Herd-Year-Season Clustering

This is the authors' abstract. We don't add key points for this paper.

Clustering, a comparison group definition procedure, set the beginning and end of a group within a herd such that all cows with a similar calving date were in the same comparison group. The procedure had two parameters: maximum interval length and desired minimum number of observations in a cluster. Variants of clusters were compared with groups with fixed seasons (December to April and May to November) and were evaluated according to within-group variance and effective number of daughters. Analyses were done on first lactation records of cows in the Northeast calving between December 1979 and September 1988 (1,280,474 first lactations of five breeds, 10,386 sires) were analyzed. For all clustering variants, the undesirable influence on accuracy of a sire's evaluation from increasing variation within comparison group was compensated for by increased effective number of daughters. Clustering parameters that yielded comparison groups with 15 to 40 observations were optimal. Clusters spanning several years may lead to biased evaluations because of a time trend in production. The clustering procedure was implemented in the Northeast AI Multiple Trait Sire and Cow Evaluation. Maximum interval length was 365 d for obtaining a minimum of 15 observations in a comparison group.