Re: [R] Lack of independence in anova()

From: JRG <>
Date: Thu 07 Jul 2005 - 05:50:07 EST

On 6 Jul 2005 at 12:30, Douglas Bates wrote:

> On 7/6/05, Douglas Bates <> wrote:
> ...
> > Perhaps we could review the sequence of events here. This exchange
> > began with your sending me a message claiming that there is a bug in
> > lm or anova in R because the results of your simulation were what you
> > expected.

At the risk of further roiling the waters ...

As several have already pointed out, the "usual" F-tests in a balanced ANOVA have independent numerators but a common denominator, and hence the F-statistics cannot be independent. Is this not the basis of Kimball's Inequality, which states that the effect of the common denominator is that the simultaneous error rate cannot exceed what it would be if the tests *were* in fact independent?

In other words, you should get a simultaneous error rate for the F-tests that is lower than that under independence of test statistics. Are you?

---JRG John R. Gleason

> I meant to write "were not what you expected". I've got to learn to
> read the email messages before posting them.
> ______________________________________________
> mailing list
> PLEASE do read the posting guide! mailing list PLEASE do read the posting guide! Received on Thu Jul 07 05:55:50 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:33:17 EST