Poverty Analysis with Item and Unit Nonresponses: alternative estimators compared

Abstract

This paper assesses some methods to estimate binary response models in presence of missing data. Particular attention is given to imputation, econometric sample selection correction and propensity score estimation methods. These methods are based on assumptions whose validity can only be verified when the missing data are observed. Using the European Community Household Panel I discuss an informal test to assess the underlying assumptions of different estimation methods for poverty probability. This informal test makes use of Manski bounds for poverty probability, which can be estimated with no or weak assumptions.