On 14 May 2004 11:14:44 -0700, [EMAIL PROTECTED] (mac55) wrote:

> I have a questionnaire with about 20 questions that will result in
> ordered categorical variables.  I also have about 10 background
> questions related to demographics and background.  MY outcome is
> dichotomous.
> 
> How can I build a logistic regression model with so many ordinal
> categorical variables.  The amount of dummy variables required (about
> 3-4 per question) would make the model unstable.
 - true, if you don't have (perhaps) thousands of cases.
> 
> My question is what is the best way to proceed with this data.  Can
> they be entered as continuous variables so long as there is a linear
> relationship with the log odds?  Or should I just build a model with

I think you mean "monotonic relationship" -- and the answer
is "probably yes."  If it is a linear relation already, the answer
is "Certainly, Yes, Without a doubt, Why are you asking?"


> the background questions and adjust the 20 variables with those
> factors.  Of course this will miss a fair amount of confounding and
> interaction.
> 
> I know a lot of this is vague without the clinical
> context.....sorry......

If this is a "clinical context," I guess that I have expectations
that you are describing a questionnaire with related items,
and you probably can say more about individual items:
How reliable?  related to how many other items? how 
serious in the literature?

Is it feasible to make one or more composite scores, 
after assuming scale-steps are equal intervals?

Ten background questions to control for is also a large
number.  A composite-score strategy might help with
those, too, if they are (something like) known risk-factors.

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to