Full conjugate analysis of normal multiple traits with missing records using a generalized inverted Wishart...

Area of Science:

• Quantitative genetics
• Statistical genetics
• Animal breeding

Background:

• Multiple trait animal models are crucial for genetic evaluations, but missing records complicate analysis.
• Markov chain Monte Carlo (MCMC) methods are commonly used but can be computationally intensive with missing data.
• Existing algorithms like data augmentation can suffer from high autocorrelation and slow convergence.

Purpose of the Study:

• To present a novel MCMC algorithm (FCG) for sampling exchangeable covariance matrices in animal models with missing records.
• To develop an algorithm that avoids sampling missing error terms, thereby improving computational efficiency.
• To reduce the number of samples required for convergence compared to existing methods.

Main Methods:

• Developed a new MCMC algorithm (FCG) based on a conjugate form of the inverted Wishart density.
• Assumed normal priors for fixed effects and breeding values, and inverted Wishart priors for covariance matrices.
• The algorithm avoids sampling missing residuals, directly sampling the environmental covariance matrix (R0).

Main Results:

• The FCG algorithm eliminates correlations between sampled missing residuals and the R0 matrix.
• Demonstrated a dramatic reduction in sample autocorrelation for lags 1 to 50 in a dataset with extensive missing records.
• Achieved a 2.5 to 7 times increase in effective sample size, significantly reducing samples needed for convergence compared to data augmentation.

Conclusions:

• The FCG algorithm offers a computationally efficient alternative for analyzing multiple trait animal models with missing data.
• This method substantially improves MCMC convergence rates and reduces computational burden.
• The FCG algorithm is particularly advantageous for datasets with complex missing data patterns.