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

Abstract This paper attempts a replication of the Cornwell and Rupert (1988) study—hereafter CR. The CR study investigated the efficiency gains in a returns to schooling example by applying alternative sets of instrumental variables estimators for panel data regressions proposed by Hausman and Taylor (1981), Amemiya and MaCurdy (1986), and Breusch, Mizon, and Schmidt (1989). Corrections on the CR data set lead to changes in the legitimate set of instruments, when the time dummies are excluded from the regression, and to much lower empirical gains in efficiency than those reported in CR. If the time dummies are retained in the wage equation, the experience coefficient is not estimable by the within regression, and the empirical gains in efficiency from using the IV procedures are not limited to the time‐invariant education coefficient.