Hi Rick,
Thanks for your response. I generated the padded time series exactly how you suggested and it looks right. However, for my 1st level analysis, the stats still look weird.
Here is my 3dDeconvolve command:
3dDeconvolve \
-input $runs -jobs 24 \
-polort A \
-ortvec MoPar_demean_pad.1D demean \
-ortvec MoPar_derv_pad.1D derv \
-censor censor_pad.1D \
-num_stimts 2 \
-stim_times 1 trials.1D 'TENT(2,6,3)' -stim_label 1 trial \
-stim_times 2 catch_trials.1D 'TENT(2,6,3)' -stim_label 2 catch \
-gltsym 'SYM: trial[1..2]' -glt_label 1 sound \
-overwrite -bucket trial.stats \
-x1D xmat.1D -x1D_uncensored uncensor.xmat.1D
I have checked that the motion parameters are padded properly, they seem to be. The censor file includes all of the padded volumes so that they are not included in the analysis - those seem accurate as well. I decided that TENT would be a better model so that I can choose which TRs to include in my t test. As a reminder, the stim is presented during 2 seconds of silence, followed by 3 2-second TRs.
The stats file has a very pronounced striping throughout the brain. This is not the case in my previous model where I did not mean pad and just inserted a design matrix using -ortvec. I have attached 2 images. This is of a sound>no sound contrast and it is threshold at a ridiculously stringent value (p<3.-39). Please refer to my previous post for an image that looks correct given our hypotheses.