Dear Gang,

I understand the difficulty of carrying out a linear test in 3dMEMA.

Overall, we are satisfied with the results of 3dRegAna, but our struggle is that we cannot use all of our subjects because they all do not have a sufficient number of trials to generate a reliable beta for each of the three conditions. The result is that the 3dRegAna analysis has a different number of subjects than the other comparisons in the study. This makes it difficult to compare results across comparisons because one does not know if any differences across comparisons are driven by the different numbers of subjects within each comparison.

My suggestion was to code a 3dDeconvolve vector that is zeros for all TRs, except when conditions 1, 2, or 3 occur. In that case, the vector would code condition 3 as "3", condition 2 as "2", and condition 1 as "1". Then I would convolve this with an HRF and carry out a traditional analysis within 3dDeconvolve. Finally, I would extract the beta and t value for this SINGLE vector for each participant and use them in a 3dMEMA one-group, t-test.

Is there a problem with coding the vector this way, i.e., with 1s, 2s, and 3s? Or do these values need to be mean justified within 3dDeconvolve, resulting in the problematic -1, 0, 1 coding? Would it eliminate this problem if I mean justified the vector AFTER convolving with the HRF?

Christine