In general, if it is less than 10, it is ok. However you may check the data, which factor has high correlation with that factor. If an outlier exists? I.e do some analyses before make decision.
In general, if it is less than 10, it is ok. However you may check the data, which factor has high correlation with that factor. If an outlier exists? I.e do some analyses before make decision.
I would be concerned about anything greater than 4 (https://onlinecourses.science.psu.edu/stat501/node/347 "The general rule of thumb is that VIFs exceeding 4 warrant further investigation, while VIFs exceeding 10 are signs of serious multicollinearity requiring correction."). I've even seen some people recommend 2.5...
In chapter 3 of the book "An Introduction to Statistical Learning with Applications in R", it is said that "The smallest possible value for VIF is 1, which indicates the complete absence of collinearity. Typically in practice there is a small amount of collinearity among the predictors. As a rule of thumb, a VIF value that exceeds 5 or 10 indicates a problematic amount of collinearity" . This book was written by Gareth James , Daniela Witten , Trevor Hastie, and Robert Tibshirani.
A VIF is a useful starting point, but only a starting point. You can interpret its reciprocal, 1/VIF as 1 - R^2, where the R^2 comes from the regression of that predictor on the other predictors. Think of that 1 - R^2 as the "usable fraction" of the predictor after the contributions of the other predictors have been accounted for.
If the VIF is large enough to indicate a problematic amount of collinearity, it is helpful to know what predictors are involved in the collinearity. One way to pursue that is to use the collinearity diagnostics developed by David Belsley and described in the book Regression Diagnostics by Belsley, Kuh, and Welsch (Wiley, 1980). Those diagnostics quantify the severity of each collinearity (a set of predictors can have more than one collinearity relation) and identify the predictors involved. They have been implemented in some software packages.
- Badges
- Science topic