|
Abstract:
|
In this presentation we further discuss the approach introduced at the last JSM. We assume a model for the outcome under complete response and a model for the response probability, which, when combined, define the model holding for the outcome of a responding unit. Estimation of the model parameters is carried out by combining maximization of the likelihood obtained for the responding units with calibration equations that use known totals of the covariates. We investigate the properties of the resulting estimators and develop a new test statistic for testing which covariates to include in the model for the response probabilities. We also consider imputation of the missing outcomes and covariate values, and estimation of the population mean of the outcome, with appropriate variance estimators. The results are illustrated using simulated data and a real data set of household incomes
|
