Tolerance Levels and other Collinearity Diagnostics

[dhanssen]

Are there any plans to add the collinearity diagnostics of Belsley, Kuh, and Welsch, in particular tolerance levels, to the available equation statistics in Eviews 7? Thanks!

[EViews Gareth]

Funny you should mention that, I've spent the last couple of weeks looking into them...

[startz]

Are there any plans to add the collinearity diagnostics of Belsley, Kuh, and Welsch, in particular tolerance levels, to the available equation statistics in Eviews 7? Thanks!

I'm curious as to how you would use these?

[dhanssen]

Sorry, I haven't checked this in a bit so I didn't know there was a question. They are tests for collinearity. My workplace uses tolerance levels quite a bit. So is this on the list for Eviews 7, or is it still TBD? Thanks again!

[dhanssen]

Any word back on this? Thanks!

[EViews Gareth]

We're still contemplating.

[dhanssen]

Got it, just wanted to have something to report back. Thanks!

[JimForest]

In SAS, i believe there used to be an option to produce variance inflation factors. That would be a nice inclustion.

[startz]

In the past, EViews has not included such "diagnostics" because they are rarely valuable (which is not to say they are rarely used.) Regressions almost always have some degree of multicollinearity. This does not bias coefficients. It does not invalidate inference.

If the multicollinearity is so extreme as to make numerical computation questionable, and therefore likely the result of perfect multicollinearity, EViews already provides a warning message.

[JimForest]

Thanks for your input. I agree.

I do think it would be a nice addition in cases when you are reducing a model from a general to a specific specification. A button that would allow the output to be appended to a regression output would be great.

JF :D

[cap]

I was just looking to see if any of the diagnostics from BKW were in Eviews.

As for its usefulness, while collinearity doesn't make it incorrect to perform inference on the slope coefficients, it makes for poor quality results of this inference (eg, very wide confidence intervals). The user who performs a regression and thinks "i wonder why my slope variances are so large" and "what can I do about it" would then find use out of collinearity diagnostics.

In particular I am thinking of condition indexes and the coefficient variance decompositions.


-cap

[EViews Gareth]

They're in EViews 7