home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.med
- Path: sparky!uunet!elroy.jpl.nasa.gov!ames!agate!spool.mu.edu!news.nd.edu!mentor.cc.purdue.edu!pop.stat.purdue.edu!hrubin
- From: hrubin@pop.stat.purdue.edu (Herman Rubin)
- Subject: Re: Bashing, truth, etc.
- Message-ID: <Bzo510.23n@mentor.cc.purdue.edu>
- Sender: news@mentor.cc.purdue.edu (USENET News)
- Organization: Purdue University Statistics Department
- References: <1992Dec21.171109.2975@cnsvax.uwec.edu> <2311@hsdndev.UUCP>
- Date: Tue, 22 Dec 1992 16:12:36 GMT
- Lines: 56
-
- In article <2311@hsdndev.UUCP> rind@binoc.bih.harvard.edu (David Rind) writes:
- >>[hrubin@pop.stat.purdue.edu writes:]
- >>>I do not need any data whatever to know that saccharin has an effect on
- >>>cancer rates, or that vitamin C has an effect on longevity, or even that
- >>>the most disreputable quack treatments have an effect.
-
- >This point, which Herman Rubin makes here frequently, has its problems.
- >The assumption is that everything has some effect, however small,
- >on everything, and so given a large enough N, a statistically significant
- >effect will be demonstrated.
-
- >I think this ignores the randomness and finiteness of the universe. If
- >the N needed to show an effect is larger than the number of humans
- >that have ever lived and can ever be expected to live over the lifetime
- >of the universe, I don't think it is fair to say that we "know" that
- >a large enough N will prove an effect. Further, given chaos in
- >the universe, it is unlikely that we could ever control things in
- >such a way that we could have any likelihood that such a small effect
- >would show the same results in an experiment performed twice.
-
- The problem is even worse. For any N, there is a positive probability
- that a "statistically significant" result will occur, even if there is
- absolutely no effect; this is what is usually called the significance
- level.
-
- >That is not to say that this is not a relevant real world problem in
- >statistics. Frequently large studies do demonstrate statistically
- >significant but clinically insignificant results. I am commenting
- >more on the specific philosophical position that leads to the
- >position that we can be certain that saccharin has an effect on
- >cancer rates (either to raise or to lower them).
-
- Again, there is the confusion due to the misuse of the word "significant."
- Whether something is clinically significant is not dependent on the size
- of the sample, but on the magnitude of the effect. Even this is an
- oversimplification; it is unlikely that there is a 0-1 statement
- possible that something is or is not clinically significant. If a
- treatment is effective, the clinical, or practical, significance is
- present BEFORE the data is collected, although the statistical
- significance is not.
-
- That there is a racial effect on the secondary sex ratio is
- statistically significant, but has little, if any, practical
- significance. On the other hand, that a drug has not been shown
- to have statistically significant effects does not mean that its
- effects are not practically significant. The problem is one of
- which action to take, and statistical significance, confidence
- intervals, etc., do not provide the answer. There are procedures
- which take into account the relevant information, and they should
- be used.
-
- --
- Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
- Phone: (317)494-6054
- hrubin@snap.stat.purdue.edu (Internet, bitnet)
- {purdue,pur-ee}!snap.stat!hrubin(UUCP)
-