 7.1 software (Muthén & Muthén, 2012). This module combines the parameter estimates of the analyses of the 100 imputed data sets according to the rules of Rubin (1987) and yields appropriate estimates of their standard errors.

Results

Spherical Structure: Randomization Tests

Table 1 presents the results of the randomization tests. In our sample, each of the 36 randomization tests was significant. For the PGI-based RIASEC model, the CIs ranged from 0.76 to 1.00 (p = .02). For the eight-type model, the CIs values ranged from 0.73 to 0.90 (p = .00). Finally, for the 18-type spherical model, the CI values ranged from 0.47 to 0.60 (p = .00). Overall, the activity competence item type yielded the lowest, yet significant, correspondence with the theoretical model, with a median of .76. CIs for the other item types and their composite were similar, with medians ranging from .85 to .87.

Table 1.
Results of the Randomization Tests of Hypothesized Order Relations.

Activity Preferences Activity Competence Beliefs Occupational Preferences Composite

All Female Male All Female Male All Female Male All Female Male

RIASEC model

Predictions made 72 72 72 72 72 72 72 72 72 72 72 72

Predictions met 67 67 68 66 68 63 69 65 72 68 66 69

CI 0.86 0.86 0.89 0.83 0.89 0.76 0.93 0.82 1.00 0.89 0.83 0.92

8-Type model

Predictions made 288 288 288 288 288 288 288 288 288 288 288 288

Predictions met 273 273 263 260 260 246 274 270 265 279 280 269

CI 0.90 0.90 0.84 0.82 0.83 0.73 0.90 0.89 0.85 0.94 0.95 0.87

18-Type model

Predictions made 9,472 9,472 9,472 9,472 9,472 9,472 9,472 9,472 9,472 9,472 9,472 9,472

Predictions met 7,179 7,109 7,318 6,910 6,964 6,859 7,410 7,284 7,424 7,398 7,356 7,510

CI 0.53 0.51 0.56 0.47 0.48 0.47 0.58 0.55 0.58 0.57 0.56 0.60

Note. N = 526 (full sample), n = 350 (female subsample), n = 176 (male subsample). RIASEC = R—Realistic, I—Investigative, A—Artistic, S—Social, E—Enterprising, and C—Conventional; CI = correspondence index.

RIASEC models: p = .02. all 8-type models: p = .00. all 18-type models: p = .001.

Although the correspondence indices of the RIASEC model (CI = .92 vs. .83) and the 18-type model (CI = .60 vs. .56) were slightly higher for male participants, those of the 8-type model (CI = .95 vs. .87) were higher for female participants. Taken together, these results provided support for the validity of the model across the different gender subsamples and item types. However, the moderate CIs indicated local deviations from the theoretical model that were investigated deeper in the subsequent analyses.

Spherical Structure: Principal Components Analyses

The subsequent analyses were conducted for the gender subsamples and the full sample. As the results were congruent across gender, we only report the results for the full sample here. Separate results are available upon request. The eigenvalues and variance accounted for of the principal components analyses for the three item types and their composite (N = 526) are presented in Table 2. As expected, the first component represented a general component with overall positive loadings. In line with Prediger (1998), this general component is considered irrelevant and was hereafter ignored. Overall, the correlations between the target rotated principal components loadings and the targets were high for all three item types, yielding strong support for the validity of the three dimensions as defined by the spherical model. The agreement with the theoretical scale positions was highest for the occupational preference item and the composite (r = .91 for both), followed by the activity preference items (r = .89), and the activity competence beliefs items (r = .84).

Table 2.
Summary of the Principal Components Analyses.

C1 C2 C3 C4

Activity preferences

Eigenvalue 5.30 3.32 2.50 1.31

VAF 0.29 0.18 0.14 0.07

∑ VAF 0.29 0.48 0.62 0.69

Activity competence beliefs

Eigenvalue 6.52 3.15 1.81 1.13

VAF 0.36 0.17 0.10 0.06

∑ VAF 0.36 0.54 0.64 0.70

Occupational preferences

Eigenvalue 6.87 3.07 2.58 1.36

VAF 0.38 0.17 0.14 0.08

∑ VAF 0.38 0.55 0.70 0.77

Composite

Eigenvalue 5.75 3.79 2.67 1.37

VAF 0.32 0.21 0.15 0.08

∑ VAF 0.32 0.53 0.68 0.75

Note. N = 526. C1 = general component. C2 = People–Things. C3 = Ideas–Data. C4 = prestige. VAF = variance accounted for.

