Non linear regression with categorical dependent variable.


Fortran must die
I have a dependent variable which is a successful or unsuccessful closure. My predictor is how many months (I also ran years) of experience a counselor who handles that case has. The results are very statistically significant, but the effect size is very small (the odds ratio is .962 roughly very close to 1 although of course the CI does not include 1). I think the cause is that the effect is non-linearity. Early in a counselor's career more experience helps them do better with a case, later in their career it probably does not - they get jaded or think about retirement (pay is awful) and the cases difficult).

I don't know a good way to do non-linear effects for a categorical dependent variable. Good would include simple here since I have not worked a lot with non-linear methods (I have in theory not in practice much). Simple descriptive that get at this would be fine - non-linear regressions would be ok. The people I will present to will not even know logistic regression let alone splines knots and the like.
ive never used proc gampl although i always plan to.

I feel like the easiest answer is just to break the years of experience into a set of categories which reflect the relationship between experience and results expressed above. I guess that selection of how to break up is arbitrary, but so be it as long as it meets your needs.


Fortran must die
I don't know that proc but I will try to do it. But this has to be done by tomorrow so I will go with fed2's suggestion in the short run.


Less is more. Stay pure. Stay poor.
Proc gampl is exactly like proc logistic but you tell it which variable may be nonlinear. It takes literal 30 seconds to figure out. If plots look near linear use logistic reg.