On the testing of correlated effects with panel data

Abstract

Using orthogonal deviations of the variables, correlated effects biases are regarded as misspecification biases due to the exclusion of relevant variables in a standard regression model. Following this approach, Hausman- and Chamberlain-type tests of correlated effects are obtained as Wald tests in an extended model estimated by OLS, and robust generalisations are suggested. Alternative estimators which introduce restrictions in the regression of the effects on the explanatory variables are proposed. Finally, the paper extends the results to dynamic models. © 1993.