Studies on changes of estimated breeding values of U.S. Holstein bulls for final score from the first to second crop of daughters
The purpose of this study was to find ways of reducing changes of sire predicted transmitting ability for type’s final scores (PTATs) from the first to second crop of daughters. The PTATs were estimated from two datasets: D01 (scores recorded up to 2001) and D05 (scores recorded up to 2005). The PTAT changes were calculated as the difference between the evaluations based on D01 and D05. The PTATs were adjusted to a common genetic base of all evaluated cows born in 1995. The single-trait (ST) animal model included the fixed effects of the herd–year–season–classifier, age by year group at classification, stage of lactation at classification, registry status of animals, and additive genetic and permanent environment random effects. Unknown parent groups (UPGs) were defined based on every other birth year starting from 1972. Modifications to the ST model included the usage of a single record per cow, separate UPGs for first and second crop daughters, separate UPGs for sires and dams, and deepened pedigrees for dams with missing phenotypic records. Also, the multiple-trait (MT) model treated records of registered and grade cows as correlated traits. The mean PTAT change, for all of the sires, was close to zero in all of the models analyzed. The estimated mean PTAT change for 145 sires with 40 to 100 first crop and ≥200 second crop daughters was −0.33, −0.20, −0.13, −0.28, and −0.12 with ST, only first records, only last records, updated pedigrees, and allowing separate parent groups (PGs) for sires and dams after updating the pedigrees, respectively. The percentages of sires showing PTAT decline were reduced from 74.5 (with ST) to 57.3 by using only the last records of cows, and to 56.4 by allowing separate UPGs for sires and dams after updating the pedigrees. Though updating of the pedigrees alone was not effective, separate UPGs for sires together with additional pedigree was helpful in reducing the bias.
KeywordsBreeding value Final score Fluctuation U.S. Holstein
Changes in sires’ predicted transmitting ability (PTA), which is half of the estimated breeding value, from the first to second crop of daughters evaluation have been a credibility concern to the dairy industry. A sire’s initial progeny evaluation, based on the first crop of daughters, should be an unbiased estimate of later evaluations based on first and additional second crop daughters. Some changes in PTA are expected due to additional daughter information. Selective reporting and misidentification of daughters could also result in larger changes than expected. The former computing limitations resulted in the usage of only basic pedigrees, whereas deeper pedigrees were available. Parental selection, female selection, non-randomness in sire sampling, soundness of evaluation procedures (Bolgiano et al. 1979), and preferential treatment practices (Kuhn et al. 1994; Kuhn and Freeman 1995) have long concerned the dairy industry as well. Parental selection can be partially accounted for by using complete pedigrees (Kennedy and Sorensen 1988) and data on all candidates for selection (Henderson 1975). Although animal model evaluations account for differential mating of sires, preferential treatment to daughters of bulls or to potential bulls and dams cannot be accounted for (Kuhn et al. 1994). The model fitting random effects for preferential treatment was found to be ineffective in accounting for bias, since the accurate estimation of variances for preferential treatment is not feasible (Kuhn et al. 1999). Generally, preferential treatment is practiced in second or later evaluations due to an outstanding first record. In these cases, bias due to preferential treatment can be reduced by using only the first classification record of cows (Jamrozik and Schaeffer 1988). However, quantifying the amount of bias with field data is very difficult (Weigel and Lawlor 1994).
With regard to type traits, changes can occur due to reclassification. This implies that popular bulls are put at a disadvantage when the second crop of daughters, which have not yet been reclassified, enter the evaluation (Lohuis and Schaeffer 1995). The disadvantage is greater if the reclasses are to higher scores only, as in Canada, and if the mates were reclassified more than the average. Treating type traits as longitudinal (Uribe et al. 2000; Tsuruta et al. 2004) can potentially improve the accuracy of the evaluations; however, it is limited by only higher later scores in the Canadian system and few later scores in the US system. Improper accounting for heterogeneity of variances across subclasses, inaccurate age and stage of lactation adjustment factors, differences between registered and grade cows, and variation among classifiers’ choice could be other identified sources of bias (Schaeffer et al. 1978; Powell and Norman 1986; Bonaiti et al. 1993; Togashi et al. 2004). Although many studies investigated the causes of changes in PTA for milk, fat, and protein from first to second crop evaluations, few studies looked at changes in the predicted transmitting ability for types (PTATs). The objective of the present study was to find ways of reducing changes of sires’ PTA for final score from first to second crop evaluations.
