Hi there,
I posted earlier about a design question and received a helpful response - but wanted to confirm one portion of it (I posted on the original thread, but I think it's getting lost in the mix).
The response to testing an interaction of post-pre differences between two conditions was the following:
"2) then in Setup.Conditions define your four conditions (e.g. pre_treatmentA, pre_treatmentB, post_treatmentA, and post_treatmentA) and associate each with the corresponding functional scan (this association may be common across all subjects or it may be different for different subjects, for example if the order of the two treatments was randomized across subjects).
3) last, when you get to second-level analyses, select AllSubjects in the 'subject effects' list, then select your four conditions in the 'Conditions' list and enter a vector [-1 1 1 -1] in the 'Between-conditions contrast' field. That will evaluate the interaction between your two factors (i.e. (post_treatmentA - pre_treatmentA) - (post_treatmentB - pre_treatmentB) )"
So, just to be clear...
Post-Treatment A = -1
Pre-Treatment A = 1
Post-Treatment B = 1
Pre-Treatment B = -1
Is this correct? I'm confusing myself I think.
Thanks so much!
Right, exactly. This will typically be tested using the default "two-tailed" statistics so the two approaches:
Post-Treatment A = -1
Pre-Treatment A = 1
Post-Treatment B = 1
Pre-Treatment B = -1
vs.
Post-Treatment A = 1
Pre-Treatment A = -1
Post-Treatment B = -1
Pre-Treatment B = 1
would produce exactly the same results.
In general, you can think of any desired contrast in a "functional" form, e.g.
Post-TreatmentA - Pre-TreatmentA > Post-TreatmentB - Pre-TreatmentB
and then simply rearrange the terms to put them all in the same side, e.g.
(Post-TreatmentA - Pre-TreatmentA) - (Post-TreatmentB - Pre-TreatmentB) > 0
and simplifyy that until you get a value multiplying each individual condition, e.g.
+1*Post-TreatmentA -1*Pre-TreatmentA -1*Post-TreatmentB +1*Pre-TreatmentB > 0
and the those values will be the corresponding elements of your contrast vector.
Hope this helps
Alfonso
Originally posted by jrobinson:
Hi there,
I posted earlier about a design question and received a helpful response - but wanted to confirm one portion of it (I posted on the original thread, but I think it's getting lost in the mix).
The response to testing an interaction of post-pre differences between two conditions was the following:
"2) then in Setup.Conditions define your four conditions (e.g. pre_treatmentA, pre_treatmentB, post_treatmentA, and post_treatmentA) and associate each with the corresponding functional scan (this association may be common across all subjects or it may be different for different subjects, for example if the order of the two treatments was randomized across subjects).
3) last, when you get to second-level analyses, select AllSubjects in the 'subject effects' list, then select your four conditions in the 'Conditions' list and enter a vector [-1 1 1 -1] in the 'Between-conditions contrast' field. That will evaluate the interaction between your two factors (i.e. (post_treatmentA - pre_treatmentA) - (post_treatmentB - pre_treatmentB) )"
So, just to be clear...
Post-Treatment A = -1
Pre-Treatment A = 1
Post-Treatment B = 1
Pre-Treatment B = -1
Is this correct? I'm confusing myself I think.
Thanks so much!