Modeling growth performance of heterogeneous rabbits in a pastured system using nonlinear, spline and random regression models

Accurate modeling of growth trajectories is critical for decision-making in rabbit production, particularly under low-input systems. However, limited studies have evaluated the comparative performance of growth models for rabbits reared in such systems. This study evaluated the growth performance of heterogeneous rabbits using three modeling approaches. Weekly body weight records of 343 rabbits from birth to 20 weeks of age across four parities were analyzed. Four classical nonlinear models (Gompertz, Logistic, von Bertalanffy, and Richards) were fitted using the NLIN procedure of SAS®. Further, the REG procedure of SAS® was used to fit four linear spline models (SP3 to SP6) while four random regression models with varying variance-covariance structures (Compound Symmetry, Unstructured, Autoregressive, and Toeplitz) were also fitted to the growth data using the MIXED procedure of SAS®. There were significant effects of parity (<i>p</i> < 0.05) on growth, with kits from later parities consistently achieving higher body weights than those from earlier parities. Among nonlinear models evaluated, the von Bertalanffy model had the lowest AIC, ΔAIC, Akaike weight and BIC values and was therefore selected as the best fit model. However, all spline regression models performed poorly and consistently over-estimated body weight at all ages. Random regression analysis also showed that the Unstructured model had the best performance in terms of goodness-of-fit tests evaluated. The findings suggest that nonlinear models especially the von Bertalanffy and random regression models with flexible covariance structures provide more accurate and biologically interpretable fits than spline models for growth prediction in heterogeneous rabbits.