The influence of outliers on a model for the estimation of crossbreeding parameters for weaning weight in a beef cattle herd

Abstract

Data on 17348 weaning weight records from a beef cattle crossbreeding operation were used to determine the effect of outliers on regression coefficients. Different criteria were used for detecting potential influential points. Eliminating a small number (932 or 5.4%) influential points resulted in the improvement of the model fitted. The R2 values increased from 41% to 49% while the mean square error was reduced from 672.9 to 500.4. The use of diagnostic statistics for detecting influential observations is recommended before any analysis is performed.


South African Journal of Animal Science Vol.32(3) 2002: 164-170