Model durations in event-related fMRI

Dear experts, dear community,

I was wondering what the general opinion is about modelling the durations of your conditions in an event-related experimental design.

From my understanding, you would – traditionally – not model the durations and use stick functions, at least that is what most tutorials/guides say that I found.
But I’ve seen it done differently a couple of times now, and it also seems sensible: you just give the model more information to work with by also modelling the durations.
I haven’t dug into the spm code yet, but I assume the functions used for modelling are the same no matter which duration is given as input, or does spm indeed do something categorically different in the case of a duration of 0?

My results vary considerably depending on whether I modelled the duration.
My experiment has events that are 1-3s long, with a SOA of 1-5s, I worked with SPM for the GLM.

Thanks a lot in advance!

I don’t think there is one size fits all to this question: how you decide to model things depends very much on what the underlying processes are given your task (or more precisely what you think they are). Note also that your assumptions and the model that flows from it may be valid for some regions but not others.

Passing a non zero duration to SPM will mean that it will be convolving (with the HRF) square waves, rather than stick functions.

If you do not have good enough theoretical reason to favor one model over the other, then as you are using SPM then you can turn to the MACS toolbox to compare 2 or more models and see which one is best for each voxel.

You might also be interested in these links:

Thank you both for the suggestions and references!
I am certainly going to try the toolbox and think about our assumptions.