help > Whole Network
Showing 1-4 of 4 posts
Display:
Results per page:
Feb 15, 2021  06:02 PM | Maximilian Schulz
Whole Network
Dear Dr. Zalewsky,

I have one more question. Is it possible to first calculate and visualize all existing connections of the matrix by applying a contrast vector of [1,0] to a design matrix with for example the group and the age. The idea behind this is to first show all existing connections and then show the connections that change with the predictors.

Best regards

Max
Feb 15, 2021  11:02 PM | Andrew Zalesky
RE: Whole Network
Hi Max,

yes - this is possible.

You may want to consider the one-sample test in this situation, which will detect connections with a weight that is greater than zero. Null hypothesis is that the connection weight is zero.

Note that what you are proposing is commonly referred to as network thresholding in the literature and there are many approaches (statistical and non-statistical) that have been developed to achieve this task.

Andrew
Originally posted by Maximilian Schulz:
Dear Dr. Zalewsky,

I have one more question. Is it possible to first calculate and visualize all existing connections of the matrix by applying a contrast vector of [1,0] to a design matrix with for example the group and the age. The idea behind this is to first show all existing connections and then show the connections that change with the predictors.

Best regards

Max
Feb 16, 2021  11:02 AM | Maximilian Schulz
RE: Whole Network
Thanks Andrew, that helps,

yes which contrast vector would be useful for a one sample test? As for thresholding, I read in this forum that it is better to use the raw matrices. Normally I use the proportional thresholding to exclude spurious connections from the BOLD data; are these excluded in the toolbox by weak significance values anyway or is it better to treshold the matrices in advance?

Max
Feb 16, 2021  10:02 PM | Andrew Zalesky
RE: Whole Network
Hi Max,

the contrast would just be 1 for a one-sample.

Note that there are very many different strategies for thresholding matrices and using a one-sample t-test might not be optimal for many applications. I would probably not use a t-test to threshold. Check out this talk for a summary of thresholding methods:  https://youtu.be/MZEQR7gKfNE

Andrew

Originally posted by Maximilian Schulz:
Thanks Andrew, that helps,

yes which contrast vector would be useful for a one sample test? As for thresholding, I read in this forum that it is better to use the raw matrices. Normally I use the proportional thresholding to exclude spurious connections from the BOLD data; are these excluded in the toolbox by weak significance values anyway or is it better to treshold the matrices in advance?

Max