What would be the most truthful way to draw inferences in a multilevel cell biology dataset?

I hardly ever use R, so am not sure how to reply. The error structure is essentially of the form brain/well/cell but how much of this is relevant will depend on the design. Still, that’s the error structure. Is there a way in R to put that in the error structure and put treatment is as a fixed effect. In that case cannot the same model be used for all cases and cannot the design matrix be left to decide how the standard errors are generated? Would this work? :
aov_general = aov(y ~ treatment + Error(brain / well/cell),
data = senn)
The answers this would give would then depend on the level at which treatment was varied.

Probably the best thing would be for me to generate sone data and analyse them in Genstat. Then others could show what it would look like in R. This will take a little time.

well my interest in genstat was piqued by the blog post you linked to when I read it back when first posted but now I think I’ll download a trial copy and explore it, especially to compare analyses between genstat, R, and something like SAS/JMP.

A recent interesting editorial on this topic (Neuron journal): PDF

2 Likes