At 10:16 AM 8/11/00 -0400, David C. Howell wrote:
>I agree with Bruce Weaver on this. Traditionally we do not worry about
>familywise error rates for simple effects, but instead evaluate each at
>alpha = .05.
>Dave Howell
of course, the broader issue is why .05? this totally ignores whether type
II error is of import ...
it seems to me ... just like for any statistical test ... to decide what
value for alpha to use ... AS ONE'S INITIAL SCREENING DEVICE OR IN SOME
POSTHOC DECISION ... then, one must have decided ... in relation to type
II error ... that .05 (or whatever) is sensible to employ ... in some cases
it is ... in many cases it is not ... and in EVERY case ... it is not a
given nor any inherently correct "default" value
what we traditionally do and what makes sense to do ... are not necessarily
(actually, rarely) the same
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================