A comparison and validation of two surface ship readiness models

Abstract

Two models are used by the U.S. Navy to predict surface ship readiness: the Surface Ship Resources to Material Readiness Model (SRM) and the Surface Ship Inventory to Material Readiness Model (SIM). This thesis examines both models, in order to validate the model fit and to determine whether the two models predict significantly different levels of readiness for a given data set using both cross validation and jackknife procedures. Examination of the models reveals that there are numerous insignificant predictor variables in the models. Normality assumptions made on the non-linear regression are not proper. Additionally, the performance of both the SRM and the SIM at the ship level is poor. However, once aggregated to the fleet level, prediction performance improves drastically. Analysis of the jackknife confidence intervals indicate that the SRM and SIM predict significantly different levels of readiness. While the SIM performs slightly better than the SRM, one has to consider the marginal cost associated with the more complex SIM for model selection. Finally, use of reduced models and model modifications such as use of Poisson regression are recommended