Accounting for variance heterogeneity in French dairy cattle genetic evaluation

Abstract

A linear mixed model assuming heterogeneous residual variances and known constant variance ratios was applied to the analysis of milk, fat, and protein yields, and fat and protein contents in the French Holstein, Montbeliarde, and Normande dairy cattle populations. The method was based on a log-linear model for residual variances. This log-linear model included a region-year fixed effect and a herd-year random effect with a within-herd autocorrelation. The estimates of this correlation coefficient varied from 0.64 to 0.92, according to trait and breed. Residual standard deviation for yields doubled over the last 20 years, whereas differences across regions were more limited and decreased over time. This model was compared with the homogeneous model and with the multiplicative model of Meuwissen et al. (1996) [Meuwissen, T.H.E., De Jong, G., Engel, B., 1996. Joint estimation of breeding values and heterogeneous variances of large data files. J. Dairy Sci. 79, 310–316]. Both heteroskedastic models provided very similar results. Accounting for heterogeneous variances had important consequences on cow ranking. Consequences on AI bull ranking were very limited, although the method affected estimated genetic trend and within birth year estimated breeding values (EBV) variability.