topic Re: Proc adaptivereg and nominal dependent variables in Statistical Procedures

Proc adaptivereg and nominal dependent variables

Plikis — Wed, 25 Nov 2015 08:40:05 GMT

Hello! This is my first post, so I hope I've come to the right place.

I have a rather simple question, which is does PROC ADAPTIVEREG have the capability to model nominal (multiple valued, non-ordinal like the name of a country, a color, etc.) variables, and if so, how would I be able to do it?

This manual contains a ouple of references (pictured in the top half of the attached picture), however no examples or statements to tell the procedure that my predicted value should be nominal. Also, this manual curiously omits (pictured in the bottom half of the attached picture) ADAPTIVEREG in the list of procedures capable of dealing with multinomial data.

I've tried everything, but in the cases of categorical variables SAS keeps telling me that the dependent variable has to have two values. Is there any way to make this work?

P.S. I have an assignment specifically on the application of MARS in specific fields, so unfortunately simple multinomial regression analysis isn't an option for me.

Re: Proc adaptivereg and nominal dependent variables

plf515 — Wed, 25 Nov 2015 15:13:08 GMT

No, ADAPTIVEREG cannot work with nominal dependent variables that have more than two categories. In the documentation for the MODEL statement, it notes the distributions that are available: Binomial, normal, Gaussian, inverse Gaussian, Poisson and negative binomial.

You could, however, run multiple ADAPTIVEREGs with the binomial distribution, each time restricting your data to two levels of the dependent variable. From what I can tell, this would be roughly equivalent to what would hapeen if the multinomial option were available.

Re: Proc adaptivereg and nominal dependent variables

Plikis — Wed, 25 Nov 2015 17:03:28 GMT

Unfortunately, running it for every category is not quite feasible given that I have to work with a hundred categories. I'll just have to narrow down my scope of my analysis to make it work.

Thanks for the answer!

Re: Proc adaptivereg and nominal dependent variables

plf515 — Thu, 26 Nov 2015 00:22:13 GMT

If you have 100 levels then you would have a mess even if ADAPTIVEREG let you do it in one PROC. In fact, you could run a loop across levels and then if you wanted, write a data step to combine outputs. You could make it look just like a single ADAPTIVEREG would look. And I think it would be pretty much uninterpretable.