Hi Gang, hi all,

We're starting another study that's very similar to the one described here, and we're considering jittering the rating stimulus a little this time. Because we want to be sure that the subject's rating reflects his/her experience of the painful stimulus and not the memory of that experience, we would probably not want to jitter the timing by more than is absolutely necessary, and we would prefer not to use any jitter at all unless it will make a big difference in the analysis. I've been playing around with 3dDeconvolve -nodata, trying to get a feeling for the effect of the jitter we might introduce, and I'm not sure how to interpret the sample standard deviations I'm getting. I've tried modelling the 20s-long thermal stimulus with BLOCK and with CSPLIN regressors, as well as an EXPR regressor that looked like a sawtooth (i.e., "/|"-shaped), and I tried both BLOCK and CSPLIN for the rating stimulus. (There's also a countdown stimulus 3s before the run begins, which I modelled with BLOCK). The only thing that I'm absolutely certain of is that if we want to use CSPLIN regressors for both the thermal and rating stimuli, then we must jitter the rating period in order to avoid collinearity in the design matrix. Short of that, though, I don't really know what normalized standard deviation I would want to aim for in the design matrix as output from 3dDeconvolve -nodata, and I don't know what sorts of cases would be most valuable to test if I were going to feed 3dDeconvolve some made-up data. Do you have any advice for me?

Isaac Schwabacher