We have an issue with our data and we were hoping to seek some advice. We have an event-related paradigm where behavioral responses were classified as either impulsive, restrained, control, or invalid. The invalid trials are either attributed to computer error or participants making incorrect responses on control items. Nonetheless, not all participants have invalid trials to model, so we're having an issue at the 3dDeconvolve stage. We've been trying to figure out the best approach to handle these trials for the participants who do have invalid data.

Our question is, do all participants have to have the same number of input timeseries in 3dDeconvolve? In other words, is it statistically sound to have 4 task regressors for some participants, and only 3 for others? So the only participants who would have the invalid regressor would be those who actually had invalid data.

If using different numbers of regressors isn't appropriate, could you advise another approach that may work better?