Random regression in comparison with finite-dimensional models for estimation of genetic parameters for growth traits in goats

The application of the random regression model in comparison with finite-dimensional models (univariate and multivariate animal models) for genetic parameter estimation of growth traits in goats was evaluated in this study. A total of 2888 body weight records from 875 animals, recorded from birth to yearling age were used. All models included direct additive genetic and maternal genetic effects as a random effect in addition to fixed effects. Random regression model (RRM) was fitted with different orders (1st – 3rd) of Legendre polynomials and accounted for both homogeneous and heterogeneous residual variance. The best-fitting RRM had a polynomial of three orders for both random effects. The direct heritability estimate obtained via RRM was moderate to high, while it varied from 0.00 ± 0.08 to 0.36 ± 0.10 in finite dimensional models. A lower standard error of heritability and genetic correlation estimates was observed with RRM compared to multivariate (MUV) and univariate (UNI) analysis. Likewise, high accuracy and reliability of breeding value estimates are obtained via RRM, whereas the accuracy for MUV and UNI animal models were moderate and low to moderate, respectively. Based on standard errors, accuracy, and reliability of estimates, RRM seems versatile for genetic evaluation of growth traits of goats. However, the MUV animal model is the best-fitting model, according to the information criteria values. Thus, for small and less frequently measured data set, multivariate animal model seems good. Further studies with large and frequently measured body weight data sets may help ensure random regression’s applicability and differentiate it from finite-dimensional models.