Semiparametric Estimation for Generalized Linear Models with Missing Covariates

• Activity Number: 445
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Yinan Fang* and Jae-kwang Kim
• Companies: Iowa State University and Iowa State University
• Keywords: Semiparametric Estimation ; Missing Covariates ; Monte Carlo EM ; Fractional Imputation

Abstract

Parameter estimation for generalized linear models (GLM) with missing covariates is commonly encountered in practice. If we assume a fully parametric model for the joint distribution, Monte Carlo EM algorithm based on parametric fractional imputation of Kim (2011) can be easily applied to obtain the maximum likelihood estimates for the regression parameters. In the semi-parametric approach, no parametric model assumption is made on the conditional distribution of the missing covariates given the other observed covariates. Instead, we use Kernel-based nonparametric regression estimator to obtain a nonparametric distribution of the conditional distribution. The resulting estimator is called semiparametric maximum likelihood estimator (SMLE) and can be implemented by semiparametric fractional imputation. Some asymptotic properties of the SMLE are discussed. Results from a limited simulation study are also presented.