Materials and methods
Conformation final scores of Holsteins were obtained from the Holstein Association USA, Inc. The description of the classification system is available at http://www.holsteinusa.com/programs_services/classification.html. PTATs were estimated from two overlapping datasets: final scores recorded up to 2001 (D01) and records taken up to 2005 (D05). D01 comprised 7.7 million records from 4.6 million cows and D05 comprised 4 million records from 5.1 million cows.
Characteristics of original and updated pedigrees for D01a and D05a as used in the genetic analyses
No. of parents known
Original pedigree D01
Updated pedigree D01
Number of parents whose pedigrees were amended across gender and datasetsa in particular classes of parentage information change
Class of parentage information change
No. of dams
No. of sires
Number of sires and cows occurring as parents but without pedigree information in original and updated pedigree files for the two datasetsa
No. of parents with no pedigree information
Variance component estimation
Variance components were estimated from three subsets of D05 sampled randomly to include 50,000 records in each subset. Subset A included final score records from 1993 to 1996, subset B from 1997 to 2000, and subset C from 2001 to 2004.
While the official statistical model is similar to that in Misztal et al. (1993), the most important differences in this paper are the lack of herd by sire interaction, which is an artificial effect to reduce the PTA of natural sires, and the stage of lactation by time. Adding these effects in the current study did not make noticeable differences for the purpose of this study.
Fixed effect of the ith management group (herd–year–season–classifier)
Fixed effect of the jth age–year group at classification
Fixed effect of the kth stage of lactation at classification
Fixed effect of registry status (l = 1 for registered animals and 2 for grade animals)
Additive genetic random effect of animal m
Random permanent environment effect of cow m.
Unknown parent groups (UPGs) were defined based on every other birth year starting from 1972. Table 3 shows the total numbers of dams and sires with UPGs starting from birth year 1972. Amending pedigrees (for D05) reduced the number of dams for whom UPGs were created from 956,455 to 115,187 and increased the number of sires from 33 to 4,547.
The MT model that considered the final scores of registered and grade cows as two separate but correlated traits accounted for the same effects as the ST model, except that the registry status effect was omitted. Adjustment for heterogeneity of variances within management group was also considered in the models as in Weigel and Lawlor (1994). The EM-REML method was used for the (co)variance components estimation.
Prediction of breeding values
SEPUPG – ST with separate UPGs for the first and second crop of daughters, common for both sexes within crop, original pedigree
SEPUPG-SD – ST with separate UPGs for sires and dams, original pedigree
ST-F – Restricting the records to first records, original pedigree, joint UPGs
ST-L – Restricting the records to last records, original pedigree, joint UPGs
ST-PED – Inclusion of additional pedigrees for dams with missing parentage, joint UPGs
ST-PED-SUPG-SD – Additional pedigrees plus separate UPGs for sires and dams
ST-PED-F – Additional pedigrees plus using only first records, joint UPGs
ST-PED-L – Additional pedigrees plus using only last records, joint UPGs
MT – Multiple-trait model assuming final scores of registered and grade cows are two separate correlated traits, original pedigree, joint UPGs
Sires born in recent years were expected to have their initial evaluation, based on the first crop of daughters only, by 2001, and those returning to service were expected to have their later evaluations, based on the first and additional second crop of daughters, by 2005. Thus, for common sires in D01 and D05, the PTATs computed using D01 were considered to be first crop evaluations (PTAT-01) and those predicted using D05 were second crop evaluations (PTAT-05). All PTATs were adjusted to the genetic base of all evaluated cows born in 1995 and compared for the stability of genetic evaluations from the first to second crop evaluation. The genetic base of cows in 1995 means that all PTATs were shifted so that the PTAT of a cow born in 1995 averaged 0. PTAT change was calculated as the difference between base adjusted PTAT-01 and PTAT-05.
