Does the applicability of VIF vary based on the parameter estimation process? For example, can I use VIF to check for multicollinearity if the parameters in my logistic regression are estimated using the Maximum Likelihood method or non-linear least squares for example?

When I look at the VIF formula it appears to me that it should be applicable for any regression, linear or non-linear given that the dependent variable itself isn't in the actual equation. However, surely the parameter estimation process plays a part in whether is multicollinearity or not?

  • 1
    $\begingroup$ What do you mean by the estimation process? If you are including variable selection then the VIF can be affected by the variables you exclude. But if you are just talking about estimation of parameters I don't see it. $\endgroup$ Nov 20, 2017 at 17:31
  • 1
    $\begingroup$ @MichaelChernick Yes, I actually just meant estimation of parameters. I guess my doubts arise as I am reading about a logistic regression model for which the model creator says "The VIF was not estimated for the Cross Moment Matrix because non linear least squares were used to estimate the regression" and I'm not sure if this is a correct statement. $\endgroup$
    – Jojo
    Nov 20, 2017 at 17:39
  • $\begingroup$ multicolinearity/vif in logistic regression: stats.stackexchange.com/questions/100476/…, stats.stackexchange.com/questions/421344/…, stats.stackexchange.com/questions/474964/… $\endgroup$ Jan 22, 2021 at 1:08