Hello AFNIS,

I've been working on an analysis for a group and am having a lot of trouble replicating some published results. Briefly, we have two behavioural measures (A, B) that might amplitude modulate some blocks and we want to see if this is true.

I originally did my analysis with AM2 (looked something like this):

3dDeconvolve \\
    -input {input_data} \\
    -mask {datadir}/ea/{sub}_anat_EPI_mask_MNI.nii.gz \\
    -ortvec {datadir}/ea/{sub}_motion.1D motion_paramaters \\
    -polort 4 \\
    -num_stimts 1 \\
    -stim_times_AM2 1 {datadir}/ea/{sub}_block-times_ea.1D \'dmBLOCK\' \\
    -stim_label 1 empathic_accuracy \\\
    -bucket {datadir}/ea/{sub}_glm_1stlevel.nii.gz \\

And a line from the timing file looks like this:

13.14*0.44,4.23529411765:170.00 234.03*0.67,6.57534246575:146.00 430.35*0.02,6.48648648649:111.00

So in this sample case I have 3 blocks with varying duration, and two married paramaters.

The outputs from AM2 look reasonable to me. However it seems like all of the variance in the data gets eaten up by the not modulated predictor. We don't see much effect in either of the parametically modulated beta weights. The papers we are trying to replicate, I believe, used SPM, and make no mention of explicitly modeling the non-modulated response. My suspicion is that at least some of the effect found in this previous work is actually non-modulated in nature.

To test this, I thought I would use AM1. I was under the impression that this would run the exact same analysis, without including a non-modulated predictor. However when I run this (exact same command, simply using AM1 instead of AM2), I only get 1 set of beta weights! I'm not sure what is going on.

To run this analysis, do I need to run two seperate AM1 stim-times? Or am I missing something more fundamental?

Also, I know what I'm doing is bad -- I just want to see whether this is what stands between us and reproducing the old results... Thanks so much for your help!

Best,
Joseph