Recreate a graph with two mirrored histograms

[ABohyn]

I have seen the following graph in this paper: https://doi.org/10.1093/ejcts/ezy167
It shows the distribution of propensity scores for two groups, highlighting the patients that have been matched using the score.

Imagine that I have a data set DATA with three variables:

• MATCHED: 0/1 variable to identify the matched patients
• PS: propensity scores for all the patients
• GROUP: variable that defines if the patients are in the ON-PUMP or OFF-PUMP group.

How could I replicate this graph using PROC SGPLOT ? 

[PaigeMiller]

Google finds an answer quickly

https://blogs.sas.com/content/graphicallyspeaking/2013/11/21/comparative-histograms/

and other solutions are found via Google as well

[Ksharp]

Check Rick.Wicklin 's blog:

Overlay a curve on a histogram in SAS - The DO Loop

/*
That would be better if you post some real data.
*/
data have;
call streaminit(123);
do group='ON-PUMP','OFF-PUMP';
 do id=1 to 200;
  matched=rand('bern',0.6);
  ps=rand('uniform');
  group2=group;
  output;
 end;
end;
run;


ods select none;
proc sgplot data=have;
histogram ps/group=group scale=count nbins=20;
ods output sgplot=sgplot1;
run;

proc sgplot data=have(where=(matched=1));
histogram ps/group=group2 scale=count nbins=20;
ods output sgplot=sgplot2;
run;
ods select all;

data sgplot1;
 set sgplot1;
 if BIN_PS_GROUP_GROUP_SCALE_cou__GP='ON-PUMP' then 
 BIN_PS_GROUP_GROUP_SCALE_cou___Y=-BIN_PS_GROUP_GROUP_SCALE_cou___Y;
 keep BIN_PS_GROUP_GROUP_SCALE_cou__GP BIN_PS_GROUP_GROUP_SCALE_cou___Y BIN_PS_GROUP_GROUP_SCALE_cou___X;
run;
data sgplot2;
 set sgplot2;
 if BIN_PS_GROUP_GROUP2_SCALE_co__GP='ON-PUMP' then 
 BIN_PS_GROUP_GROUP2_SCALE_co___Y=-BIN_PS_GROUP_GROUP2_SCALE_co___Y;
 keep BIN_PS_GROUP_GROUP2_SCALE_co__GP BIN_PS_GROUP_GROUP2_SCALE_co___Y BIN_PS_GROUP_GROUP2_SCALE_co___X;
run;
data want;
 merge sgplot1 sgplot2;
run;


proc format;
picture fmt
low-0='00009';
run;
proc sgplot data=want noautolegend;
format BIN_PS_GROUP_GROUP_SCALE_cou___Y fmt.;
styleattrs datacolors=( darkgreen lightgreen) AXISEXTENT=data;
/*styleattrs datacolors=( CX006837 CXC2E699) AXISEXTENT=data;*/
 vbarparm category=BIN_PS_GROUP_GROUP_SCALE_cou___X 
          response=BIN_PS_GROUP_GROUP_SCALE_cou___Y /
          barwidth=1 group=BIN_PS_GROUP_GROUP_SCALE_cou__GP outlineattrs=(color=grey) nofill;
 inset 'OFF-PUMP'/textattrs=(color=lightgreen) position=top;

 vbarparm category=BIN_PS_GROUP_GROUP2_SCALE_co___X 
          response=BIN_PS_GROUP_GROUP2_SCALE_co___Y /
          barwidth=1 group=BIN_PS_GROUP_GROUP_SCALE_cou__GP outlineattrs=(color=grey);
 inset 'ON-PUMP'/textattrs=(color=darkgreen) position=bottom;

xaxis values=(0 to 1 by 0.2) type=linear display=(nolabel) offsetmin=0.05 offsetmax=0.05;
yaxis label='n of subject' offsetmin=0.1 offsetmax=0.1;
run;

[Rick_SAS · Accepted solution]

This type of graph is sometimes called a butterfly plot.

Your example is slightly more complicated than the one in the blog post, "A butterfly plot for comparing distributions,"

but I think that article is a good place to start.