Improvement in accuracy using records lacking sire information in the animal model
Four alternative methods were examined with computer simulated data to improve accuracy of animal model genetic evaluations by including records lacking sire identification. Methods 1 and 2 assumed genetic values of cows missing sire identity were population and management group average, respectively. Methods 3 and 4 accounted for genetic values through producing abilities estimated as random and fixed effects, respectively. Correlations between true and estimated management group effects and breeding values of cows and sires were used as measures of estimation accuracy. Alternative methods were examined to determine 1) optimum, minimum management group size, 2) increases in estimation accuracy of alternative methods relative to the conventional method of discarding records lacking sire identity, 3) the effects on accuracy of missing sire identity for lower true breeding value sires, and 4) the potential to use different alternative methods in herds of varying size, proportion of cows sire identified, and level of variation.
Management group effects were estimated more accurately as minimum management group size increased (3 to 6 to 9), but breeding values were less accurate. Accuracies of alternative methods slightly exceeded those of the conventional method for all estimated effects and all minimum group sizes.
Accuracies of alternative and conventional methods were compared in 60 population with 250 sires and averages of 11,139 cows with 23,849 records. Alternative methods were always more accurate than the conventional method for estimating group effects. Methods 1 and 3 were uniformly more accurate in estimating breeding values of cows, and estimated breeding values of sires more accurately in 55 and 54, respectively, of 60 populations. Increases in accuracy were largest for method 3, but small for all methods.
Intentionally omitting identity for daughters of sires with low breeding value reduced accuracy of estimation for breeding values but not for group effects. However, alternative methods were more accurate than the conventional method. Alternative methods were relatively most accurate for estimating breeding values in small herds having high variance and low proportions of sire identified cows. Method 3 had uniformly highest accuracy but method 1 often was similar with less computing cost.