Hi,
 
My study design includes 3 between-subjects factors (2Groups, Age, and questionnaire).
 
I was wondering what would be the right contrast if i want to test Group x Age interaction effects while at the same time controlling for the questionnaire data across subjects.
 
Thanks in advance
 
Best,
Giannis Lois
 
 

Hi Giannis,

If you have the following covariates: Group1 (1/0 indicating subjects in first group), Group2 (1/0 indicating subjects in second group), Age, and Questionnaire, you would need to first define two additional covariates: "Age1" (age of subjects in first group, 0 values for subjects in second group), and "Age2" (age of subjects in second group, 0 values for subjects in first group). Then, in the subject effects list you would select Group1, Group2, Questionnaire, Age1, Age2, and in the between-subjects contrast field enter the contrast [0 0 0 1 -1]. That will implement your desired Group x Age interaction (differences in the association between functional connectivity and age between the two groups) while controlling for the association between functional connectivity and questionnaire (jointly across both groups). 

Hope this helps
Alfonso

Originally posted by Giannis Lois:

Thanks!
 
One more question: Should i use the original age values or the demeaned ones after subtracting the mean age of both Groups?
 
Best,
Giannis

Dear Giannis

For these analyses (group by age interaction) it does not matter whether you center the age covariates or not (you will be getting the exact same results either way). If, on the other hand, you are planning to look at the main group effects (average difference in connectivity between your two groups comparing them at the same fixed age-level), then yes, you should center the two age covariates by subtracting from each the same desired target age-level (e.g. the average age across all your subjects).

Hope this helps
Alfonso


Originally posted by Giannis Lois:

Hi Alfonso,

What is the difference between running two t-tests in conn for a between-subjects analysis (i.e., placebo and tx) with one covariate (e.g., age) (vector = 0 0 1 -1) vs one f-test in conn (vector = [0 0 1 0; 0 0 0 1]) and then making t-contrasts in SPM where/when a significant effect of interaction is observed?

Thanks,
Patrick

Hi Patrick

I am not sure if I am interpreting correctly (please let me know otherwise), the first analysis (selecting placebo, tx, age_placebo, and age_tx and entering a [0 0 1 -1] contrast) is looking at the age by group interaction (different associations between connectivity and age in the two groups), while the second analysis (selecting the same between-subject effects and entering a [0 0 1 0; 0 0 0 1]) is looking at the main effect of age across any of the two groups (an association between connectivity and age in any of the two groups). If you display the latter analysis results in SPM and define there a new contrast ([0 0 1 -1]) the results should be identical to those obtained in the first analysis. One way to potentially increase the sensitivity of these analysis would be, as you describe, to look at the age by group interaction only in those areas where you find a significant main effect in either one of the two groups (in this way you reduce the total search space and minimize the required multiple-comparison correction). You can do that, for example, by taking the latter analysis into SPM, creating there the new interaction contrast [0 0 1 -1] and masking it with your original contrast ([0 0 1 0;0 0 0 1]). Compared to the original interaction test, this latter test will result in exactly the same uncorrected statistics, but the family-wise corrected statistics should be more sensitivity (lower p-values) due to the decreased total search volume.

Hope this helps
Alfonso
Originally posted by Patrick McConnell: