[BioC] Testing main effects and interactions in edgeR without contrasts

[Ryan C. Thompson]

On 12/13/2013 02:25 AM, David Shuker [guest] wrote:
> We note that testing whether the two contrasts associated with a three-level main effect - a, b, c - are significant is not the same question as testing whether the main effect itself is significant
Are you sure? My understanding is that these two are precisely the same 
thing. What is the difference?

> We would like outputs and tests along the lines of glm() or anova(), not "anova-like" tests that use the results of one or more contrasts to ask similar, but not exactly the same, questions of the data.
 From everything I've read, the only difference between say, limma's 
"anova-like" test and an actual ANOVA is the use of empirical Bayes 
squeezed variances that share information between genes. As far as I 
know, limma's tests are otherwise equivalent to t-tests/ANOVA tests. 
Similarly, edgeR's glmLRT is the same as a classical glm likelihood 
ratio test in every way except for the use of squeezed dispersions. And 
I think you can even use un-squeezed dispersions by calling estimateDisp 
with prior.df=0.

>
> We apologise in advance if we have either missed something obvious in the user guide or relevant discussion somewhere online. If edgeR does not have this functionality, we will probably use edgeR in conjunction with other packages.
>
> As with all the community, we remain in awe of the work the edgeR developers have done, and thank them in advance.
>
> Best wishes,
>
> Dave Shuker
>
>   -- output of sessionInfo():
>
> NA
>
> --
> Sent via the guest posting facility at bioconductor.org.
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor