Evaluation of the generalized likelihood ratio test of hypotheses on the genetic control of continuous traits</n

Abstract

Studies of genetic control in plants are carried out to characterize genetic effects and detect the existence of a major gene and/or genes of minor effects (polygenes). If the trait of interest is continuous, the likelihood can be constructed based on a model with mixtures of normal densities. Once exact tests are not evident with such models, the likelihood ratio test is generally used, using the chi-square approximation. This work aimed at evaluating such test statistic using computer simulation. Data sets were simulated using generations typical in plant studies, under two conditions of null hypothesis, without a major gene, and without polygenes. The power of the test was evaluated with both types of genes present. Different sample sizes and values of heritability were considered. Results showed that, although the empirical densities of the test statistic departed significantly from a chi-square distribution, under null hypotheses, there was a reasonable control of type I error, with a significance level of 5%. The power of the test was generally high to detect polygenes and major genes. Power is low to detect a major gene only when it explains a low fraction of genetic variation.