Re: A simple question on Bonferroni Correction

On Tue, 30 Aug 2005 03:56:11 EDT, "sine_arc@xxxxxxxxxxx"
<sine_arc@xxxxxxxxxxx> wrote:

> Bonferroni correction assumes a joint null hypotheses, which means
> it allows a common joint inference for all hypotheses tested. Am I
> right? [break]]

No, I think you are wrong. Here is the way that I use the
language: An overall ANOVA is what tests the "joint null"
and so you can't conclude about the individual differences.

Bonferroni keeps a larger (say, 5%) nominal test size by
using much smaller (5%/k) cut-offs, for individual test decisions.

> In that case, it is probably not relevant to use it when I want to
> detect which comparisons are significantly different. For e.g., in an
> ANOVA on different varieties of drugs, if we apply Bonferroni
> correction, we can only infer whether induction of drugs show a
> significant difference (say compared to a control condition) but we
> cannot infer which of the drugs was found to be significantly
> different.

- And for Control versus Others, it is advisable to have a
larger N for Control, and to use the specific set of contrasts
designed for one-versus-many. Dunnet's?

> I would appreciate if you can share your views regarding the above. Thanks

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.