The motivation for the HCIA stems from a well-documented gap in existing career-interest measurement. The dominant framework in vocational psychology, John Holland’s RIASEC model, sorts interests into six broad themes—Realistic, Investigative, Artistic, Social, Enterprising, and Conventional—usually arranged in a hexagon where adjacent types are assumed to be more similar than opposite ones. That parsimony is useful for summarizing general orientation, but it can mask enormous within-domain heterogeneity. The broad Artistic domain, for example, lumps together visual arts, music, performing arts, creative writing, and design, which are distinct majors with distinct course demands. Research on so-called basic interests has shown that narrower, content-homogeneous scales provide incremental information beyond broad RIASEC themes, and one study found that interest congruence measured at the basic-interest level predicted college-major satisfaction better than congruence measured at the broad level. For students choosing among academic programs that look similar at a distance but differ substantially in learning content, that specificity matters.

To build the HCIA’s category structure, the researchers conducted an iterative content-mapping process that integrated several large classification systems. On the educational side, they drew on Taiwanese academic-field classifications—including the fifth revised Standard Classification of Academic Fields, which organizes fields into 11 broad fields, 27 narrow fields, 93 categories, and 174 detailed categories—and cross-checked coverage against UNESCO’s International Standard Classification of Education Fields of Education and Training, known as ISCED-F, which spans 11 broad fields, 29 narrow fields, and roughly 80 detailed fields. On the occupational side, they used the 2018 U.S. Standard Occupational Classification, with its 867 detailed occupations nested in 459 broad occupations, 98 minor groups, and 23 major groups, along with the O*NET-SOC 2019 taxonomy. Crucially, the authors emphasize that these taxonomies served as content-mapping frameworks for coverage and naming, not as psychological theory; the interpretation of the final categories is grounded in vocational-interest theory, basic-interest research, and hierarchical models of interests, including Gati’s hierarchical model and the more recent Comprehensive Assessment of Basic Interests, or CABIN, which organizes 41 basic-interest scales under eight broad dimensions.

The resulting HCIA framework comprises nine broad categories—Arts, Services, Law and Politics, Natural Sciences, Social Sciences, Human Sciences, Business, Engineering, and Physical Fitness—each containing four subcategories. Item development began with a pool of 245 items written to describe the essential features of learning contents and work activities rather than relying on the names of courses or occupations, a choice intended to reduce the influence of students’ differential familiarity with specific labels. An item representing politics, for instance, reads “Exploring how national government works,” while one representing tour-guide work describes “Leading tours and explaining scenic spots, dealing with problems in traveling.” A panel of three experts in educational psychology and psychometrics reviewed the pool for relevance, clarity, developmental appropriateness, representativeness, and conceptual overlap, retaining 228 items for pilot testing. After psychometric screening, the final version retained exactly four items per subcategory—two course-related learning-activity items and two occupational-activity items—yielding the 144-item instrument with balanced content representation.

Perhaps the most distinctive technical feature of the HCIA is its response format. Rather than asking students to rate items on a traditional Likert scale, the assessment employs the Visual Analogue Scale for Rating, Ranking, and Paired-Comparison, or VAS-RRP, a computerized format in which students drag item icons onto a continuous response line. The 144 items are assembled into 24 testlets of six items each, displayed on a single screen so that all six can be compared simultaneously. Because the system prevents two icons from occupying exactly the same position, students must make relative preference judgments rather than assigning identical ratings to everything—a built-in forcing function against indiscriminate responding. The 36 subcategories were treated as treatments in an incomplete-block design, with the allocation of items to testlets generated by an algorithmic procedure based on a D-optimality criterion, ensuring that each subcategory appeared exactly four times across the assessment. Scoring converts each icon’s horizontal position into a proportional 0-to-100 coordinate relative to the scale’s endpoints, a normalization that removes the influence of screen resolution and display size on the measurement metric.

