Modeling Binary Panel Data with Nonresponse

This paper studies modeling of nonignorable nonresponse in panel surveys. A class of sequential conditional logistic models for nonresponse is considered. Model-based maximum likelihood estimation and imputation are used for estimating population proportions. Various models are evaluated, and comparisons are made with traditional methods of weighting and direct data imputation. Two cases are considered, (i) the population rate of participation in the 1989 Norwegian Storting election and (ii) estimation of car ownership in Norway in 1989 and 1990.