Estimation of variance components for cow and parity effects from test-day yields

The initial step in implementation of a US test-day model includes estimation of cow and parity test-day variances needed to calculate lactation stage, age, and pregnancy effects. Single-trait repeatability models were fitted, and variance components were estimated for milk, fat, and protein test-day yields using Method R and a preconditioned conjugate gradient (PCG)equation solver because of large data sets (0.7 to 7.7 million records). Data were obtained for calvings since 1990 for Brown Swiss and Jerseys and for Holsteins from California, Pennsylvania, Texas, and Wisconsin. A minimum of three observations were required per subclass for herd test date and milking frequency. Three parity groups were defined: first, second, and later. Test-day data were adjusted for environmental effects of age, calving season, and milking frequency. Estimated breeding values (EBV)were expressed on a daily basis. To assess effect of adjustments, data also were analyzed without correction. For adjusted data, variance ratios (residual divided by variance of effect)within parity were similar across breeds, subpopulations, and samples: 1.5 to 1.8 for milk, 3.0 to 4.3 for fat, and 1.8 to 2.3 for protein. Variance ratios across parities ranged from 3.5 to 6.8 for milk, 8.7 to 17.6 for fat, and 5.5 to 9.4 for protein. Adjustment for EBV reduced both cow genetic and nongenetic variances. Variance ratios for permanent environment within parity from unadjusted data were nearly identical to those from adjusted data. For unadjusted data, heritabilities ranged from 0.19 to 0.30 for milk, 0.13 to 0.15 for fat, and 0.17 to 0.23 for protein. Although computations took several weeks, use of Method R and a PCG solver enabled estimation of the variance components that will be used for US evaluations based on a test-day model.