help
help > Between-condition contrast weighting in gPPI
Jan 20, 2021 06:01 AM | Sara Kim
Between-condition contrast weighting in gPPI
Dear all,
I am somewhat new to fMRI and would be grateful for some insight into contrast weighting in 2nd-level gPPI (bivariate regression) analyses.
The data used an event-related design with six within-subjects conditions. I had thought that within-subject contrasts reflect linear combinations of regression coefficients–for example:
• Conditions selected: Cond1, Cond2
• Between-condition contrast: [0.5 0.5]
• Model: y = 0.5*BetaCond1 + 0.5*BetaCond2 + 0*BetaCond3 + 0*BetaCond4 + 0*BetaCond5 + 0*BetaCond6
• Interpretation: Regions showing greater connectivity during Cond1 and Cond2, relative to an implicit baseline
However, I see in the SPM.mat file that the actual model does not use a weighting of 0.5 but rather sqrt(0.5):
• Conditions: Cond1, Cond2
• Contrast: [0.5 0.5]
• SPM.mat description: 0.707107*Cond1 + 0.707107*Cond2
This is true when I take the difference between task conditions as well:
• Conditions: Cond1, Cond2
• Contrast: [1 -1]
• SPM.mat description: 0.707107*Cond1 - 0.707107*Cond2
And in case it is relevant, the weighting is perhaps proportional to the number of conditions included in the contrast:
• Conditions: Cond1, Cond2, Cond3, Cond4
• Contrast: [0.25 0.25 0.25 0.25]
• SPM.mat description: 0.5*Cond1 + 0.5*Cond2 + 0.5*Cond3 + 0.5*Cond4
I must be missing something key in gPPI, or perhaps contrasts more generally, and haven't been able to find an answer. Any help would be very much appreciated!
Thank you,
Sara
I am somewhat new to fMRI and would be grateful for some insight into contrast weighting in 2nd-level gPPI (bivariate regression) analyses.
The data used an event-related design with six within-subjects conditions. I had thought that within-subject contrasts reflect linear combinations of regression coefficients–for example:
• Conditions selected: Cond1, Cond2
• Between-condition contrast: [0.5 0.5]
• Model: y = 0.5*BetaCond1 + 0.5*BetaCond2 + 0*BetaCond3 + 0*BetaCond4 + 0*BetaCond5 + 0*BetaCond6
• Interpretation: Regions showing greater connectivity during Cond1 and Cond2, relative to an implicit baseline
However, I see in the SPM.mat file that the actual model does not use a weighting of 0.5 but rather sqrt(0.5):
• Conditions: Cond1, Cond2
• Contrast: [0.5 0.5]
• SPM.mat description: 0.707107*Cond1 + 0.707107*Cond2
This is true when I take the difference between task conditions as well:
• Conditions: Cond1, Cond2
• Contrast: [1 -1]
• SPM.mat description: 0.707107*Cond1 - 0.707107*Cond2
And in case it is relevant, the weighting is perhaps proportional to the number of conditions included in the contrast:
• Conditions: Cond1, Cond2, Cond3, Cond4
• Contrast: [0.25 0.25 0.25 0.25]
• SPM.mat description: 0.5*Cond1 + 0.5*Cond2 + 0.5*Cond3 + 0.5*Cond4
I must be missing something key in gPPI, or perhaps contrasts more generally, and haven't been able to find an answer. Any help would be very much appreciated!
Thank you,
Sara
Threaded View
Title | Author | Date |
---|---|---|
Sara Kim | Jan 20, 2021 | |
Sara Kim | Feb 9, 2021 | |