Figures 2 and 3 display the empirical versus the theoretical scale positions for the upper and lower hemisphere for each of the three item types and their composite. The theoretical model was not perfectly reproduced although the majority of the scales were close to their theoretical positions. In order to identify meaningful aberrations from theory, we used the Euclidean distances (Δ_E) between the empirical and the theoretical scale positions.

Figure 2.
Theoretical versus empirical scale positions: Activity item types. ▪ = theoretical positions, • = empirical positions on displayed hemisphere, and º = empirical positions on opposite hemisphere.

Figure 3.
Theoretical versus empirical scale positions: Occupational preferences and composite. ▪ = theoretical positions, • = empirical positions on displayed hemisphere, and º = empirical positions on opposite hemisphere.

Table 3 presents the Euclidean distances for the three item types and their composite. Given our cutoff value of Δ_E = 0.765, we only found three noticeable deviations from the theoretical model: The Influence scale for the activity competence beliefs item type shifted away from the upper pole of the sphere in the Data and People direction (Δ_E = 1.11). The Financial Analysis scale for the activity preferences item type shifted away from the People pole toward the Things pole (Δ_E = 0.90). Likewise, the Financial Analysis scale for the activity competence beliefs item type shifted from the People pole toward the Things pole (Δ_E = 0.99). Noticeably, not one scale exceeded our cutoff for the composite score and the occupational preference item types.

Table 3.
Euclidean Distances Between Theoretical Scale Positions and Target Rotated Loadings.

Scale Activity Preferences Activity Competence Beliefs Occupational Preferences Composite

Social facilitating .36 0.45 .13 .33

Managing .26 0.52 .11 .20

Business detail .46 0.72 .53 .53

Data processing .59 0.33 .19 .31

Mechanical .39 0.43 .43 .39

Nature/outdoors .28 0.33 .42 .25

Artistic .15 0.28 .15 .24

Helping .07 0.15 .12 .03

Social sciences .63 0.75 .50 .60

Influence .52 1.11 .58 .40

Business systems .37 0.52 .38 .41

Quality control .35 0.04 .70 .27

Manual work .46 0.31 .19 .34

Personal service .70 0.57 .62 .62

Financial analysis .90 0.99 .58 .65

Science .58 0.60 .29 .47

Construction/repair .25 0.39 .48 .28

Basic service .57 0.68 .41 .75

Note. Euclidean distances exceeding the cutoff are printed in boldface.

Convergent Validity: Monotrait–Heteromethod Correlations

Although the GIST is well established in Germany, it was advisable to validate the circumplex structure in our sample. We therefore conducted a randomization test for the GIST in the full sample. This test yielded a CI of .74 (p = .02), thus supporting the circumplex structure in our sample. However, this CI was lower than that of the composite PGI RIASEC model (CI = .89), indicating a better fit for the latter.

Table 4 displays the monotrait–heteromethod correlation matrix of the RIASEC scales calculated from the PGI composite scores with the RIASEC scales of the GIST. The correlation structure was very similar for all item types and across gender subsamples (correlation matrices for the other item types are available on request). The diagonal elements of the monotrait–heteromethod matrix serve as indicators for the validity of the respective scales. We found strong evidence for the convergent validity of the Realistic (r = .63, p < .001), Investigative (r = .66, p < .001), Artistic (r < .79, p < .001), Social (r = .77, p < .001), and Enterprising (r = .61, p < .001) scales. The only interest domain that yielded a higher correlation with a different scale than its counterpart was the PGI Conventional scale, which correlated higher with the GIST Enterprising scale (r = .50, p < .001) than with the GIST Conventional scale (r = .32, p < .001).

Table 4.
Monotrait–Heteromethod Correlation Matrix for the RIASEC Scales.

Scale GIST-R GIST-I GIST-A GIST-S GIST-E GIST-C

PGI-R .63* .34* −.14 −.24* −.04 .08

PGI-I .42* .66* .17* −.02 −.06 .00

PGI-A .10 .11 .79* .24 .10 −.10

PGI-S −.06 −.10 .47* .77* .39* .03

PGI-E .08 −.14 .17 .40* .61* .50*

PGI-C .45* .29* −.23* −.25* .12 .32*

Note. Corresponding scales are printed in boldface. R = realistic, I = investigative, A = artistic, S = social, E = enterprising, C = conventional; PGI = Personal Globe Inventory (Composite); GIST = General Interest Structure Test.

p < .001.

