Approaches investigated to correct for preferential treatment

Simulated data were used to investigate the effectiveness of three approaches to correct for preferential treatment in the genetic evaluation of US dairy cattle, The methods tested included power transformations applied to the phenotypic records, fitting a random preferential treatment effect to suspect records that were defined according to magnitude of residual, and a two-group mixture model. Transformation of records with a power of 0.1 reduced bias to zero, but had a very adverse effect on ranking. Fitting a random effect for preferential treatment in the model for genetic evaluation was effective in reducing bias, given an appropriate variance for the preferential treatment effect, but records with preferential treatment were typically identified only 45 to 60% of the time. Reductions in bias brought about by the mixture model were small but were accomplished without negative effects on ranking. Although not yet ready for use in practice, results for the effect of random preferential treatment and the two-group mixture model were favorable enough to warrant further development of these methods as possible approaches to correcting for preferential treatment in genetic evaluation.