Thanks for saying exploratory ideas and research are ok in statistics. Thank you for sharing your thoughts and teaching.

My biggest problem with VIF is: Models can mostly deal well with collinearity if the sample size is large (not with perfect collinearity, off course). The formula for the VIF does not include the sample size at all, so rules of thumb such as VIF < 10 should really define for which sample sizes they are (kind of) valid. Am I wrong?

Only so much you can ever do about multicollinearity?

@QuantPsych

2 жыл бұрын

I'm not understanding your question....

@thejll

2 жыл бұрын

@@QuantPsych Hrm. You can determine whether you have MC, but you cannot stop the effect it has on the uncertainty of model coefficients.

Also, multicollinearity only inflates the standard errors of the variables that are collinear. If we're only interested in the effect of one variable on some outcome (i.e. a treatment effect in an experimental design) and multicollinearity is present in some of the control variables, then multicollinearity shouldn't be a big deal (since it won't impact your inferences related to your variable of interest). Wooldridge has an interesting twitter thread on this and reiterates some of the points made here. twitter.com/jmwooldridge/status/1483493723233259527 Great video!

@QuantPsych

2 жыл бұрын

Excellent point! I should have said that in the video :) And thanks for the twitter thread link.