Genetic evaluation of US dairy cattle in the presence of preferential treatment
Data were simulated according to the USDA animal model for the purpose of determining bias in female and sire predicted transmitting abilities (PTA) caused by preferential treatment (PT). Bias in female PTA ranged from six to 39 percent of the PT effect and varied according to number of records on the PT cow, whether all or only second and later records received PT, and which, if any, relatives received PT. Bias in sire PTA ranged from 10 to 77 percent of the PT effect and depended largely on proportion of daughters receiving PT and, to a lesser extent, total number of daughters. Distribution of daughters across herds also affected bias in sire PTA. As more PT daughters were placed in a single herd, bias decreased because a larger portion of the PT effect was accounted for by the management group and sire by herd interaction effects;Biases found were large. Thus, preliminary research was conducted on possible methods to correct for PT. Three approaches were examined: power transformations applied to the phenotypic records, fitting a random PT effect in the model for genetic evaluation, and a two-group mixture model. Transforming records with a power of.1 reduced bias to zero but had a very adverse effect on ranking. The power transformation cannot be recommended for use in practice;The random PT and mixture model approaches, however, both hold promise as a means to correcting for PT in genetic evaluation. When PT records were identified, the random PT effect reduced bias to nearly zero. Further development of the random PT approach should focus on improving the identification step. PT records were identified, at best, 50 to 60 percent of the time;The mixture model brought about only a small reduction in bias, but this reduction was achieved without any loss in mean true transmitting ability of the top 5% of cows. The shortcoming was the conditional probabilities of PT for PT records. Even in the best case, average probability was only.27. Utilizing the correct variance of residuals may be adequate to improve calculation of these conditional probabilities.