2003 G.R. Wiggans,* P.M. VanRaden, and J.L. Edwards Animal Improvement Programs Laboratory Agricultural Research Service, USDA, Beltsville, MD Standardization of lactation records for variance of Mendelian sampling to reduce bias in evaluations of bull dams
ADSA Phoenix (2) G.R. Wiggans 2003 Holstein AI Bulls Sampled in 1995
ADSA Phoenix (3) G.R. Wiggans 2003 Holstein AI Bulls Sampled in 1995
ADSA Phoenix (4) G.R. Wiggans 2003 Holstein AI Bulls Sampled in 1995
ADSA Phoenix (5) G.R. Wiggans 2003 Top 10 Holstein elite cows and active AI bulls May 2003 EBV (kg) TraitBullsCows Milk Fat88122 Protein72104
ADSA Phoenix (6) G.R. Wiggans 2003 Possible reasons for bias Preferential treatment Build up over generations as biased daughters contribute to their dams Incomplete standardization of variances Inadequate adjustment of outlier production records Heritability increase from 25 to 30% in 1997
ADSA Phoenix (7) G.R. Wiggans 2003 Efforts to account for bias Heterogeneous variance adjustment based on phenotypic variance (since 1991) Floor and ceiling at 4 SD on production records (since 1997)
ADSA Phoenix (8) G.R. Wiggans 2003 Efforts to understand bias Estimation of bias in embryo- transfer cows Investigation of distribution of records within herd (search for bimodal distributions) Merchandisable cows Contemporaries with low yields
ADSA Phoenix (9) G.R. Wiggans 2003 Bimodal herd First-lactation milk yield deviations Deviation Density
ADSA Phoenix (10) G.R. Wiggans 2003 Plan Calculate variance of Mendelian sampling using male ancestor mean to avoid possible bias from dam evaluations Calculate variance within herd and 5 yr of first calving groups Adjust records by ratio of population SD to herd SD
ADSA Phoenix (11) G.R. Wiggans 2003 Method During iteration Male ancestor mean =.5 (EBV sire + EBV MGS ) Mendelian sampling = EBV – male ancestor mean
ADSA Phoenix (12) G.R. Wiggans 2003 Method Variance of Mendelian sampling Lower for recent cows with fewer records or first record in progress Expected variance obtained from sum of data collection ratings Population variance gets weight = 20 cows when combining with herd variance
ADSA Phoenix (13) G.R. Wiggans 2003 Method Ratio = population variance / herd variance.5 < ratio < 2 Further limits proportional to herd mean High yield: upper limit < 2 Low yield: lower limit >.5 Ratio frozen after 40 iterations
ADSA Phoenix (14) G.R. Wiggans 2003 Analysis Holstein calvings analyzed with and without standardization of variance of Mendelian sampling to assess predictive ability Separate evaluations for pre-1997 calvings compared with corresponding evaluations for February 2003
ADSA Phoenix (15) G.R. Wiggans 2003 Results Correlation 1 Regression coefficient Standardized? TraitNoYesNoYes Milk Fat Protein Between pre-1997 and February, 2003
ADSA Phoenix (16) G.R. Wiggans 2003 Results Bias Mean EBV (kg) 1 Standardized? TraitNoYesNoYes Milk– 75.8– Fat– 3.3– Protein– 2.5– Top 100 cows
ADSA Phoenix (17) G.R. Wiggans 2003 Effect of adjustment Correlations higher Regression of EBV on parent mean was closer to 1 Bias lower Mean EBV of top 100 cows reduced Different cows (20%) in top 50,000 for PTA milk Largest drop in EBV was 1324 kg
ADSA Phoenix (18) G.R. Wiggans 2003 Future developments Submit national evaluations standardized for variance of Mendelian sampling to Interbull for September 2003 test evaluation Possibly enhance standardization method by imposing floor and ceiling on yield deviations during iterations
ADSA Phoenix (19) G.R. Wiggans 2003 Conclusions Cow evaluations overestimate eventual evaluations of sons Standardization for variance of Mendelian sampling improves prediction of progeny evaluations