topic Re: Can Proc SGPlot generate the slope value of a regression line? in Graphics Programming

Can Proc SGPlot generate the slope value of a regression line?

dhappleby — Fri, 22 Feb 2013 15:26:53 GMT

I'm using proc sgplot with the reg option to generate a number of related plots. Is there a way to get proc sgplot to generate the slope of the regression line for me, so that I can compare the slopes of different plots? Or do I have to go run PROC REG? separately to have the system calculate the slope?

Re: Can Proc SGPlot generate the slope value of a regression line?

Reeza — Fri, 22 Feb 2013 16:39:10 GMT

GPLOT will, but I'm not sure about sgplot

Re: Can Proc SGPlot generate the slope value of a regression line?

GraphGuy — Fri, 22 Feb 2013 16:45:37 GMT

Here's an example using gplot's "regeqn" option.

I'll have to let someone else answer whether sgplot has such an option.

proc sort data=sashelp.class out=foo;

by sex;

run;

symbol1 value=dot interpol=rl;

proc gplot data=foo;

by sex;

plot height*weight / regeqn;

run;

Re: Can Proc SGPlot generate the slope value of a regression line?

Quentin — Fri, 22 Feb 2013 18:33:37 GMT

Since PROC REG calls SGRENDER in the background when you have ODS graphics on, I had hoped it would be easy to get the slope on the graph from PROC REG.

I tried with:

proc reg data=sashelp.class plots=FitPlot(stats=all);
  model height=weight;
  ods select FitPlot;
run;

And while it gives lots of statistics, slope isn't one of them.

I suppose you might be able to get it from reg if you mucked about with Stat.REG.Graphics.Fit.

Re: Can Proc SGPlot generate the slope value of a regression line?

dhappleby — Fri, 22 Feb 2013 18:41:59 GMT

Thanks for all of your help and suggestions.

Since I already had all of my "proc sgplot" code written out, and since sgplot apparently does NOT have a way to generate the slope, I just did what I needed in a brute force manner as shown here:

For each plot of interest, I used proc sgplot to get my plot with confidence intervals:

title "Scatter Plot with Confidence Interval - PMC Non-Hispanics, Dx = Normal";

proc sgplot data=allpmc3;

where short_dx = "Normal";

reg y=mmsetotal x=frstotal / cli ;

yaxis values= (0 to 30 by 5);

xaxis values = (0 to 55 by 5);

run;

quit;

Then I did a Proc Rec using 'noprint', but with the OUTEST keyword to create a dataset containing the needed data:

proc reg data=allpmc3 noprint outest=PMC_norm;

where short_dx = "Normal";

model mmsetotal = frstotal;

run;

quit;

Then I modified the dataset created by OUTEST so that it provided the info in the format I needed:

data PMC_norm2;

set PMC_norm;

rename frstotal = slope;

group = "PMC Normal";

keep group intercept frstotal;

run;

Finally, once I had repeated the above steps for multiple plots and groupings, I just put all my datasets together so that I could print a nice neat table of intercepts and slopes:

data eq_params;

length group $12;

set metab_ad2 metab_mci2 metab_norm2

pmc_ad2 pmc_mci2 pmc_norm2;

run;

proc print data=eq_params;

run;

Re: Can Proc SGPlot generate the slope value of a regression line?

Rick_SAS — Mon, 25 Feb 2013 15:03:42 GMT

You can do the following:

(1) call PROC REG,

(2) create a macro variable with the parameter estimate (slope), and

(3) call PROC SGPLOT and use the INSET statement to put the slope on the graph. No need to modify any templates:

For an example, see SAS/STAT(R) 9.3 User's Guide