In order to elaborate on whether this inconsistency contests the validity of the PGI or can be attributed to the GIST, we took a closer look at the items of the Enterprising and Conventional scales of the GIST by means of a principal components analysis. We found that those items of the GIST Conventional scale that had no significant correlations with the PGI Conventional scale defined a third component. Excluding these items from the GIST Conventional scale increased the correlation with the PGI Conventional scale (r = .48, p < .001), rendering the difference between the two correlations insignificant. The excluded items were rather unspecific (e.g., “Doing work that requires precision and perseverance”), thus providing a reason to attribute the source for the incongruity to the GIST rather than the PGI.

Gender Differences

Figure 4 displays the y-standardized regression coefficients of the gender mean differences for the three item types. In order not to overburden the figure, we decided to omit the gender differences for the composite score because they roughly correspond to the average of the presented effects. Gender was dummy coded (1 = male, 2 = female*), which means that regression weights of positive sign indicate higher means in female participants. The 18 scales are presented in descending order, progressing from the upper pole (Influence), through the higher prestige scales, the eight basic scales, and the lower prestige scales toward the lower pole (Manual Work). At each level of prestige, we started with the top left scale (see Figure 1), progressed counterclockwise from People to Data and Things to Ideas and back again to People. The x-axis is two-fold, indicating (1) each of the 18 scales and (2) the scale positions relative to the poles of Prediger’s axes.

Figure 4.
Standardized gender differences across the 18 spherical scales of the Personal Globe Inventory. • = activity preferences, ▴ = activity competence beliefs, squares = ▪; IN = Influence, FA = Financial Analysis, BS = Business Systems, SC = Science, SS = Social Sciences, SF = Social Facilitating, MA = Managing, BD = Business Detail, DP = Data Processing, ME = Mechanical, NA = Nature/Outdoors, AR = Artistic, HE = Helping, BAS = Basic Service, QC = Quality Control, CR = Construction/Repair, PS = Personal Service, MW = Manual Work, P = People, D = Data, T = Things, and I = Ideas.

For all three item types, the same pattern emerged: On average, female participants scored higher on scales that are closer to the People pole and male participants scored higher on scales that are closer to the Things pole on the People–Things axis. Gender differences decreased and became insignificant the closer a scale was located to either pole of the Data–Ideas axis. Across all item types, the largest negative difference was found for the Data Processing scale ( ${\hat{β}}_{σ_{Y}}$ = −.85 to −.65, p < .001). For both activity item types, the largest positive differences were found for the Personal Service scale ( ${\hat{β}}_{σ_{Y}}$ = .68 to .71, p < .001), whereas, for the occupational preference item type, the strongest positive difference was found for the Helping scale ( ${\hat{β}}_{σ_{Y}}$ = .59, p < .001). The only result that was not in line with our expectations was the negative effect on the Financial Analysis scale for all three item types ( ${\hat{β}}_{σ_{Y}}$ = −.45 to −.22, p < .05). Note that this scale was also identified as a source of misfit from the theoretical model in the previous section, demonstrating a sizeable shift from its theoretical position (close to the People pole) toward the Things pole. In this respect, the moderate negative regression coefficients are consistent with the empirically determined position on the interest sphere.

The question of whether the Prestige axis is related to gender was approached by inspecting the average gender effects at the five different levels of prestige. For the occupational preference item type, no mean effect was significantly different from zero (upper pole: ${\bar{β}}_{σ_{Y}}$ = −.05, higher prestige: ${\bar{β}}_{σ_{Y}}$ = −.07, intermediate prestige: ${\bar{β}}_{σ_{Y}}$ = −02, lower prestige: ${\bar{β}}_{σ_{Y}}$ = .12, and lower pole: ${\bar{β}}_{σ_{Y}}$ = .09). The only significant mean effects were the lower poles for the activity preference items (upper pole: ${\bar{β}}_{σ_{Y}}$ = −.07, higher prestige: ${\bar{β}}_{σ_{Y}}$ = −.12, intermediate prestige: ${\bar{β}}_{σ_{Y}}$ = .04, lower prestige: ${\bar{β}}_{σ_{Y}}$ = .09, and lower pole: ${\bar{β}}_{σ_{Y}}$ = −.33, p < .001) and activity competence beliefs items (upper pole: ${\bar{β}}_{σ_{Y}}$ = −.06, higher prestige: ${\bar{β}}_{σ_{Y}}$ = −.10, intermediate prestige: ${\bar{β}}_{σ_{Y}}$ = .02, lower prestige: ${\bar{β}}_{σ_{Y}}$ = −.05, and lower pole: ${\bar{β}}_{σ_{Y}}$ = −.51, p < .001). Taken together, these results indicate that gender is related to Prediger’s axes but not to the Prestige axis.

