home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Newsgroups: bit.listserv.sas-l
- Path: sparky!uunet!gatech!paladin.american.edu!auvm!uvvm!klassen
- References: <9209091735.AA28587@slc10.INS.CWRU.Edu>
- Message-ID: <92253.111111KLASSEN@UVVM>
- Date: Wed, 9 Sep 1992 11:11:11 PDT
- Reply-To: Melvin Klassen <KLASSEN@UVVM.BITNET>
- Sender: "SAS(r) Discussion" <SAS-L@UGA.BITNET>
- Comments: Warning -- original Sender: tag was NETNEWS@UVVM.UVIC.CA
- From: Melvin Klassen <KLASSEN@UVVM.BITNET>
- Subject: Re: numbers in SAS
- Lines: 25
-
- "Paul A. Thompson" <pat@PO.CWRU.EDU> says:
- >Length is really quite simple.
- >
- >1. Don't attempt to save space with reals. Real numbers must be
- > (IMEVHO) represented in length 8 variables (default).
- > Shorter variables will lead to difficulties down the line.
- >
- I don't agree with these statements.
- If you use "LENGTH 4" to store all real numbers,
- you can store "integer" numbers up to 16,777,215.
- (Oops! It won't hold Michael Milliken's salary ($50M) without round-off. :-)
-
- If you use "LENGTH 4" under VM/CMS or MVS/XA,
- you will be chopping the fractional-part of a floating-point number
- after the 24th bit, i.e., introducing round-off errors in the order
- of only 1 part in 2**25 (32,554,432).
- Rhetorical question: How many researchers have data which is accurate
- to this many digits?
- Rhetorical answer: If you're the 1-in-a-million who does, use "LENGTH 5".
- Second R-question: How many researchers would spot any difference in any
- table produced by any statistical procedure in SAS,
- especially when the numbers in the table are only
- displayed to 7 decimal digits?
- Second R-answer: Only those researchers with **extreme** numerical
- instability or auto-correlation in their datasets.
-