[BioC] Re: limma and affy (fwd)

[Gordon Smyth]

At 09:20 AM 24/06/2003, Phguardiol at aol.com wrote:
>Gordon,
>
>if I have 4 groups instead of 3 and no missing values shall I change 
>df=3+eb... into df= 4+eb... ? ie: df=number of groups+... ? of course I d 
>have to change the contrast matrix

If you have no missing values, then the degrees of freedom contributed from 
the fit will be (number of arrays) - (number of groups). But please use

    df = fit$df + eb$df

as this is always correct. fit$df is the usual residual degrees of freedom 
from one-way analysis of variance model for each gene. eb$df is the degrees 
of freedom contributed by Bayesian smoothing of the residual standard 
deviations.

>>Do you have any missing values? If you do, df = fit$df + eb$df would be 
>>better.
>>-> I guess you mean df = fit2.... ?

It doesn't matter, 'fit' and 'fit2' have the same residual degrees of 
freedom. Changing the contrasts or changing the parametrization doesn't 
change the residual standard error estimates.

>>As we discussed in a previous email, lines 19-21 just can't go together. 
>>M is a matrix, A is just a vector.
>>-> yes that was a mistake but I did correct it in my last command file
>
>>Now you are in the realm of the R language itself rather than LIMMA 
>>specifically.
>
>-> yes and I need to learn a lot there !
>
>thanks a lot
>Philippe

Gordon