Hi all, in my experiment we tested participants in 6 conditions (3 thermal stimulation x 2 force of contraction) randomized over 4 runs, total of 48 trials. I computed the 1st level analysis with nilearn, so I get one map per condition per run per participant. I am now developping the 2nd level analysis with nilearn and I am wondering which is the best option to build it. I would like ask which is the difference between inputing in the 2nd level analysis the output of the fixed effect function (which gives one map per participant for each experimental condition) vs adding the runs as regressor in the design matrix of the 2nd level analysis?
Thanks for your help.
I would rather input fixed effects statistics. I think that this is simple yet powerful and we don’t have mixed effects models to take into account several levels of variance.
HTH,
Bertrand
I assume that both thermal stimulation and force of contraction are within-individual factors. Effect estimates from multiple runs can be effectively incorporated into the population level without aggregation by using a hierarchical model. This approach has the advantage of more accurately characterizing the variance-covariance structure.
This modeling approach is supported by the AFNI program 3dLMEr, using a specification for two within-individual factors of A and B such as:
A*B + (A*B | subject)
For additional details, refer to the 3dLMEr documentation.
Gang Chen
1 Like