Implicit baseline condition - rest and/or task condition?

Dear experts,

I’m a little confused by what and how to model baseline conditions in fMRI.
What is meant by the authors when it says that one of the task conditions was modelled as implicit baseline condition? Was is not included as regressor in the first level GLM, and if so did they then include the rest condition instead? Some say that one of the task conditions was the baseline but then at second level they compare that condition with another, how would that work?

Let’s say we have a task with three conditions and we want to know whether activation during C1 is higher than C2 relative to the baseline condition C3, how would one proceed ideally?

Thank you in advance for your input and patience with a newbie.


IIUC your question, what people mean when they “take a condition as a baseline” is that they indeed do not model it, so that all effects estimated will take that condition as an implicit baseline. For instance, is c3 is not modeled (and is occurring whenever neither C1 nor C2 is occurring), then the estimated effect of C1, will be the C1-C3 effect, same for C2.
If you’re interested in C1-C2 you should not worry about the baseline, given that C1-C2 is not affected by the baseline choice.


thank you for your reply and sorry for my delay in responding.
If I understand you correctly, when authors say ‘the betas estimated activity versus an implicit baseline C3’ the design matrix would have only included C1 and C2, but not C3 and also not the ISI? The contrast of interest was indeed only C1-C2, if that would not have been affected by the baseline condition, why wouldn’t they have modelled all conditions of the experiment?
Another follow up question: if C3 and ISI are not modelled, can one truly say that C3 is the baseline condition, wouldn’t the baseline be also effected by the ISI?

thank you, for sticking with a newbie =)

The common procedure is not to model ISI. Modeling C3 explicitly is indeed the best thing to do. In practice, it has limited impact on C1-C2 (at least if C1, C2 and C3 do not overlap temporally): It captures a bit of variance corresponding to the difference between the true baseline and C3, which can increase the sensitivity of the statistical tests performed on C1-C2. In my experience, this is a rather small effect.

1 Like