I have multiple runs data for one task. Of course I want to use all the data to execute 1st level analysis.
I referenced this tutorial:
9.5.8. Simple example of two-session fMRI model fitting
I saw the design matrix and found the design matrixs in this tutorial don’t model the run effect comparing with SPM design matrixs.
Just a example:
SPM by Nipype:
Do I need add it into design matrix by myself? Or it is contained by fixed effect model?
For my unstanding, if i model the difference of mean between different runs, it will be a mixed effect model?
Am I understanding this correctly?
In Nilearn, the data from different runs are fit independently, and a fixed-effects analysis is run later. So, there is no need (and actually no way) to model session effects.
Does that answer your question ?
@bthirion in your opinion, is there any advantage/disadvantage in modelling the different runs independently as opposed to modelling session effects in the same design matrix? Is there a rationale for the latter procedure?
The should not have a large effect on the results.
- In terms of computation, having an aggregated model makes the fit more costly, and hence is suboptimal computationally.
- On the other hand, you may estimate noise autocorrelation slightly better (but SPM12 is biased anyhow IIRC), and this is secondary anyhow.