Nested Logit Analysis of Missing Response Observations

Abstract

We propose a new estimation technique to deal with missing response variables in the context of a nested multinomial logit model. Survey data often have a significant number of incomplete or missing responses. If such data are systematically missing (i.e., not missing at random) and if such observations are deleted from the analysis, biased sample selection results. We apply our new method to the empirical analysis of determining job loss status.