Hi Gang,
I'm still having troubles with SEM, I had put it aside for a couple weeks, but now it has my full attention.
I tried both the rescaling (using 1deval) and censoring (using 1d_tool.py), and I am still getting an output from 1dSEM that requires a threshold of over 1000. This doesn't make sense to me in terms of the mean changing 1000 standard deviations from the mean.
One thing I didn't mention, which I saw came up in another thread. We used a rapid event related design, which you mention is not ideal for SEM. Using the same data set, we have a context-dependent connectivity analysis. Is it the case that CDCA is a more robust test for rapid-event related design? Maybe we shouldn't be using this data set in a SEM analysis?
Cheers,
Steve