home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!europa.asd.contel.com!paladin.american.edu!auvm!MIS.CORP.MCCAW.COM!PETERK
- Message-ID: <9209110203.AA04692@mis.corp.mccaw.COM>
- Newsgroups: bit.listserv.sas-l
- Date: Thu, 10 Sep 1992 19:02:54 -0700
- Reply-To: Peter Kretzman <peterk@MIS.CORP.MCCAW.COM>
- Sender: "SAS(r) Discussion" <SAS-L@UGA.BITNET>
- From: Peter Kretzman <peterk@MIS.CORP.MCCAW.COM>
- Subject: RE: numbers in SAS
- Comments: To: SAS-L@uga.cc.uga.edu
- Lines: 27
-
- Paul A. Thompson writes on Thu, 10 Sep 1992 18:18:15
-
-
- So my conclusion is that consistent lengths produce consistent
- results. The problem is NOT THE PROBLEM OF THE IMPERFECT
- REPRESENTATION of decimals in a binary coding - this is a given. The
- real problem is DIFFERING LEVELS of IMPERFECT REPRESENTATION. If you
- use different lengths your imperfect representations will be
- different. If you use the same length all around, your imperfect
- representations will be consistent,and USUALLY yield the correct
- result.
- (end of quoted material)
-
-
- I think Paul has given an excellent summary of the issue. The only
- thing I would add, given some of the postings on this thread, would
- be to stress again the concept that _all_ variables (whereever they
- may come from) within the DATA step take on an internal length of 8.
- Their assigned length ONLY BECOMES A FACTOR WHEN THEY ARE STORED in
- the output data set, at which point shorter length variables are
- truncated. If any subsequent steps retrieve those observations,
- these shorter length variables are then padded (internally, in
- memory) back to length 8, then are truncated again on output.
-
- Peter Kretzman
- McCaw Cellular Communications, Inc.
- (206) 828-1344
-