Diagnostic and reduction of the influence of multicollinearity in the estimation of genetic additive and non-additive effects in multibreed population of cattle (Bos taurus x Bos indicus) / Diagnóstico e redução da influência da multicolinearidade na estimação de efeitos genéticos aditivos e não-aditivos em uma população de bovinos compostos (Bos taurus x Bos indicus)

AUTOR(ES)
DATA DE PUBLICAÇÃO

2009

RESUMO

The genetic additive and heterozygosity effects are important in the genetic evaluation of multibreed populations. When there is strong linear relation between the explanatory variables, the regression coefficients have large standard errors and are sensitive to changes in the data set and to the addition or removal of explanatory variables in the model. The alternative used to try to reduce this problem was to apply the method of ridge regression - RC, which could allow for the estimation of more stable coefficients of direct and maternal breed additive effects of genetic origin and heterozygosity in relation to those obtained by the method of least squares QM . The objective is to analyze the data of birth weight - PESNAS, weaning - PESDES, the scrotal perimeter 390 days - CE and scoring for the muscularity 390 days - MUSC of cattle compounds Montana Tropical r, with different racial compositions NABCs, obtained in several Brazilian farms on of animals born from 1994 to 2008. The model included additive and non-additive effects. The degrees of multicollinearity were obtained through the value of the variance inflation factor - V IF, the index conditions - IC and by proportional decomposition of Variance. The ridge parameters were obtained from the multiplication of a constant to the ratio of the VIF from each covariate and the highest VIF. For each explanatory variable, the ridge trace was used to verify that the estimated coefficients were stabilized using the ridge parameter. Two different methods were applied: i) the effects were estimated by least squares; ii) the effects were estimated by ridge regression. For each response variable the number of colinearities was identified, their degrees and the variables involved in each. The covariates used in the model for birth weight participated in a strong colinearity and four other weak colinearities; for weaning weight and muscle score for 390 days, there were two strong relations of dependency and three almost weak, while for the perimeter scrotal 390 days it was observed three strong and three weak colinearities. The ridge regression coefficients method was considered better than that of least squares for all factors. The V IFs average for PESNAS, PESDES, CE and MUSC reduced from 15.5, 16, 17.5 and 23.9 to 5.8, 5.3, 5.7 and 5.1 respectively, after using the RC. The standard errors of the estimators decreased providing estimates more stable than those obtained by least squares. Only for A covariate on the response variable weight at birth the solutions obtained by QM and RC differ in direction, where the other ones differed only in magnitude.

ASSUNTO(S)

component of variance análise de regressão e de correlação componentes de variância mínimos quadrados. square least. melhoramento genético animal cattle regression analysis and correlation animal breeding bovinos

ACESSO AO ARTIGO

Documentos Relacionados