When there is no linear relationship among predictor
variables at all, they are said to be orthogonal. The lack of
orthogonality among the predictor variables usually is not strong
enough to affect the analysis or the ability of the entire number of
predictors to predict the response variables. In other words, the
lack of orthogonality does not diminish the usefulness of the
model, at least within the sample data used to find the regression
coefficients. However