I would like to do the equivalent of running a Proc Summary with about 10 CLASS variables, getting totals of my VAR variable at every dimension combination, and also the number of distinct levels of each class variable at each class-var combination.

proc summary data=orig; class c b a; var traffic; output out=cube sum=traffic(rename=(_type_=uniqclas)) ndistinct=na nb nc;

Unfortunately there is no such NDISTINCT statistic.

One can then do the following:

proc sql; create table na as select *, count(distinct(a)) as na from cube(where=(uniqclas=1)) union select *, count(distinct(a)) as na from cube(where=(uniqclas=3)) group by b union select *, count(distinct(a)) as na from cube(where=(uniqclas=5)) group by c union select *, count(distinct(a)) as na from cube(where=(uniqclas=7)) group by b,c ;

create table nb as select *, count(distinct(b)) as nb from cube(where=(uniqclas=2)) union select *, count(distinct(b)) as nb from cube(where=(uniqclas=3)) group by a union select *, count(distinct(b)) as nb from cube(where=(uniqclas=6)) group by c union select *, count(distinct(b)) as nb from cube(where=(uniqclas=7)) group by a,c ;

create table nc as select *, count(distinct(c)) as nc from cube(where=(uniqclas=4)) union select *, count(distinct(c)) as nc from cube(where=(uniqclas=5)) group by a union select *, count(distinct(c)) as nc from cube(where=(uniqclas=6)) group by b union select *, count(distinct(c)) as nc from cube(where=(uniqclas=7)) group by a,b ; quit;

One can then merge these back to cube after appropriately revising uniqclas. However, my problem with this aproach is that with 10 class vars, some of which have many (say 100-200) distinct values, while I'm able to do the Proc Summary within my resource limits, the SQL takes a very long time.

Is there a better, faster way ?

Thanks!

/===========================================\ | Aaron L. Dukes | Consultant Statistician and SAS Developer | | Breakthru Systems Development, Inc. | | bsd01@sprynet.com \===========================================/