Multiple-breed genetic inference using heavy-tailed structural models for heterogeneous residual variances

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Statistical Genetics

Background:

• Existing multiple-breed genetic models account for heterogeneous genetic variances between beef cattle breeds.
• These models can be extended to incorporate residual heteroskedasticity, meaning residual variances differ across observations.

Purpose of the Study:

• To extend multiple-breed genetic models to include residual heteroskedasticity, specified by fixed and random effects.
• To evaluate different residual distributions (Gaussian, Student's t, Slash) and error specifications (homoskedastic vs. heteroskedastic) for analyzing postweaning gain records.

Main Methods:

• Utilized a Markov chain Monte Carlo (MCMC) animal model implementation.
• Analyzed 22,717 postweaning gain records from a Nelore-Hereford population.
• Compared models with Gaussian, Student's t, and Slash residual densities under both homoskedastic and heteroskedastic error assumptions.

Main Results:

• The heteroskedastic Student's t error model was the best fit, indicating heavier-tailed distributions and heterogeneous variances are crucial.
• Breed group heterozygosity and calf sex were significant sources of residual heteroskedasticity.
• Residual variance was lower in F1 animals (0.70 times purebreds) and males (1.13 times females).
• Contemporary group effects significantly contributed to residual heteroskedasticity (CV = 0.72).
• Genetic variance estimates differed substantially between Nelore and Hereford breeds depending on the model's homoskedasticity assumption.

Conclusions:

• Accounting for residual heteroskedasticity and using robust error distributions (like Student's t) is vital for accurate variance component and genetic parameter estimation in multi-breed cattle populations.
• Ignoring these factors can lead to biased estimates, impacting breeding program decisions.