NITRC: Graph Theory GLM (GTG) MATLAB Toolbox: help

help

8 Subscribers

help > how to write design matrix for permutation analysis

Showing 1-4 of 4 posts

how to write design matrix for permutation analysis

Hi Dr. Spielberg,

Thank you for developing this tool!
I'm having some difficulty at the run permutation analysis stage.

If the format of connectivity matrix is n(ROI)*n(ROI)*number of subject*2(condition), how should I write the design matrix and contrast vector in the stage of run permutation analysis to contrast the graph properties of 2 conditions.

I am looking forward to hearing from you. Thank you for your help.
xiuling

RE: how to write design matrix for permutation analysis

You don't have to account for the repeated measure in the design matrix - the toolbox will recognize that there is a repeated measure because of the dimensions of your input data and automatically set up between, within, and (if needed) between x within interactions. Therefore, you should just set up your design/contrast matrices as if you did not have a repeated measure and those effects will be automatically computed.

RE: how to write design matrix for permutation analysis

Dear Dr. Spielberg,

Forgive my ignorance, I meet the same questions and cannot fully understand what you replied. So, can we give an example：

If the connectivity matrix is 90*90*5*2 , now we set a design matrices:
(5 subjects: sub1/sub2/sub3/sub4/sub5; 2 conditions: A/B )

we plan to compare A > B, should predictors be set like this?

predictors: 1 1 1 1 1
options: Contrasts
contrast: [1 0] (group is 1, intercept is 0)
A < B contrast: [-1 0]

What do these two columns represent? And how should I set the predictors and contrast to find the correction between A > B and scale score?
Best wishes,
Wenlong

Attachment: fig1.png

RE: how to write design matrix for permutation analysis

Originally posted by li wenlong:

The program will recognize that you have a repeated measure, because the last dimension of your input matrix is 2 (not 1). Therefore, you don't have to explicitly do anything beyond setting up your input matrix correctly. In other words, the toolbox will automatically compute between, within, and (if needed) between x within interactions. Therefore, you should just set up your design/contrast matrices as if you did not have a repeated measure and those effects will be automatically computed.

If you just want to compare the repeated measure, just create a variable that is one column of all ones and enter use that as your predictor. This will setup a within-subjects comparison (equivalent to a paired t-test) that tests whether the mean of one condition differs from the mean of the other. You don't need a separate 'group' predictor, unless you have different groups. The contrast would just be 1. You also don't need to do the -1 contrast, unless you want a 1-tailed test (you will be given the 1-tailed and 2-tailed p-values).