topic Re: SAS Proc Reg plots overlay problem in Statistical Procedures

SAS Proc Reg plots overlay problem

BDORR — Wed, 26 Jun 2019 18:00:52 GMT

Hey Folks,

I am running a SAS Proc Reg procedure and producing prediction plots using the Plots syntax. I have a control and 3 treatment levels that I would like to overlay each regression and their respective CLM CLI. I can get the plots separately fine but when I use overlay it builds CLM CLI around all the regression lines not each individually. How can I correct this? Pgm below, plots attached.

proc sort data=MeanTotTRlat; by treatment;

proc reg data=MeanTotTRlat plots=predictions (x=LatR );
var LatR2;
model TotTR = LatR LatR2;by treatment;
plot overlay;
run;quit;

Re: SAS Proc Reg plots overlay problem

Rick_SAS — Thu, 27 Jun 2019 12:54:48 GMT

A few thoughts:

1. SAS introduced ODS statistical graphics way back in version 8, so you should upgrade to ODS. The newer graphs have many advantages over the older graphs.

2. There is a difference between BY-group analysis by treatment and using treatment as a CLASS variable. If you put Treatment on a CLASS statement, the overlay will happen automatically. As a bonus, you can compare the treatment effects in the statistical output.

3. You can switch from PROC REG to PROC GLM to solve two problems: (A) PROC GLM enables you to use the quadratic variable LatR*LatR directly without having to create it in a DATA step, and (B) PROC GLM supports the CLASS statement.

Putting these all together, the following statements provide what you want

ods graphics on;
proc glm data=MeanTotTRlat plots=FitPlot;
class treatment;
model TotTR = latR LatR*LatR treatment;
run;

Re: SAS Proc Reg plots overlay problem

BDORR — Thu, 27 Jun 2019 14:08:29 GMT

Rick,

Thanks for the reply. A more elegant solution for sure for the issue I had (your code below). I get all four regression lines on my graph. However, I still don't get the CLM on my regression lines. Even when I use the Plots=INTPLOT(CLM) option.

Any Suggestions?

proc glm data=MeanTotTRlat plots=fitplot;
class treatment;
model TotTR = latR LatR*LatR treatment;
run;

Re: SAS Proc Reg plots overlay problem

Rick_SAS — Thu, 27 Jun 2019 15:02:05 GMT

I apologize for not answering that portion of your question.

First, a confession: I simplified the call by using PLOTS=FITPLOT, which is sort of a generic specification. The actual name of the plot you are creating is the ANCOVAPLOT, as stated in the documentation for the PLOTS= option

To get only CLM, use PLOTS=ANCOVAPLOT(CLM)

To get only CLI, use PLOTS=ANCOVAPLOT(CLI)

Because you want both, you can use

proc glm data=MeanTotTRlat plots=ancovaplot(limits);

Be aware that the four lines with confidence bands might overlap and the graph might look crowded.

Re: SAS Proc Reg plots overlay problem

BDORR — Thu, 27 Jun 2019 15:05:43 GMT

Perfect, thanks!