Estimação de componentes de (co)variância e de tendências genéticas em populações simuladas

Abstract

Alternative methods for the estimation of genetic trends for weight at 550 days (P550) in 10 herds, submitted to 20 years of selection were studied. The familiar relationship between animals and the heterogeneity of (co)variance by the inclusion of genetic (co)variance between the year of birth of the animals were contemplated. For the construction of the (co)variance structure a model of multiple regression from the estimated components for each generation of selection, obtained through multi-character analysis for Bayesian inference, was used. The determination coefficient (R²) was high with mean of 0.90272 and standard deviation of 0.03936. This indicates that genetic (co)variance components among years can be estimated precisely by multiple regression using the components estimated in each generation. The genetic trends were estimated for 8 years of selection, with two conditions (C1 and C2) applied to the genetic evaluation. In C1 the evaluation considered all the animals until 20th year of selection and in C2 only the animals that were included on the data set used to make the estimation of the genetic trend were considered. The genetic trend was obtained by the adjustment of regression equations, alternatively, through the methods of Ordinary Least Squares (OLS), Weighted Least Squares (WLS) and Generalized Least Squares (GLS), this last one under two forms, one considering homogeneity of variance and another considering heterogeneity of variance throughout the selection years. A quadratic behavior of the genetic values as a function of the year of birth of the animals was verified for the 10 herds studied in the first condition and for herds 1, 4, 8 and 9 in the second condition. For C1 the WLS and GLS methods, although allowing greater inclusion of information in the model, were not able to detect the genetic changes with great precision, the OLS being the method that allowed a greater approach to the annual averages of the real and predicted genetic values. For C2 the GLS method, considering heterogeneity of variance throughout the years of birth of the animals, was closer to the real genetic changes.