How about split your data into two subsets, and rank the one without the particular value first. And for the other one, assign a number 1-10 to x1 based on obs number, then concatenate them back to one data:

data a c; set original; if x ne *** then output a; else output c;

proc rank data=a out=b group=10; var x; ranks x1;

data c; set c; x1=mod(_n_,10);

data b; set b c;

Ya Huang

> -----Original Message----- > From: Tzachi Zach [mailto:zach@ssb.rochester.edu] > Sent: Thursday, August 23, 2001 4:37 PM > To: SAS-L@LISTSERV.UGA.EDU > Subject: proc rank equal-sized groups > > > Hi all, > > I have a data set of 20,000 oservations and i would like to > Group rank it > into 10 equal-sized groups (more or less). The problem is > that there is a > value that appears in about 7,000 observations. Proc rank > assigns all these > observations into one decile, resulting in an uneven distribution of > observations. I would like to spread these 7000 across a few > deciles, as > long as all the 10 deciles are equal-sized. > > The code i used was: > > proc rank data=a out=b groups=10; var x; ranks x1; > run; > > > Any ideas? > > Thanks, > Tzachi >