Bartlett-corrected tests for normal linear models when the error covariance matrix is nonscaiar

Abstract

This paper provides Bartlett corrections to improve likelihood ratio tests for heteroskedastic normal linear models when the error covariance matrix is nonscaiar and depends on a set of unknown parameters. The Bartlett corrections are simple enough to be used algebraically to obtain several closed-form expressions in special cases. The corrections have also advantages for numerical purposes because they involve only simple operations on matrices and vectors.

Keywords:

• Bartlett correction
• heteroskedastic model
• likelihood ratio test
• maximum likelihood estimate
• normal linear model

Additional information

Notes on contributors

Elisete C. Q. Aubin