John Fox wrote:
Dear Tal,
I didn't have time to look at all this yesterday.
Since aov() doesn't do what I typically want to do, I guess I've not paid
much attention to it recently. I can see, however, that you appear to have
specified the error strata incorrectly, since (given your desire to compare
to Anova) the within-block factors are nested within blocks. Something like
npk.aovE <- aov(value ~ N*P*K + Error(block/N*P*K), npk.long)
should be closer to what you want, and in fact produces all of the sums of
squares, but doesn't put all of the error terms together with the
corresponding terms; thus, you get, e.g., the test for N but not for P and
K, even though the SSs and error SSs for the latter are in the table. By
permuting N, P, and K, you can get the other F tests. I suspect that this
has to do with the sequential approach taken by aov() but someone else more
familiar with how it works will have to fill in the details. I wonder,
though, whether you've read the sections in Statistical Models in S and MASS
referenced in the help file for aov.
Does it not help if you use Error(block/(N*P*K))? I suspect you're being
bitten by operator precedence, of the same making as the fact that 1/2*2
is 1, not 0.25.
-pd
--
O__ ---- Peter Dalgaard Øster Farimagsgade 5, Entr.B
c/ /'_ --- Dept. of Biostatistics PO Box 2099, 1014 Cph. K
(*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalga...@biostat.ku.dk) FAX: (+45) 35327907
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
