Statistical power of KH and SH tests?

[Richard.Zander at MOBOT.ORG]

Peter:

Not being able to reject a null hypothesis NEVER means you must accept it.
You simply can't reject it. Thus, if your null is that the trees are not
significantly different, not being able to reject the null does not mean
that the trees are generated identically, merely that they are not generated
differently. Regarding "one of these tests" make sure you are not skirting a
multiple test problem (requiring Bonferroni correction) by fishing for a
best p-value.

My understanding is that beta is 1 minus alpha.



-----Original Message-----
From: Peter Werner [mailto:pgwerner at SFSU.EDU]
Sent: Thursday, May 13, 2004 11:37 AM
To: TAXACOM at LISTSERV.NHM.KU.EDU
Subject: [TAXACOM] Statistical power of KH and SH tests?


I was wondering if anybody could point me to literature that would
explain how to estimate beta (the probability of making a Type II
error) for a given Kishino-Hasegawa or Shimodaira-Hasegawa test. I'm
wondering about cases where one of these tests results in a p-value of
greater than 0.05 resulting in the acceptance of the null hypothesis
that two trees are not significantly different - I want to know the
probability that in that case I may be failing to reject a false null
hypothesis.

BTW, I'm aware of the paper by Goldman, et al (2000) pointing out the
problems of using a KH test to compare a tree derived through maximum
likelihood with an alternate tree. I'm using it in an exploratory way
and would only use an SH test or the more computationally difficult
SOWH test for reporting of any such tree comparisons.

Peter Werner
Graduate Student, Mycology
San Francisco State University