Diagnosis of multicollinearity: Assessment of the condition of correlation matrices used in genetic studies

Data from samples of bean plant genotypes, assessed in a trial carried out in the agricultural year 1991/92 in Vicosa, MG, were used to study multicollinearity. Different methods to diagnose the problem were applied to correlation matrices. The adverse consequences of multicollinearity become evident as it increases, because the variance associated with the estimates of the parameters increases simultaneously, making them less reliable. For the diagnosis of multicollinearity, the informal methods give only general guidance, and a combination of methods based on the estimation of the variance inflating factors and variance splitting allows not only quantification of the intensity of multicollinearity but also identification of the variables involved.