Department of Epidemiology & Biostatistis, Bioinformatics Presents:

Bin Nan, PhD, Professor in Statistics, UC Irvine

It has been shown that the complete case analysis, by eliminating observations with covariate values below the limit of detection, yields valid estimates for regression coefficients but loses efficiency. Substitution methods are biased, and the maximum likelihood method relies on parametric models for the unobservable tail probability, thus may suffer from model misspecification. To obtain robust and more efficient results, Dr. Nan proposes a semiparametric pseudo-likelihood approach and a two-stage estimation procedure for the regression parameters using an accelerated failure time model for a single covariate subject to limit of detection, where the conditional distribution of the covariate with limit of detection is estimated prior to maximizing the likelihood function for the regression parameters. When there are two or more covariates subject to their limits of detection, the implementation of the two-stage semiparametric method becomes much more difficult: This added challenge will also be discussed.