Formulating the Brogden Classification Framework as a Discrete Choice Model.

摘要

The Brogden optimal classification framework measures potential classification benefits of predictors by assigning applicants to the jobs that will maximize predicted performance subject to job quota constraints. Current implementations of Brogden's framework do not include classification policy constraints (e.g., cut scores and gender restriction), applicant preferences, or the impact of other classification tools available to the Army (e.g., monetary incentives to channel applicants to particular job training). To accommodate elements of real world classification systems and thereby better inform operational problems, this research reformulated Brogden's classification framework using discrete choice modeling. We specified a mixed multinomial logit model for classification that is mathematically equivalent to a multivariate normal based implementation of Brogden's framework. We also proposed an empirical or sample based method for classification analysis based on the multinomial logit (MNL) model that can accommodate personnel classification policy constraints, such as cut scores and gender restriction, and is robust to the functional form and distribution of the criterion estimates. Illustrative example applications of the MNL classification model showed the expected effects of policy constraints, with practical implications for the analysis results.