Tests for Unbalanced Error-Components Models under Local Misspecification

Abstract

This paper derives unbalanced versions of the test statistics for first-order serial correlation and random individual effects summarized in Sosa-Escudero and Bera (2001, Stata Technical Bulletin Reprints, vol. 10, pp. 307–311), and updates their xttest1 routine. The derived test statistics should be useful for applied researchers faced with the increasing availability of panel information where not every individual or country is observed for the full time span. The test statistics proposed here are based on ordinary least-squares residuals and hence are computationally very simple.