The first thing that can be done is to find pairwise correlations between all X variables. If two variables are perfectly uncorrelated, the correlation is zero, one suggesting a perfect correlation. In our case, a correlation of > 0.9 must sound an alarm for multicollinearity.
Another method to detect multicollinearity is the Variance Inflation Factor (VIF). VIF estimates how much the variance of the estimated regression coefficients is inflated compared to when Xs are not linearly related. The way to estimate VIF work the following way:
- Create an auxiliary regression for each X, such as
X1 = A0 + A*1 X2 + A*2 X3 + …
Or how much the X1 regresses using the other independent variables. - Estimate R-squared from the regression model
- VIF (X1) = 1/(1-R21)
- As a rule of thumb, a VIF of > 10 suggests that X is redundant in the original model.