help > Two-sample ttest
Showing 1-3 of 3 posts
Sep 9, 2016 10:09 AM | Anika Sierk - Charite Berlin
Two-sample ttest
Dear all,
I have a very basic question. I am using NBS for the first time on my structural DTI data. I would like to run a two-sample t-test. I am contrasting 2 groups - a patient and a healthy control group. We do not have an a prioiri hypothesis on the differences in connectivity between these two groups as no previous study has looked at structural connectivity in that particular patient group.
We have 46 patients in total. In the connectivity matrix ( that we created with ExploreDTI), the third dimension corresponds to the subject number. They are not chornically ordered, that is, the first 23 patients do not necessarily belong to the patient group and the following 23 are healthy controls. Instead, they are mixed up, so that my design matrix looks as follows (to simplify only for 8 subjects):
1 0
0 1
1 0
1 0
0 1
1 0
1 0
0 1
My questions refers to
1) the contrast matrix: I put either [1 -1] or [-1 1]. Is that correct? Or do I need to put [1 0] or [0 1], respectively?
2) the Threshold. Which appropriate threshold would you suggest (apart form being able to play around with it a bit, if reasonable)?
Many thanks already for taking the time to help,
Cheers,
Anika
I have a very basic question. I am using NBS for the first time on my structural DTI data. I would like to run a two-sample t-test. I am contrasting 2 groups - a patient and a healthy control group. We do not have an a prioiri hypothesis on the differences in connectivity between these two groups as no previous study has looked at structural connectivity in that particular patient group.
We have 46 patients in total. In the connectivity matrix ( that we created with ExploreDTI), the third dimension corresponds to the subject number. They are not chornically ordered, that is, the first 23 patients do not necessarily belong to the patient group and the following 23 are healthy controls. Instead, they are mixed up, so that my design matrix looks as follows (to simplify only for 8 subjects):
1 0
0 1
1 0
1 0
0 1
1 0
1 0
0 1
My questions refers to
1) the contrast matrix: I put either [1 -1] or [-1 1]. Is that correct? Or do I need to put [1 0] or [0 1], respectively?
2) the Threshold. Which appropriate threshold would you suggest (apart form being able to play around with it a bit, if reasonable)?
Many thanks already for taking the time to help,
Cheers,
Anika
Sep 10, 2016 01:09 AM | Andrew Zalesky
RE: Two-sample ttest
Hi Anika,
For the design matrix you have described, you should use [-1 1] and [1 -1] as your contrasts. One contrast will test for an increase in connectivity in patients, while the other will test for a decrease in connectivity in patients.You should select "t-test" as the statistical test.
The contrasts [0 1] and [0 -1] are for designs that include a column 1', which you are not using.
Regarding the threshold, I suggest trying a few different values - perhaps 2, 2.5, 3 and 4.It is important to understand how the network changes as a function of the threshold.
Regarding Explore DTI, it is important to seed a large number of streamlines to build your connectivity matrices. You should be seeding at least 1 million streamlines for each patient, otherwise the connectivity measures tend to have too much sampling variability.
Good luck with your analysis,
Andrew
Originally posted by Anika Sierk:
For the design matrix you have described, you should use [-1 1] and [1 -1] as your contrasts. One contrast will test for an increase in connectivity in patients, while the other will test for a decrease in connectivity in patients.You should select "t-test" as the statistical test.
The contrasts [0 1] and [0 -1] are for designs that include a column 1', which you are not using.
Regarding the threshold, I suggest trying a few different values - perhaps 2, 2.5, 3 and 4.It is important to understand how the network changes as a function of the threshold.
Regarding Explore DTI, it is important to seed a large number of streamlines to build your connectivity matrices. You should be seeding at least 1 million streamlines for each patient, otherwise the connectivity measures tend to have too much sampling variability.
Good luck with your analysis,
Andrew
Originally posted by Anika Sierk:
Dear all,
I have a very basic question. I am using NBS for the first time on my structural DTI data. I would like to run a two-sample t-test. I am contrasting 2 groups - a patient and a healthy control group. We do not have an a prioiri hypothesis on the differences in connectivity between these two groups as no previous study has looked at structural connectivity in that particular patient group.
We have 46 patients in total. In the connectivity matrix ( that we created with ExploreDTI), the third dimension corresponds to the subject number. They are not chornically ordered, that is, the first 23 patients do not necessarily belong to the patient group and the following 23 are healthy controls. Instead, they are mixed up, so that my design matrix looks as follows (to simplify only for 8 subjects):
1 0
0 1
1 0
1 0
0 1
1 0
1 0
0 1
My questions refers to
1) the contrast matrix: I put either [1 -1] or [-1 1]. Is that correct? Or do I need to put [1 0] or [0 1], respectively?
2) the Threshold. Which appropriate threshold would you suggest (apart form being able to play around with it a bit, if reasonable)?
Many thanks already for taking the time to help,
Cheers,
Anika
I have a very basic question. I am using NBS for the first time on my structural DTI data. I would like to run a two-sample t-test. I am contrasting 2 groups - a patient and a healthy control group. We do not have an a prioiri hypothesis on the differences in connectivity between these two groups as no previous study has looked at structural connectivity in that particular patient group.
We have 46 patients in total. In the connectivity matrix ( that we created with ExploreDTI), the third dimension corresponds to the subject number. They are not chornically ordered, that is, the first 23 patients do not necessarily belong to the patient group and the following 23 are healthy controls. Instead, they are mixed up, so that my design matrix looks as follows (to simplify only for 8 subjects):
1 0
0 1
1 0
1 0
0 1
1 0
1 0
0 1
My questions refers to
1) the contrast matrix: I put either [1 -1] or [-1 1]. Is that correct? Or do I need to put [1 0] or [0 1], respectively?
2) the Threshold. Which appropriate threshold would you suggest (apart form being able to play around with it a bit, if reasonable)?
Many thanks already for taking the time to help,
Cheers,
Anika
Sep 10, 2016 06:09 AM | Anika Sierk - Charite Berlin
RE: Two-sample ttest
Dear Andrew, many thanks for your helpful response!