Modelos estatísticos empregados para descrição do crescimento de codornas de corte

Abstract

Non-linear regression models were evaluated with Brody, Gompertz, Logistic, MorganMercer-Flodin (MMF), Richards and von Bertalanffy functions in the analysis and interpretation of the growth curve for broiler quails. Further, 30,410 body weight registers were used at 1, 7, 14, 21, 28, 35 and 42 days old of male and female quails for the Genetic Improvement Program of the Universidade Federal dos Vales do Jequitinhonha e Mucuri (UFVJM). Models comprised fixed and mixed modalities. Brody function failed to converge efficiently, whereas the Gompertz model proved to be the most adjusted when fixed models based on Aikake (AIC) and Bayesian (BIC) Information Criteria were taken into consideration. When the asymptotic body weight, called parameter A, is randomized, the Gompertz, Logístico, MMF and von Bertalanffy models had a decrease in residual standard deviation estimates and an increase for AIC and BIC. In the case of randomized parameter K, only Richards function converged, whilst no model converged with A and K simultaneously randomized. Richards had the lowest AIC and BIC rates among the mixed models, featuring problems in the Hessian matrix resulting in the estimation of unusable parameters, perhaps due to the multi-co-linearity among the parameters. Mixed non-linear models reduced prediction bias when the residual standard deviation was reduced. The MMF model fitted best to describe the growth curve of broiler quails.