Comparison of Collinearity Indices for Linear Models in Agricultural Trials

Abstract

The deleterious consequences of collinearity in linear regression on the precision of estimators of regression coefficients and the interpretability of the fitted model are widely recognized. In this study, we compare several methodologies for assessing collinearity in linear models and explore the effect of outliers on collinearity. The robustness of collinearity measures (individual and overall) is validated through two detailed Monte Carlo simulation study which also considers the effect of outliers on collinearity indices. The methods are illustrated with two real-world agricultural and fish morphology l data sets to show potential applications. The results do not provide any evidence for an effect from outliers on collinearity identification using the collinearity indices (individual and overall). The FG and Fj collinearity indices more robust as both sample size and collinearity degree increase. The VIF (individual measure) had a better performance on the fitted model with a greater number of parameters.