Re: st: Can Spearman's rho be used to measure of the degree of association between two binary variables ?

[Richard Williams <richardwilliams.ndu@gmail.com>]

At 01:56 AM 5/21/2012, Maarten Buis wrote:
> On Mon, May 21, 2012 at 12:19 AM, Marcos Vinicius wrote:
> > I was conducting a multicollinearity diagnostic analysis for a 
> logistic regression using spearman correlation and VIF. Important 
> detail:All the covariates are binary variables.
>
> Multicollinearity is never a problem, see e.g.:
> <http://www.stata.com/statalist/archive/2010-07/msg00675.html>, so
> there is nothing to diagnose. If you want to inspect the association
> between binary covariates I would look at a table of odds ratios.
>
> -- Maarten

I would have to disagree with that a bit. Sometimes multicollinearity 
might reflect a mistake on the researchers part. For example, your 
model includes education, income, and then you decide to include this 
SES measure you find at the end of the codebook. If SES was computed 
using income and education, you may have extreme or even perfect 
multicollinearity.

Or, suppose you have a categorical variable, and you create dummies 
out of it. If some categories have extremely small Ns (e.g. 2 cases) 
you will get near-perfect collinearity. You may have to combine 
categories or else drop some cases.

Suppose, too, that you have several items that basically measure the 
same concept. You may be better off creating a scale from the items 
or constraining them all to have the same effects.

I don't think I have ever seen it happen with Stata, but there might 
be situations where multic makes it difficult for the model to 
converge. If so, doing things like centering a variable before you 
square it might help.

If you happen to be at the design stage of the study and you are 
worried about multic, you may wish to collect a larger sample as 
larger samples will reduce the standard errors.

I do think the problem is exaggerated. But, the researcher should be 
aware that they may have done something stupid, that there may be 
better ways to set the problem up, and that they may be able to avoid 
the problem in the first place when they design their study.

Also, I would discourage simply dropping variables that seem to be 
causing you problems, as that could lead to specification error, 
which may be an even worse problem.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/