Bartlett corrections in heteroskedastic t regression models

Abstract

This paper gives a general Bartlett correction formula to improve likelihood ratio tests in heteroskedastic t regression models where both location and dispersion parameters vary with the observations. The correction covers many important and commonly used models and can be viewed as an extension of the results in Cordeiro [1993. Bartlett corrections and bias correction for two heteroscedastic regression models, Comm. Statist. Theory Methods 22, 169–188] and Botter and Cordeiro [1997. Bartlett corrections for generalized linear models with dispersion covariates, Comm. Statist. Theory Methods 26, 279–307]. We present some Monte Carlo investigations of Bartlett corrections that show that this approach has better performance than the classical likelihood ratio tests even under degrees of freedom misspecification.