[BioC] Factorial experiment: No degree of freedom in maanova

[Marcelo Laia]

Dear,

I try to solve the "No degree of Freedom" problem after the Naomi's
suggestion. Thank you very much!

I found that:

1. with interactions in the formula, there aren't DF for analysis in
both, fixed and mixed model (with Samples in random factor) for four
factors (Var, ind, Trat and Time).

2. without interactions in the model are possible to calculate all
factors (Var, ind, Trat and Time) in both methods, mixed and fixed.

2. without Samples in the model (fixed model) is it possible to
calculate only interactions. No DF for single factors.

Now, I am on a dilemma: Do the analysis of the four factors in a mixed
model (or fixed, both works better) and, after to include the
interactions, do the analysis using fixed model (without
Samples)?!?!?!  Do I violating statistic's laws here?!?!

This is not a question 100% related to Bioconductor, but, here, there
are persons that, I think, could have, in the past, problems like this
one ( I hope :)   ).

Any comments is very welcome!

Thank you very much.

Marcelo


> 2008/2/20, Naomi Altman <naomi at stat.psu.edu>:
>
> > I have not used MAANOVA, but I think the problem
>  >  is probably that you included Sample in your model.
>  >
>  >  --Naomi
>  >
>  >
>  >  At 01:03 PM 2/19/2008, Marcelo Laia wrote:
>  >  >Hi,
>  >  >
>  >  >We have a factorial experiment like this:
>  >  >
>  >  >2 Varieties (resistant and susceptible)
>  >  >2 Conditions (with inductor and with H2O)
>  >  >2 Treatments (inoculated and non inoculated)
>  >  >4 Times points
>  >  >3 biological repetitions
>  >  >96 arrays (one color)
>  >  >
>  >  >We start to run the analysis on maanova package.
>  >  >
>  >  >In our design file design.txt there are these columns:
>  >  >
>  >  >Array   Dye     Var     ind     Trat    Time    Sample
>  >  >1       1       Res     H2O     Inoc    6h      1
>  >  >(...)
>  >  >96      1       Susc    Ind     N_inoc  48h     3
>  >  >
>  >  >We had been adjusted these model:
>  >  >
>  >  >anova.full.mix <- fitmaanova(data=leonardo.raw.averaged,
>  >  >+ formula=~Var+ind+Trat+Time+Var:ind+Var:Trat+Var:Time+
>  >  >+ ind:Trat+ind:Time+Trat:Time+Sample, random=~Sample)
>  >  >
>  >  >When we try to run the matest function,
>  >  >
>  >  >ftest.full.Var.mix = matest(leonardo.raw.averaged,anova.full.mix,
>  >  >+ term="Var",n.perm=100,test.method=rep(1,3))
>  >  >
>  >  >  we got a "No degree of freedom" error.