Avaliação de carcaças ovinas da raça santa Inês : uma abordagem multivariada

Abstract

Multivariate analysis corresponds to a set of methods that simultaneously use information from all response variables in the interpretation of the data set, taking into account the correlations between them. Among the sets of multivariate methods we can highlight factor analysis, multiple regression analysis and multivariate discriminant analysis. These methods have in common the objective to reduce the dimensionality of the data. Considering that a large number of variables are used in the carcass evaluation, there is the possibility that many of them contribute little to the discrimination of the evaluated individuals, as they are redundant due to the high correlations. Based on the above, the objective was, through factor analysis, to evaluate the adequacy of factor analysis and rotational methods; establish a multivariate model using two complementary multivariate statistical techniques, Factor Analysis and Stepwise Multiple Regression to predict tissue composition (muscle, bone and fat) and; distinguish groups of sheep as a function of conformation score and carcass finish and identify the variables that most contribute to differentiation, through multivariate discriminant analysis. Information from 122 sheep of the Santa Inês breed was used, comprising 24 carcass characteristics: Empty body weight (EBW); Body weight at slaughter (BWS); Hot carcass weight (HCW); Could carcass weight (CCW); Biological yield (BY); Hot carcass yield (HCY); Cold carcass yield (CCY); External carcass length (EXL); Internal carcass length (ICL); Rump width (RW); Chest width (CW); Rump perimeter (RP); Leg perimeter (LP); Chest perimeter (CP); Leg (LEG); Loin (LOI); Ribs (RIB); Neck (NEC); Blade Shoulder (BLA); Leg yield (LEGY); Loin yield (LOIY); Rib yield (RIBY); Neck yield (NECY); Blade Shoulder yield (BLAY). The sphericity test of Bartlett’s and KMO (Kaiser-Meyer-Olkin) indicated that the carcass variables evaluated were adequate for the application of factor analysis. Among the rotations tested, the Varimax orthogonal rotation presented the simplest structure for interpreting the constructed factors. The use of latent variables from factor analysis in multiple regression models eliminates the problem of multicollinearity of the explanatory variables, thus improving the accuracy of the interpretation of results by proposing a better adjustment of the mathematical model. However, the values of the coefficients of determination (R²) were moderate for the proportion of muscle and total fat and low for the proportion of bone, indicating that more adequate independent variables should be used to better predict the proportion of tissues in Santa Inês sheep. In the multivariate discriminant analysis, the variables with the highest discriminatory power for carcass conformation scores were CCW, EXL and NEC and for carcass finishing were BWS, EXL and CP. Multivariate discriminant analysis proved to be efficient to allocate the animals to their original groups (carcass scores).