Development of a model for estimation of breeding values for beef cattle in Germany .1. Modelling the herd effect

The analysis of modelling the herd effect for estimation of breeding values in beef cattle was based on field data of Charolais. Traits of interest were birth weight, weaning weight and yearling weight Models considering herd and herd-year effects as fixed and random as well as models with a random herd-year-season effect were compared with each other. Criteria for the comparison were log-likelihood, residual variance, accuracy of prediction of breeding values and correlation of breeding values of sires based on two randomly divided data sets. Hierarchical model comparison showed significance of the interaction between herd and year for the fixed as well as for the random effect model as indicated by the highest log-likelihoods. When the interaction between herd and year were additionally considered in the fixed effect model the accuracy of estimation of breeding values decreased from 0.67 to 0.63 (0.67 to 0.64) and the correlation between breeding values of two randomly divided data sets decreased from 0.55 to 0.25 (0.52 to 0.36) for weaning weight (yearling weight). In contrast, when the interaction between herd and year were additionally fitted in the random effect model, the accuracy of estimation of breeding values increased from 0.67 to 0.72 (0.68 to 0.71) and the correlation between breeding values of two randomly divided data sets increased from 0.56 to 0.62 (0.56 to 0.67) for weaning weight (yearling weight). These indicated the desirability of the random model, but the concern was discussed that this model may result in biased estimation of breeding values due to non-random distribution of genetic potential over herd-years. However, a conclusion based on weighing increased In accuracy and correlation between breeding values with the possible biased estimation of breeding values as well as the indication of random distribution of genetic potential over all herds; resulted in the recommendation of using the interaction of herd-year effects as random.