Re: st: Logistic regression and log-transformation of covariates

[uatz@mail.uni-mannheim.de]

Hi Jase,

I'd suggest you use some logit postestimation command to calculate the exact marginal effects for your hdrinkers variable.
(For Stata 11 -margins- and you may want to read [R] Obtaining margins of derivatives of responses (a.k.a. marginal effects))

I suspect you interpretation is not precise because such an interpretation is only valid for infinitesimal (very small) changes in x and you have discrete steps.

If you want to refresh your odds ratio-knowledge, read the Stata FAQ:
http://www.ats.ucla.edu/stat/mult_pkg/faq/general/odds_ratio.htm

Hope this helps,
Ulrich



Am 30.04.2010 um 10:39 schrieb Jason Ferris:

> Hi,
> I have a simple(?) model with a binary outcome (drinker 1=problematic
> drinker 0=no problematic drinker).  These people are asked about the
> number of heavy drinkers they know (hdrinkers).  This is skewed to the
> right (as would be expected with count data of this type) - and does not
> include zero given that my sample of people know at least 1 heavy
> drinker.
> 
> Running the model without transforming the hdrinkers variable 
> Results in the estat gof being highly significant and the linktest
> suggesting a poor model
> 
> If I transform hdrinkers using a natural log ln(hdrinkers)
> The results of estat gof are not signitificant and the linktest passes.
> 
> How do I interpret the results
> Where the OR for ln(hdrinkers) is 1.8?
> 
> Do I simply interpret this as 'for every 1%' increase in hdrinker?
> Rather than the case if it was not log transformed 'for every 1 unit'
> increase in hdrinker?
> 
> Cheers 
> 
> Jase
> 



Department of Small Business Research & Entrepreneurship
Institute for Small Business Research
University of Mannheim
L9, 1-2
68161 Mannheim
uatz@mail.uni-mannheim.de


*
*   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/