|
Activity Number:
|
320
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Stat. Sciences
|
| Abstract - #301169 |
|
Title:
|
Model Choice in Animal Breeding: A Case Study of Litter Size in Pigs
|
|
Author(s):
|
Daniel Gianola*+ and Daniel Sorensen and Rasmus Waagepetersen
|
|
Companies:
|
University of Wisconsin, Madison and Danish Institute of Agricultural Sciences and Aalborg University
|
|
Address:
|
Dept. of Animal Sciences, Madison, WI, 53706,
|
|
Keywords:
|
model choice ; Bayes factor ; predictive checks ; MCMC ; animal breeding ; litter size
|
|
Abstract:
|
Increasing prolificacy in pigs via selection is important. Four hierarchical models (effects of season and type of insemination, genetic and permanent environmental effects of sow) were fitted to 10,000 litter size records from 82 herds: (1) homoscedastic residual variance; (2) structured heteroscedasticity by parity and type of insemination; (3) as in (2), plus genetic control of the variance; (4) as in (3) plus environmental female-specific control of variance. Posterior predictive checks of residuals were used to build models. Bayes factors, posterior Bayes factors and the deviance information criterion were used for comparisons. Models (3) and (4) were favored. The 95% credible interval of the correlation between genetic values affecting prolificacy and residual variance was (-.79; -.43). Genetic selection should lead to larger and more uniform litters. Models with different plausibility were hardly distinguishable in predictive ability. Models with residual variance under genetic control led to different breeding decisions. A simple model may be adequate in some context while, yet, fail to address data features accounted for by more complex specifications.
|
