Comparison of different legendre and B-spline random regression models of lamb's estimate variance components for average birth weight per lambing in Mehraban sheep

The present study was carried out to compare different random regression models to estimate variance components of lamb's average birth weight per lambing (ABWLL) in Mehraban sheep. The data were 5,559 ABWLL records of 2,244 Mehraban ewes. The random regression models consisted of namely, flock-year-season of lambing as fixed effect, a fixed regression to fit average trajectory of the population and two random regressions to fit random additive genetic and permanent environmental effects. The models had linear and quadratic B-Spline or quadratic or cubic Legendre functions, all with heterogeneous residual variances. Variance components were estimated using Average Information algorithm of Restricted Maximum Likelihood (AI-REML). According to Akaike and Bayesian information criteria, the model BS212 with quadratic, linear and quadratic B-Spline functions for fixed regression and random regressions of additive genetic and permanent environment was considered as the best model to fit the data. Using the BS212 model, the highest and lowest heritabilities for ABWLL were estimated for 12 months of age (0.74) and 31 to 66 months of the ewe age (0.03), respectively and coefficients of permanent environment were close to 0 in all ages. The estimated values in middle ages were more accurate than initial or final ages. The results of the present study showed that the records obtained on initial lambings probably could have higher importance for breeding programs.