> I wonder if testing for a linear test using a conjunction of 2 t-tests is a more strict test
> than testing for a linear trend in 3dRegAna? For example, is the 3dMEMA conjunction
> analysis method only identifying the largest linear increases, whereas the 3dRegAna
> method has a less strict criterion?
Yes the conjunction approach is indeed very restrictive from the following perspective: It requires that both contrasts be statistically significant as well as monotonically ascending or descending. On the other hand, the 3dRegAna approach only needs at least one contrast being signifiant. In addition, it does not require both contrasts being in the directionality. For example, a linear trend can be established in the 3dRegAna with the following scenario: condition 2 > condition 1, condition 3 < condition 2 but condition 3 > condition 1.
3dMEMA is better in accounting for varying reliability among the effect estimates, but now I feel that conjunction analysis does not seem to be a good idea. Do you see more clusters in each of the two contrasts from 3dMEMA? I don't know if this is feasible or reasonable: find all those clusters (both positive and negative) from each of the two contrasts (those clusters don't have to be statistically significant for both contrasts), and identify the trend based on their relative magnitudes. See if this would reconcile with the 3dRegAna approach. Keep in mind the scenario mentioned above would be considered as a linear trend: condition 2 > condition 1, condition 3 < condition 2 but condition 3 > condition 1.
Gang
Edited 1 time(s). Last edit at 04/10/2014 04:48PM by Gang.