topic Re: Question on VIF and P value in SAS Programming

Question on VIF and P value

subhadip_kumar — Tue, 25 Aug 2020 03:44:53 GMT

if in a linear regression model p values are less than 0.05 and VIF is less than 2 what will be the correct answer. I'm confused with below the answers and don't know which one is correct?

collinearity is not a problem since all VI values are less than 10

collinearity is not a problem since all P values are less than 0.05.

Re: Question on VIF and P value

PaigeMiller — Tue, 25 Aug 2020 15:15:58 GMT

collinearity is not a problem since all VI values are less than 10

A usual "rule of thumb" that I have heard is that VIF values should be less than 3. I have never heard 10, but maybe that's what some people think and I haven't read those articles. In any event, asking if "collinearity is not a problem" assumes we know what "problem" means, and that's not defined. Furthermore, collinearity has a big impact on some estimation methods, and less of an impact on other estimation methods, and you don't say what method you are using.

I would say that if you have real data that is not from a designed orthogonal experiment, then collinearity is present and you should examine its effect. I have never seen real data that is not from a designed orthogonal experiment that doesn't have collinearity, and you should be concerned.

You cannot determine the effect of collinearity by looking at p-values. In fact, collinearity works the other way, it can make the p-values questionable.

Re: Question on VIF and P value

Ksharp — Tue, 25 Aug 2020 12:19:21 GMT

Did you also try CORRB option in MODEL statement ?

I would pick up the first one .