Reaction times in Variable Epoch design

AFNI Message Board
1

Dear Afni experts,
I have a design consisting of responses to stimuli. I have reaction times available for each response. I wanted to use a variable epoch design with a single regressor for responses consisting of variable length blocks: ‘BLOCK(RT)’, where RT = is reaction time at each response.

As it is, 3dDeconvolve admits providing stimulus files of responses at, say, four time points “3, 10, 16, 21”, where the HRF could be modeled by a Gamma Variate or BLOCK(d), all of the same parametric specification. If the reaction times at these time points were “1.2, 2.1, 0.4, 1.1”, is there a way of admitting a single regressor with blocks “BLOCK(1.2), BLOCK(2.1), BLOCK(0.4), BLOCK(1.1)” at times “3, 10, 16, 21”?

Thank you.

Best,
Chintan

2

Hi Chintan,

Use dmBLOCK or dmUBLOCK for that. To do so, marry the
durations to the onset times with a colon, then use
AM1 and specify the preferable basis function. The
timing file might look like:

3:1.2 10:2.1 16:0.4 21:1.1

  • rick
3

Hi Rick,

I have a question related to that. What if there is one (or more than one) missing value (ie, subject did not respond to certain trial)? Should the 1D file be
3:1.2 10:2.1 16 21:1.1 (no reaction time–RT for time point 16)

Also, to make sure I did it right. Should the semi-complete command be:
-stim_times_AM1 1 xxx.1D ‘dmUBLOCK(1)’
in which case I only modulate the duration but not the amplitude, depending on the actual RT

Finally, AM2 seems to be able to internally auto de-mean (subtract mean). For AM1 and RT, does one need to manually de-mean RT or simply use raw RT?

Thanks a lot!
Jerry

4

What if there is one (or more than one) missing value (ie, subject did not respond to certain trial)? Should the 1D file be
3:1.2 10:2.1 16 21:1.1 (no reaction time–RT for time point 16)

No, this is not right. First, the program will not run if the same number of extra values are not “married” to each time value – it will complain.

Second, I don’t think there is an obvious way to impute a value for this missing data. In my opinion, if the subject did not react to a stimulus, then something different happened in that trial in the subject’s brain, and so such trials should be broken out into a separate stimulus class – that is analyzed with a “typical” stimulus response (perhaps ‘BLOCK(d)’ where d=average of actual responses). By putting such stimuli into a separate class, the data from those intervals won’t contaminate the results computed from the “correct” or “typical” trials.