A Methodology for Comparison of Attrition Forecasting Models.

摘要

This analysis develops a methodology for comparing forecasting models and examines in detail the ability of available attrition/success models to adequately predict the service performance of nongraduate Marines. Available models will almost always select graduates over nongraduates; therefore, it is possible that these models cannot optimally select from among the remaining nongraduates -- some of whom the Marine Corps will be forced to accept. The data was disaggregated and the attrition of graduates and nongraduates was examined separately. The results of this analysis indicate that using a separate attrition/success model developed especially for nongraduates will not significantly improve the accuracy and reliability of prediction. The conclusion, therefore, is that existing models, which deal with an aggregated population, are both appropriate and sufficient. (Author)