Skip to main content
ARS Home » Research » Publications at this Location » Publication #146894


item Vanraden, Paul

Submitted to: Interbull Annual Meeting Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 3/2/2003
Publication Date: N/A
Citation: N/A

Interpretive Summary: Methods to evaluate dairy cattle longevity and fertility traits were compared in theory and using simulated data. Genetic effects are estimated more accurately from repeated records than from single records. Time until pregnant or time until culled can be modeled as a series of binomial (yes / no) observations that are repeated only if the cow is not yet pregnant or culled. This practice results in more observations, but each binomial observation contains less information than does a single length of time measure. Data were simulated using 500 sires with 10 to 1000 daughters each. Means and standard deviations were obtained for 10 replicates. Heritability was found to be 3.8% for culling rate and 10.6% for number of lactations. The estimated breeding value for number of lactations and the binomial culling rate from the same data were correlated by 99.1%. Accuracy of the two estimated breeding values was nearly identical, but reported reliabilities differ slightly. In the binomial analysis, a cow that survives many years is assumed to contribute more information to her own and to her sire¿s longevity evaluation than a cow culled after just one lactation because her repeated records provide more information about the culling rate. Including all relatives is more important when heritabilities are low. The results suggest that accurate estimates of breeding values can be obtained with linear models even though data are not normally distributed.

Technical Abstract: Longevity and fertility traits often are based on binomial observations that are repeated only if the cow is not yet pregnant or culled. Time until pregnant or time until culled are not normally distributed, but theory indicates that linear models can extract all information from such data. With binomial models, cows with more observations are assumed to provide more information about the success rate parameter. Although reported reliabilities differ slightly for binomial models as compared with productive life models, estimated breeding values were correlated by 0.991 and had identical true accuracies as obtained from simulated data. For traits with low heritability, information from progeny-tested maternal brothers can lead to larger benefits from animal models as compared to sire models.