[19128] in s-news-athena
Re: [S] Pr(F) in anova()
daemon@ATHENA.MIT.EDU (Dave Krantz)
Tue Sep 7 14:21:58 1999
Date: Tue, 7 Sep 1999 14:17:47 -0400 (EDT)
From: Dave Krantz <dhk@paradox.psych.columbia.edu>
Message-Id: <199909071817.OAA23693@paradox.psych.columbia.edu>
To: dbm3@po.cwru.edu
Cc: dhk@paradox.psych.columbia.edu, s-news@wubios.wustl.edu
What must be happening in this case is that the added terms decrease
the estimated random error, which appears in the denominator of the
F tests. The sums of squares and mean squares for old model terms
should not change in a sequential anova table; but the residual sum
of squares necessarily goes down as you add terms, and the residual
mean square CAN go down (if the added terms account for more variance
than they cost in degrees of freedom for error).
I'm not sure what use you want to make of the F and p values in a
sequential anova table. I find them nearly useless, when the design
is unbalanced, since the model comparisons involved are seldom the
ones that I am interested in.
Dave Krantz
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
