Submitted to: Journal of Dairy Science
Publication Type: Abstract Only
Publication Acceptance Date: February 28, 2005
Publication Date: July 24, 2005
Citation: Van Raden, P.M., Tooker, M.E. 2005. Effect of herd by sire interaction variance on genetic evaluations [abstract]. Journal of Dairy Science. 88(Suppl. 1):353.
Records from multiple daughters of a bull in the same herd have had reduced weighting in USDA evaluations since 1967 by adjusting for herd by sire interaction. Interaction variance (c2) defined as a percentage of phenotypic variance was initially estimated to be 14%. In 1997, c2 was reduced to 10%, and heritability was increased from 25 to 30% in the animal model. Recent c2 estimates have been lower, but higher c2 limits differential management or fraud by individual herd owners. Official November 2004 predicted transmitting abilities (PTA) for Holstein protein yield with c2 = 10% were compared with PTA computed with reduced c2. Repeatability of protein yield was held constant at 0.55 by increasing permanent environmental variance as c2 was decreased to 5 and 1%. Correlations of official PTA with PTA with reduced c2 were high: 0.999 for c2 = 5% and 0.993 for c2 = 1% for active artificial-insemination (AI) bulls and 0.999 and 0.994, respectively, for cows born since 1998. Reliability of PTA increased as c2 decreased: 88.6% for c2 = 10%, 89.2% for c2 = 5%, and 90.0% for c2 = 1% for bulls and 49.3, 49.4, and 49.4% for cows. Mean PTA decreased slightly: 18.4, 18.3, and 18.2 kg for bulls and 8.2, 8.1, and 8.0 kg for cows. Individual PTA changes were small for active bulls and moderate for non-AI bulls. Of the top 100 bulls for protein, 98 had no PTA change; only 2 bulls had changes of 1 kg when c2 was reduced from 10 to 5%. When c2 was reduced from 10 to 1%, 92 of the top 100 bulls had no PTA change; changes for the other 8 bulls were 1 or 2 kg. Of 2.3 million recent cows, PTA of only 20 cows when c2 = 5% and 1256 cows when c2 = 1% changed by >4 kg; those cows were sired by non-AI bulls with many daughters in one or a few herds. Use of a relationship matrix among herd by sire or herd by animal interactions could be helpful but was not tested. Reduction of c2 to 5% would facilitate AI companies obtaining more daughters per herd in large cooperating herds. However, somewhat less protection would be provided against chance or fraud when sampling was limited to a few herds.