Estimation of variance components using Bayesian and frequentist inferences considering simulated data under heterogeneity of variance

A genome of 3.000 centimorgans was simulated for a single quantitative trait governed by 900 loci with two alleles per locus. According to the genomic structure proposed, 1,500 males and 1,500 females constituted the base population which was used to form two (small and large) initial populations. Two types (only additive genetic variance and both additive genetic and environmental variances) and three variability levels (high, medium and low) of heterogeneity of variances were inserted in the initial populations. Variance components were estimated by Bayesian inference via Gibbs Sampling using three different levels of priors (non-informative, slightly informative and informative) and by REML. The best estimates of variance components were obtained with large populations. In small populations, the individual analyses for different variability levels presented problems related to variance components estimation due the small size of subpopulations. Both methods presented similar results for variance components when non-informative priors were used in Bayesian inference. Increasing the level of a priori information improved the estimates of variance components by Bayesian inference, especially in small populations.