Tests in covariance selection models
Poul Svante Eriksen · VBN Forskningsportal (Aalborg Universitet) · 1996
Consider the likelihood ratio test between two nested covariance selection models. It is shown that the distribution of the test statistic raised to the power 21n can be approximated by a product of independent beta distributions. Furthermore, conditions ensuring exactness of the approximation is given, and for this case the test statistic is shown to be independent of the maximum likelihood estimator under the null hypothesis. A simulation study reveals that the approximation is much superior to the usual chi-square approximation for small and moderate sample sizes.