A simulation study of estimates of a first passage time distribution for a censored semi-Markov process.

Abstract

This thesis reports on a simulation study of parametric and nonparametric estimators of a first passage time distribution for a censored semi-Markov process. Four estimators are proposed and compared; Maximum Likelihood Estimator, Renewal Equation Estimator, Asymptotic Renewal Estimator, and the Kaplan-Meier Estimator; the last three estimators are nonparametric. For the particular semi-Markov process studied, the Kaplan-Meier estimator of the first passage times appears to be the best for small times and the Asymptotic Renewal estimator appears to be the best for large times. The Maximum Likelihood estimator is sensitive to incorrect model assumptions. All the estimators are sensitive to censoring.