Attenuating effects of preferential treatment with Student-t mixed linear models: a simulation study

Preferential treatment of cows in four herds of a multiple ovulation and em bryo transfer scheme under selection was simulated. Prevalence and amount o f preferential treatment depended on a function correlated with true breedi ng value. Three mixed effect linear models were compared in terms of their ability to handle preferential treatment: the classical Gaussian model, a m odel with multivariate t-distributed errors clustered by herd, and a model with independent t-distributed errors. In the models with t-distributed err ors, both the scale parameters and the degrees of freedom were considered u nknown. A Bayesian analysis was carried out for all three models via the Gi bbs sampler, and posterior means were used to infer about genetic variance, herd-year effects, breeding values and realised response to selection. Per formance over repeated sampling was assessed via Monte Carlo mean squared e rror. In the absence of preferential treatment; the three models had a simi lar performance. When preferential treatment was prevalent and strong, the univariate t-model was the best; hence, the Gaussian assumption for the err ors was clearly inappropriate. It appears that some robust linear models ca n handle preferential treatment of animals better than the standard mixed e ffect linear model with Gaussian assumptions. (C) Inra/Elsevier, Paris.