Non-normality in carcass quality measurements and effects on the genetic evaluation of sheep

Abstract

In order to estimate genetic parameters for four traits related to carcass quality in sheep (live weight, ultrasonic fat and muscle depth and in vivo conformation, assessed at 21 weeks of age), the validity was tested of the assumptions of homogeneity of variance and normality of residuals. Animals were of the Bluefaced Leicester breed, the most prevalent crossing sire breed used in the UK. The fit of the model and the distribution of the residuals were investigated when fitting a mixed (animal) model after applying Box–Cox power transformations to deal with scale-related non-normality. For ultrasonic fat depth, the log-transformation was shown to improve normality and decreased the correlation between flock means and standard deviations. The other traits did not show sufficient improvement to justify a transformation. For ultrasonic fat depth, the regression of both genetic and residual variance on sample means obtained by bootstrapping (sampling flocks) was positive whereas the regression of heritability on sample means was negative. The log-transformation made the association weaker and the regression coefficient changed sign for genetic variance only. The four traits all had moderately high heritabilities. Log-transforming ultrasonic fat depth caused the heritability to decrease from 0.43 to 0.30. Conformation had the lowest heritability (0.20) but had high positive correlations with the other traits, suggesting that improving conformation could lead to more muscling but may also increase the amount of fat in the carcass. The log-transformation had a major effect on ranking of individuals based on estimated breeding values, particularly for high selection intensities.