Missing Observations on a Variable: When Do the Listwise Deletion and Indicator Approaches Work?

Abstract

Econometric analyses widely use the listwise deletion approach - which is also called the complete case method - and the indicator approach for models with censored regressors and regressors with missing observations. Nevertheless, the existing literature provides few studies which examine the statistical properties of these methods and they suggest the listwise deletion method generates smaller bias than the indicator method. Using the Monte Carlo simulations this study shows sometimes the listwise deletion estimates have larger bias than the indicator estimates when missingness on a regressor is correlated with unobserved error terms and the value of the regressor. These conditions indicate that greater care is warranted in interpreting the estimates under each approach than the existing literature and the applications which generally employ these methods.

Keywords: Missing Data, Unobserved Error Terms, Censored regressors, Listwise Deletion, Dummy Indicator

JEL Classification: C01, C13, C15, C31