Merit of outliers for milk yield as indicators of accuracy of genetic evaluations of sires.

To determine whether the accuracy of the genetic evaluations of sires can be assessed by the presence of extreme daughter records, we studied herd-years with records from first-crop daughters of 217 Holstein bulls that were not sampled by artificial insemination (AI) organizations but that later entered AI. The presence of outliers for standardized milk yield was determined within herd-year. Outliers were defined as records exceeding 1.5 interquartile ranges below the 25th percentile or above the 75th percentile. Herd-years were separated into two groups based on whether or not an outlier daughter record was present for an AI bull that had initially been sampled through non-AI. Herd-years without daughter outliers from those bulls were divided into herd-years with 1) no daughter outliers from any bull, 2) only negative daughter outliers from other bulls, 3) only positive daughter outliers from other bulls, or 4) negative and positive daughter outliers from other bulls. Herd-years with daughter outliers from AI bulls initially sampled through non-AI were divided into herd-years with 1) only negative daughter outliers, 2) only positive daughter outliers, 3) positive daughter outliers from those bulls and negative daughter outliers from other bulls, or 4) both negative and positive daughter outliers. The relationship between the frequency of outlier classes and a change in the Modified Contemporary Comparison genetic evaluations (the difference between the last available second-crop evaluation and the next to the last first-crop evaluation) was examined with logistic regression. For AI bulls that were initially sampled through non-AI and having evaluations that decreased > or = 386 kg, 9% of herd-years had positive first-crop daughter outliers and negative daughter outliers from other bulls; 38% had no outliers. For bulls with evaluations that increased > or = 194 kg, comparable percentages were 2 and 53%.