|Wright, Janice - Jan|
Submitted to: Interbull Annual Meeting Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 5/14/2000
Publication Date: N/A
Citation: Interpretive Summary: A common assumption of genetic evaluation models is homogeneity of (co)variances, however this assumption often may be incorrect across time or herds. To correct for this, adjustments can be made. The statistical model currently used in type trait evaluations for breeds other than Holsteins does not make a correction for heterogeneous variance. This study modified the existing algorithm to allow integration of this adjustment within the calculation procedures. This method is preferred over pre-adjustment because it accounts for genetic or other (co)variances among observations and allows for more flexibility. Addition of the adjustment slowed convergence somewhat. A comparison of how animals ranked using the new methodology versus the existing programs showed only minimal change. Highest ranking animals, especially cows, were most affected. The new model is theoretically better and should result in improved, less biased rankings of animals and greater fairness in genetic evaluation of type traits.
Technical Abstract: Although a common assumption of genetic evaluation models is homogeneity of (co)variances, this assumption often may be incorrect across time or herds. Data can be adjusted to stabilize (co)variances by contemporary group before evaluation, and this strategy is used for some yield and type evaluations but not currently for U.S. genetic evaluation of type traits for breeds other than Holstein. Most type research has focused on phenotypic heterogeneity of (co)variances for final score. Heterogeneity of (co)variances for Jersey linear and final scores was investigated using February 2000 genetic evaluations. Only first appraisal scores during first lactation from records that included all traits were studied. Three classes of contemporary groups were created based on number of cows for that herd and appraisal date: 5 to 15, 30 to 55, and 100 and more. In each class, contemporary groups were separated into high (above class mean) and low (below class mean) final score subclasses. The 6 data sets contained appraisal information from 21,024 to 23,692 cows. (Co)variance components were estimated using expectation-maximization restricted maximum likelihood and canonical transformation. Across all traits and independent of herd size, phenotypic variances tended to be higher in low-scoring contemporary groups; mean differences in variance were around 30%. Similar or even larger differences existed for genetic variances, but those differences were not as consistent across traits and contemporary group size. Variance differences led to differences in estimated heritabilities: large and high-scoring contemporary groups showed on average around 15% lower heritabilities. Results indicated that phenotypic and genetic (co)variances for Jersey type traits are heterogeneous.