Kwame,

Similar to Barry's suggestion you could also do something like:

data Ds1; input id aw; cards; 37 3 37 6 37 13 37 14 37 18 39 7 39 10 39 12 39 17 ;

data ds2 (keep=id aw difw); set Ds1; by id; hold_difw=ifn(first.id, (.), dif(aw)); last_aw=lag(aw); if hold_difw < 3 and not(missing(hold_difw)) then do; hold_aw=aw; call missing(difw); aw=last_aw; output; difw=hold_difw; aw=hold_aw; output; end; run;

HTH, Art --------- On Tue, 12 Aug 2008 15:40:24 -0700, barry.debenham@TALK21.COM wrote:

>On Aug 12, 4:19 pm, aso...@HOTMAIL.COM (Kwame Asom) wrote: >> Could someone help me with the SAS code to obtain ds2,please? Ie. to >> retain the ids and the aws, that will give me difw less than 3? I HAVE >> 100 DIFFERENT IDS, ONLY 2 ARE SHOWN. >> *Initially, I used the ifn and the dif functions to arrive at difw in ds1. >> This is the code I used to obtain difw: My ultimate aim is to arrive at >> ds2 below. >> Data w1; >> Set w; >> BY ID; >> Difw=ifn(first.id, (.), dif(aw)); >> Run; >> Ds1: >> Id aw difw >> 37 3 . >> 37 6 3 >> 37 13 5 >> 37 14 1 >> 37 18 4 >> 39 7 . >> 39 10 3 >> 39 12 2 >> 39 17 5 >> >> Ds2: >> Id aw difw >> 37 13 >> 37 14 1 >> 39 10 >> 39 12 > > > >An alternative might be to read the data twice in the same data step. >If I can assume that the data is sorted as in your example. > >Then:- > >Data W1(keep=id aw); > Set w(keep=id aw > rename=(aw=awf id=idf ) > firstobs=1); > Set w(keep=id aw > rename=(aw=awl id=idl ) > firstobs=2); > If ((awl - awf) lt 3) and (idf = idl) then > do; > aw = awf; > id = idf; > output; > aw = awl; > id = idl; > output; > end; >Run; > >This reads the data twice and renames the variables each time. Since >the first and second obs are read into the PDV at the same time you >can compare the difference directly. Combine this comparison with a >test to make sure that the id is the same across both observations >then output data for both of the records that meet the required >conditions. > >N.B. - This code is untested. > >If you haven't solved this yet I hope this code proves useful. > >Regards, > >BPD