A VIF = 1 indicates no collinearity, whereas increasingly higher values suggest increasing multicollinearity. The approach suggested by Zuur et al. (2010) is to calculate VIFs for each parameter in the model, and if they are larger than some cutoff, sequentially drop the predictor with the largest VIF, recalculate, and repeat until all values are below the cutoff (they suggest a cutoff of 2). VIFs are especially nice for dealing with collinearity of interaction terms.
“In that case, even a VIF of 2 may cause nonsignificant parameter estimates, compared to the situation without collinearity” (Zuur et al., 2010, p.9)
“Multicollinearity among independent variables for each logistic model was tested by checking the variance inflation factor (VIF). Any variable with a VIF that exceeded 4 was excluded from the model; no variable was detected with VIF greater than 4” (Pan and Jackson, 2008, p.423).
Various recommendations for acceptable levels of VIF have been published in the literature. Perhaps most commonly, a value of 10 has be recommended as the maximum level of VIF (e.g., Hair, Anderson, Tatham, & Black, 2014). The VIF recommendation of 10 corresponds to the tolerance recommendation of .10 (i.e., 1 / .10 = 10). However, a recommended maximum VIF value of 5 (Ringle et al., 2015) and even 4 (e.g., Pan & Jackson, 2008) can be found in the literature. It would appear that researchers can use which ever criterion they wish to help serve their own purposes.
“Because no formal cutoff value or method exists to determine when a VIF is too large, typical suggestions for a cutoff point are 5 or 10” (Trevor and Surles, 2002, p.393).
“A common cutoff threshold is a tolerance value of .10, which corresponds to a VIF value of 10. However, particularly when samples sizes are smaller, the researcher may wish to be more restrictive due to the increases in the standard errors due to multicollinearity” (Hair et al., 2014, p.200).
References
Pan, Y, & Jackson, R. T. (2008). Ethnic difference in the relationship between acute inflammation and and serum ferritin in US adult males. Epidemiology and Infection, 136, 421-431.
Zuur Alain F., Ieno Elena N., Elphick Chris S. (2010). A protocol for data exploration to avoid common statistical problems. Methods in Ecology and Evolution, Volume 1, Issue 1, pages 3–14.
Trevor A. Craney, Surles James G. (2002), Model-Dependent Variance Inflation Factor Cutoff Values, Quality Engineering, Volume 14, Issue 3, p.391-403.