A cluster analytic approach to interest measurement with the ACT Interest Inventory
The study attempted to identify, replicate and validate subgroups of high school students based on similarity of their responses to items of the ACT-Interest Inventory. A data set collected in 1973 and in 1976-1979 by the American College Testing Program/Institute for Demographic and Economic Studies was used for this study. Random samples were selected and clustered with two hierarchical cluster analysis techniques (Ward's method and average linkage method). Average linkage did not yield a meaningful structure. A six-group solution was selected from Ward's method results for further study. Evidence for replicability of the clusters across samples was sought. Five of six groups appeared consistently across subsamples. The groups did not match the six themes suggested by Holland's hexagon. No Enterprising or Conventional groups were found. Strong sex differences appeared: Investigative and Realistic groups were at least two-thirds male; Artistic groups were predominantly female; and only Social groups were gender balanced. Replicability of the clusters was also studied by assigning members of two subsamples to clusters based on the structure of a third subsample. Two-thirds of members of these subsamples were accurately assigned;Demographic variables selected from an MANOVA on a fourth subsample were analyzed to explore differences between the clusters. An overall MANOVA was highly significant but variables hypothesized to be related to clusters (job values, family variables, socioeconomic factors, and job satisfaction variables) were not significant. Remaining variables were selected empirically to validate cluster structures generated by two other samples. Significance was found for sets of variables in two domains: seven demographic variables (especially gender) and seven values/preferences variables. A discriminant analysis found sex differences were so powerful that other differences were somewhat overshadowed;Sex differences were examined by separately clustering male and female subsamples. Females formed six groups and males formed nine groups. Significant differences also emerged on the demographic and values/preferences variables by gender. When creating groups by clustering, it was clear gender had a differential impact;The literature has raised questions about the choice of items or scales for generating clusters. Two additional clusterings using scales were undertaken to consider this issue. Ward's method yielded similar results with both items and scales. Average linkage performed better with scales but still had a tendency to chain, thereby creating one large group. When examined by discriminant procedures, two subsamples of Ward's clusters produced different results. Limitations of these findings and suggestions for future research were discussed.