Maximum Empirical Likelihood Estimation for Nonignorable Missing Data Problems

• Activity Number: 584
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Korean International Statistical Society
• Author(s): Jing Qin*+ and Zhong Guan
• Companies: National Institutes of Health, BRB and Indiana University South Bend
• Keywords: Constrained estimation ; Empirical likelihood ; Non-ignorable missing data ; maximum likelihood method ; survey sampling ; AIDS applications

Abstract

Missing response data problem is ubiquitous in survey sampling, medical, social science and epidemiology studies. It is well known that non-ignorable missing is the most difficult missing data problem to deal with. In this paper we study a semiparametric non-ignorable missing data problem, where the missing probability is assumed to be known up to some parameters, but the regression model for the responsevariable conditioning on the covariate is not specified. By employing the empirical likelihood method we can construct constrained empirical likelihood. Moreover the semiparametric likelihood ratio statistic can be used to test whether the missing of some of the responses is non-ignorable or completely at random. A real AIDS trial data set is used for illustration. Analysis result indeed shows that the missing data of CD4 count around 2 years are non-ignorable. The naive sample mean based on observed data only is biased.