Discussion

This study aimed to make a contribution toward the cross-cultural validation of the spherical model of vocational interests, as introduced by Tracey and Rounds (1996). To this end, we translated the PGI (Tracey, 2002) into German and distributed it to a sample of university students in Germany. Results from randomization tests and principal components analyses with orthogonal target rotations provided strong empirical evidence for the structural validity of the model. Analyses of gender differences and associations of the PGI RIASEC scales with the GIST further supported the construct validity of the instrument. However, we observed some inconsistencies, which shall be addressed in the following paragraphs.

Structural Validity

We carried out randomization tests for the three item types in the male and female subsamples as well as in the full sample. Of the three item types, the activity competence belief items displayed the largest deviations from the theoretical model. This finding is in line with other studies (e.g., Long et al., 2005; Šverko, 2008). Differences between the other item types and their composite were negligible. Although there were moderate differences in fit for the RIASEC and the 8-type model between male and female participants, the differences in the spherical 18-type model were only marginal.

Results of the principal components analyses provided further support for the validity of the dimensions that constitute the spherical model. Moreover, the target rotated component loadings allowed us to identify specific locations of deviations from the theoretical model. Across all item types, we found three sizeable deviations. For the activity preference and the activity competence belief items, the Financial Analysis scale exceeded our cutoff criterion. According to the theoretical model, this scale should be located in the People–Data quadrant of the coordinate system defined by Prediger’s dimensions. In our sample, the scale shifted toward the Things pole. This scale is assumed to measure “Interest in working directly with customers on their finances” (Tracey, 2002, p. 121). A closer look at the item phrasing revealed that the direct interaction with customers is only implicitly inherent to the items. Further revision of the instrument might modify the respective items to put stronger emphasis on the social interaction with customers.

Another noticeable deviation was the shift of the Influence scale for the activity competence belief items in the People and Data direction toward the position of the Managing scale. This finding indicates that these scales are not as distinguishable as they are supposed to be for this particular item type. Both scales consist of items that emphasize activities that imply leading or supervising groups and high-stake decision-making. Further research is needed to see whether this effect is unique for the population of university students or applies to other populations as well.

Gender Differences and Convergent Validity

For the greater part, the correlations between the PGI and the GIST scales were in line with our expectations. Corresponding RIASEC scales yielded the highest correlation coefficients with one exception: The correlation between the two Conventional scales was moderately lower than the correlation between the PGI Enterprising and the GIST Conventional scale. Further analysis of the items of the GIST Enterprising and Conventional scales by means of a principal components analysis identified the reason in the items of the GIST rather than the PGI.

Mean differences of gender across the 18 PGI scales were in line with gender differences reported by other authors (e.g., Darcy, 2005; Tracey, 2002). On average, female participants scored higher on interest scales located closer to the People pole, whereas male participants scored higher on interest scales located closer to the Things pole. These results are in line with the findings of Lippa (1998), who recognized that gender differences in vocational interests tend to concentrate on the People–Things axis.

Furthermore, we found evidence that this relationship is consistent across different item types and different levels of prestige, and that gender is not related to the Prestige axis of the PGI. The only significant differences were found for the lower pole of the activity item types. However, this pole is indicated by only one scale, Manual Work, which was also found to have shifted slightly toward the Things pole of the Things–People axis. Closer inspection of the activity items of the Manual Work scale revealed that they can be described as rather male activities (e.g. “operating a woodworking machine”), whereas the occupational preference items were mostly neutral with respect to gender. As a consequence, this effect might be attributed to gender differences along Prediger’s content axes rather than the Prestige axis.

Limitations and Directions for Future Research

Despite the strong support that was found for the spherical model, this study has some limitations. First, we validated the PGI only in a certain population, namely, German university students. Therefore, this does not imply that the abovementioned results are generalizable across other vocational groups (e.g., high school students, employed, self-employed, and unemployed people). The analyses of the spatial representation of the spherical model were only explorative and, with the exception of the randomization tests, were not supported by statistical significance tests. Further methodical work is needed to elaborately test the spherical model by means of powerful confirmatory research methods. Finally, other correlates of the interest domains and axes should be included in future analyses in order to strengthen the convergent validity.

