Binary models with misclassification in the variable of interest

In this paper we propose a general framework to deal with datasets where a binary outcome is subject to misclassification and, for some sampling units, neither the error-prone variable of interest nor the covariates are recorded. A model to describe the observed data is for-malized and eficient likelihood-based generalized method of moments (GMM) estimators are suggested. These estimators merely require the formulation of the conditional distribution of the latent outcome given the covariates. The conditional probabilities which describe the error and the nonresponse mechanisms are estimated simultaneously with the parameters of inter-est. In a small Monte Carlo simulation study our GMM estimators revealed a very promising performance.