help
help > RE: Regarding between-subject contrast error
May 9, 2023 01:05 AM | Abhishek Patil - Vellore Institute of Technology
RE: Regarding between-subject contrast error
Hello Alfonso,
Thank you for your quick response.
In Analysis (2), I am examining the effect of Patients and Controls in Groups A and B separately, rather than the overall Patients > Control effect.
As you have mentioned previously, the interaction variable can be constructed, which I have done now.
I just have one additional issue about this.
For example, which one should I use when looking at effects of Group A - Control participants?
Patients*GroupA, Controls*GroupA, Age*GroupA, Gender*GroupA, Education*GroupA and contrast [0 1 0 0 0 0] or
Controls*GroupA, Age*GroupA, Gender*GroupA, Education*GroupA and contrast [1 0 0 0 0]
Should the results differ in these two cases?
Regards,
Abhishek
Originally posted by Alfonso Nieto-Castanon:
Thank you for your quick response.
In Analysis (2), I am examining the effect of Patients and Controls in Groups A and B separately, rather than the overall Patients > Control effect.
As you have mentioned previously, the interaction variable can be constructed, which I have done now.
I just have one additional issue about this.
For example, which one should I use when looking at effects of Group A - Control participants?
Patients*GroupA, Controls*GroupA, Age*GroupA, Gender*GroupA, Education*GroupA and contrast [0 1 0 0 0 0] or
Controls*GroupA, Age*GroupA, Gender*GroupA, Education*GroupA and contrast [1 0 0 0 0]
Should the results differ in these two cases?
Regards,
Abhishek
Originally posted by Alfonso Nieto-Castanon:
Hi Abhishek
The analysis (1) (Patients>Control, controlling for age/gender/education) would use the model:
subject-effects: Patients, Controls, Age, Gender, Education
between-subjects contrast: [1 -1 0 0 0]
Analysis (2) (if I am interpreting correctly you would like to compute the evaluate the same patients>controls effect, but now restricted to only subjects within the groupA set), would use the model:
subject-effects: Patients*GroupA, Controls*GroupA, Age*GroupA, Gender*GroupA, Education*GroupA
between-subjects contrast: [1 -1 0 0 0]
and similarly for analysis (3) (patients>controls within the GroupB subset of subjects)
subject-effects: Patients*GroupB, Controls*GroupB, Age*GroupB, Gender*GroupB, Education*GroupB
between-subjects contrast: [1 -1 0 0 0]
note: you can easily create all those ____*groupA and ____*groupB variables, in the Setup.Covariates (2nd-level) tab using the 'Covariate tools -> create interaction of selected covariates' menu. For example you may select 'Patients', 'Controls', 'GroupA' and 'GroupB' and then select 'create interaction of selected covariates' to have CONN automatically create the Patients*GroupA, Patients*GroupB, Controls*GroupA, Controls*GroupB variables (and then repeat the same thing but now selecting age, GroupA,GroupB to create the age-by-group interactions, etc.)
Hope this helps
Alfonso
Originally posted by Abhishek Patil:
The analysis (1) (Patients>Control, controlling for age/gender/education) would use the model:
subject-effects: Patients, Controls, Age, Gender, Education
between-subjects contrast: [1 -1 0 0 0]
Analysis (2) (if I am interpreting correctly you would like to compute the evaluate the same patients>controls effect, but now restricted to only subjects within the groupA set), would use the model:
subject-effects: Patients*GroupA, Controls*GroupA, Age*GroupA, Gender*GroupA, Education*GroupA
between-subjects contrast: [1 -1 0 0 0]
and similarly for analysis (3) (patients>controls within the GroupB subset of subjects)
subject-effects: Patients*GroupB, Controls*GroupB, Age*GroupB, Gender*GroupB, Education*GroupB
between-subjects contrast: [1 -1 0 0 0]
note: you can easily create all those ____*groupA and ____*groupB variables, in the Setup.Covariates (2nd-level) tab using the 'Covariate tools -> create interaction of selected covariates' menu. For example you may select 'Patients', 'Controls', 'GroupA' and 'GroupB' and then select 'create interaction of selected covariates' to have CONN automatically create the Patients*GroupA, Patients*GroupB, Controls*GroupA, Controls*GroupB variables (and then repeat the same thing but now selecting age, GroupA,GroupB to create the age-by-group interactions, etc.)
Hope this helps
Alfonso
Originally posted by Abhishek Patil:
Greetings!
Thank you for this amazing toolbox.
This might be a very basic question related to task-fMRI design:
I have two groups of participants: Patients and Control with covariates Age, Gender and Education level.
I have two more sub-groups which include dividing participants with respect to a score and named as Group A and Group B.
I have added covariates defined as GroupA_age, GroupB_age, GroupA_education, GroupB_education, GroupA_gender and GroupB_gender. I have mean centred the age and education covariates.
Furthermore, in this work, I am using ROI-to-ROI analysis and want to understand the differences in connectivity between the two analyses, namely weighted GLM and gPPI.
To begin, I'd like to know which between subject contrasts I should employ in my second level analysis to comprehend:
1. Patients > Control in terms of functional connectivity.
2. Only Group A effect, and
3. Only Group B effect.
In addition, when I try to choose GroupA and related covariates, I receive a warning suggesting that the design may contain an invalid model.
Thank you in advance
Regards,
Abhishek
Thank you for this amazing toolbox.
This might be a very basic question related to task-fMRI design:
I have two groups of participants: Patients and Control with covariates Age, Gender and Education level.
I have two more sub-groups which include dividing participants with respect to a score and named as Group A and Group B.
I have added covariates defined as GroupA_age, GroupB_age, GroupA_education, GroupB_education, GroupA_gender and GroupB_gender. I have mean centred the age and education covariates.
Furthermore, in this work, I am using ROI-to-ROI analysis and want to understand the differences in connectivity between the two analyses, namely weighted GLM and gPPI.
To begin, I'd like to know which between subject contrasts I should employ in my second level analysis to comprehend:
1. Patients > Control in terms of functional connectivity.
2. Only Group A effect, and
3. Only Group B effect.
In addition, when I try to choose GroupA and related covariates, I receive a warning suggesting that the design may contain an invalid model.
Thank you in advance
Regards,
Abhishek
Threaded View
| Title | Author | Date |
|---|---|---|
| Abhishek Patil | May 4, 2023 | |
| Alfonso Nieto-Castanon | May 5, 2023 | |
| Abhishek Patil | May 9, 2023 | |
| Alfonso Nieto-Castanon | May 9, 2023 | |
| Abhishek Patil | May 17, 2023 | |
| Abhishek Patil | May 10, 2023 | |