Where is the likelihood ratio test powerful for detecting two component normal mixtures?

Biometrics. 1993 Sep;49(3):907-15.

Abstract

We compare the power of the likelihood ratio test, the Engelman-Hartigan test, two outlier tests, four goodness-of-fit tests, and eight tests of normality to detect a mixture consisting of two components that are normally distributed with different means but equal variances. We consider the entire range of mixing proportions pi, 0 < pi < 1. For pi > .85 or pi < .15, overall Fisher's skewness statistic is best with Filliben's probability plot correlation coefficient test somewhat less powerful. A combined skewness and kurtosis test, the Anderson-Darling test, and the likelihood ratio test are also competitive. For .35 < pi < .65, the Engelman-Hartigan test is best. For other mixing proportions, the likelihood ratio test is best. For situations in which the preferred test had power 50% or more, the power of the likelihood ratio test is also above 50% and within 15 percentage points of the preferred test.

Publication types

  • Comparative Study
  • Research Support, U.S. Gov't, P.H.S.

MeSH terms

  • Alleles
  • Biometry / methods*
  • Evaluation Studies as Topic
  • Genetics, Medical / statistics & numerical data*
  • Genetics, Population
  • Humans
  • Likelihood Functions*
  • Models, Genetic