Retention Analysis Model (RAM) For Navy Manpower Analysis

Abstract

In the first year of our Retention Analysis Modeling project, we began developing the modelling approach by performing the following analyses: Describe the retention models used to analyze policy levers affecting reenlistment rates, including the simple reenlistment model, the Average Cost of Leaving (ACOL) model, and the Dynamic Retention Model (DRM); critically assess the advantages and disadvantages of these models; indicate what would be needed to obtain more credible estimates from future models; discuss the problems with a one-size-fits-all and one-moment-in-YOS-fits-all mid-career bonuses; Make recommendations for tools to address the questions above, in both setting bonuses in real time and predicting expected impacts in the longer term. One of our key findings is that existing models suffer from potentially large biases affecting the estimates. The sources of the biases include: reverse causality because a lower reenlistment propensity would lead to higher bonuses; measurement error in correctly coding the bonus considered by the sailor at the time a decision was made; and excess supply because, sometimes, more sailors want to reenlist than are allowed to reenlist. To minimize biases and maximize their efficacy, we will use the increased computing capacity of modern high-performance computing (HPC) cluster servers, and models will be designed from the ground-up to leverage this increased power; we will incorporate non-monetary incentives, other personalized incentives, and measures of service member quality; we will incorporate new and more detailed data of individual’s socio-economic and professional status in the estimation.