Dear Gang,
You wrote: > "I should mention that I turned 3dRegAna into a within-subjects regression by adding 15 vectors, one for each subject. I'm quite impressed by you tricking 3dRegAna into such analysis!"
I should reiterate that it was Doug Ward (as you know, the author of 3dDeconvolve and 3dRegAna) who made this suggestion to me early in the 2000s.
You also wrote: "I wonder that -1, 0, and 1 for the condition coding would work equally well as 1, 2, and 3. "
It turns out that 3dRegAna does not care about the specific numbers that are used to code conditions and seems to care only about the spacing between the numbers. For example, the results are the same when using -1, 0, and 1 as with 1, 2, and 3. Does this mean that our earlier analyses looking for linear trends across the three conditions were actually not equivalent to t-tests between the first and third condition? That is, are our earlier test for linear increases using 3dRegAna still valid?
Thank you for your other suggestions for testing changes across 4 conditions in my other experiment.
Christine