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!VTVM1.BITNET!WLTPIRIE
- Message-ID: <STAT-L%92110623252454@VM1.MCGILL.CA>
- Newsgroups: bit.listserv.stat-l
- Date: Fri, 6 Nov 1992 23:19:41 EST
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: Walt pirie <WLTPIRIE@VTVM1.BITNET>
- Subject: Chi-square
- Lines: 53
-
- Following the discussion below: I had decided not to join the
- original discussion on this, and in the meantime, several notes
- came in giving various ways of computing "exact" probabilities
- for the values given in the first note. ( I don't have those
- any more) giving probability values of the order of e-7, e-8,
- etc.
- Phil Spector's note is much more relevant than those so-called
- probabilities from the approximating chi-squares. In fact I think
- even from his note, one could be mislead into thinking those
- probability values could be used to comment on which of the
- original statistics were more significant than others. With
- differing degrees of freedom, even an ordering of the pro-
- babilities would be completely unreliable that far out in the
- tails.
- Walt.
-
- In response to the following query:
-
- > I have used the PROBCHI function in SAS
- > (probability = PROBCHI (value, df)) to estimate the probability
- > of each chi square value. I've run into the problem that the
- > function cannot produce a probability value precise enough to
- > distinguish between the significance of chi square values.
-
- Please realize that values of chi-square probabilities as small as the
- ones you are talking about don't really have any quantitative value.
- The behavior of a statistic which is claimed to have a chi-square
- distribution (which in all likelihood is obtained only asymptotically),
- cannot be so delicately measured, especially in the very far tails,
- where the values which were reported lie.
-
- This is *not* a flame, just a gentle reminder about the way that
- probability values, especially ones corresponding to the tails of
- a distribution, should be used.
-
- - Phil Spector
- Statistical Computing Facility
- Department of Statistics
- UC Berkeley
- spector@stat.berkeley.edu
-
- |==================================|============================|
- | Walter R. Pirie | |
- | Department of Statistics | |
- | Virginia Tech | |
- | Blacksburg, VA 24061-0439 | |
- | | |
- | Tel. 703-231-5441 | |
- | Bitnet WLTPIRIE@VTFVM1 | |
- | Telnet WLTPIRIE@VTVM1.CC.VT.EDU | |
- | | |
- | Fax 703-231-3863 | |
- |==================================|============================|
-