Time Invariant Variables in the Mundlak and Hausman-Taylor Panel Data Models

This paper shows that the classic augmented Mundlak (1978) regression yields the between estimator for time-invariant variables. It is well known that the estimates of the time varying variables yield the fixed effects estimates. While the latter are consistent for this correlated random effects model, the between estimates are not. The between estimator is consistent only when the Hausman (1978) test does not reject the null based on between versus fixed effects. An alternative modified Hausman and Taylor (1981) estimator is proposed. Monte Carlo experiments are performed to compare various estimators under a correlated random effects as well as a purely random effects model.