help
help > RE: Interactions with covariates in CONN
Apr 2, 2015 01:04 AM | Alfonso Nieto-Castanon - Boston University
RE: Interactions with covariates in CONN
Dear Lars,
If I am interpreting correctly, the analysis in (1) is perfectly correct, but the interaction analyses in (2) should be implemented in this way instead:
a) create two new second-level covariates, one named 'covariate_patients' containing the actual values of your covariate for patients only, and 0's for control subjects, and another named 'covariate_controls' containing the actual values of your covariate for controls only, and 0's for your patients
b) then select 'patients', 'controls', 'covariate_patients', and 'covariate_controls', and enter a contrast [0 0 1 -1] (for the positive interaction test; i.e. higher association between connectivity and your covariate in patients) or [0 0 -1 1] (for the negative interaction test, i.e. higher association between connectivity and your covariate in controls)
Also remember to extract the actual connectivity values and/or perform post-hoc analyses to get the interpretation of the interaction effects right (e.g. a positive interaction could indicate higher -more positive- association between connectivity and your covariate in patients, but it could also indicate a higher -less negative- association, so you typically want to look at the signs of the individual associations, in controls and patients separately, within any significant areas found in the interaction analyses, to get the correct interpretation of your results)
Hope this helps
Alfonso
Originally posted by Lars Michels:
If I am interpreting correctly, the analysis in (1) is perfectly correct, but the interaction analyses in (2) should be implemented in this way instead:
a) create two new second-level covariates, one named 'covariate_patients' containing the actual values of your covariate for patients only, and 0's for control subjects, and another named 'covariate_controls' containing the actual values of your covariate for controls only, and 0's for your patients
b) then select 'patients', 'controls', 'covariate_patients', and 'covariate_controls', and enter a contrast [0 0 1 -1] (for the positive interaction test; i.e. higher association between connectivity and your covariate in patients) or [0 0 -1 1] (for the negative interaction test, i.e. higher association between connectivity and your covariate in controls)
Also remember to extract the actual connectivity values and/or perform post-hoc analyses to get the interpretation of the interaction effects right (e.g. a positive interaction could indicate higher -more positive- association between connectivity and your covariate in patients, but it could also indicate a higher -less negative- association, so you typically want to look at the signs of the individual associations, in controls and patients separately, within any significant areas found in the interaction analyses, to get the correct interpretation of your results)
Hope this helps
Alfonso
Originally posted by Lars Michels:
Dear CONN users
I have two groups (controls and patients) and computed 1 -1 to see in which regions controls do show higher connectivty than patients.
Now, I tried to incoperate covariates in two ways in CONN
1) regress it out, i.e. 1 -1 0 (to get regions corrected for the influence of the covariate)
2) looking at interactions.
a) 1 -1 1 (positive interactions: which regions do show an interaction with this covariate for the contrast 1 -1),
b) 1 -1 -1 (negative interactions)
However, for some of my covariates, I see the identical output for all three types of analyses (see attached png)
Any clue why this is the case?
Thanks a lot
Dr. Lars Michels
I have two groups (controls and patients) and computed 1 -1 to see in which regions controls do show higher connectivty than patients.
Now, I tried to incoperate covariates in two ways in CONN
1) regress it out, i.e. 1 -1 0 (to get regions corrected for the influence of the covariate)
2) looking at interactions.
a) 1 -1 1 (positive interactions: which regions do show an interaction with this covariate for the contrast 1 -1),
b) 1 -1 -1 (negative interactions)
However, for some of my covariates, I see the identical output for all three types of analyses (see attached png)
Any clue why this is the case?
Thanks a lot
Dr. Lars Michels
Threaded View
| Title | Author | Date |
|---|---|---|
| Lars Michels | Apr 1, 2015 | |
| Alfonso Nieto-Castanon | Apr 2, 2015 | |