The psychometric evaluation proceeded in stages. A pilot study with 1,550 Grade 11 students from northern Taiwan, aged 16 or 17, was conducted between mid-December 2018 and late January 2019. Item analysis flagged ten items with item-total correlations below the prespecified 0.40 discrimination threshold for removal or revision, and Cronbach’s alpha coefficients for the nine broad categories all exceeded 0.88 in the pilot data. The formal validation study, run from early May to late June 2020, initially collected 3,308 response records from Grade 11 students across Taiwan; after screening out 52 incomplete records and 116 records completed in under seven minutes—far shorter than the typical 25-minute completion time and indicative of careless responding—3,140 valid records remained. A separate subsample of 238 students retook the assessment approximately one month later to establish temporal stability, and a third cohort of 555 students provided scores on both the HCIA and an earlier Holland-based instrument, the Situation-Based Career Interest Assessment, for convergent comparison.

The results were largely favorable. In the formal validation sample, Cronbach’s alpha ranged from 0.75 to 0.94 for the 36 subcategories and from 0.86 to 0.96 for the nine broad categories, while composite reliability coefficients derived from confirmatory factor analysis ranged from 0.76 to 0.95 and 0.84 to 0.98, respectively. Test-retest coefficients over the one-month interval ranged from 0.73 to 0.91 for subcategories and 0.85 to 0.91 for categories. Average variance extracted, a measure of convergent validity, ranged from 0.64 to 0.96 across levels, comfortably exceeding the conventional 0.50 threshold. The researchers compared four confirmatory factor analysis models: a nine-factor direct model, a testlet-adjusted version of it, a correlated 36-factor subcategory model, and the hypothesized testlet-adjusted second-order model in which items load on 36 subcategory factors that in turn load on nine broad categories. The correlated 36-factor model fit best empirically, but the hypothesized second-order model also showed acceptable fit—CFI of 0.969, RMSEA of 0.061—supporting the intended hierarchical organization, with item loadings ranging from 0.49 to 0.91 and second-order loadings from 0.45 to 0.98.

External validity evidence came from correlations between HCIA broad-category scores and the RIASEC scores of the Holland-based SCIA. The pattern was theoretically coherent: Arts correlated most strongly with Artistic interests (r = 0.62), Natural Sciences with Investigative (r = 0.62), Engineering with both Investigative and Realistic (r = 0.54 and 0.52), Law and Politics and Business with Enterprising and Conventional, and Social Sciences with Social (r = 0.52). Some categories showed more differentiated, cross-domain patterns—Services correlated with both Artistic and Social interests, likely reflecting its inclusion of fashion, hospitality, and tourism—suggesting the HCIA captures structure that a six-type inventory would blur. Supplementary score-level analyses of differential item functioning across gender, school ownership, and geographic region found that metric-like effects were generally negligible, while scalar-like effects were more frequent for gender, appearing to reflect genuine mean-level differences in interest profiles rather than measurement bias, though the authors caution these were preliminary score-level checks rather than full item-level invariance tests.

The authors are careful to frame what the study does and does not establish. The confirmatory findings support the HCIA’s proposed measurement structure, not the superiority of hierarchical models over circumplex ones, nor a direct validation of Gati’s original hard-science versus soft-science partition. Limitations include the focus on Taiwanese students, the absence of predictive validity evidence—future longitudinal work should test whether HCIA scores forecast track choices, major satisfaction, and persistence—and the lack of discriminant validity analysis. The study also notes that the original administration did not undergo prospective IRB review, and that device-related variability in the drag-and-drop format, though mitigated by proportional scoring, warrants further study. Still, the practical promise is clear: by letting students and counselors zoom from broad academic directions down to specific subcategory profiles, the HCIA extends hierarchical interest assessment to an earlier educational stage, potentially giving teenagers facing high-stakes streaming decisions a far more detailed map of where their interests might take them.

Subject of Research: Development and psychometric validation of a hierarchical career-interests assessment for senior high school educational tracking

Article Title: Constructing the hierarchical career-interests assessment (HCIA) for educational tracking

Article References: Sung, Y.-T., & Hsiao, M.-T. (2026). Constructing the hierarchical career-interests assessment (HCIA) for educational tracking. Current Psychology, 45(17), Article 1495. https://doi.org/10.1007/s12144-026-09871-3

Image Credits: AI Generated

DOI: 10.1007/s12144-026-09871-3

Keywords: career-interest assessment, vocational interests, psychometrics, educational tracking, HCIA, RIASEC model, confirmatory factor analysis, high school students, Taiwan, ISCED-F, test reliability, academic major selection