Date of this Version
A multivariate linear model was used to estimate sire variance and covariance components and residual variance components for first lactation milk yield and logarithms of yield at three herd production levels using Restricted Maximum Likelihood with the Expectation- Maximization algorithm. Data for four separate analyses were 305-d, mature equivalent first lactation milk records from cows sired artificially in the northeastern United States that freshened in 1970, 1971, 1976, and 1984. Respective numbers of records for each year were 42,618, 40,207, 33,581, and 34,196. Corresponding numbers of sires were 298, 289, 305, and 313. Herd production level was defined by mean yield of all cows freshening in same herd-year-season.
For untransformed records sire and residual components of variance increased as mean increased, both within and between years. Correlations between sire effects at different production levels were all above .85. Heritabilities increased as production level increased. These results indicate that it may be necessary to account for heterogeneous genetic and environmental variance in sire evaluations.
For logarithms of yield, sire components of variance were similar for each of the three production levels within a year. Residual components for logarithms decreased as production level increased. Change in variance from one production level to another was considerably more for logarithms than for untransformed yields.