Random herd curves in a test-day model for milk, fat, and protein production of dairy cattle in the Netherlands

With random regression models, genetic parameters of test-day milk production records of dairy cattle can be estimated directly from the data. However, several researchers that used this method have reported unrealistically high variances at the borders of the lactation trajectory and low genetic correlations between beginning and end of lactation. Recently, it has been proposed to include herd-specific regression curves in the random regression model. The objective was to study the effect of including random herd curves on estimated genetic parameters. Genetic parameters were estimated with 2 models; both included random regressions for the additive genetic and permanent environmental effect, whereas the second model also included a random regression effect for herd x 2-yr period of calving. All random regressions were modeled with fourth-order Legendre polynomials. Bayesian techniques with Gibbs sampling were used to estimate all parameters. The data set comprised 857,255 test-day milk, fat, and protein records from lactations 1, 2, and 3 of 43,990 Holstein cows from 544 herds. Genetic variances estimated by the second model were lower in the first 100 d and at the end of the lactation, especially in lactations 2 and 3. Genetic correlations between d 50 and the end of lactation were around 0.25 higher in the second model and were consistent with studies where lactation stages are modeled as different traits. Subsequently, estimated heritabilities for persistency were up to 0.14 lower in the second model. It is suggested to include herd curves in a random regression model when estimating genetic parameters of test-day production traits in dairy cattle.