A Generalized Missing-Indicator Approach to Regression with Imputed Covariates

This paper considers estimation of a linear regression model using data where some covariate values are missing but imputations are available to fill-in the missing values. The availability of imputations generates a trade-off between bias and precision in the estimators of the regression parameters. The complete cases are often too few, so precision is lost, but filling-in the missing values with imputations may lead to bias. We provide the new Stata command gmi which allows handling such bias-precision trade-off using either model reduction or model averaging techniques in the context of the generalized missing-indicator approach recently proposed by Dardanoni et al.(2011). If multiple imputations are available, our gmi command can be also combined with the built-in Stata prefix mi estimate to account for the extra variability due to the imputation process. The gmi command is illustrated with an empirical application which investigates the relationship between an objective health indicator and a set of socio-demographic and economic covariates affected by substantial item nonresponse.