Related Experiment Videos
Multicollinearity in canonical correlation analysis in maize.
B M Alves1, A Cargnelutti Filho2, C Burin1
1Departamento de Fitotecnia, Universidade Federal de Santa Maria, Santa Maria, RS, Brasil.
Genetics and Molecular Research : GMR
|April 1, 2017
Summary
Multicollinearity in canonical correlation analysis can overestimate results. Eliminating variables effectively addresses this issue in maize (Zea mays L.) crop studies.
Area of Science:
- Agricultural Science
- Biometrics
- Genetics
Background:
- Multicollinearity poses challenges in statistical analyses, potentially inflating canonical coefficients.
- Canonical correlation analysis (CCA) is used to explore relationships between variable sets.
Purpose of the Study:
- To assess multicollinearity effects in CCA for maize (Zea mays L.) crop data.
- To compare CCA with and without variable elimination for managing multicollinearity.
Main Methods:
- Evaluated 76 maize genotypes across three experiments using a randomized block design.
- Measured 29 variables: 11 agronomic, 12 protein-nutritional, and 6 energetic-nutritional.
- Performed multicollinearity diagnosis (variance inflation factor, condition number) and CCA with/without variable elimination.
Main Results:
- CCA without variable elimination overestimated canonical coefficient variability when multicollinearity was present.
- Variable elimination proved effective in mitigating multicollinearity's impact on CCA.
Conclusions:
- Multicollinearity significantly affects CCA results in maize breeding.
- Variable elimination is a recommended strategy to ensure reliable CCA outcomes in the presence of multicollinearity.