Comparação de modelos de regressão aleatória para estimação de parâmetros genéticos em caprinos leiteiros

Random regression and Legendre polynomial (LP) of different orders were used for modeling the genetic, permanent environmental and residual variances of test day milk yield in dairy goats. The models included the fixed effects of contemporary group, age of dam at kidding as a covariate and the fixed regression of LP for the average lactation curve and the additive genetic, permanent environmental and residual as random effects. According to the values of the logarithm of the likelihood function, AIC and BIC using higher orders of LP (fifth order for the genetic effect and seventh order for the permanent environmental effect) improved the fitting of the models. The model with four classes of residual variances provided the best fit. The eigenvalues of the (co)variances matrix among the regression coefficients suggested the possibility of reducing the dimension of the models. The estimates of genetic variances and genetic and permanent environmental correlations for test day milk yields obtained from polynomials of higher orders are not biologically expected. The LP of fifth order for the addictive genetic and seventh order for the permanent environmental effects was the best fitted model. However, a LP of fourth order for the addictive genetic and of sixth order for the permanent environmental effects may be considered as a more parsimonious model for the estimation of variances of test day milk yield in dairy goats, by random regression.