Well, yes all your points are valid, and we have tried to figure out ways to deal with them. We are looking for changes in mental state, and the long periods are necessary to see what we want to see. We did a pilot with an ABAB design and saw some changes which we predicted, but it was not optimal for achieving the state we wanted. The analysis of the individual data with the new design is similar to what we saw before, which gives us some confidence that despite the unusual design we are getting real data. Also, the blocks are counterbalanced,
and we have short 1 minute "transition" periods between the blocks and at the end of the scan in which they look at a flashing checkerboard, which we are using to assess drift effects, etc.(didn't include that in my last posting to keep things simple).

As for using the Deconvolve plugin to plot the fit - I was not sure how to interpret the results. The red line follows the time course fairly well, but not completely. The r2 values for the regressors of interest are generally better than the r2 values for the motion regressors (-4 or -6 vs -2), and the r2 for the full model is better than -10 in many voxels that are activated. Also, when I remove one of the regressors of interest and re-run the plugin, the r2 values for the full model and the other regressors of interest drop considerably (eg r2 of -8 went to -1). Is it correct to interpret this as meaning the data fit my paradigm better than the motion? What exactly should I look for in the Deconvolve results to determine whether the data is any good? As a skeptical reviewer, what would you want to see to convince you that the findings are legit?

Thanks!
Sara