Modeling career counselor decisions with artificial neural networks: Predictions of fit across a comprehensive occupational map

Aptitude test scores from high school freshmen (N = 335) served as the basis for career recommendations using an adaptation of Gottfredson's (1985, 1986) Occupational Aptitude Patterns Map. Probabilistic neural networks were used to model recommendations to 12 occupational clusters made by a career counselor on the basis of scores on 12 aptitude tests. In cross-validation, mean overall accuracy of neural networks (.80) approached but did not surpass that of discriminant function analysis (.84). Mean kappa was approximately the same for both methods at .42 for neural networks and .43 for discriminant analysis. However, analysis of various types of errors and accurate hit rates formed by crossing rater (recommend, not recommend) and model (recommend, not recommend) suggested differential strengths and weaknesses of the two multivariate methods. The results demonstrate the potential for neural network-based test interpretation systems as resources in career assessment. (C) 1999 Academic Press.