> Should I then not worry about overfitting? I have some strange looking
> iresp output, and I wanted to rule out that possibility.
Yes, indeed you should be concerned about overfitting, which is the downside when fitting the signal with 'liberal' basis functions such as TENT or CSPLIN due to their little constraint on the hemodynamic responses. Anything in the signal not properly accounted for (e.g., head motion) might lead to overfitting. Assuming you've already tried the 'conservative' approach (GAM or BLOCK), one alternative is to adopt some 'moderate' basis functions like SPMG3 or SPMG2.
Gang