topic SAS Enterprise model transformation in SAS Enterprise Guide

SAS Enterprise model transformation

ryanh — Mon, 18 Apr 2016 20:40:53 GMT

Hello all,

Very new undergraduate SAS user here - trying to bring myself up to speed on some procedures.

What are the procedures for transforming variables in SAS Enterprose? In my case, I have a dataset for an internship that is giving me a curvinlinear "residual by predicted" graph after running a linear regression on my data. I want to transform the variables via a model in order to have them become more linear. What are my options

Here is a very short dataset that is representative of what I am trying to transform

x: 10 15 20 30 40 50 60

y (dep): 700 550 400 270 190 130 100

right now it is giving me a "smiley face" and I am looking for a model to transform it and the what I have to do to the variables.

Re: SAS Enterprise model transformation

TomKari — Mon, 18 Apr 2016 21:17:33 GMT

I was taught to try the log of one or both variables first. In your case, log(y) seems to come out nicely linear.

Tom

Re: SAS Enterprise model transformation

ryanh — Mon, 18 Apr 2016 21:20:46 GMT

Tom, thanks.

So in that case would I just change the input data to reflect logs of y instead of the given values and still run as a linear regression? or do I run it as a different kind of regression?

Sorry for the naive questions, I am very new at this and trying to teach myself as I go before I take an in-depth class on it next semester.

Re: SAS Enterprise model transformation

Reeza — Mon, 18 Apr 2016 23:12:46 GMT

FYI the first course on statistics with SAS is free if you're interested.

Search SAS Analytics U community forum and see the right hand menu.

Re: SAS Enterprise model transformation

TomKari — Tue, 19 Apr 2016 00:55:38 GMT

I'm on somewhat thin ice here, but fortunately @Reeza is watching this also, so she'll put me straight if I stray.

Yes, just use log(y) instead of y, and run your linear regression.

Re: SAS Enterprise model transformation

Reeza — Tue, 19 Apr 2016 01:29:04 GMT

@TomKari Is correct.