I previously believed that CSPLIN was a "drop-in replacement" for TENT, as it says in the documentation. I interpreted this to mean that replacing TENT with CSPLIN would not change the coefficients fit in the model; but would change the way the iresp option generated output: namely, the outputs would be smoother if the output TR was smaller than the spacing of the coefficients. Of course my experience confirms the latter expectation, but I've just noticed that there seem to be some subtle differences in the way the design matrix is constructed. The attached image shows a blown-up graph of some of the regressors from two models that differ only in that the top one uses CSPLIN and the bottom one uses TENT. Notice that the regressor in the CSPLIN model has a pre-undershoot, but is otherwise almost identical to the TENT regressor.

What's up with that? Why isn't it the same? It seems like a very small difference that isn't likely to lead to different results, but I still would like to understand what it is about.

Note that the bottom trace is from a dmBLOCK stimulus, which is why it's fatter.

Thanks!
-dave