help
help > RE: Interaction: treatment x symptomscores
Sep 1, 2016 08:09 PM | Alfonso Nieto-Castanon - Boston University
RE: Interaction: treatment x symptomscores
Dear Julian,
Yes, that is a perfectly valid and sensible procedure (I am assuming here that the cluster/ROIs that you are extracting from in the last REX step correspond to those found in analysis (A)). The key why this is valid, in case anyone questions potential nonindependence or "double dipping" issues in these analyses, is that your two analyses (A) and (B) are testing orthogonal effects. The only potential suggestion I might add would be to try in (A) to use instead: Treatment, Placebo, ScoresTreatmentCentered, ScoresPlaceboCentered [1 -1 0 0]. That will still look at the same main effect of treatment but it could be slightly more sensitive particularly in cases where the variability associated with score differences may be large compared to the between-group differences.
Hope this helps
Alfonso
Originally posted by Julian Roessler:
Yes, that is a perfectly valid and sensible procedure (I am assuming here that the cluster/ROIs that you are extracting from in the last REX step correspond to those found in analysis (A)). The key why this is valid, in case anyone questions potential nonindependence or "double dipping" issues in these analyses, is that your two analyses (A) and (B) are testing orthogonal effects. The only potential suggestion I might add would be to try in (A) to use instead: Treatment, Placebo, ScoresTreatmentCentered, ScoresPlaceboCentered [1 -1 0 0]. That will still look at the same main effect of treatment but it could be slightly more sensitive particularly in cases where the variability associated with score differences may be large compared to the between-group differences.
Hope this helps
Alfonso
Originally posted by Julian Roessler:
Dear Colleagues
I have a seamingly simple question for the second level seed2voxel analysis, where we are interested inthe effects of a treatment in relation to a numeric symptom score. So we have two groups: treatment & placebo. For each subject there is a score i.e. from 0-30.
we would first calculate:
(A) the contrast for the main effect of treatment: Treatment > Placebo [1-1]. Using the same seed we then calculate
(B) the contrast for the interaction: Treatment, Placebo, Scores for only the Treatment Group, Scores for only the Placebo Group [0 0 1 -1].
With the Rex toolbox, we then load the main effect results (from contrast (A)). Then in order to see, weather there is an effect of interaction (treatment x symptoms) we would load the SPM file of the interaction statistics (the spm.mat from (B)) and extract the results on a cluster level. As such, we then are able to interpret whether the aberrant connectivity due to treatment is affected by the sypmtomscores.
Is this procedure a valid approach?
Warm regards,
Julian
I have a seamingly simple question for the second level seed2voxel analysis, where we are interested inthe effects of a treatment in relation to a numeric symptom score. So we have two groups: treatment & placebo. For each subject there is a score i.e. from 0-30.
we would first calculate:
(A) the contrast for the main effect of treatment: Treatment > Placebo [1-1]. Using the same seed we then calculate
(B) the contrast for the interaction: Treatment, Placebo, Scores for only the Treatment Group, Scores for only the Placebo Group [0 0 1 -1].
With the Rex toolbox, we then load the main effect results (from contrast (A)). Then in order to see, weather there is an effect of interaction (treatment x symptoms) we would load the SPM file of the interaction statistics (the spm.mat from (B)) and extract the results on a cluster level. As such, we then are able to interpret whether the aberrant connectivity due to treatment is affected by the sypmtomscores.
Is this procedure a valid approach?
Warm regards,
Julian
Threaded View
| Title | Author | Date |
|---|---|---|
| Julian Roessler | Aug 31, 2016 | |
| Alfonso Nieto-Castanon | Sep 1, 2016 | |
| Julian Roessler | Sep 3, 2016 | |