Random regression models with different residual variance structures for describing litter size in swine

Abstract

The objective of this work was to compare random regression models with different residual variance structures, so as to obtain the best modeling for the trait litter size at birth (LSB) in swine. One thousand, seven hundred and one records of LSB were analyzed. LSB was analyzed by means of a random-regression, single-characteristic animal model. The fixed and random regressions were represented by continuous functions over the farrowing order, adjusted by third-order Legendre’s orthogonal polynomials. To obtain the best modeling for the residual variance, variance heterogeneity was assumed by means of 1 to 7 classes of residual variance. The general analysis model included a contemporary group; the fixed regression coefficients for modeling the population’s average trajectory; the random regression coefficients of the direct additive genetic effects both of the litter and of the animal’s permanent environment; and the residual random effect. The likelihood-ratio test, Akaike’s information criterion, and Schwarz’s Bayesian information criterion appointed the model that considered variance homogeneity as being the one that provided the best adjustment to the data used. Overall, the heritabilities obtained were close to zero (0.002 to 0.006). Regarding the permanent environment proportion, different magnitudes were observed for the farrowing order: increasing from the 1st (0.06) to the 5th (0.28) orders and decreasing from there to the 7th order (0.18). The common litter effect presented low values (from 0.01 to 0.02). The use of residual variance homogeneity was more suitable for modeling variances associated to the trait litter size at birth in this data set.