Dear experts,

I know this has been brought up before, but it seems like the experimental situation is always a little different, so I thought I'd ask it here. I'm having trouble with missing EVs in specific runs.

I'm trying to analyze an experiment with 3 runs per subject. Each run has 3 EV's (A, B, C) based on post-scanner behavior (memory test). Ultimately I'm interested in the contrast of A>B. I've been performing the traditional 1st Level on each run, then combine at 2nd level for a within-subject FE analysis, and at 3rd level for a between-subjects ME analysis. 

However, a handful of subjects run into the problem where for 1 run they have no trials for one of the EV's (always the same EV = B). So far I've been  excluding those runs, such that a handful of subjects contribute only 2 runs instead of 3 to the second-level FE analysis, each with normal EVs.

However, It still seems strange to loose the data for the run without B, since the information in A (and C) could still be relevant, and since the EVs are based on post-scan response, it seems unfair to penalize the subject who had B's distributed such that none occured during a specific run.

Would it make sense to perhaps pass up COPE images for each single-run EV up to the 2nd level FE - building the contrast there? Essentially could I build an A>B contrast, where A is averaging 3 COPEs representing EV A (from each run), and B is averaging 2 COPEs representing EV B (only from the 2 runs that have it)? Would that work?

It seems to me that that would preserve the most amount of data, but I'm not sure how kosher it is.

I'd appreciate any help you all have.

Thanks,

Jared

Graduate Student

Sleep and Neuroimaging Lab

Department of Psychology
University of California, Berkeley