Date:         Wed, 2 Aug 2000 09:46:59 -0400
Reply-To:     "Sorkin, John" <SorkinJ@GRC.NIA.NIH.GOV>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "Sorkin, John" <SorkinJ@GRC.NIA.NIH.GOV>
Subject:      Multiple t-tests and Type-I error
Comments: To: "J.Groeneveld@ITGROUPS.COM" <J.Groeneveld@ITGROUPS.COM>
Comments: cc: "patrickm@MAIL.NEWCASTLE.EDU.AU" <patrickm@MAIL.NEWCASTLE.EDU.AU>
Content-Type: text/plain; charset="iso-8859-1"

Jim,

Please help me clarify my thinking. You suggest that adjustment for multiple
comparisons is only necessary when performing a series of non-independent
tests. I believe your example was
 a-b, b-c, c-a.
In this case, I agree as the third comparisons is a linear combination of
the first two comparisons:
c-a=-(a-b + b-c).

You also state that in the case of a series of independent comparisons,
a-b, a-c, a-d, a-e
no adjustment is needed.

I was taught that adjustment for multiple comparisons is needed when
performing multiple independent tests as the probability of a type-I error
when performing N tests is N*0.05 assuming the tests are preformed at a 5%
significance level. (To simplify the disucssion I am ignoring the
probability of making a type-II error.) Have I missed something?

Thank you,


John Sorkin

P.S. I have sent a copy of this message to Patrick. I invite Patrick to help
me clarify my thinking.