Classification and ROC post GLM with family (binomial) and link (logit)

[Daniel Sutcliffe]

Hello,

I have ran a logistic regression using the GLM framework:

glm nonzero $newvars if s1==1 ,family(binomial) link(logit)

Unfortunately, I didn't realise that some of the commands you might use after using logit rather than GLM cannot be used, for example estat classification and lroc.

I was wondering if anyone had any resources or code that would help me replicate the outputs I would get from using the post estimation commands I would get after using logit.

I have looked at the post estimation commands you can use post GLM and none seem to be what I am looking for.

[Rich Goldstein]

not directly, but you can use the "predict" post-estimation command and the use (1) ROC to get the curve and the area; my guess is that you want -rocfit- but maybe you want -rocreg-; note that each has post-estimation commands; and (2) choose your cutoff (as with estat classification (but recall that the default cutoff is 0.5)) and use tabulate'; see

h roc

which describes the full set of 6 (sub)commands