[BioC] Limma: how to calculate ordinary p-values omitting the eBayes() command

[Gordon Smyth]

>Date: Tue, 12 Jul 2005 16:52:34 +0200
>From: "Groot, Philip de" <philip.degroot at wur.nl>
>Subject: [BioC] Limma: how to calculate ordinary p-values omitting the
>         eBayes() command
>To: <bioconductor at stat.math.ethz.ch>
>
>Hello all,
>
>Gordon Smith helpen me a while ago with calculating regularised p-values
>(paired t-test) using Limma.

Smyth!

<code deleted>

>Now I want to investigate the effect of the regularisation, so I also
>want to calculate the ordinary p-values.

The Limma User's Guide explains in a couple of places how to obtain 
ordinary t-statistics, for example on page 38.

 > tstat.ord <- fit$coef / fit$stdev.unscaled / fit$sigma
 > pvalue.ord <- 2 * pt( abs(tstat.ord), df=fit$df.residual, lower.tail=FALSE)
 > pvalue.ord.adj <- p.adjust(p.value.ord, method="fdr")

Why do you want to investigate the effect of regularisation? If you don't 
understand the statistical procedures very well, it isn't a good idea to 
second-guess the methods. It is better to choose a well-respected method 
and stick to it. I'm not saying that you're doing this, but I suspect there 
is a tendency for some people to try lots of ad hoc alternatives until they 
get a set of p-values which suits their pet hypothesis, and this is very 
bad science!

Gordon

>Regards,
>Philip