Robust Bayesian inference based on birth, weaning and yearling weight data in Brangus beef cattle using normal/independent distributions

Growth data are often assumed to be normally distributed, but inferences are not robust when the data distributions exhibit heavy tails and outliers. Symmetric heavy-tailed normal/independent distributions are viable alternatives to normal distributions to model the densities of residuals. We compar...

Full description

Saved in:
Bibliographic Details
Published in:Journal of animal science 2018-12, Vol.96, p.141-142
Main Authors: Peters, S, Sinecen, M, Kizilkaya, K, Yildiz, M, Garrick, D, Thomas, M
Format: Article
Language:English
Subjects:
Bayesian analysis
Beef
Beef cattle
Birth
Bovidae
Cattle
Childbirth & labor
Computer simulation
Conditional probability
Data processing
Distribution
Error analysis
Goodness of fit
Heritability
Markov analysis
Markov chains
Mathematical models
Monte Carlo simulation
Multivariate analysis
Outliers (statistics)
Parameter estimation
Statistical inference
Weaning
Weight
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Growth data are often assumed to be normally distributed, but inferences are not robust when the data distributions exhibit heavy tails and outliers. Symmetric heavy-tailed normal/independent distributions are viable alternatives to normal distributions to model the densities of residuals. We compared estimates of genetic parameters fitting symmetric heavy-tailed normal/independent distributions (Normal, Student's-t and Slash) for residuals in multivariate analysis of birth, 205-day and 365-day weights in beef cattle. Weight and average daily gain (ADG) records of heifers (3/8 Brahman, Bos indicus x 5/8 Angus, Bos taurus; n = 719) were analyzed. Models included fixed effects of contemporary group and age of dam and random effect of animal. Inference was conducted using a Bayesian framework via Markov chain Monte Carlo sampling. The goodness of fit of the models was done by comparing the deviance information criteria (DIC). These indicated that Student's-t model was the best fitting model for the analysis of growth data but there was no significant difference among models for analysis of ADGs. Results revealed that posterior means for degrees of freedom were 8.4 and 18.1 in Student's-t error model and 2.6 and 6.9 in Slash error model for weight traits and ADG. Posterior means of heritabilities for weight traits were higher in Student's-t (0.29, 0.43 and 0.48) when compared to Normal (0.27, 0.39 and 0.46) and Slash (0.28, 0.38 and 0.46) models. However, similar heritability estimates were obtained from the Normal, Student's-t and Slash models in the analysis of ADG from birth to 205-day weights, and from birth to 365-day weights. Evaluation of 95% posterior probability interval showed that genetic correlations were significant between 205-day and 365-day GW, between birth-205-day-ADGs and birth- 365-day-ADG, between 205-day-365-day-ADG and birth-365-day-ADG. Conclusively, all three models of weight traits, the Student's-t model outperform the normal model.
ISSN:0021-8812
1525-3163