home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!wupost!gumby!yale!yale.edu!jvnc.net!netnews.upenn.edu!netnews.cc.lehigh.edu!ns1.cc.lehigh.edu!den0
- From: den0@ns1.cc.lehigh.edu (DEAN E. NELSON)
- Newsgroups: bit.listserv.stat-l
- Subject: Re: Bonferroni adjustments for annual results...
- Message-ID: <1992Jul22.204729.150339@ns1.cc.lehigh.edu>
- Date: 22 Jul 92 20:47:29 GMT
- Organization: Lehigh University
- Lines: 33
-
- In article <STAT-L%92071910460529@VM1.MCGILL.CA>, ECOOCH@PENNSAS.UPENN.EDU (Evan
- Cooch) writes:
- >
- > So, I end up with a large table, one row for each year, with the SS, F and P
- >values noted. Say I have 30 years. In your opinion, do the individual year
- >significances need to be adjusted for the number of years presented?
- >
- >I have my opinion on this (and I won't tip my hand yet). One approach is to
- >ignore the issue and present both within year probabilities, and the
- >Bonferroni significances of same. Obviously, this begs the issue of whter you
- >need to adjust in the first palce or not.
-
- The Bonferroni adjustment comes from an inequality, and thus ensures an
- overall error rate will be below some specified alpha, but it could be a lot
- lower, and hence it is notoriously conservative (unless you are doing safety
- studies, in which case it is liberal).
-
- A confounding factor in your case is whether the relation of X and Y from years
- near each other are correlated, which I believe makes the Bonferroni even
- more conservative.
-
- We once did some simulations comparing the bonferroni with other real and
- heuristic methods and found it to be highly conservative under the range
- of alternatives (in the specific experimental design we were looking at)
- we tried.
-
- So, even if philosophically you buy the multiple comparisions argument,
- experience would seem to point you to another form of adjustment.
-
- --
- Dean E. Nelson den0@lehigh.edu (Internet)
- (215)-758-4988 (That is d-e-n-zero, not the letter O)
-
-