Refinement and Extension of Shrinkage Techniques in Loss Rate Estimation of Marine Corps Officer Manpower Models

摘要

This thesis is a continuation of previous work to apply modern multiparameter estimation techniques to the problem of estimating attrition rates for a large number of small inventory cells in manpower planning models used by the U.S. Marine Corps. The main advances involve the promising introduction of empirical Bayes (non-constant shrinkage) techniques, recognition of the non symmetric nature of the errors with a response to this, and some insight into all aggregation plans that should help provide greater stability for the estimation methods. In addition, the roles of some middle level methodological choices are explored.