Resumen
- The objective of this study was to analyze buffalo growth based on body weight, Longissimus dorsi muscle area (AOL), and fat deposition over the hip (FOH) using random regression mixed models of first (FORRM) and second order (SORRM), each with nine different variance structures. Ten measurements for each trait were taken on 26 animals during the first performance test (93 days test plus 23 d adaptation period) developed for buffaloes in Colombia. Computations were performed using the lme procedure of the R program. Preliminary analyses determined that an SORRM was appropriate for body weight (BW) and FOH and an FORRM was suitable for AOL. The maximum likelihood ratio (MLR), the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) were used to compare models. The best models were an SORRM with homogeneous residual variances for BW, an FORRM with heterogeneous animal residual variances for AOL, and an SORRM with heterogeneous residual variances among farms times an exponential function of age for FOH. Heterogeneity of residual variances was likely due to environmental differences among farms, and to genetic differences among buffaloes not accounted for by FORRM and SORRM. Fixed intercepts with the best models for each trait were 227 ± 7.90 kg for BW, 34.8 ± 0.99 cm2 for AOL, and 4.19 ± 0.229 mm for FOH. Fixe linear regression coefficients were 1.29 ± 0.073 g / d for BW, 0.0584 ± 0.0042 cm2 / d for AOL, and 0.0035 ± 0.0032 mm / d for FOH. The fixed quadratic regression coefficient indicated that BW rate decreased after one year of age whereas FOH rate continued to increase until the end of the test. Random regression coefficients suggested that there was considerable variability among trait curves for individual buffaloes, particularly for FOH. The evaluated residual variance structures did not eliminate completely the heterogeneity of variances.