ARS | Publication request: Impact on Calving Ease Evaluations of Excluding Herds with Abnormal Distribution of Scores

Submitted to: Journal of Dairy Science
Publication Type: Abstract
Publication Acceptance Date: March 23, 2004
Publication Date: July 25, 2004
Citation: Van Tassell, C.P., Wiggans, G.R., Thornton, L.L.M. 2004. Impact on calving ease evaluations of excluding herds with abnormal distribution of scores [abstract]. Journal of Dairy Science. 87(Suppl. 1):84-85.

Technical Abstract: Threshold model sire and maternal grandsire (MGS) calving ease evaluations for the United States were calculated after excluding two sets of 10% of the data. Herds were excluded based on chi squared goodness of fit test. To determine the first set of herds to be excluded, frequencies of observed (O) records were determined by parity (1 vs. 2+) and difficulty score (1-5). Expected (E) numbers of records were calculated as the total number of calvings by parity multiplied by the fraction of difficulty scores across all herds in each parity. Goodness of fit values were calculated as the sum of (O-E)squared/E across parity and difficulty score. The 299 herds failing this test tended to be large. The second set of 8605, mostly small herds, was identified using frequencies in place of counts. Evaluations were calculated using data from all years or only calvings before 1999. Pedigree indexes (0.5*sire + 0.25*MGS) were calculated from two earlier data sets (abnormal distribution herds included or excluded). Correlations were calculated between pedigree index and solutions from evaluations for bulls with at least 30 calvings in the complete data. Correlations were higher between solutions from sets with abnormal distribution herds excluded. For the first set of excluded herds, for sire solutions, the increase was from .579 to .599 for 25,067 bulls, and for MGS, from .598 to .617 for 21,221 bulls. For the second set, the increases were .004 greater. Differences between evaluations from early and complete data were also examined. Variances of these differences and mean and maximums of the absolute values of differences tended to be larger when the first set of abnormal distribution herds were excluded. These differences are likely due in part to the reduction in total data represented in the predicted breeding values. The differences tended to be smaller when the second set was excluded. These improvements indicate that the evaluations should better predict future performance when data from herds with abnormal distributions is excluded.