Date:         Sun, 17 Jan 2010 20:12:17 -0500
Reply-To:     Peter Flom <peterflomconsulting@mindspring.com>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject:      Re: proc mixed: multicolinearity
Comments: To: Ching <kcwong5@GMAIL.COM>
Content-Type: text/plain; charset=UTF-8

Ching <kcwong5@GMAIL.COM> wrote

>
>I'm testing the possible multicollinearity among the fixed effects in
>proc mixed model with the option corrb in the model statement but
>wondering whether I should specify method=reml or method=ml?
>
>Your help is greatly appreciated. Have a good week.

Corrb shows correlations. Correlations are neither necessary nor sufficient conditions for collienarity.

The proper collearity diagnostics are condition indexes, but, AFAIK, they are only available in SAS through PROC REG.  Since collinearity is an effect of only the independent variables, you can often "fake" the diagnostic by using PROC REG; this works, e.g. where the main analysis is PROC LOGISTIC or some such.

But I am not sure about mixed models.  You could check for collinearity at each time point, or each cluster; if it existed there, that would be a problem.  But I am not sure if this is the ideal solution.

Peter

Peter L. Flom, PhD
Statistical Consultant
Website: http://www DOT statisticalanalysisconsulting DOT com/
Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
Twitter:   @peterflom