enter image description here

I have a dataset with 57 independent variables, many of which are highly correlated with each other. I calculated the VIF numbers and plotted them against the standard errors of the estimated coefficients.

I know that the more the variable is correlated, the higher the standard error should be, but it does not look like that in the plot.

Does it mean that I can ignore multicollinearity?

I am not trying to predict anything through my model. I am only trying to get the coefficient estimates and explain them.

  • 1
    $\begingroup$ You're comparing apples with oranges here. The inflation happens to the SE when the collinear variable is included in the model. It's not about how all the SEs in the model being big. For example, yes, there is not high VIF values with high SE in this graph, but the right sides dots could have been much lower should their collinear counterparts are removed. $\endgroup$ Jul 17, 2018 at 19:58
  • 3
    $\begingroup$ Are you saying you created a single model using all 57 predictors and that what you've plotted are the resulting numbers for VIF and for (coefficient standard error)/coefficient? $\endgroup$
    – rolando2
    Jul 17, 2018 at 22:14


Browse other questions tagged or ask your own question.