Austin Chapter of the American Statistical Association

Samiran Sinha
Informative Missing Data in Stratified Retrospective Studies

We will discuss how to handle informative missing data in highly stratified retrospective studies, such as in matched case-control studies. In these studies exposure information is collected conditional on the outcome of our interest, therefore, not the response, rather predictor variable is missing for some study units. We propose a full likelihood based approach for dealing with informative missing values in highly stratified retrospective studies. The small sample properties of the estimators are studies through small scale simulations, and also compared with some existing methods. Lastly we apply the proposed method on a matched case-control data from the Los Angeles study of endometrial cancer.

Key words: E-M algorithm; Generalized exponential family; Los Angeles endometrial cancer data; Matched case-control data; Selection model.