Hi all,
I've searched the mailing lists, but still have some residual questions about this issue. I'd like to detect regions that show increased activation during error trials versus correct trials. I have three different conditions, and I want to allow for the possibility that the error responses, and the regions involved in processing those responses, are different during these three conditions. Since we're dividing our error trials into three subsets, there are some subjects with no incorrect trials for one of the conditions.
This poses problems with setting up the first-level design, and the higher-level group analysis. I know that I can set up an empty EV for subjects who are missing error trials, but is that equivalent to a one-column custom EV that contains all zeros? Is it necessary to use the empty EV option from the Feat GUI, or can I simply create my own empty EV? (The second option is easily scripted.)
When running the higher-level group analysis, FLAME1+2 should deweight subjects with a very high varcope for a specific condition. Does this effectively ignore a subject with an empty EV, or do I need to set up the higher-level analyses manually excluding subjects with empty EVs in the lower-level? Stating this question differently, what is the consequence of including three subjects with an empty EV, wrt to the resulting statistical significance of the group average of contrasts including these empty EVs? Are these individuals better excluded (with the concession that the sample is no longer representative of the entire population)?
Thanks in advance for your answers,
--Greg
____________________________________________________________________
Greg Burgess, Ph.D.
Research Associate, Institute of Cognitive Science
University of Colorado - Boulder
Email: [log in to unmask]
|