Efficient estimation with panel data: An empirical comparison of instrumental variables estimators

Abstract Hausman and Taylor (1981) proposed an efficient instrumental variables estimator for panel data regression models where the individual effects may be correlated with some of the regressors. Amemiya and MacCurdy (1986) and Breusch, Mizon, and Schmidt (1986) have suggested instrumental variables estimators that are potentially more efficient than the Hausman and Taylor estimator. We address the empirical question of how large these efficiency gains might be. In our returns to schooling example noticeable efficiency gains are limited to the coefficients of time‐invariant endogenous variables. This is important, however, since education does not vary over time in our sample.