topic Fitplot with Multiple Regression? in Graphics Programming

Fitplot with Multiple Regression?

BenPorter — Sun, 19 Jan 2014 03:32:16 GMT

Hello,

I understand fitplot is not an option with Proc Reg when you have a multi-variable model. But what do I do if I want to plot the fitted values vs the actual values? Is there any work around that provides a similar look/feel? I used the output function to create a new dataset with the predicted values and tried plotting those (see code below), but when I tried to plot the fitted (yhat) values as a series the line was erratic (see attached).

ods graphics on;

proc sgplot data = fitted_model;

scatter X =x1 Y=Y /markerattrs =(color = blue);

series X =x1 Y=yhat /lineattrs =(color = red);

run;

quit;

Re: Fitplot with Multiple Regression?

PGStats — Sun, 19 Jan 2014 04:06:34 GMT

If you want actual by fitted (which is not a FITPLOT but a OBSERVEDBYPREDICTED plot), seems like you should be plotting

scatter X =Yhat Y=Y / markerattrs =(color = blue);

and no series.

Re: Fitplot with Multiple Regression?

PGStats — Sun, 19 Jan 2014 04:39:20 GMT

If you want to show the linear relationship between x1 and Y then you should sort your data BEFORE doing the regression :

proc sort data=myData; by my01variable x1; run;

proc reg data=myData ...;

....

run;

ods graphics on;

proc sgplot data = fitted_model;

scatter X =x1 Y=Y / markerattrs =(color = blue);

series X =x1 Y=yhat / group=my01variable lineattrs =(color = red);

run;

where my01variable is the dummy variable that generates the two lines in the x1-Y relationship.

Re: Fitplot with Multiple Regression?

BenPorter — Sun, 19 Jan 2014 19:58:18 GMT

THANK YOU!!! Didn't realize I had to sort it...was driving me NUTS!!!!