Most test-day models used in genetic evaluations of dairy cattle define contemporary groups (CG) as the herd-test-date effect. Fitting this effect as fixed may minimize prediction bias, but requires a minimum number of observations per CG to simultaneously maximize the effective number of observations and minimize the residual error and prediction error variance. Nearly 4 million test-day records from the Portuguese Holstein database of 238,271 cows calving in 1,330 herds from 1994 through 2006 were used to evaluate the effect of clustering CG from small herds based on the similarity of their production environments. Principal component analysis was used to summarize 14 descriptive variables in 5 eigenvectors that explained 88% of the total variation. Based on the distance matrix, 2 different approaches were applied to group the herds. For each approach, 4 data sets were built having at least 3, 5, 10, or 15 observations per CG, respectively. For the data sets of group A, all herds, with or without the required number of observations per CG, were used in the clustering process. For the data sets of group B, only herds without the minimum number of observations were candidates to form clusters. All data sets were analyzed by an autoregressive test-day animal model fitting a fixed herd test date in a multiple-lactation setting, and results were compared with the current clustering procedure used in the Portuguese genetic evaluations. The data set from group B, with a minimum of 3 records per CG, was the one that provided the highest accuracy of prediction and the smaller within-CG variance, revealing a better fit for the data. This procedure also preserved the original herd structure of the database, better maximizing the number of herd groups. Correlations among EBV, rank, prediction error variance, and accuracies of prediction for this data set were high (0.97, 0.97, 0.85, and 0.82, respectively), suggesting that no major reranking is to be expected.

Effects of Clustering Herds with Small-Sized Contemporary Groups in Dairy Cattle Genetic Evaluations / J. Vasconcelos, F. Santos, A. Bagnato, J. Carvalheira. - In: JOURNAL OF DAIRY SCIENCE. - ISSN 0022-0302. - 91:1(2008 Jan), pp. 377-384. [10.3168/jds.2007-0202]

Effects of Clustering Herds with Small-Sized Contemporary Groups in Dairy Cattle Genetic Evaluations

A. Bagnato
Penultimo
;
2008

Abstract

Most test-day models used in genetic evaluations of dairy cattle define contemporary groups (CG) as the herd-test-date effect. Fitting this effect as fixed may minimize prediction bias, but requires a minimum number of observations per CG to simultaneously maximize the effective number of observations and minimize the residual error and prediction error variance. Nearly 4 million test-day records from the Portuguese Holstein database of 238,271 cows calving in 1,330 herds from 1994 through 2006 were used to evaluate the effect of clustering CG from small herds based on the similarity of their production environments. Principal component analysis was used to summarize 14 descriptive variables in 5 eigenvectors that explained 88% of the total variation. Based on the distance matrix, 2 different approaches were applied to group the herds. For each approach, 4 data sets were built having at least 3, 5, 10, or 15 observations per CG, respectively. For the data sets of group A, all herds, with or without the required number of observations per CG, were used in the clustering process. For the data sets of group B, only herds without the minimum number of observations were candidates to form clusters. All data sets were analyzed by an autoregressive test-day animal model fitting a fixed herd test date in a multiple-lactation setting, and results were compared with the current clustering procedure used in the Portuguese genetic evaluations. The data set from group B, with a minimum of 3 records per CG, was the one that provided the highest accuracy of prediction and the smaller within-CG variance, revealing a better fit for the data. This procedure also preserved the original herd structure of the database, better maximizing the number of herd groups. Correlations among EBV, rank, prediction error variance, and accuracies of prediction for this data set were high (0.97, 0.97, 0.85, and 0.82, respectively), suggesting that no major reranking is to be expected.
Autoregressive test-day model; Cluster analysis; Contemporary group; Principal component analysis
Settore AGR/17 - Zootecnica Generale e Miglioramento Genetico
gen-2008
Article (author)
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2434/55557
Citazioni
  • ???jsp.display-item.citation.pmc??? 3
  • Scopus 12
  • ???jsp.display-item.citation.isi??? 11
social impact