Prediction of dairy bull fertility from field data: A comparison of linear and threshold models and effects of matings with unknown outcomes

M.T. Kuhn and J.L. Hutchison

Animal Improvement Programs Laboratory, Agricultural Research Service, USDA, Beltsville, MD 20705-2350

2007 J. Dairy Sci. (?)
© American Dairy Science Association, 2007. All rights reserved.
Individuals may download, store, or print single copies solely for personal use.
Do not share personal accounts or passwords for the purposes of disseminating this article.


        The objectives of this research were to compare linear and threshold models for US phenotypic evaluation of bull fertility, based on field data, and to determine the consequences of matings with unknown outcomes (unconfirmed matings). The effect of errors in positive pregnancy exams was also investigated. Both simulation and field data were used to address these objectives, but simulation was the primary method used for comparison. Linear model predictions compared in simulation included 1) using the true (binary) data, 2) setting all last services on file to success, 3) unconfirmed matings deleted, 4) unconfirmed matings deleted with additional herd edits, and 5) a 10% error rate in positive pregnancy exams. The 3 threshold predictors compared in simulation corresponded to options 1, 2, and 4 used for the linear model. In simulation, predictions were compared to true values using, primarily, correlations and mean differences. Two general types of comparisons were done with field data: 1) evaluations from linear and threshold models were simply compared to each other (correlations and mean differences) and 2) predicted conception rates (CR) were compared to observed future yr CR, where the future yr data was not included in estimation.
        Results with both simulated and real data showed no advantage of threshold over linear model evaluations. In simulation, correlations with true values were identical for the 2 models and mean differences were actually slightly closer to 0 for the linear model. With real data, the 2 models provided nearly identical evaluations; the largest difference between the linear and threshold model predictions, across all 4,790 bulls evaluated, was 0.8%. Linear and threshold models can be used indifferently for prediction of US service sire CR.
        Unconfirmed matings account for only 2% of all breedings in the US. Deleting unconfirmed matings provided better evaluations than assuming all last services to be successes. The effect of deleting unconfirmed matings was small and largely negligible. Across all bulls, the loss in accuracy, relative to knowing all true outcomes, was only 0.3%. Young bulls have a larger percentage of unconfirmed matings than older bulls because the frequency of unconfirmed matings is higher in the most recent yr and young bulls have a larger proportion of their breedings in the most recent yr. Even young bulls, though, had only a 1% loss in accuracy, relative to knowing all true outcomes. Furthermore, bias due to deleting unconfirmed matings was negligible because the effect on the mean is partitioned largely into fixed effects. Excluding unconfirmed matings will provide satisfactory predictions. Pregnancy check errors resulted in only a slight additional loss of accuracy and no bias. While methods to incorporate unconfirmed matings are not necessarily to be discouraged, they will provide at most only slight improvements to US bull fertility evaluations.

(Key words: bull fertility, conception rate, threshold model, artificial insemination)