Abstract
Simulated data were generated by a Monte Carlo computer simulation to study the effects of unbalanced data on sire evaluation and estimate of variance components. The simulated data included 40 sires and 100 herds, and subclass size was assumed to be distributed by Poisson distribution with mean observations of five. The degree of unbalanced data was set up according to the percentage of deleted subclasses: 0 to 90%. Estimation of variance components and sire effect prediction were carried out by i-MINQUE. The percentages of subclass deletion showed little effect on the estimated std. dev. (standard deviation) of sire effect. However, error std. dev. increased significantly as the percentage of subclass deletion increased. Estimated heritability was almost constant against the change of the deletion rate; whereas, the correlation between the predicted values and the expected values of sire effect decreased as the percentage of subclass deletion increased. This declining trend was becoming clear when the rate of deletion was more than 70%, and when heritability was less then 0.3. Consequently, if the target value of the correlation was more than 0.8, data filled with more than 10% subclasses was considered to be desirable with 0.3 heritability, data with more than 20% was desirable with 0.2 heritability and data with more than 30% was desirable with 0.1 heritability. Then the pattern of subclass deletion was analyzed to clarify the desirable data structure. The results showed that the types of deletion pattern had no effect on estimates of sire std. dev. and heritability; whereas, the correlated deletion pattern had accuracy as high as the random deletion pattern. Even if accuracy was compared between deletion patterns of selection, positive or negative correlated deletion patterns had relatively higher accuracy for the prediction of sire effect.
