Determining characteristic groups to predict Army attrition

The Office of the Deputy Chief of Staff, Personnel (ODCSPER), is charged with managing the Army's military strength levels and forecasting future strength levels for planning purposes. ODCSPER is reformulating its Enlisted Loss Inventory Model (ELIM), which projects losses of first-term enlisted personnel. These projections in turn are passed to a program which is designed to maintain the Army's strength as closely as possible to prescribed levels. These projections are based on characteristic groups, a set of sub-groups of recruits who are similar in terms of sex, education level, term of service and mental category; the presumption has been that attrition rates ought to be different between groups. However in recent years ELIM projections have been unsatisfactory. This study used Classification and Regression Tree methodology (CART) to generate improved c-groups for predicting not only first-term attrition but also early-term behavior and re-enlistment. The most important variables by which to create these groups turn out to be race and gender. Generally white women have the lowest term completion and re-enlistment rates; those for non-white women and white men are similar; and those for non-white men are the highest.