First insightful results concerning the role of people’s preference for prestige in vocational choices have been provided by Leung et al. (2014). Future research should aim to explore the relevance of this particular dimension in the analyses of interest congruence and outcomes such as satisfaction, tenure, and performance. Although extensive research has been conducted on this topic, many questions are left open, some of which might be answered by considering the explanatory power of the prestige dimension.

	Activity Preferences	Activity Competence Beliefs	Occupational Preferences	Composite
RIASEC model
Predictions made	72	72	72	72	72	72	72	72	72	72	72	72
Predictions met	67	67	68	66	68	63	69	65	72	68	66	69
CI	0.86	0.86	0.89	0.83	0.89	0.76	0.93	0.82	1.00	0.89	0.83	0.92
8-Type model
Predictions made	288	288	288	288	288	288	288	288	288	288	288	288
Predictions met	273	273	263	260	260	246	274	270	265	279	280	269
CI	0.90	0.90	0.84	0.82	0.83	0.73	0.90	0.89	0.85	0.94	0.95	0.87
18-Type model
Predictions made	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472
Predictions met	7,179	7,109	7,318	6,910	6,964	6,859	7,410	7,284	7,424	7,398	7,356	7,510
CI	0.53	0.51	0.56	0.47	0.48	0.47	0.58	0.55	0.58	0.57	0.56	0.60

	C1	C2	C3	C4
Activity preferences
Eigenvalue	5.30	3.32	2.50	1.31
VAF	0.29	0.18	0.14	0.07
∑ VAF	0.29	0.48	0.62	0.69
Activity competence beliefs
Eigenvalue	6.52	3.15	1.81	1.13
VAF	0.36	0.17	0.10	0.06
∑ VAF	0.36	0.54	0.64	0.70
Occupational preferences
Eigenvalue	6.87	3.07	2.58	1.36
VAF	0.38	0.17	0.14	0.08
∑ VAF	0.38	0.55	0.70	0.77
Composite
Eigenvalue	5.75	3.79	2.67	1.37
VAF	0.32	0.21	0.15	0.08
∑ VAF	0.32	0.53	0.68	0.75

Scale	Activity Preferences	Activity Competence Beliefs	Occupational Preferences	Composite
Social facilitating	.36	0.45	.13	.33
Managing	.26	0.52	.11	.20
Business detail	.46	0.72	.53	.53
Data processing	.59	0.33	.19	.31
Mechanical	.39	0.43	.43	.39
Nature/outdoors	.28	0.33	.42	.25
Artistic	.15	0.28	.15	.24
Helping	.07	0.15	.12	.03
Social sciences	.63	0.75	.50	.60
Influence	.52	1.11	.58	.40
Business systems	.37	0.52	.38	.41
Quality control	.35	0.04	.70	.27
Manual work	.46	0.31	.19	.34
Personal service	.70	0.57	.62	.62
Financial analysis	.90	0.99	.58	.65
Science	.58	0.60	.29	.47
Construction/repair	.25	0.39	.48	.28
Basic service	.57	0.68	.41	.75

Scale	GIST-R	GIST-I	GIST-A	GIST-S	GIST-E	GIST-C
PGI-R	.63*	.34*	−.14	−.24*	−.04	.08
PGI-I	.42*	.66*	.17*	−.02	−.06	.00
PGI-A	.10	.11	.79*	.24	.10	−.10
PGI-S	−.06	−.10	.47*	.77*	.39*	.03
PGI-E	.08	−.14	.17	.40*	.61*	.50*
PGI-C	.45*	.29*	−.23*	−.25*	.12	.32*

Footnotes

Acknowledgments

We thank Simon Grund for his advice concerning the handling of missing data in this study.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Armstrong

P. I.

Day

S. X.

McVay

J. P.

Rounds

(2008). Holland’s RIASEC model as an integrative framework for individual differences. Journal of Counseling Psychology, 55, 1–18. doi:10.1037/0022-0167.55.1.1

Bergmann

Eder

(2005). Allgemeiner Interessen-Struktur-Test mit Umwelt-Struktur-Test (UST-R)—Revision [General Interest Structure Test and Environmental Structure Test—Revision]. Göttingen, Germany: Beltz.