PTAT changes were calculated for two sets of sires: ‘ALL’ including all of the sires with daughters in both D01 and D05 and ‘SELECT’ comprising 145 sires with 40 to 100 daughters in D01 and ≥200 daughters in D05.
Results and discussion
Estimates of (co)variances and heritability
Estimates of variance components and their ratios from single-trait (ST) models
Estimates of the residual variance decreased from subset A to subset C. Permanent environmental variance estimated from subset C was higher than from the other two subsets of data. Heritability estimates from subsets A and B were similar (0.31), but were reduced (0.26) in subset C. The differences between variances estimated from different subsets of data may also be due to changes in the structure of field data, i.e., changing the definition of final scores or changes in the base definition.
Estimates of (co)variances and their ratios from multiple-trait (MT) animal models
Predicted transmitting abilities
Averagesa of predicted transmitting ability for types (PTAT) for two datasets and studied models
ALLc (n = 177,451)
SELECTc (n = 145)
Estimatesa of PTAT changes from first to second crop evaluations for different models
ALL (n = 177,451)c
SELECT (n = 145)c
The inclusion of additional pedigrees for dams with missing parentage (ST-PED) reduced the bias (−0.28) by very little compared to ST. Allowing separate UPGs for sires and dams together with additional pedigrees (ST-PED-SUPG-SD) greatly reduced the mean PTAT change to −0.12. The addition of pedigrees increased the number of sires contributing to UPGs from 33 to 4,547 (in D05) and allowed more accurate estimations of genetic groups for sires (Table 3). Because most of the contributions to these UPGs are from maternal grand sires, these UPGs act as a proxy for the average merit of dams of test bulls. The results of the MT model showed that the mean PTAT change was larger for registered cows (−0.40) than for grade cows (−0.24), although the changes are large. Smaller changes in grade cows’ PTATs could be due to the lower additive genetic variances in grades.
Percentage of sires showing PTAT changes
Percentage of SELECTa sires (n = 145) showing decline in PTAT from the first to second crop for different models
% sires showing bias between
0 to −0.5
−0.5 to −1
Using additional pedigrees (ST-PED) slightly reduced the percentage of sires showing bias from 74.5% to 72.1%, of which 32.0% were showing a change below −0.5. Only a slight improvement was possibly due to low connectivity through those dams and possibly larger misidentifications. The percentage of sires showing changes below −0.5 was reduced to 17.7% by ST-PED-L and to 17.0% by ST-PED-F. The accuracy of evaluation methods based on the first classification records only depends on how accurate the classifiers are in judging a cow at a young age. Restricting data to only the most recent classification reduces the bias but still retains it in the second crop and may actually cause even more bias than reducing it. Using single records per cow would also reduce the participation in the classification program.
The percentage of sires showing bias was smallest (56.4%) with separate UPGs for sires and dams together with additional pedigrees (ST-PED-SUPG-SD). Only 12.2% of sires showed a change below −0.5.
Declines in the predicted transmitting ability (PTA) for final score from the first to second crop of daughters were caused indirectly by increased fraction of grade cows and, subsequently, inadequate accounting for the merit of mates. Bias is not reduced by considering separate unknown parent groups (UPGs) for first and second crop daughters, additional pedigrees alone, or by considering differences in registered and grade animals. Use of the first records reduces the bias but also the accuracy of evaluations. Using only the last records reduces the decline but is less efficient with amended pedigrees. Reduction of the decline requires separate UPGs for sire and dams, along with additional pedigrees for dams.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- Bonaiti B, Boichard D, Barbat A, Mattalia S (1993) Problems arising with genetic trend estimation in dairy cattle. In: Proceedings of the Open Session of the Interbull Annual Meeting, Aarhus, Denmark, August 1993. Bulletin No. 8, pp 65–72Google Scholar
- Kennedy BW, Sorensen DA (1988) Properties of mixed model methods for prediction of genetic merit. In: Proceedings of the Second International Conference on Quantitative Genetics, Raleigh, North Carolina, USA, May/June 1987, pp 91–103Google Scholar
- Lohuis M, Schaeffer LA (1995) Prelude’s type evaluation: facts. Holstein Journal. Available online at: http://cgil.uoguelph.ca/pub/articles/prelude2.html. Accessed 17 October 2010