Dimorfismo sexual de cães da raça Labrador Retriever no Brasil utilizando estatística multivariada

Abstract

The aim of this study was the morphological characterization of the Labrador Retriever in Brazil by morphometry, and the accreditation of different multivariate techniques in the study of sexual dimorphism of these dogs. Were measured 74 animals, 47 females and 27 males. The dogs met the following requirements: be adult animals, aged two years or more; being breeder or array; being registered at AKC or CBKC; and no noticeable signs of pregnancy or lactation. Were measured 30 quantitative biometric characteristics related to the morphology of the head, trunk and front and hind limbs. The sexual dimorphism was analyzed using multivariate statistical techniques, were they the principal component analysis and discriminant analysis. Both the descriptive data and multivariate analyzes were analyzed using the software Statistica 6.0. The principal component analysis was processed in two ways, using all variables and performing a pre-selection of more correlated variables based on graphic dispersion of the correlation of the 30 original variables with the first three principal components, which together explained 50% of total variation. A discriminant analysis was performed with 30 variables and also with the five variables most correlated with the first component (PC1), in order to classify new individuals. The average height at withers for both sexes were lower than the minimum values of the tracks described as ideal by the AKC, while for CBKC only females were lower than the ideal minimum value according to Student's t-test (p <0 , 05). The principal component analysis with 30 variables was able to identify the sexual dimorphism existent in animals and reduce the 30 original variables to three principal components. When processed with the pre-selected variables, the analysis continued efficient in to demonstrate the sexual dimorphism and was able to improve the reduction to two components. The CP1 was the most representative of the analysis (30 variables and pre-selection), and this component is highly correlated with variables related to the size of the animal. The discriminant analysis of Anderson was able to discriminate the two populations (males and females), both for 30 variables as for the five variables most correlated with the CP1. However, the larger number of variables reduces the misclassification probability. The functions with five variables can be used to classify other dogs of the breed by sex, with an error of about 6.75%.