home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!NIOSHE2.EM.CDC.GOV!DKW4
- Return-Receipt-To: <dkw4@NIOSHE2.EM.CDC.GOV>
- Encoding: 150 TEXT
- X-Mailer: Microsoft Mail V3.0 (beta-2)
- Message-ID: <2B354C38@router.em.cdc.gov>
- Newsgroups: bit.listserv.stat-l
- Date: Fri, 18 Dec 1992 14:12:00 EST
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: dkw4@NIOSHE2.EM.CDC.GOV
- Subject: Re: Tutorial Software (fwd)
- Lines: 136
-
- I forwarded J. Philip Miller's comments about the dangers of using
- spreadsheet packages with ill-conditioned data to some of the industrial
- hygienists I work with. What follows is a response from one of them.
-
- Dr. Miller, could you answer these questions directly if you've got the
- time/energy? If not, I'll take my own best stab 'cause I tend to concur
- with your opinion. I'd also be happy to hear from anyone else on this.
-
- Thanks very much -
-
- Deanna Wild
-
- *-----------------------------------------------------------------*
- Deanna Wild
- Statistician
- National Institute for Occupational Safety and Health
- 4676 Columbia Parkway MS: R-4
- Cincinnati, Ohio 45226
-
- INTERNET: DKW4@NIOSHE2.EM.CDC.GOV
- *-----------------------------------------------------------------*
-
-
-
- ------------------------------------------------------------------------------
- FORWARDED FROM: Wild, Deanna K.
- FROM: Zaebst, Dennis D.
-
- TO: dkw4 CINC/NIOSHE2 DATE: 12-18-92
- TIME: 12:21
- CC: Hines, Cynthia J.
- Kawamoto, Melody M.
- Waters, Martha A.
- Zaebst, Dennis D.
-
- SUBJECT: Re: Tutorial Software (fwd)
- PRIORITY: R
- ATTACHMENTS:
-
- Deanna:
-
- I hope I'm not stirring up a hornet's nest, but here goes...
-
- Can you explain any of the background, or expand on exactly what Dr. Miller
- is saying?
-
- As one who has used Lotus, and subsequently, Quattro Pro and Excel for
- straightforward statistics for years, I am a little perplexed by this guy's
- response that teaching students to do statistical analyses on a spreadsheet
- is "worrisome". Is he saying that, specifically, using the multiple
- regression routines in spreadsheets gives inaccurate results? Does he have
- data which demonstrate this? What does he mean by "ill-conditioned data"?
- If the results are inaccurate, how inaccurate are they? When should they be
- avoided, or should they never be used?
-
- By the way, have you looked at the list of statistical functions and
- commands built into Excel 4.0? All I can say is, WOW! But can we use them
- safely?
-
- I guess the issue boils down to how well tested is the software under a wide
- variety of situations (data sets and analyses)? Is SAS better tested than
- Quattro Pro, Excel, or for that matter, Statgraphics, SPSS, S-plus, or CSS
- Statistica? What do we really know about the comparative accuracy of
- results? I would think that the accuracy would partly depend on the software
- itself, but we also have to remember that all these programs are running on
- the same microprocessor with the exact same instruction set built into it
- (80x86).
-
- Is this a really serious issue, or are these people splitting hairs about the
- 15th decimal place, or something in between? (Can we get Dr. Miller to
- respond directly to the above questions)?
-
- Sorry about all the questions. Just what you need just before the holidays,
- eh?
-
- Thanks,
-
- Dennis Z.
- ------------------------------------------------------------------------------
- REPLY FROM: Zaebst, Dennis D. FROM: Wild, Deanna K.
-
- TO: Kawamoto, Melody M. CINC/NIOSHE1 DATE: 12-11-92
- Sussell, Aaron L. CINC/NIOSHE1 TIME: 15:13
- Zaebst, Dennis D. CINC/NIOSHE1
-
- CC:
- SUBJECT: Re: Tutorial Software (fwd)
- PRIORITY: R
- ATTACHMENTS:
-
- Just one additional comment (from a statistical discussion group on Internet)
- about using spreadsheets to do statistical analyses... It showed up today
- and seemed too coincidental to ignore...
-
- Deanna
-
- ------------------------------------------------------------------------------
- FORWARDED FROM: Wild, Deanna K.
- DATE: Dec 11 04:08:29 1992 -00:00 relative to GMT
- FROM: "J. Philip Miller" [A=ATTMAIL; S=Miller"; G="J.; I="P; DDA.TYPE=ID; DDA.
- VALUE=internet(b)WUBIOS.WUSTL.EDU(b)phil]
-
- SUBJECT: Re: Tutorial Software (fwd)
- IPMessageID: internet3460408260
-
- TO: Multiple recipients of list STAT-L [A=ATTMAIL; S=STAT-L; G=Multiple; I=Mr;
- DDA.TYPE=ID; DDA.VALUE=internet(b)VM1.MCGILL.CA(b)STAT-L]
-
- PRIORITY:
- > In article <9212100544.AA23083@leven> brobinso@LEVEN.APPCOMP.UTAS.EDU.AU
- > writes:
- > >Wot about a simple spreadsheet? All you have to do is get them to set up
- > >the required columns and program the formulae. I haven't seen a spreadsheet
- > >yet that does regression and correlation, even though nearly all have mean
- > >and SD.
- >
- > I think using a spreadsheet is a very good suggestion, but I'm a little
- > perplexed by the last sentence: 123 has done multiple regression in every
- > version I've used, as has Quattro Pro. They don't do them via @ functions,
- > instead they're there as commands. In Quattro Pro 1.0 it's in the Tools
- > Advanced Math menu. I forget where 123 put it, and don't have recent
- > versions of either package to look at.
-
- I don't want to open up a heated discussion that was previously held here, but
- I feel I must point out that many of us who specialize in computational
- statistics have some problems with this. The primary reasons for this is that
- the "obvious" formulae are problematic with respect to numerical accuracy for
- ill conditioned problems. Thus the idea that we should teach students to do
- something which may result in wrong answers is worrysome.
-
- -phil
-
- --
- J. Philip Miller, Professor, Division of Biostatistics, Box 8067
- Washington University Medical School, St. Louis MO 63110
- phil@wubios.WUstl.edu - Internet (314) 362-3617 [362-2694(FAX)]
-