Observations on sire evaluation with categorical data using heteroscedastic mixed linear models

Area of Science:

• Animal breeding and genetics
• Quantitative genetics
• Statistical modeling

Background:

• Accurate sire genetic evaluation is crucial for genetic improvement in livestock.
• Mixed linear models are commonly used for sire genetic evaluation.
• Assumptions about residual variance homogeneity can impact model performance.

Purpose of the Study:

• To compare the sire ranking ability of three mixed linear models.
• To investigate the effect of residual variance homogeneity assumptions on sire evaluation.
• To assess the impact of weighting for unequal variances on selection response.

Main Methods:

• Simulated half-sib progeny data were used for analysis.
• Three mixed linear models with differing variance assumptions were tested.
• Realized response to 20% truncation selection was estimated for sire evaluations.

Main Results:

• Weighting for unequal residual variances reduced the apparent prediction error variance.
• However, this weighting impaired the ability of best linear unbiased prediction (BLUP) to identify superior sires.
• Model performance varied based on the assumptions of residual variance.

Conclusions:

• Ignoring heterogeneity of residual variances may be preferable for sire ranking in certain scenarios.
• Theoretical predictions from threshold models align with the observed results.
• Careful consideration of variance assumptions is necessary for accurate genetic evaluations.