Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: question for listserve


From   "Kieran McCaul" <Kieran.McCaul@uwa.edu.au>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: question for listserve
Date   Sat, 13 Jun 2009 06:10:19 +0800

If, as you say, X3 is associated with both z and y, then X3 is potentially confounding the relationship between z and y.  If this is the case, when X3 is added to the model, the odds ratio for z will change.  

So I think what you are asking is how to test this confounding effect.  The change in the odds ratio for z in model 1 compared to the odds ratio for z in model2.  Is that right?

If so, you don't test this.  It is a confounding effect, a bias.  What you need to determine is whether or not the change is important.

If the odds ratio for z in model 1 was 2.00 and in model 2 it was 1.98, I would say that X3 is not exerting much of a confounding effect (assuming z is a dichotomous variable).  If, however, the odds ratio for z in model 2 was 1.50, I would have evidence that X3 was an important confounder because it has resulted in, what I consider to be, a large change in the odds ratio for z.

Note, this does not apply if X3 is in the causal pathway between z and y.  In this case, X3 is not a confounder but an intermediate variable and adjusting for X3 would bias your estimate of the effect of z.



______________________________________________
Kieran McCaul MPH PhD
WA Centre for Health & Ageing (M573)
University of Western Australia
Level 6, Ainslie House
48 Murray St
Perth 6000
Phone: (08) 9224-2701
Fax: (08) 9224 8009
email: Kieran.McCaul@uwa.edu.au
http://myprofile.cos.com/mccaul 
http://www.researcherid.com/rid/B-8751-2008
______________________________________________
If you live to be one hundred, you've got it made.
Very few people die past that age – George Burns


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of scho0600@umn.edu
Sent: Friday, 12 June 2009 8:44 PM
To: statalist@hsphsun2.harvard.edu
Subject: st: question for listserve

I am trying to determine the statistical significance of the change in a 
parameter estimate for a predictor variable in a logistic regression when 
another predictor is added In other words, I have two models;

model 1: logit y on z x1 x2

model 2: logit y on z x1 x2 x3. 

I want to check if the association between y and z is mediated in part by 
x3. I have already determined that z is associated with x3 and that y is 
associated with x3.

I have tried to do this with the suest command but I do not trust the 
results since the two regressions are not run on independent samples.

I would be grateful for any guidance from anyone on how to perform a valid 
statistical test the change in the parameter estimate for z when x3 is 
added to the regression.

John T. Schousboe
University of Minnesota
*
*   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/


*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index