Bernaards

C. A.

Jennrich

R. I.

(2005). Gradient projection algorithms and software for arbitrary rotation criteria in factor analysis. Educational and Psychological Measurement, 65, 676–696. doi:10.1177/0013164404272507

Darcy

M. U. A.

(2005). Examination of the structure of Irish students’ vocational interests and competence perceptions. Journal of Vocational Behavior, 67, 321–333. doi:10.1016/j.jvb.2004.08.007

Darcy

M. U. A.

Tracey

T. J. G.

(2007). Circumplex structure of Holland’s RIASEC interests across gender and time. Journal of Counseling Psychology, 54, 17–31. doi:10.1037/0022-0167.54.1.17

Day

S. X.

Rounds

(1998). Universality of vocational interest structure among racial and ethnic minorities. American Psychologist, 53, 728–736. doi:10.1037//0003-066X.53.7.728

Enders

C. K.

(2010). Applied missing data analysis. New York, NY: Guilford Press.

Etzel

J. M.

Nagy

(2016). Students’ perceptions of person-environment fit: Do fit perceptions predict academic success beyond personality traits? Journal of Career Assessment, 24, 270–288. doi:10.1177/1069072715580325

Gottfredson

G. D.

(1996). Prestige in vocational interests. Journal of Vocational Behavior, 48, 68–72. doi:10.1006/jvbe.1996.0006

10.

Guttman

(1954). A new approach to factor analysis: The radex. In Lazarsfeld

P. F.

(Ed.), Mathematical thinking in the social sciences (pp. 258–348). Glencoe, IL: Free Press.

11.

Hansen

J. I.

Collins

R. C.

Swanson

J. L.

Fouad

N. A.

(1993). Gender differences in the structure of interests. Journal of Vocational Behavior, 42, 200–211. doi:10.1006/jvbe.1993.1014

12.

Hedrih

(2008). Structure of vocational interests in Serbia: Evaluation of the spherical model. Journal of Vocational Behavior, 73, 13–23. doi:10.1016/j.jvb.2007.12.004

13.

Holland

J. L.

(1997). Making vocational choices: A theory of vocational personalities and work environments (3rd ed.). Odessa, FL: Psychological Assessment Resources.

14.

Holtrop

Born

M. P.

de Vries

R. E.

(2015). Relating the spherical representation of vocational interests to the HEXACO personality model. Journal of Vocational Behavior, 89, 10–20. doi:10.1016/j.jvb.2015.04.003

15.

Hubert

Arabie

(1987). Evaluating order hypotheses within proximity matrices. Psychological Bulletin, 102, 172–178. doi:10.1037/0033-2909.102.1.172

16.

Knapp

R. R.

Knapp

(1974). The California occupational preference system interest inventory, technical manual. San Diego, CA: EdITS.

17.

Lee

Lawson

K. M.

McHale

S. M.

(2015). Longitudinal associations between gender-typed skills and interests and their links to occupational outcomes. Journal of Vocational Behavior, 88, 121–130. doi:10.1016/j.jvb.2015.02.011

18.

Leong

F. T. L.

Austin

J. T.

Sekaran

Komarraju

(1998). An evaluation of the cross-cultural validity of Holland’s theory: Career choices by workers in India. Journal of Vocational Behavior, 52, 441–455. doi:10.1006/jvbe.1997.1637

19.

Leung

S. A.

Zhou

E. Y.-F.

K. P.

Tracey

T. J. G.

(2014). The use of interest and competence scores to predict educational choices of Chinese high school students. Journal of Vocational Behavior, 84, 385–394. doi:10.1016/j.jvb.2014.02.010

20.

Lippa

(1998). Gender-related individual differences and the structure of vocational interests: The importance of the people-things dimension. Journal of Personality and Social Psychology, 74, 996–1009. doi:10.1037/0022-3514.74.4.996

21.

Long

Adams

R. S.

Tracey

T. J. G.

(2005). Generalizability of interest structure to China: Application of the personal globe inventory. Journal of Vocational Behavior, 66, 66–80. doi:10.1016/j.jvb.2003.12.004

22.

Meir

E. I.

(1973). The structure of occupations by interests—A small space analysis. Journal of Vocational Behavior, 3, 21–31.

23.

Muthén

L. K.

Muthén

B. O.

(2012). Mplus user’s guide (7th ed.). Los Angeles, CA: Author.

