Page Banner

United States Department of Agriculture

Agricultural Research Service

Title: Methods to Combine Information from Separate Sources

Author
item Vanraden, Paul

Submitted to: Journal of Dairy Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: December 30, 2000
Publication Date: N/A

Interpretive Summary: Various statistical methods allow animal breeders to get the most accuracy from their available data. Although the use of more data can provide even higher accuracy, data from all traits, all places, and all times may not be available in the same database. In addition, exact statistical methods that work well for smaller data subsets may not work at all with a large, combined data set. Approximate combination of estimates from separate data sets can be achieved through a selection index. Then better models can be applied to each data set and evaluations combined to include information from all sources. Selection index methods were used to combine single-trait genetic evaluations of dairy cattle into approximate multitrait evaluations for productive life (longevity) and combine single-country rankings into multicountry rankings for yield traits. Accuracy of multitrait evaluations for productive life was higher than for single-trait evaluations by an average of 7% for recent bulls and 3% for recent cows. Selection index methods may allow current multitrait across- country evaluations for bulls to be improved and to be extended to cows, which would improve efficiency of international evaluation of dairy cattle.

Technical Abstract: Separate estimates of breeding value (BV) can be combined using meta-analysis if combined analysis of all data is not possible or efficient. Computation is fast but not exact if reliabilities of separate estimates are approximate, extent of overlap of data sets is unknown, or selection has occurred across data sets. Selection index methods were used to combine single-trait evaluations into approximate multitrait evaluations for productive life and combine single-country rankings into multicountry rankings for yield traits. Parent evaluations were included in the data and combined before progeny evaluations to avoid iteration. Some information is lost because foreign progeny contribute to domestic parents but not to domestic grandparents. Exchange of sire and dam evaluations provides closer connection between national and international evaluations and may be more accurate than the current sire-maternal grandsire model used internationally. Correlations of the two evaluation methods were about .99 for 35,414 bulls from 8 countries. Estimated BV of each bull was adjusted separately for information from foreign parents and progeny. Reliabilities of the animal, its sire, and its dam were used to determine how much information came from parents versus from progeny and records. Multitrait reliabilities for productive life were higher than single-trait reliabilities by a mean of 7% for recent bulls and 3% for recent cows. Selection index methods may allow current multitrait across-country evaluations for bulls to be improved and to be extended to cows.

Last Modified: 8/1/2014
Footer Content Back to Top of Page