|Analysis of Censored Survival Times with Missing Covariates
Using the EM Algorithm
Josh Horstman, M.S. Candidate, Department of Statistics, Colorado State University.
Monday, November 3, 2008
9:30 a.m. 223 Weber
We consider the use of the EM algorithm for maximum likelihood estimation in
the analysis of censored survival times, both with and without missing covariate values.
After attempts to complete the EM algorithm analytically fail, the EM gradient method
of Lange (1995) is invoked to complete the M step. Variance estimation follows using
the SEM algorithm of Meng and Rubin (1991). In the case of missing covariates, we
employ the Monte Carlo EM algorithm as used by Ibrahim, Chen, and Lipsitz (1999)
with a simple bootstrap method for variance estimation. Monte Carlo simulation is
based upon a Gibbs sampler using several sampling mechanisms including rejection
Geof Givens, Adviser
Dan Cooley, Committee Member
David Thompson (Department of Atmospheric Science)