GLM for pre and post test design

1

Dear experts,

The experiment consists of 5 stages: the first run collects resting-state data, the second run is a pre-task measurement, the third run is a learning phase, the fourth run is a post-task measurement, and the last run collects resting-state data again.

The scanning protocol are consistent across all runs, and the tasks in the second and fourth runs are the same.If I want to compare brain activity before and after the learning phase in both resting-state and task conditions. How should I model and fit the general linear model (GLM)? Should I fit a separate GLM for each run? Additionally, the third run’s learning phase is long, and I want to separate the early and late parts of this phase to compare differences between early and late learning stages. How should I set up the GLM for those?

Thanks,
Yang

2

Dear Yang,
Indeed, I think that you want to build a GLM for each run, and then do relevant statistics to compare pre-and post-task activity, e.g. using a paired t test.
The only exception is the resting state, for which GLM coefficient do not make sense. My best advice is to extract a connectome from a fixed set of regions in the corresponding runs, and then compare statistically connectome differences across runs.
Regarding you question on the third run, you should indeed create regressors for early and late parts of the learning phase. For which you probably want to create an events file in which you define the onset and offset of each phase of the experience. Then you’ll be able to create a design matrix that represents this properly.
Home that this makes sense.
Best,
Bertrand

3

Hi Bertand,

Thanks for your reply!
The third run is a boxcar design and has the same task throughout the run. For early and late parts of the learning phase, should I fit one GLM to them? e.g. Y=β1x+β2x+β3+...+ε. β1 for early stage, β2 for middle stage and β3 for late. Because the task is the same in each stage, will the three GLM coefficients be homogeneous? So is it necessary to create a GLM for each of the three stages as well?

Best,
Yang

4

I think that having single GLM with regressors \beta_1, \beta_2, \beta_3 is fine.
Best,
Bertrand

1 Like