James,
1) If the number of time points per condition are not many, it probably does make sense to use the pre-detrended data. I can imagine that, if you have a few time points, detrending within 1dSVAR would be problematic because it may remove some BOLD signal.
2) > I get very different results for the LR overidentification test for different subjects. How should I interpret these--i.e.,
> is high x-squared bad and if so how bad is "too bad"? is there some cutoff I should have in mind?
The LR test compares two models, one is the restricted model (e.g., constrained by the instantaneous effect structure) and the other is unrestricted (no instantaneous effects). So the chi-square test is similar to the omnibus F-test in an ANOVA, indicating the significance of the instantaneous structure.
Gang
Edited 1 time(s). Last edit at 05/15/2015 04:16PM by Gang.