GMACE variance estimation


  • Peter G Sullivan Canadian Dairy Network


genomics, international evaluation, GMACE, robust, variance estimation


Estimation of genomic variances for GMACE is similar to the estimation of genetic variances for MACE, and is based on REML equations that include a prediction error variance term.  For GMACE, the prediction error variances must be approximated, and this approximation was improved in the present study to better reflect statistical covariance between the national GEBV of young bull and MACE EBV of parents.  In previous estimation, some parental information was being excluded, allowing a simple approximation to work well, but the simple approximation did not work well after including all parental information.  Use of any approximation can bias genomic variance estimates, and such bias would adversely affect conversions of genomic information from the evaluation scales of GEBV to non-GEBV countries.  This bias was minimized by scaling, to eliminate across-country average difference between estimated variance from GMACE relative to MACE.  This "MACE-neutral" scaling of genomic variance estimates does not affect bull comparisons among GEBV countries, because it does not alter relative genomic variances.  However, it should improve comparisons between GEBV and non-GEBV countries.  The new estimates of genomic variances were very similar to previous estimates from the same data.  Some individual estimates changed, but the rankings of countries from high to low variance were nearly identical as before.  As such, all GMACE bull rankings were nearly identical to previous rankings, with correlations higher than 0.997 for all 5 traits studied and all country scales.  Standard deviations of GMACE predicted breeding values were also very similar, within 1% of the previous in almost all cases.