An application of survival analysis methods to the study of Marine enlisted attrition

Abstract

This thesis is an application of survival analysis methods to study first term enlisted attrition from the Marine Corps. The data comprise over 99 percent of all enlisted accessions into the Marine Corps between 1 October 1983, and 31 August 1988. A large percentage of the observations are censored, thus motivating the use of survival analysis techniques. The enlistees are categorized by three covariates: education credential, Armed Forces Mental Group and presencenon-presence of a moral waiver. The attrition behavior of the enlistees is then examined to identify which covariate classifications are associated with premature attrition. The majority of the findings concerning the effects of the covariates on attrition are consistent with published results from previous military attrition studies. Two findings of the thesis, though, are perhaps new. First, the attrition behavior of alternate high school credential holders varied significantly according to credential type. Second, the relationship between aptitude and attrition behavior appears to have weakened in recent years. The thesis also provides an opportunity to evaluate the uncommon practice of using survival analysis methods to examine military attrition. The results are promising as the survival analysis methods prove to be both accurate and efficient. Graphical plots of survivor function estimates provide an easily understood illustration of attrition behavior. The use of log- linear regression to model military attrition shows potential as a desk-top tool for conducting informal analyses.