Heterogeneity of variances for milk and fat yield across different climatic regions of Iran and its effects on re-ranking of Holstein cows

In most animal models applied for genetic evaluation of animals, it is assumed that residual and genetic variances are homogeneous in different environments and genotypes, but in situations where data are obtained from different environments or at different times, this assumption may be unrealistic and inaccurate. However, numerous studies have reported heterogeneous genetic, residual, and phenotypic variances for production traits based on different factors such as geographical region, production level and herd. Since the selection decisions in animal breeding rely on estimated breeding values, ignoring the fact that variances are heterogeneous has consequences, including inappropriate selection, loss of expected genetic progress and decreased efficiency of breeding programs. The aim of this study was to investigate the homogeneity or heterogeneity of variances for milk and fat yield in different climatic regions of Iran as well as the impact of various methods of data transformation on reducing or possible elimination of variance heterogeneity and its effects on changing the rank of the top animals in the population of Holstein cows in Iran.
In this study milk and fat yield records which were collected from 1983 to 2014 by the Animal Breeding Center and Promotion of Animal Products of Iran were used. Records of milk and fat yield were adjusted for 305-d lactation and two times milking. Data were grouped based on the Domarten method and Bartlett’s test was used for studying the heterogeneity of variances. The transformation methods of Logarithmic, Square root and Box-Cox were applied for removing heterogeneity. Estimated breeding values (EBVs) for animals were estimated with adjusted and unadjusted data and the Spearman's rank correlations between EBVs were computed to assess the re-ranking of best sires and cows under different percentages of selection.
Results of Bartlett’s test were significant for milk and fat yields in the first three lactations which indicated the heterogeneity of variances. Applying Box-Cox data transformation resulted in homogenous variances milk and fat yields in the first and third lactations, respectively, and square root and logarithmic methods did not remove completely the heterogeneity, but they reduced partly non uniformity. Heritability of milk yield and fat yield after data transformation varied from 0.114 to 0.223 and 0.161 to 0.193, respectively, which were slightly higher than that raw data. Re-ranking of common animals in the top 1% of dams was more than 5% of top sires. Box-Cox transformation method resulted in homogenous variances and excluded top 5% and 2% of sires and dams for milk yield and fat yield from the routine list, respectively.
Based on the current results, it can be concluded that variance components for milk and fat yields according to different geographic areas in Holstein cow population is heterogeneous and heterogeneity of variance affects on genetic evaluation and re-ranking of top sires and dams. Box-Cox transformation method is more appropriate than other methods because of removing heterogeneity of variances as well as having the highest rank correlation and minimal changes in the rank of animals.