Dear FSL users,
I have a question regarding empty EVs for which I’m hoping someone can provide some suggestions.
I have a dataset in which subjects played an economic trust game with 3 partners. Participants could choose to trust or not trust; decisions to trust a partner could result in the partner reciprocating (positive outcome) or defecting (negative outcome). I am interested in examining the outcome phase of the task for each partner. This design lends itself to some subjects having empty EVs in some runs. For example, while all subjects included in this analysis experience each possible condition at some point during the experiment (i.e., reciprocation and defection with each partner), there are certain runs in which a subject may not experience a given condition (i.e., a partner may not reciprocate in a given run), leading to empty EVs.
If I were to run a contrast of say, reciprocate > 0 [1 0], or reciprocate - defect [1 -1], and include runs of subjects that had empty EVs, I understand that this would lead to empty contrasts, since FSL would not be able to estimate anything for runs in which a subject had not experienced a given condition.
However, in error, I had conducted a weighted contrast (e.g., partner 1 reciprocate > other partners reciprocate; [2 -1 -1]) including some runs of subjects that had empty EVs, and this contrast ran successfully with no errors, and was not empty. Now in theory, this shouldn’t have produced any activation, yet it seems to have done so in a region I would have expected to be active. I know this is likely problematic, and so my question is why FSL was able to estimate something when empty EVs were included if the contrast was weighted, but not otherwise? Is there something I’m missing regarding the math here or is this a possible bug?
Any clarification would be much appreciated.