This article shows that the presence of incidental parameters in Baltagi and Griffin's (International Economic Review, 29 (1988), 745-53) generalized error-components model implies there is no guarantee feasible and true GLS estimators have the same asymptotic distribution. The article then considers a related stratified error-components model that has no incidental parameters. In this model, unobserved heterogeneity occurs through variances changing across strata. The article provides an EM algorithm for calculating estimates of the model's parameters without assuming observations can be classified into strata, derives a bootstrap test for identifying the number of strata, and applies the new methods in an example. Copyright 2003 By The Economics Department Of The University Of Pennsylvania And Osaka University Institute Of Social And Economic Research Association.
