SELECTION CORRECTIONS FOR PANEL-DATA MODELS UNDER CONDITIONAL MEAN INDEPENDENCE ASSUMPTIONS

Some new methods for testing and correcting for sample selection bias in panel data models are proposed. The assumptions allow the unobserved effects in both the regression and selection equations to be correlated with the observed variables; the error distribution in the regression equation is unspecified; arbitrary serial dependence in the idiosyncratic errors of both equations is allowed; and all idiosyncratic errors can be heterogeneously distributed. Compared with maximum likelihood and other estimators derived under fully parametric assumptions, the new estimators are much more robust and have significant computational advantages.