topic Re: PROC GAM: Additive suboption with class statement in Statistical Procedures

PROC GAM: Additive suboption with class statement

yasyuki_k — Thu, 20 Nov 2014 04:15:33 GMT

Hi, experts!

I need your help about GAM procedure.

If I use the ADDITIVE suboption with class statement to get the additive component plots, estimates for the plots aren't reasonable.

Is there any solution?

proc gam data=dataset plots=component(additive);

class x0;

model y= param(x0) spline(x1) spline(x2) / link=Identity dist=normal method=GCV;

run;

Re: PROC GAM: Additive suboption with class statement

SteveDenham — Thu, 20 Nov 2014 19:27:28 GMT

What do you mean that the plots aren't reasonable? Can you give an example of what happens, and what you would expect?

Steve Denham

Re: PROC GAM: Additive suboption with class statement

yasyuki_k — Fri, 21 Nov 2014 01:22:49 GMT

When model only included continuous variable,

estimates of additive component in the data from ods output SmoothingComponentPlot were like left table.

On the other hand, when model include categorical variable, estimates were like right table.

Re: PROC GAM: Additive suboption with class statement

SteveDenham — Fri, 21 Nov 2014 13:37:00 GMT

It appears that the right hand column is giving the difference from the reference level of x0, which in the default case would be the last level. Perhaps you could add the value of the last level (or its mean/median if the class collapses several values) to get values more like those on the left.

Steve Denham