DID YOU KNOW? DTIC has over 3.5 million final reports on DoD funded research, development, test, and evaluation activities available to our registered users. Click HERE
to register or log in.
The Use of Survival Analysis in the Prediction of Attrition in Large Scale Personnel Flow Models
NAVAL POSTGRADUATE SCHOOL MONTEREY CA DEPT OF OPERATIONS RESEARCH
Pagination or Media Count:
Survival analysis methods are applied to the problem of estimating attritions for the ELIM module of the Armys computerized personnel flow models. The use of the Cox proportional hazards model resulted in considerable computational overhead and the discrete proportional odds version i.e., logit model is examined instead. Since this model is supported by the generalized linear model software systems, some closely related models are considered and compared, specifically the linear, probit, and complementary log-log. The existing exponential smoothing method is also included. There are a number of important findings. First there is need to standardize the measures of performance. The present study uses average magnitude relative error for sets of forecast cells. Second, there is considerable variability in the forecastability of the attrition behavior of soldiers in the various personnel partitions i.e., partitions made by cross classifying by education, mental group, contract term, etc. The logit, probit, and exponential smoothing models perform about equally well within the partitions when used without covariates and focusing on the involuntary losses. The inclusion of covariates in the generalized linear models appears to have a disturbing effect and the relative errors are larger than before. These errors have distinctive patterns, which at present, are unexplained.
APPROVED FOR PUBLIC RELEASE