Abstract
The appealing but complex Hausman and Taylor (1981) random effects
(instrumental variable) estimator requires prior knowledge that certain explanatory
variables in a panel are uncorrelated with the latent group effects. The purpose of
this examination is to outline a tractable variable pretest that facilitates the initial
sorting of regressors as likely exogenous or endogenous. The variable pretest
proposed herein builds on the pretest estimator suggested by Baltagi et al (2003) by
providing the necessary foundation for regressor identification. Extensions are
suggested for the two-way error components construct.
Keywords: Panel data, Random effects, Variable pretest, Hausman-Taylor.
JEL Classification: C12, C13, C23.