Hello everyone,
@ Donald just in case: The script from Wager lab is where we started from, https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;e9d214b.1606 . As stated, I haven't looked at the script since then any more, but the idea is to have one regressor reflecting the average response / fixed "on" period (e.g. based on mean RT) and another capturing the variablity of the "on" periods (e.g. based on trial-specific RTs - mean RT). You can do so by creating a regressor with certain onsets and mean RT and another regressor with the same onsets and trial-specific RT.
Whether to orthogonalize or not is basically another issue. You could also just subtract the time course of the RT-derived regressor from that of the fixed-duration regressor.
With regard to whether to just modulate the amplitude via a PM regressor as provided in the GUI, or whether to go with a regressor taking into accout the length of the "on" periods, it's worth looking at microtime settings, as brought up at https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;f303e94.1503 . With the default settings and e.g. a repetition time of 2 s the temporal resolution of the final stimulus function = onsets and "on" periods is in a resolution of 2 s / 16 = 0.125 s, thus you discretize to possibly just a few steps. This is not the case with a parametric modulator working on the amplitude only. This can be solved by adjusting the microtime settings though.
This is insofar interesting as a regressor constructed from variable "on" periods (but not a PM regressor) has the same "amplitude units" as the fixed "on" period regressor, which can be useful when contrasting these two - if it's really just one neural process that varies in time, then the estimates should be identical for these two (leaving aside possible issues due to orthogonalization). If the estimate for the variable-duration regressor is larger (smaller) than that of the fixed-duration regressor then the modulation by duration is larger (smaller) than *what one would expect based on the durations*. I'm not aware of any paper that has tested this, but it might be interesting in some instances.
Best
Helmut
|