Erro tipo I, poder do teste e intervalos de confiança com técnicas de reamostragem em modelos de herança genética utilizando misturas de normais

Abstract

In plant breeding programs, it is common to carry out studies to investigate the nature of allelic interactions and if a major gene and/or polygenes a represent. If the trait under consideration is continuous and a major gene is present, the probability distribution is a mixture of densities, generally Gaussian. The likelihood ratio test can be used to test the existence of polygenes and/or major gene. Data simulation was used to evaluate such test, accounting for the type I error rate and the power of the test. Results showed that the test kept type I error rates under the nominal value, being somewhat conservative. Test power was high to detect polygenes and the major gene inmost cases, but not to detect the major gene when it explained a small ration of phenotypic variation. Therefore,the likelihood ratio test was excellent, in general. Another aim of this study was to obtain interval estimates of the parameters using the resampling techniques of bootstrap and jackknife. The bootstrap was implemented using the methods of percentiles and BCa (which has a bias correction and an acceleration parameter). It could be seen that all resampling confidence intervals were adequate, agreeing with the conclusions of the likelihood ratio test with regard the choice of the model.