multiple logit odds ratio

[Nick Bornschein]

Hello,

I was trying to calculate odds ration in an example for myself, and first I had no problem with it. Here is an example:

clear
* data and preparation for the example
use http://www.stata-press.com/data/r15/sysdsn1.dta
keep if site < 3
drop if insure == 3

* calculation
logit ppd2 i.insure, or        // OR = 28.44898
margins insure, expression(exp(xb()))
matrix list r(table)
mat m = r(table)
display m[1,2] / m[1,1]        // OR = 28.44898

So via margins and using the matrix I could easily calculate the OR of insure. But using another independed variable in the model, the calculation has differences:

logit ppd2 i.insure i.site, or                // insure OR = 29.51753, site OR = 0.6142358
margins insure site, expression(exp(xb()))
matrix list r(table)
mat m = r(table)
* insure OR
display m[1,2] / m[1,1]        // OR = 29.517533
* site OR
display m[1,4] / m[1,3]        // OR = 0.61423582

I know that the difference is small and so I tried an interaction mode, too:

logit ppd2 i.insure##i.site, or                // insure OR = 30.28333, site OR = 0.6234737
margins insure site, expression(exp(xb()))
matrix list r(table)
mat m = r(table)
* insure OR
display m[1,2] / m[1,1]        // OR = 29.674937
* site OR
display m[1,4] / m[1,3]        // OR = 0.6018451

Finally, there must something to be done when calculating the OR via margins, but I don't get it. Hope you have an idea.

-Nick

[Nick Bornschein]

Sorry, gad the delete a false entry. The question is still a question.

[Nick Bornschein]

Ok, now I got it. Took me some time to find out...