@Ronein:
On second thought, if your score values are limited-range integers (say, from 0 to 100, as assumed below) and you would like to do everything in a single step, you can use a key-indexed table:
data want (keep = score n p rename=(n=No_customers p=PCT_From_Total)) ;
array f [0:100] _temporary_ ;
do until (end) ;
set have end = end ;
f [score] + 1 ;
end ;
t = sum (of f [*]) ;
do score = lbound (f) to hbound (f) ;
n = f [score] ;
if nmiss (n) then continue ;
p = divide (n, t) ;
tp + p ;
output ;
end ;
score = . ;
n = t ;
p = tp ;
output ;
format p percent7.1 ;
run ;
However, if your score values are arbitrary-type, arbitrary-range, you still can do it in one step, but then the key-indexed table needs to be replaced with a hash table. It makes code somewhat lengthier but also makes no assumptions about data:
data want (keep = score n p rename=(n=No_customers p=PCT_From_Total)) ;
dcl hash h () ;
h.definekey ("score") ;
h.definedata ("score", "n") ;
h.definedone () ;
do until (end) ;
set have end = end ;
if h.find() ne 0 then n = 1 ;
else n + 1 ;
h.replace() ;
end ;
dcl hiter i ("h") ;
do while (i.next() = 0) ;
t + n ;
end ;
do while (i.next() = 0) ;
p = divide (n, t) ;
output ;
tp + p ;
end ;
score = . ;
n = t ;
p = tp ;
output ;
format p percent7.1 ;
run ;
By the way, I do like the TABULATE solution by @andreas_lds (and like this proc in general) but think that you need an output data set.
Paul D.
