01-31-2017 10:06 PM
I am fitting the conditional logistic model using PROC GENMOD and I want to output the robust standard error for the estimated odds ratio instead of the log(odds).
I know the ods output GEEEmpPEst can output the empirical standard error on the log scale, but I am not sure whether ods output ESTIMATES will output the empirical SE on the OR scale.
07-03-2017 02:48 PM
The "L'Beta Estimate" and "Standard error" values, and the "L'Beta Confidence limits" in the ESTIMATE statement results are on the logit scale (log[p/(1-p)]) when fitting a logistic model. For an ESTIMATE statement that defines a difference in two logits, using the EXP option provides the odds ratio estimate and its confidence limits. Equivalently, you can use an LSMEANS statement along with the DIFF, ODDSRATIO, and CL options. For example, using the GEE example in the Getting Started section of the GENMOD documentation, adding these statements provides the odds ratio estimate and confidence interval for comparing the two cities:
estimate 'city' city 1 -1 / exp;
lsmeans city / diff oddsratio cl;
However, a standard error on the odds ratio scale is not available.