[BioC] No degree of freedom in maanova

[Marcelo Laia]

2009/3/13 Juan Pedro Steibel <steibelj at msu.edu>:
> Hello Marcelo,
> I have some experience with MAANOVA and I would be happy to help, but I can
> not find the original email in the BIOC list, could you please forward it to
> me?
> Thanks
> Juan P. Steibel

Dear Juan,

Thank you for your help. Here is the link:
https://stat.ethz.ch/pipermail/bioconductor/2008-February/021247.html
and all two messages that I receive related on the thread. I'm not
currently using maanova, only these two guys. Please, could you help
they?

I suggest to keep the discussion on list, because it could be useful
for someone on the future!

Thank you very much!

Marcelo


2009/3/11 Goutham Vemuri <goutham at chalmers.se>:

Hello Marcelo,

I saw the problem that you had with Maanova on the mailing list and I would
like to request some help from you on how you solved it. I have the exact
problem and I am not able to get the interaction effects.

I have 2 strains (Reference and mutant), 2 carbon sources and 2 kinds of
nutrient starvation, in duplicate (16 conditions). So, my design file is

Array   Strain  Csource Limit   Sample
1       REF     GLU     C       1
2       REF     GLU     C       2
3
.
.
16      MUT     ETH     N       2

The model I used was

anova.full.mix <- fitmaanova(ciao.raw, formula = ~Strain + Csource +
Limitation + Sample + Strain:Csource + Strain:Limitation +
CSource:Limitation, random = ~Sample, method = "REML", verbose=TRUE)

When I ran the matest as

Csource.int.mix <- matest(ciao.raw, anova.full.mix,
term="Strain:Csource",n.perm=1,test.method=rep(1,2))

I get the error that there is no DF to do the test. After reading the
thread, I could not understand how you fixed the problem to get the
interaction effects. Any help is very much appreciated. Thanks.

Goutham


2009/3/13 Kim Kultima <kim.kultima at medsci.uu.se>:
Dear Marcelo,

I have read your thread in Biocunductor concering no degree of freedom in
maanova in factorial designs. I have the exact problem you describe in the
tread. In the last thread I see that you solved the problem:

I already solve that problem. I found that is possible to do the
analysis step-by-step (with interactions and without interactions -
for main effects).


Does this mean that you solved the problem by analysing first main effects,
followed by interaction effects, separately?
Is that ok to do?

Do you know why this problem appear in maanova?

I hope you have the time to answer my questions.

Regargs,

Kim Kultima

2009/3/16 Kim Kultima <kim.kultima at medsci.uu.se>:
Dear Marcelo,

Thank you for your help. Here is the link:
https://stat.ethz.ch/pipermail/bioconductor/2008-February/021247.html
I am currently trying to solve the problem by using the nlme library
instead. Unfortunaley this library is not at all created for use with
microarray data so things has to be re-structured. I have not come up
to an solution why it does not work with maanova. I have a three way
interaction also, which can not be tested for. I have tried different
linear models where it worked, so it should work. I wonder if there is
a bug somewhere in the maanova library or similarily.

Regards,

//Kim