Dear Andrew,
I have a query about NBS set-up and want to confirm if my interpretation. After reading the manual & a lot of posts on this forum, I am not certain if my method is correct.
My hypothesis: I am looking if there is a difference between healthy controls and patients, in functional connectivity (source EEG -> imaginary coherence).
My doubts: I don't know if my contrast is set-up correctly; and therefore if my interpretation of the results is correct.
Design Matrix: My design matrix is as shown. I have 17 controls, 22 patients - so its 39 rows in total and 2 columns.
| 1 | 0 |
| 1 | 0 |
| 1 | 0 |
| x17 | |
| 0 | 1 |
| 0 | 1 |
| 0 | 1 |
| x22 |
Statistical Test: t-test
Contrast: [1 1] - im not interested in increase/decrease just the difference in functional connectivity between controls and patients.
Threshold: 0.5 -to- 22 - I have ran the code for a wide range of threshold values, but even at t=22 I have a large network of significant connections, which to me seems unusual.
Connectivity matrices: These are 68x68 ROI matrices, 39 total - again set up as control 1-17 (total17), patient 18x39 (total22). Diagonal is also '1', just as in the schizophrenia example.
Node Coordinates & Node Labels: separate files, both load in well. Tripled checked these are correct.
Exchange blocks: N/A - not used.
Permutations: 5000
Significance: 0.05
Method: Network-Based Statistics (NBS)
Component Size: I tried both: extent & intensity.
Question 1: Is my design matrix correct? From the manual and some of the posts on this forum I believe it is suitable for my hypothesis.
Question 2: What about the contrast? I've set it to [1 1] because im just looking for a difference. However, maybe contrast should be [1 0]?
Question 3: Is my "t" threshold unsually high? At t=22,
contrast[1 1] - I get a p-value of 0.035, 55 edges, 39 nodes all
over the brain (on intensity setting), but at 22.5 nothing is
significant anymore; seems like a very sharp drop aswell.
Thank you for your time!
Vyte
Hi Vyte,
I think that your contrast should be either [1 -1] or [-1 1]. Using a contrast of [1 1] does not seem correct.
I'm not sure what you mean by just looking for a difference. If you are not interested in the sign of the difference, you could select the F-test.
Your design matrix looks ok.
Yes - t=22 is extremely high. I recommend a threshold between 2-8.
Andrew
Originally posted by Vyte Jan:
Dear Andrew,
I have a query about NBS set-up and want to confirm if my interpretation. After reading the manual & a lot of posts on this forum, I am not certain if my method is correct.
My hypothesis: I am looking if there is a difference between healthy controls and patients, in functional connectivity (source EEG -> imaginary coherence).
My doubts: I don't know if my contrast is set-up correctly; and therefore if my interpretation of the results is correct.
Design Matrix: My design matrix is as shown. I have 17 controls, 22 patients - so its 39 rows in total and 2 columns.
1 0 1 0 1 0 x17 0 1 0 1 0 1 x22
Statistical Test: t-test
Contrast: [1 1] - im not interested in increase/decrease just the difference in functional connectivity between controls and patients.
Threshold: 0.5 -to- 22 - I have ran the code for a wide range of threshold values, but even at t=22 I have a large network of significant connections, which to me seems unusual.
Connectivity matrices: These are 68x68 ROI matrices, 39 total - again set up as control 1-17 (total17), patient 18x39 (total22). Diagonal is also '1', just as in the schizophrenia example.
Node Coordinates & Node Labels: separate files, both load in well. Tripled checked these are correct.
Exchange blocks: N/A - not used.
Permutations: 5000
Significance: 0.05
Method: Network-Based Statistics (NBS)
Component Size: I tried both: extent & intensity.
Question 1: Is my design matrix correct? From the manual and some of the posts on this forum I believe it is suitable for my hypothesis.
Question 2: What about the contrast? I've set it to [1 1] because im just looking for a difference. However, maybe contrast should be [1 0]?
Question 3: Is my "t" threshold unsually high? At t=22, contrast[1 1] - I get a p-value of 0.035, 55 edges, 39 nodes all over the brain (on intensity setting), but at 22.5 nothing is significant anymore; seems like a very sharp drop aswell.
Thank you for your time!
Vyte
Hi Andrew,
Thank you very much for your message and feedback.
I changed my hypotheses and I have been re-running the data using [-1 1] and [1 -1]. The results now make a lot more sense and the t-score is max 6.
Thank you a lot for the advice, I finally understand the results a lot more and can start to analyse the data :)
Healthy Regards,
Vyte