> I want to report the efficiency of two model designs using 3dDeconvolve.
Do the two designs have the same experiment parameters such as number of time points, number of conditions, number of trials per condition, etc.?
> What metrics do I use to compare the models?
Efficiency is tied to a specific test in a model. Suppose that you have three conditions/tasks, and in the end there are a few possible tests such as b1 - b2, b1 - b3, and b2 - b3. The efficiency for b1 - b2 might be different from the one for b1 - b3.
> Do I use the signal +baseline condition number (higher the better) and
> the matrix inverse average error (the lower the better)?
The efficiency might be better characterized through the normalized standard error. See discussion here:
[
afni.nimh.nih.gov]
Gang