home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Subject: (no subject given)
- Path: sparky!uunet!gatech!destroyer!sol.ctr.columbia.edu!howland.reston.ans.net!paladin.american.edu!auvm!UTDALLAS.EDU!WIORKOW
- Message-ID: <93Jan8.094752cst.15577@utdallas.edu>
- Newsgroups: bit.listserv.stat-l
- Date: Fri, 8 Jan 1993 09:28:22 -0600
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: wiorkow@UTDALLAS.EDU
- Lines: 32
-
- The reason that statisticians are often not put on TV shows or even panels
- is that we have this nasty habit of being faithful to our profession and to th
- e truth. About six years ago I was on a panel of the National Academic of Scie
- nces studying the effect of highway speed limitations on mortality and accident
- rates. I was the only statistician (although a well trained economist also was
- on the panel), the rest of the folks were ones with a vested interest in the a
- rea including representatives of state highway departments, and many people who
- had written on the relationship. We reviewed many studies, none of which show
- ed any relationship of accident or death rates with average speed. What did sh
- ow up was a relationship between the rates and the variability in speeds. In o
- ther words if everyone went 85 mph on the interstates there would be fewer acci
- dents than if some went 55 and others went 65. The two of us were unsuccessful
- in convincing the other 15 or so members on emphasizing this point. (Science o
- n panels is done by majority). I would be most interested in seeing the data o
- n which the "secondary smoke" report is based. This is just the kind of issue
- that the ASA should review to see if in fact the "unbiased" researchers are at
- least doing the statistical inference correctly. I was particularly intrigued
- by the statment to wit that several of the studies did not show significance si
- nce they were too small. We all know that the underlying theory of statistical
- inference indicates that any difference, no matter how small in effect, will be
- detected if the sample sizes are large enough. The real question has to do wi
- th what the probabilities of contracting disease are. I also noted that this w
- as never stated. Instead we got figures on number of deaths (3000 as I recall)
- out of a population of 300,000,000 gives a probability in a single year of abou
- t 1 in 1000. This strikes me as sufficiently small to not burden out society w
- ith the costs of administering and policing this issue when those same funds co
- uld be used to help people in slums exposed to much greater risk factors. But
- of course that is a much harder problem.
-
- John J. Wiorkowski
- Professor of Statistics
- University of Texas at Dallas
-