Asymptotic Analysis of the Bayesian Likelihood Ratio for Testing Homogeneity in Normal Mixture Models
share
A normal mixture is one of the most important models in statistics, in theory and in application. The test of homogeneity in Normal mixtures is used to determine the optimal number of its components, but it has been challenging since the parameter set for the null hypothesis contains singular points in the space of the alternative one. Although in such a case a log likelihood ratio does not converge to any chi-square distribution, there has been a lot of research on cases that employ the maximum likelihood or a posterior estimators. We studied the test of homogeneity based on the Bayesian hypothesis test and theoretically derived the asymptotic distribution of the marginal likelihood ratio in the following two cases: (1) the alternative hypothesis is a mixture of two fixed normal distributions with an arbitrary mixture ratio, (2) the alternative is a mixture of two normal distributions with localized parameter sets. The results show that the log likelihood ratios are quite different from regular statistical model ratios.
READ FULL TEXT
