11/4/2016

Post date: Nov 9, 2016 7:56:55 PM

Title: Estimation in the Semiparametric Accelerated Failure Time Model with Missing Covariates: Improving Efficiency through Augmentation.

Speaker: Jon Steingrimsson, Department of Biostatistics, JHU

[Abstract]

We considers linear regression with missing covariates and a right censored outcome. We mostly focus on the case-cohort design even though the results hold for more general missingness mechanisms. We consider a class of augmented estimating equations focusing on improving efficiency over the simple but inefficient Horvitz-Thompson type estimators. We summarize asymptotic properties of the class of estimators, identify the most efficient estimator within that class, and give guidance for calculating the augmented estimator in practice. Finite sample performance is evaluated via simulations as well as by analyzing data arising from the Wilms' tumor study. Furthermore, we discuss how the ideas presented connect to the parallel literature on augmentation for the proportional hazard model. This is joint work with Professor Rob Strawderman (University of Rochester).