Hello I am currently working with a 2x2x2 experiment and I’m hoping to conduct some PPI analyses. I’m becoming a bit stuck on a few bits however.
Our EVs are:
EH (easy task, high pain),
HH (hard task, high pain)
EL (easy task, low pain)
HL (hard task, low pain)
Across conditions one and two (repeated measures).
*****
I am interested in looking at the main effects of pain and task (initially as a 2x2 design). The thing I’m not 100% certain about is whether we should run separate PPIs for these different main effects? Specifically, what I’m not sure about is whether the different designs we could use are equivalent….
E.g. main effect of pain is (EH+HH)-(EL+HL) we can write this in a single
regressor with a weight of 1 whenever the temp is (H)igh, and -1 whenever the temp is (L)ow….
type onset duration weight
EL 20 30 -1
HL 90 30 -1
HH 160 30 1
EL 230 30 -1
EH 300 30 1
…..
…..
This is our PSYCHological regressor
We would then need another regressor like this:
type onset duration weight
EL 20 30 1
HL 90 30 1
HH 160 30 1
EL 230 30 1
EH 300 30 1
…..
…..
to span “parameter space”
We would then add in a single PPI (Interaction) regressor using the PHYSiological signal recorded from the PAG
I.e. the PPI is PSYCH .* PHYS
We run the model and get a stats map for the PPI regressor for each subject and each study day. We can then average the PPI maps across all the condition 1 and condition 2 to see if they “look” different, and then run a paired t-test to see if they actually are different.
Would this also need to be done for the main effect of task? Or are these equivalent?
Also another question, I’m no expert in this, but I guess we’ll struggle to see an effect….the reason being that even though we've used a factorial design, the bits of signal that are not accounted for by either regressor are pretty small, and then you ask whether the difference between 2 small bits of signal is significantly different…. it might be that we are underpowered to detect this. However, are allowed to be more lenient when reporting PPI results for this very reason? If so, are there any particular significant thresholds used?
Any advice would be much appreciated!
Thanks
|