topic Re: Proc Adaptivereg Parameter Estimates Interpretation in Statistical Procedures

Proc Adaptivereg Parameter Estimates Interpretation

JonB_ — Thu, 14 May 2015 23:02:56 GMT

Hi all,

I've recently been playing around with a few different types of predictive models and finally got around to learning about MARS. Little did I know that SAS recently implemented the algorithm. I've been reading through the SAS documentation on Proc Adaptivereg and found what SAS produces for parameter estimates:

source: https://support.sas.com/documentation/onlinedoc/stat/121/adaptivereg.pdf

How to I interpret this? MARS centers around basis functions that change at knots. I see the knots the algorithm found, but how do I interpret the coefficients? I think the "Parent" column is throwing me off particularly. I think you need to add the coefficients of the basis function you're interpreting with the parent basis function to get the coefficient for that basis function. is that right? What do you do in the case where the parent basis function was eliminated? how does this work with class variables?

Re: Proc Adaptivereg Parameter Estimates Interpretation

plf515 — Fri, 15 May 2015 10:28:44 GMT

I find that the best way to interpret all the coefficients is graphically. I demonstrate one way to do this in a paper I presented at the latest SGF. The presentation is here and the paper is here. I also cover how to interpret the numerical output.

Re: Proc Adaptivereg Parameter Estimates Interpretation

Rick_SAS — Fri, 15 May 2015 14:30:01 GMT

You might find the DETAILS=BASIS option helpful. See the example in Kuhfeld and Cai (2013).

Re: Proc Adaptivereg Parameter Estimates Interpretation

JonB_ — Fri, 15 May 2015 15:45:41 GMT

This is exactly what I needed. When all the basis functions are spelled out it becomes a little clearer whats going on.

Thanks.