Modeling final scores in US Holsteins as a function of year of classification using a random regression model

Abstract Conformation final scores in Holsteins were used to assess genetic changes over the years due to various factors such as selection and changes in trait definition. The model included management group, age group, and stage of lactation as fixed effects; additive genetic effects with random regressions on year of classification using Legendre polynomials with order from linear to cubic; and residual effects assuming heterogeneous variances. Two sets of simulated data were used to test the feasibility of variance component estimation in situations where the definition of the trait of interest changes continuously over time. Estimated variances from the simulated data sets were unbiased. Empirical tests involved 30,041 records of cows with single records scored in 1981–1999. Heritability estimates and additive genetic variances from field data decreased while residual variances increased over time. Differences among estimates of variance components from linear, quadratic and cubic random-regression models were small. Genetic correlations among final scores at years of classification estimated with the multiple-trait model that treated different groups of years as separate traits and with linear, quadratic and cubic random-regression models decreased from 1.0 to a minimum of 0.91, as the distance between the years increased. Although there were no significant differences among estimates of variance components from random-regression models, genetic correlations between different years estimated with higher order random-regression models were closer to those with the multiple trait model that treated different group of years as separate traits. Genetic changes in a trait over time can be studied with a random-regression model.