 |
 |
Dear SPMers, I'm in the process of designing a rapid event-related fMRI experiment, and a few questions have emerged that I would like to address before collecting the data. I apologize for the lengthy description that follows, but I believe it's necessary background info for my questions. The simplest version of the experiment would include two types of "standard" events (let's call them A and B, duration 750 ms, modeled as single events) and two types of "mismatch" events (C,D, duration 750 ms, modeled as single events). In the current version of the design, individual events would be presented in mini blocks of 8-10 repetitions, with SOAs between consecutive events selected randomly between [3.25 4.25 5.25] seconds. The mini blocks would be separated by a 7-second baseline interval. In one type of mini block, events of type A have an 80% probability of occurrence, whereas events of type C/D have a 20% probability of occurrence. The same goes for the other type of mini block, where events of type A are replaced with events of type B. The order of presentations of the two types of mini blocks would be randomized across the acquisition sessions. The contrasts of interest would span the regressors modeling the "mismatch" events: C vs D within mini blocks of type A, C vs D within mini blocks of type B, and the interaction term (C vs D) within A vs. (C vs D) within B. I have simulated a number of design matrices, playing around with the SOAs and the presentation order. When examining the design orthogonality for a design matrix with the above parameters, I get the following correlations between regressors (all remaining correlations are negligible): Type A - (Type C within A) = 0.06 Type A - (Type D within A) = 0.06 Type B - (Type C within B) = 0.04 Type B - (Type D within B) = 0.06 so the regressors are nearly orthogonal. My hypothesis revolves around a difference between mismatch events of type C and D within A, this difference being larger than that between C and D within B. I also expect a difference in the level or response to conditions A vs. B in the regions of interest. Given this, should I be concerned about the (small) correlation left between the regressors of interest? Could significant differences between the "mismatch" events be partly accounted for by differences in the level of response to the context (A or B) in which they occur? Many thanks to those who actually managed to read through the whole thing! Looking forward to your feedback. Best, Giovanni