A comparative simulation study of the small sample powers of several goodness fit test

Abstract

Eight goodness of fit tests are compared with respect to their simulated small sample power of detecting an inbreeding alternative to the Hardy-Weinberg null hypothesis. The Pearson's x 2 test is found to be most powerful, and the small rample levels of this test are close to the nominal (x 2) significance levels. The use of conditional expectations, rather than expected frequencies based on ML estimates, increases the power and improves thc x 2 fit to the true significance level. The small sample powers are also compared to the asymptotic (Pitman) pourer, based on the noncenlral x 2 distribution.