home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!gatech!paladin.american.edu!auvm!QMRELAY.MAIL.CORNELL.EDU!PAUL_VELLEMAN
- Message-ID: <STAT-L%92072718035136@VM1.MCGILL.CA>
- Newsgroups: bit.listserv.stat-l
- Date: Mon, 27 Jul 1992 18:05:37 U
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: Paul Velleman <paul_velleman@QMRELAY.MAIL.CORNELL.EDU>
- Subject: Fixed regressions
- Lines: 23
-
- Subject:
- Fixed regressions
- John J. Wiorkowski says of Lotus that "They have obviously fixed the bug in
- the algorithm which existed in Ver. 2.1." based on one small trial.
- First: I don't believe that there really was a bug, just a poor choice of
- algorithm. (For those of you who have been around from the start of the
- spreadsheet debates, this is just one more example of how tricky it can be to
- get the algorithm right.)
- Second: If they have changed the algorithm, it might still not be stable
- against other challenges, or against the same challenge carried just a bit
- further. Have they documented their algorithm? Do we know what it will do when
- (not if) it breaks down? (*All* algorithms for regression break down somewhere.
- The trick is knowing when and failing gracefully.) Is the behavior of the
- algorithm dependent on the floating point chip?
- Finally: Just one more plug. If you want to do accounting, by all means use a
- spreadsheet. If you want to do statistics, you will be *much* better off
- getting a statistics package designed and written by professional
- statisticians. There is much science and a bit of art behind the choice and
- implementation of algorithms and a wide enough array of packages to choose from
- to please almost anyone.
- -- Paul Velleman
- (Claimer: As the developer of the Data Desk exploratory data analysis package
- for the Mac, I am biased in favor of professionally developed software.)
-