Multiple imputation methods for the missing covariates in generalized estimating equation

This paper discusses the missing covariates problem in the generalized estimating equation (GEE) model. Estimates by various multiple imputation techniques (MI) are examined and compared to the sample average imputation method (SA) through simulations and an example. The simulation results show that, under the correct model specification, the MI estimators have negligible bias and have fairly similar efficiencies as the SA estimator. A practical advantage of the MI estimates is that the standard errors can be more easily computed than the SA estimates.