24.

Nagy

Trautwein

Lüdtke

(2010). The structure of vocational interests in Germany: Different methodologies, different conclusions. Journal of Vocational Behavior, 76, 153–169. doi:10.1016/j.jvb.2007.07.002

25.

Prediger

D. J.

(1982). Dimensions underlying Holland’s hexagon: Missing link between interests and occupations? Journal of Vocational Behavior, 21, 259–287. doi:10.1016/0001-8791(82)90036-7

26.

Prediger

D. J.

(1998). Is interest profile level relevant to career counseling? Journal of Counseling Psychology, 45, 204–211.

27.

Roe

(1956). The psychology of occupations. New York, NY: Wiley.

28.

Rounds

(2014). The nature and power of interests. Current Directions in Psychological Science, 23, 98–103. doi:10.1177/0963721414522812

29.

Rounds

Tracey

T. J.

(1996). Cross-cultural structural equivalence of RIASEC models and measures. Journal of Counseling Psychology, 43, 310–329.

30.

Rubin

D. B.

(1987). Multiple imputation for nonresponse in surveys. New York, NY: Wiley.

31.

Šverko

(2008). Spherical model of interests in Croatia. Journal of Vocational Behavior, 72, 14–24. doi:10.1016/j.jvb.2007.10.001

32.

Tracey

T. J. G.

(1997). RANDALL: A Microsoft FORTRAN program for a randomization test of hypothesized order relations. Educational and Psychological Measurement, 57, 164–168. doi:10.1177/0013164497057001012

33.

Tracey

T. J. G.

(2002). Personal globe inventory: Measurement of the spherical model of interests and competence beliefs. Journal of Vocational Behavior, 60, 113–172. doi:10.1006/jvbe.2001.1817

34.

Tracey

T. J. G.

Robbins

S. B.

(2006). The interest-major congruence and college success relation: A longitudinal study. Journal of Vocational Behavior, 69, 64–89. doi:10.1016/j.jvb.2005.11.003

35.

Tracey

T. J. G.

Rounds

(1995). The arbitrary nature of Holland’s RIASEC types: A concentric-circles structure. Journal of Counseling Psychology, 42, 431–439. doi:10.1037/0022-0167.42.4.431

36.

Tracey

T. J. G.

Rounds

(1996). The spherical representation of vocational interests. Journal of Vocational Behavior, 48, 3–41. doi:10.1006/jvbe.1996.0002

37.

Tracey

T. J. G.

Watanabe

Schneider

P. L.

(1997). Structural invariance of vocational interests across Japanese and American culture. Journal of Counseling Psychology, 44, 346–354. doi:10.1037/0022-0167.44.4.346

38.

van Buuren

Groothuis-Oudshoorn

(2011). Mice: Multivariate imputation by chained equations in R . Journal of Statistical Software, 45, 1–67.

39.

Warwas

Nagy

Watermann

Hasselhorn

(2009). The relations of vocational interests and mathematical literacy: On the predictive power of interest profiles. Journal of Career Assessment, 17, 417–428. doi:10.1177/1069072709339284

40.

Wilkins

K. G.

Ramkissoon

Tracey

T. J. G.

(2013). Structure of interest in a Caribbean sample: Application of the personal globe inventory. Journal of Vocational Behavior, 83, 367–372. doi:10.1016/j.jvb.2013.06.005

	Activity Preferences			Activity Competence Beliefs			Occupational Preferences			Composite
	All	Female	Male	All	Female	Male	All	Female	Male	All	Female	Male
RIASEC model
Predictions made	72	72	72	72	72	72	72	72	72	72	72	72
Predictions met	67	67	68	66	68	63	69	65	72	68	66	69
CI	0.86	0.86	0.89	0.83	0.89	0.76	0.93	0.82	1.00	0.89	0.83	0.92
8-Type model
Predictions made	288	288	288	288	288	288	288	288	288	288	288	288
Predictions met	273	273	263	260	260	246	274	270	265	279	280	269
CI	0.90	0.90	0.84	0.82	0.83	0.73	0.90	0.89	0.85	0.94	0.95	0.87
18-Type model
Predictions made	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472	9,472
Predictions met	7,179	7,109	7,318	6,910	6,964	6,859	7,410	7,284	7,424	7,398	7,356	7,510
CI	0.53	0.51	0.56	0.47	0.48	0.47	0.58	0.55	0.58	0.57	0.56	0.60