Heritability estimates of milk production with different numbers of records per sire by herd subclass.

The first-lactation milk records of 20,850 artificially sired Holstein cows and their darns were analyzed by a sire-by-herd, variance components model to determine if unequal numbers of records in the filled subclasses had any effect on the paternal half-sib correlatiou or the daughter-dam regression. The variance components were estimated with the number per subclass held constant at 1, 2, 3, 4, or 5. The records were also analyzed as deviations from herdmate averages. The analysis of deviations with one observation per subclass gave the highest he~tability estimate from daughterdam regression, .40, and the lowest from paternal half-sib correlation, .23. Analyses of deviations with two, three, four, or five observations per subclass gave approximately the same estimates of heritability from both daughter-dam regression (.31, .32, .29, and .23) and paternal half-sib correlation (.32, .31, .27, and .36). The difference in heritability estimates from daughter-dam regression (.44) and paternal half-sib emwelation (.25) reported by Bradford and Van Vleek (2) has led to several studies of the New York DHIA milk production data to explain this difference (5-7). The present study was made to determine if unequal numbers of observations in the filled subclasses of a sireby-heM, two-way classification model may affect Received for publication July 16, 1965. the paternal half-sib correlation or the daughterdam regression. Materials and Methods The data were the same as those of Van Vleck and Bradford (6), although only first-lactation records were used. Briefly, these were firstlactation (305-day, 2× , M.E.) milk records of artificially sired Holstein daughters and their dams. The records were also analyzed as deviations from herd-mate averages (3). The statistical model included sire, herd, sire-by-herd interaction, and residual effects, all considered as random variables. Components of variance associated with these effects were estimated by Method I of Henderson (4), for all data, and also for data when there were one, two, three, four, or five observations in each sire by herd subclass. The interaction component was eliminated from the model and the variance components re-estimated to compare with the estimates from the analysis by the full model. Heritability was estimated from the sire components on a within-herd basis for each analysis. Comparable estimates from daughter-dam regression were computed from the residual components of variance and eovariance. Results and Discussion The number of records, herds, sires, and subclasses for each of the analyses are given in Table 1. The daughter and dam means of deviations and mature equivalent records are also shown in Table 1. The means indicate that the daughter records are more selected, the grea£er the number of records required per subclass.