Related Experiment Videos
Comparison of restricted maximum likelihood and method R for estimating heritability and predicting breeding value
R J Cantet1, A N Birchmeier, M G Santos-Cristal
1Departamento de Producción Animal, Facultad de Agronomía, Universidad de Buenos Aires, Argentina. rcantet@mail.agro.uba.ar
Journal of Animal Science
|October 26, 2000
Summary
Restricted Maximum Likelihood (REML) provides more accurate heritability (h2) estimates and breeding value predictions than Method R, especially under selection and with incomplete pedigree data. REML demonstrated lower bias and empirical mean square error, leading to superior predictive performance.
Area of Science:
- Quantitative genetics
- Animal breeding
- Statistical genetics
Background:
- Estimating heritability (h2) and predicting breeding values (a) are crucial in animal breeding.
- Method R and Restricted Maximum Likelihood (REML) are common statistical methods for these estimations.
- Data subjected to selection and pedigree information loss can impact the accuracy of these methods.
Purpose of the Study:
- To compare the performance of Method R and REML for estimating heritability (h2).
- To evaluate the accuracy of breeding value predictions (a) using h2 estimated by Method R and REML.
- To assess the impact of selection and missing pedigree information on both methods.
Main Methods:
- A single-trait animal model was used to simulate data over 10 overlapping generations.
- Truncation selection was applied based on predicted breeding values.
- Four pedigree structures were evaluated: complete, 50% sire missing, 50% dam missing, and 50% both missing.
Main Results:
- Under selection with complete pedigree data, REML was more consistent than Method R for h2 estimation.
- Both methods showed downward bias in h2 estimates with selection and pedigree loss, but were unbiased without selection.
- REML yielded lower empirical mean square error (EMSE) and mean square prediction error (MSEP) compared to Method R.
Conclusions:
- REML is a more reliable method than Method R for estimating heritability and predicting breeding values, particularly in scenarios with selection and incomplete pedigree data.
- Method R exhibited greater variability in h2 estimates and less accurate predictions of breeding values compared to REML.
- REML-based empirical Best Linear Unbiased Prediction (EBLUP) provided more accurate breeding value predictions under selection.