I am analyzing MRI data in SPM with two conditions (A and B), each with three possible responses: remembered, not-remembered, and no response. Typically, I’d use six regressors per functional run per participant for these combinations with some variability across subjects depending on their performance.
The complexity arises when considering items learned pre-scan. After an encoding phase, we assess recall (>90% typically) to restrict analysis to learned items. Should non-learned items be a single regressor, or multiple, reflecting condition x response x pre-learning?
Additionally, the main phase also contains a few practice items—previously seen tasks in a practice phase. This adds more potential regressors, complicating tracking and inflating degrees of freedom. These items, while perhaps distinct, should largely involve the same cognitive processes as main items, it’s just that participants are familiar with them.
My dilemma: How can I account for these different subcategories? Is it valid to include the same trial in multiple regressors? For instance, placing a trial in both the unlearned items regressor and a main condition x response regressor (e.g., unlearned item in condition A and recalled).
Options I’m considering:
- Group all non-critical items into a single regressor.
- Model every possible combination of sub-categories separately.
- Model all combinations but also include them in the main regressors.
- Focus on main trials, with additional regressors for practice and unlearned items, overlapping with main regressors.
I hope this was clear. I would be happy to hear what you have to say!