topic Re: PROC QUANTREG in Statistical Procedures

PROC QUANTREG

FB1 — Fri, 17 May 2019 08:13:41 GMT

Hello everybody,

is it possible to add boundaries in proc quantreg?

Here's my code:

proc quantreg 	data=d
			algorithm=simplex outest=out ci=sparsity plots=none;
			model y= x1 x2 x3 / quantile=&q.;
			output out=out p=q res=r;
run;

I would like to add the following condiiton on y:

y>=0

Thanks a lot.

Re: PROC QUANTREG

Rick_SAS — Fri, 17 May 2019 13:01:48 GMT

Can you clarify? Do you want

1. To only build the model for observations that have y >= 0. If so, insert

WHERE y>=0;

before the MODEL statement.

2. To build a model that predicts only positive values. If so, you need to build a model that reflects that restriction. To do this correctly requires knowledge of the data and errors.

For example, to expand on (2), if all the observed Ys are positive, you can define LogY = log(Y) and then model LogY. The predictions of Y are obtained from the predictions of LogY by exponentiating, Therefore the predictions are always positive. If you are interested in learning more, read "Error distributions and exponential regression models."

Re: PROC QUANTREG

FB1 — Fri, 17 May 2019 14:16:08 GMT

Thanks for you answer.

It's option 2., I need to build a model that predicts only positive values, but my Ys observed are both positive and negative.

I will absolutely read the article you suggest.

In the meanwhile, thanks again.

Re: PROC QUANTREG

FB1 — Mon, 20 May 2019 13:20:06 GMT

Would it make sense to put a floor (to 0) on observed Ys and then build a model on log(Y) as you suggested?

Re: PROC QUANTREG

Rick_SAS — Mon, 20 May 2019 13:22:18 GMT

No, that transformation won't work because log(y) is not defined for y <= 0.

Re: PROC QUANTREG

FB1 — Mon, 20 May 2019 14:19:36 GMT

Sorry, I meant 0+