Hi,

I would appreciate some help to find out the correct design matrix for my experiment in EdgeR.
The experiment is about thermal acclimation in fish, where I have two groups 1.Control 2.Treatment. From each group I have 2 tanks, so 4 tanks in total, 2 from control and 2 from treatment. I sampled fish tissue from 6 fish from each tank. I am interested in genes which are differentially expressed between the control and treatment. However I have 2 tanks from each group but I am not interested in the difference between the tanks of same group. At the same time I have to take into account the tank effect which I could do with a nested ANOVA. So my question here is **Is there any model close to nested ANOVA that fits to my experiment in EdgeR?** I looked for different possibility in the manual but most of them are crossed models which does not fit to my experiment. Also I am a newbie here so I might have overlooked on the possibilities. Any help will be greatly appreciated. Thank you in advance.

Hi Gordon,

I'm not sure I understand. Will you explain how summing the counts within each tank is any different from just ignoring the tank effect?

Summing the counts is, in effect, averaging the fish in each tank. You then get a separate value for each tank. The tank-to-tank variation then becomes the error term. The statistical significance of the treatment effect is judged relative to the tank-to-tank variation, so the latter is not being ignored.

Nested anova does the same thing in effect. Here's an example with normally distributed data. First do an nested anova with the tank effect as random:

Now, average the values in each tank to get just four values instead of 24. Doing an anova on just the four tank averages gives the same p-value:

This averaging approach will work for any balanced repeated measures design.

Ah, I get it. Basically set the within-tank variability to zero by summing. Thanks.

Thank you very much Gordon and James for the valuable comments. It helped a lot. I understand it now :)