Generalized VIF (GVIF)

[Bruce Weaver]

Earlier today, someone told me about the "generalized variance
inflation factor" (GVIF), which is described in the following
publications:

Fox, J. and Monette, G. (1992) Generalized collinearity diagnostics.
JASA, 87, 178–183.

Fox, J. (1997) Applied Regression, Linear Models, and Related Methods.
Sage.

I find no mention of it when I search the archives of this group.
There is a package to compute it in R (http://hosho.ees.hokudai.ac.jp/
~kubo/Rdoc/library/car/html/vif.html). I'm just wondering if anyone
has written any code to compute it in SPSS.

Thanks,
Bruce

--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/
"When all else fails, RTFM."

[JKPeck]

On Jan 20, 3:34 pm, Bruce Weaver <bwea...@lakeheadu.ca> wrote:
> Earlier today, someone told me about the "generalized variance
> inflation factor" (GVIF), which is described in the following
> publications:
>
> Fox, J. and Monette, G. (1992) Generalized collinearity diagnostics.
> JASA, 87, 178–183.
>
> Fox, J. (1997) Applied Regression, Linear Models, and Related Methods.
> Sage.
>
> I find no mention of it when I search the archives of this group.
> There is a package to compute it in R (http://hosho.ees.hokudai.ac.jp/
> ~kubo/Rdoc/library/car/html/vif.html).  I'm just wondering if anyone
> has written any code to compute it in SPSS.
>
> Thanks,
> Bruce
>
> --
> Bruce Weaver

> bwea...@lakeheadu.cahttp://sites.google.com/a/lakeheadu.ca/bweaver/

> "When all else fails, RTFM."

My Japanese isn't nearly good enough to decipher that package, but
there is GVIF code in the R car package
using the vif function. That hasn't been nicely wrapped up in SPSS
AFAIK, but here's a usage example that
should get you started. It assumes that you have already installed
the R car package.

This code gets the active SPSS data, converts variables to R factors
as necessary, runs a linear model,
and then calculates and prints the GVIF information.

HTH,
Jon

get file="c:/spss17/samples/english/cars.sav".
begin program r.
library(car) # no relation to dataset!
dta = spssdata.GetDataFromSPSS()
dta$origin=factor(dta$origin)
dtalm = lm(mpg~origin+weight, data=dta)
print(summary(dtalm))
print("GVIF...")
print(vif(dtalm))
end program.

[Bruce Weaver]

JKPeck wrote:
> On Jan 20, 3:34 pm, Bruce Weaver <bwea...@lakeheadu.ca> wrote:
>> Earlier today, someone told me about the "generalized variance
>> inflation factor" (GVIF), which is described in the following
>> publications:
>>
>> Fox, J. and Monette, G. (1992) Generalized collinearity diagnostics.
>> JASA, 87, 178–183.
>>
>> Fox, J. (1997) Applied Regression, Linear Models, and Related Methods.
>> Sage.
>>
>> I find no mention of it when I search the archives of this group.
>> There is a package to compute it in R (http://hosho.ees.hokudai.ac.jp/
>> ~kubo/Rdoc/library/car/html/vif.html). I'm just wondering if anyone
>> has written any code to compute it in SPSS.
>>
>> Thanks,
>> Bruce
>>
>> --
>> Bruce Weaver
>> bwea...@lakeheadu.cahttp://sites.google.com/a/lakeheadu.ca/bweaver/
>> "When all else fails, RTFM."
>
> My Japanese isn't nearly good enough to decipher that package, but

I see there was some line-wrap, so you probably didn't get the
page I intended. Let's try that again:

http://hosho.ees.hokudai.ac.jp/~kubo/Rdoc/library/car/html/vif.html

> there is GVIF code in the R car package
> using the vif function. That hasn't been nicely wrapped up in SPSS
> AFAIK, but here's a usage example that
> should get you started. It assumes that you have already installed
> the R car package.
>
> This code gets the active SPSS data, converts variables to R factors
> as necessary, runs a linear model,
> and then calculates and prints the GVIF information.
>
> HTH,
> Jon
>
> get file="c:/spss17/samples/english/cars.sav".
> begin program r.
> library(car) # no relation to dataset!
> dta = spssdata.GetDataFromSPSS()
> dta$origin=factor(dta$origin)
> dtalm = lm(mpg~origin+weight, data=dta)
> print(summary(dtalm))
> print("GVIF...")
> print(vif(dtalm))
> end program.

I'd forgotten about calling R from SPSS. What I'll have to think
about is whether it's easier to do that, or just work in R
directly when I want to do something like this.

Thanks Jon.

--
Bruce Weaver
bwe...@lakeheadu.ca

[JKPeck]

I should also have pointed out, if you are interested in regression
diagnostics, the R-based
diagnostic plots dialog downloadable from Developer Central (pick the
R filter in Downloads).
It has a wealth of diagnostic plots and transformation statistics for
linear models.
This dialog was crreated by Prof Hans A Grüner of the Free University
of Berlin.

Regards,
Jon

[JKPeck]

Ok. That's the same car module I used in the example program above.

The nice thing about the SPSS/R integration is that you can benefit
from the strengths of both, and you get all the output in one place.
There are now 10+ R packages integrated into SPSS, complete with
dialog box interfaces and SPSS-style syntax.

-Jon