It was used 17,767 weight records of 4,210 Santa Inês breed lambs aiming to compare random regression models with different structures to model the residual variance in genetic studies of the growth curve. The fixed effects included in the analysis were contemporary group and age of the ewe at lambing. Fixed and random regressions were fitted through Legendry polynomials of orders 4 and 3, respectively. The residual variance was fitted by heterogeneous classes and by functions of variances employing ordinary polynomials and Legendry polynomials of the orders 2 to 8. The model considering homogeneity of residual variances was inadequate. Accordingly to the used criteria, the residual variance containing seven heterogeneous classes provided the best fit, although a more parsimonious one, with five classes, could be used without losses on the quality of variance fit on the data. The fit of functions of variances with any order was better than that obtained through classes. The ordinary polynomial of order 6 provided the best fit among the tested structures. The modeling of the residue interfered on the estimative of the variances and genetic parameters. In addition to changes in the classification of the reproducers, the magnitude of the predicted genetic values shows significant variations, accordingly to the fitting of the used residual variance.
animal model; breeding value; genetic parameters; hair sheep; heritability; heterogeneity of variance
