Dear Dr. McLaren,
Thanks for your interpretation. I am also interested in this area and I would like to ask three questions.
The first one: in a 2 (G, between-group) by 2 (C, within-group) design, if the main effect of G or C were found at p < 0.001 (uncorrected) level, while the interaction effect was found at p < 0.01 (uncorrected) level, could I only discuss the main effect without mention the interaction if the significant threshold was set at p < 0.001 (uncorrected)? Could I say that the difference between C1 and C2 were consistent across G1 and G2 at the p < 0.001 (uncorrected) level?
The second question: in a 2 by 2 within-group design, if I found a significant interaction effect (e.g. p < 0.001, uncorrected) at some regions and I would like to do post-hoc analysis to know the details of such interaction effect (i.e. C11-C12 and C21-C22, respectively), could I do two t-test in the inclusive mask defined by the significant interaction effect with some loose threshold (i.e. p < 0.01 than < 0.001)? I ask this is because I found sometimes the significant interaction effect with strict significance threshold appeared near to but not exactly at the regions I was interested, while some looser significance threshold for the interaction effect could involve the ROIs.
The third question: there are two kinds of analysis within ROIs. a) post-hoc t-test within the inclusive mask defined by interaction effect; b) analysis of percent-signal changes (fMRI) or mean/peak amplitude of some electrode (ERP) within ROIs. If the two kinds of analysis give different results, for example, a) significant while b) no significance, how to interpret the results?