Here's the brute force way. I.e. it doesn't take advantage of the "fast" algorithm in the Wikipedia reference. so it won't scale well against larger datasets.
The example finds the medcouple stat for variable CLOSE in sashelp.stocks for IBM stocks (N=233). If you have a test data set to confirm correctness of the algorithm, I'd suggest running it.
data have (keep=date close);
set sashelp.stocks;
where stock='IBM';
run;
proc means data=have noprint;
output out=median median=median;
var close;
run;
data vneed (keep=h) / view=vneed;
set have end=end_of_have;
if _n_=1 then set median;
array upper{2000} _temporary_; /* distance of upper obs to median*/
array lower{2000} _temporary_; /* distance of lower obs to median*/
if close>=median then do;
P+1;
upper{p}=close-median;
end;
if close<=median then do;
Q+1;
lower{q}=median-close;
end;
if end_of_have;
do i=1 to P;
do j=1 to Q;
if upper{i}=0 and lower{j}=0 then h=sign(P-1-i-j);
else h= (upper{i}-lower{j})/(upper{i}+lower{j});
output;
end;
end;
run;
proc means data=vneed median; var h;
run;
