GTL Legend Different from Plot

[djbateman]

I am trying to add a legend to my plot that does not exactly match the data that I am plotting.  Sorry for the lengthy code, but I wanted you to have all my information.  I am probably giving more than is needed.  Simply put, I want to color code the plotting markers, but I only want 1 black marker to appear in the legend with a set label.

proc format;

            value posneg

                        1='Positive'

                        2='Negative'

                        9='Non-Evaluable';

run;

data waterfall;

            input SUBJECT LESID POSNEG BESTPCT @@;

            format posneg posneg. bestpct percent9.2;

            cards;

1010     1          1          -0.269   1020     1          1          -0.235

1020     2          1          -0.043   1020     3          1          -0.167

1020     4          1          -0.03     1050     1          9          0.077

1060     1          9          .                       1060     2          1          -0.25

1060     3          2          -0.059   1060     4          1          -0.111

1060     5          1          -0.24     1070     1          2          0

1080     1          1          .                       1080     2          2          .

1080     3          1          .                       1080     4          1          .

1090     1          1          -0.125   1110     1          1          0.179

1110     2          1          -0.056   1120     1          2          0

1120     2          2          -0.211   4010     1          2          -0.022

4010     2          2          1.344    2010     1          9          0.462

2020     1          1          0.217    2030     1          1          0.4

2030     2          1          0.174    2040     1          1          0.091

2040     2          1          0.1                    2050     1          9          0.2

2050     2          1          0.917    2060     1          1          .

2060     2          1          .                       2060     3          1          .

2060     4          1          .                       2060     5          1          .

;

run;

proc sort data=waterfall;

            by posneg descending bestpct;

run;

data waterfall;

            set waterfall;

            n=_n_;

            if bestpct=. then scattermissing=0;

            label scattermissing='No Post-Baseline Assessment';

run;

/* Determine the range and tick values of the y-axis for plotting */

data tickvalues;

            do VALUE=-1 to &yaxismax. by 0.2;

                        output;

            end;

            format value 5.1;

run;

proc sql noprint;

            select value into :tickvaluelist separated by ' ' from tickvalues;

            select (ceil(max(&yvar.)/0.2))*0.2 into :yaxismax from waterfall;

quit;

I want the plot to look like this one (X's color coded based on POSNEG):

proc template;

            define statgraph barscatter;

                        begingraph;

                                    entrytitle 'Best Percent Change in Lesion Size';

                                    layout overlay / xaxisopts=(display=none) yaxisopts=(label="&ylab." linearopts=(viewmin=-1.05 viewmax=%sysevalf(&yaxismax.+.05) tickvaluelist=(&tickvaluelist.)));

                                                barchart x=n y=bestpct / group=posneg name='b';

                                                scatterplot x=n y=scattermissing / group=posneg markerattrs=(symbol=X) name='s';

                                                referenceline y=eval(coln(0.2, -0.3)) / lineattrs=(color=black pattern=dot);

                                                discretelegend 'b' 's' / across=1 autoalign=(bottomleft) location=inside;

                                    endlayout;

                        endgraph;

            end;

run;

proc sgrender data=waterfall template=barscatter;

run;

But I want the legend to look like this one (only 1 black marker for SCATTERMISSING with the "No Post-Baseline Assessment" label):

proc template;

            define statgraph barscatter;

                        begingraph;

                                    entrytitle 'Best Percent Change in Lesion Size';

                                    layout overlay / xaxisopts=(display=none) yaxisopts=(label="&ylab." linearopts=(viewmin=-1.05 viewmax=%sysevalf(&yaxismax.+.05) tickvaluelist=(&tickvaluelist.)));

                                                barchart x=n y=bestpct / group=posneg name='b';

                                                scatterplot x=n y=scattermissing / markerattrs=(symbol=X) name='s';

                                                referenceline y=eval(coln(0.2, -0.3)) / lineattrs=(color=black pattern=dot);

                                                discretelegend 'b' 's' / across=1 autoalign=(bottomleft) location=inside;

                                    endlayout;

                        endgraph;

            end;

run;

proc sgrender data=waterfall template=barscatter;

run;

[BrunoMueller]

Have a look below. Only mention the scatterplot name in the discretelegend statement and use the legendlabel option for the text of the legend label

proc template;
  define statgraph barscatter;
    begingraph;
      entrytitle 'Best Percent Change in Lesion Size';
      layout overlay / xaxisopts=(display=none) yaxisopts=(label="&ylab." linearopts=(viewmin=-1.05 viewmax=%sysevalf(&yaxismax.+.05) tickvaluelist=(&tickvaluelist.)));
        barchart x=n y=bestpct / group=posneg name='b';
        scatterplot x=n y=scattermissing /
          markerattrs=(symbol=X color=cx000000)
          name='s'
          legendlabel='No Post-Baseline Assessment'
        ;
        referenceline y=eval(coln(0.2, -0.3)) / lineattrs=(color=black pattern=dot);
        discretelegend  's' / across=1
          autoalign=(bottomleft) location=inside;
      endlayout;
    endgraph;
  end;
run;

[djbateman]

This is almost identical to the second PROC TEMPLATE I had.  It does put just a single black X in the legend, but the X's on the plot are also black.  I am trying to keep the legend the way you provided it but still have the X's to stay color coded like the bars by the POSNEG variable.

[DanH_sas]

What version of SAS are you running?

[djbateman]

I'm using SAS 9.3.

[Jay54]

I did something like this for the survival plot in this article:  http://blogs.sas.com/content/graphicallyspeaking/2014/02/09/survival-plot/

Here, I have drawn the censored markers using non-grouped scatter plot (in black), and overlaid the same with a grouped scatter plot resulting in colored markers.  Now, only the colored markers are visible in the plot.  However, I have included the non-grouped scatter plot in the legend.

Could you use a similar idea?

[djbateman]

Thank you, Sanjay!  I had tried something similar to this before, but when I overlaid the color on top of the black, it almost looked like the X's had been bolded.  I tried it again, and it wasn't too bad.  I think it was because I didn't specify an exact black color but rather used a default lighter black (dark gray or sorts).

[DanH_sas · Accepted solution]

Since you have 9.3, you might be able to use the LEGENDITEM statement to define the legend entry without having to define the additional scatter. Here is an example:

proc template;

define statgraph custom;

beginGraph;

  legendItem type=marker name='markerItem' / label='Custom Marker' markerattrs=(color=black symbol=circle) labelattrs=black;

  layout overlay;

    scatterPlot x=weight y=height / group=sex markerattrs=(symbol=circle);

    discreteLegend 'markerItem';

  endLayout;

endGraph;

end;

run;

proc sgrender data=sashelp.class template=custom; run;

[Jay54]

Yes, overplotting markers can make them look bolder due to the anti-aliasing.  To avoid that, a trick I have used before is to plot the points you don't want to see somewhere outside the graph range (and set x and y axis ranges to show only the range you need).  You will need another column for that.  Now, there is no overplotting of the markers.

Or, you can set x or y for this column as missing...but an entire column cannot be missing.  If it is, the plot will be thrown out.  Dan's suggestion will work too.