Univariate and bivariate Bayesian analysis for feed conversion of the Piau swine breed

Abstract

The objective of this work was to present alternative uni‑ and bivariate modeling procedures for the evaluation of feed conversion (FC) of the Piau swine breed, using Bayesian inference. The effects of sex and genotype on animal FC were evaluated by the Markov chain Monte Carlo (MCMC) and the integrated nested Laplace approximation (INLA) procedures. The univariate model was evaluated using different distributions for the error – normal (Gaussian), t‑Student, gamma, log‑normal, and skew‑normal –, whereas, for the bivariate model, the normal error was considered. The skew‑normal distribution was the most parsimonious model to infer on the direct response (univariate) of FC to the effects of sex and genotype, which were nonsignificant. The bivariate model was capable to identify significant differences on weight gain and feed intake in significance levels not detected by the univariate model. Moreover, it was also able to detect differences between sexes, when grouped by NN (male, 2.73±0.04; female, 2.68±0.04) and Nn (male, 2.70±0.07; female, 2.64±0.07) genotypes, and revealed greater accuracy and precision for nutritional inferences. In both approaches, the Bayesian method proves flexible and efficient for assessing animal nutritional performance.