st: RE: comparing between uneven groups using logistic regression??

["Mak, Timothy" <timothy.mak07@imperial.ac.uk>]

Hi Alison, 

I'm not sure I completely follow your question. I presume you have: 1 binary dependent variable (mental illness), 1 binary exposure variable (alcohol death, which has 100 positive and 4050 negative), and 2 covariates (age, sex). If so, I think logistic regression is quite appropriate. The choice of dependent/independent variable is slightly weird to me: Usually the supposedly causal factor is the independent variable (i.e. the exposure variable), but here, it's difficult to imagine how alcohol death might be a cause to mental illness (the other way round makes a bit more sense). Nonetheless, I wouldn't say it's wrong, since if you're simply looking at association, with no causal assumptions, you can use either as your independent variable. 

I haven't come across Chi-square test with adjustment on the effect of size. The reviewer will have to provide the reference, since I don't think it's a well known technique. In any case, there's no violation of assumption whatsoever in logistic regression if the two categories of the independent variable are unbalanced, or even sparse. (No violation assumption even if it's the dependent variable). 

Yours, 
Tim

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Alison McCarthy
Sent: 19 January 2011 01:21
To: stata
Subject: st: comparing between uneven groups using logistic regression??

Dear statalist
 
I have recently got some feedback on a technique used in a report. This report examine differences in mental illness between two (cases of alcohol related deaths) groups using logistic regression (age and gender were included as covariates). The groups were very different in size. One group had 100 persons in it, the other had 4050.  The reviewer stated that logistic regression was innapropriate and that I should consider using "Chi-square test with adjustment on the effect of size". I have been looking about for this in stata but am at a dead end. 
Can anyone provide any direction about what technique to look at? 
 
I was also thinking to adjust for the difference in the two groups using the weight option. To weight the results of a logistic by age and gender, could this be calculated by multiplying the two variables e.g. logistic ALCOHOLDEATHgroups DEPRESSION [pweights= age*gender] 
 
These are probably very silly questions, but any directions about where to start would be much appreciated.
Cheers 		 	   		  
*
*   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/