complete separation issue

complete separation issue

by Siripong -
Number of replies: 1

Dear Teacher 

I ran logistic command looking for odd ratio and p value between variable leak (leakage of anastomosis) and omenflap (intervention)  then I got the complete separation issue. How to solve it?(I try to find my fit model)

. logistic leak omenflap
note: omenflap != 0 predicts failure perfectly
omenflap dropped and 44 obs not used

Logistic regression Number of obs = 53
LR chi2(0) = 0.00
Prob > chi2 = .
Log likelihood = -14.181099 Pseudo R2 = 0.0000
------------------------------------------------------------------------------
leak | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
omenflap | 1 (omitted)
_cons | .0816327 .0424496 -4.82 0.000 .02946 .2262015
------------------------------------------------------------------------------
In reply to Siripong

Re: complete separation issue

by Mark Pletcher -

OK, last Biostat 212 Forum Post for 2015!

Weird things happen when one of your predictors is a PERFECT predictor of events (or non-events), and Stata usually responds by dropping observations and re-running the analysis.  Looks like that is what happened here.  I suggest starting with a cross tabulation of leak and omenflap, and then talking with a statistician for next steps