Hi Mike,
thanks for your great support. We need some assistance in the selection of the correct model for an experiment. Mice were unilaterally whisker-deprived, so the main effect is expected in one hemisphere (the right one, here called ipsi). We have two main groups: unilaterally deprived mice and undeprived controls. However, we have sequenced both hemispheres separately for both groups.
| Condition--- | Hemisphere |
|---|---|
| deprived | ipsi |
| deprived | contra |
| undeprived | ipsi |
| undeprived | contra |
Since the contralateral side in deprived mice is expected to show an effect as well, which in all likelihood is different from the ipsilateral effect, we don't know how to best approach the design. We are guessing that a simple ~ Condition + Hemisphere design is not what we want, because deprived contra did not undergo the same intervention as deprived ipsi (but the indirect effect of the contralateral deprivation).
Is there a way to include all samples in one design that lets us correct for the hemisphere differences that are seen in the undeprived animals, and at the same time find the effect of the treatment for both ipsi- and contralateral deprived hemispheres? Or do we have to compare ipsi to ipsi and contra to contra?
Many thanks, Sebastian
Thanks for the quick answer. Sorry for the late reply, I had exceeded my comment limit.
This is for clarification, because I am not too experienced with the interaction terms, so far I had always designed my experiments to prevent this. Is the baseline difference influenced by the unknown effect that the deprivation might have on the contralateral side? For example, three possibilities: In the deprived animals, 1) The contralateral side reacts same as the ipsi side, 2) the contra side does not react at all and remains on control level, 3) the contra side reacts inversely to the ipsi side.
We don't know which of these is true. So, in all these theoretical cases, would the design
~hemisphere + hemisphere:conditionbe able to resolve the true differences?Take a look at the diagram of the way interaction terms work in the vignette. By adding the interaction of hemisphere and condition, you allow for a different condition effect across hemisphere. I'm perhaps missing your question.
My question is more about the estimation of baseline, since in this experiment the conditions are different from the vignette example. In the vignette, there is only a combination of two influences, genotype and condition, and the premise is that expression is already different between those two, and the estimation of baseline is based on that difference. This would be perfectly applicable to our experiment, if the whisker deprivation (
condition) were performed on both sides (hemisphere). But since it is only performed on one side, it only has direct effects on the ipsi side of the brain, about its effect on the contra side we do not know (it could be any of the three described in my previous answer, or in between).The catch is, we estimate our baseline based on the hemisphere differences, but we don't know the influence that our experiment has on the contra side of the deprived animals. We could also think about it as two different interventions, directly deprived (ipsi) and indirectly deprived (contra), and this would naturally reflect in the model that we choose, giving us one more factor level. However, as you recommend in the vignette, we don't want to use a needlessly complicated design. Thus, we are not certain how to approach it; this might be a very nuanced question, and I apologise if my description is not clear.
Sorry, I might suggest that you meet with a statistician. I think there's an aspect I don't catch here, and unfortunately I have limited time on the support site and have to focus primarily on software issues.
Perfectly understandable. Thanks for your time.