Seminars

This talk considers semiparametric estimation for parameters defined by estimating equations with missing values.  When there are no surrogate for the missing values, we propose a nonparametric imputation of the missing values, which is different from the mean imputation and specially suits the generality of estimation equations, followed by an empirical likelihood inference.  When there are surrogates, a two-step method that re-weighs the original estimating equations  with weights obtained by an empirical likelihood formulation based on the surrogate and covariate information is proposed.  It is shown that the proposed methods have some desirable theoretical properties as well as promising empirical performance. Applications on real data examples will be presented.

pdf version

Coffee and Cookies at 3:30 p.m. in Room 1210 MSC