Re: Multicollinearity_Logistic regression

Posted by Rich Ulrich on Sep 07, 2020; 5:48pm
URL: http://spssx-discussion.165.s1.nabble.com/Multicollinearity-Logistic-regression-tp5739623p5739633.html

I have never used the Complex Sampling module, either.

Also, though I have paid close attention to redundant effects,

I have never paid much attention at all to VIF, except to show

someone else, "Here is a diagnostic." 

What did I look at?  - Standardized betas that are large are usually

accompanied by loadings for some other variables in the "wrong"

direction, showing a suppressor relationship. Standardized betas

that are much smaller than the simple r  can be a clue that two

variables are each accounting for the same variance, without the

reversed sign of a complete suppressor.

Of course, I always started by looking at the intercorrelations of

the set of continuous predictors -- The OP's problem has "categorical"

predictors, which the program may or may not require to be pre-coded

into dummy predictors.  Even if the procedure does not require it, it

could be useful to look at the dummies, for the correlations and for

the check for small Ns.

Since this is a Complex Sampling problem, the overall N is apt to be

large -- but the d.f.  for particular contrasts might not be so large, if

the categories are being "tested", not merely sampling elements. That

suggests to me another potential problem from having categories,

How many d.f.  are implied? And, since this is Logistic Regression, the

sample size (d.f.) for a robust analysis is maybe twice what it would be

for OLS linear regression.

--

Rich Ulrich

---

From: SPSSX(r) Discussion <[hidden email]> on behalf of Bruce Weaver <[hidden email]>
Sent: Monday, September 7, 2020 9:38 AM
To: [hidden email] <[hidden email]>
Subject: Re: Multicollinearity_Logistic regression

 

DISCLAIMER:  I have never used the Complex Samples commands in SPSS. 

If you were not using Complex Samples, the answer would be straightforward:
Estimate your model via the REGRESSION command to get tolerance & VIF.  But
inspection of the Command Syntax Reference manual shows that there is no CS
version of the REGRESSION command, there is only CSGLM.  And as far as I can
see, it does not have options to compute tolerance & VIF.  In that case, you
might have to "roll your own" so to speak.  This Support Forums thread from
4 years ago seems to suggest as much:

https://www.ibm.com/mysupport/s/question/0D50z00006PsOWuCAN/multicollinearity-test-on-spss-complex-samples-module?language=en_US

HTH.



Saima Shafique wrote
> Hello,
>
> I am working on a weighted sample and my question is how to test
> multicollinearity in binary logistic regression while using complex sample
> analysis?
> All of my IVs are categorical variables.
>
>
> Thanks,
>
> Saima
>
> =====================
> To manage your subscription to SPSSX-L, send a message to

> LISTSERV@.UGA

>  (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD