Milk yield adjusted for 305 days and for age of Holtein cows were used to verify the effectiveness of including sire-by-herd interaction in the model of analysis as an adjustment factor for heterogeneous variances between herds. The herds were classified in three classes of phenotypic standard deviation: high, medium and low. Variance component estimates were obtained by Restricted Maximum Likelihood methodology, for classes of phenotypic standard deviation, considering each one as a different trait, using a multiple trait analysis. Two different animal models were used, either fitting or not fitting sire-by-herd interaction effect.The effectiveness of including this interation in the model was tested by Likelihood ratio test. Averages and variance components for milk yield increased from low to high phenotypic standard deviation classes. Variance component estimates were smaller, when sire-by-herd interaction was fitted, however, residual variance components were similar for both models. Heritability estimates for milk yield obtained by fitting or not fitting the interaction term in the model, were similar. Genetic correlation between classes of phenotypic standard deviation were close to unity. The proportion of the phenotypic variance due to sire-by-herd interaction variance fluctuated from 2.2 to 4.4 percent. The logarithms of the likelihood functions were higher when fitting a sire-by-herd interaction than for the model not including the interaction, for all classes.
variance components; sire by herd interaction; milk yield
