Dear Gang,

I am trying to familiarize myself with the SVAR approach and I have a few questions that I was not able to find an answer for in the AFNI site:

1) Is 1dsvar scriptable, even if at the individual subject level only? It is quite a bit of work to re-enter the same specifications over and over interactively for every subject, if you have a large group.

2) In trying 1dsvar at the individual level (200-time points EPI, TR=2.35 s), I gave 1dsvar the following parameters:

- 4 ROIs
- 2nd order trend removal
- 7 confounds (6 motion parameters + global signal)
- Scale timeseries: yes
- VAR model order: 1

And I used the following instantaneous effect specification matrix:

R.IFG R.TPJ R.Hipp vPMC
R.IFG 0 NA 0 0
R.TPJ NA 0 0 NA
R.Hipp 0 0 0 NA
vPMC 0 NA 0 0

Now, if I try the option for the iterative algorithm, 1dsvar crashes out with the following error:

Numerical scheme (0: Iterative; 1: Direct)? 0
Error in solve.default(infgamma) :
system is computationally singular: reciprocal condition number = 2.51723e-21

=> Why do you think this is failing?


On the other hand, if I use the Direct approach, 1dsvar does not crash and gives me the following output:

Instantaneous effect matrix:
R.IFG R.TPJ R.Hipp vPMC
R.IFG 0.0000000 -0.3600128 0 0.0000000
R.TPJ -0.4314677 0.0000000 0 0.1770602
R.Hipp 0.0000000 0.0000000 0 -0.1272101
vPMC 0.0000000 -0.2759719 0 0.0000000

LR overidentification

data: newData
Chi^2 = 1377.982, df = 1, p-value < 2.2e-16

=> but the statistics looks awfully out of range here, and the following T-values matrix also seem to sport humongous values:


T-values for instantaneous effect matrix:
R.IFG R.TPJ R.Hipp vPMC
R.IFG 0 -3600128 0 0
R.TPJ -4314677 0 0 1770602
R.Hipp 0 0 0 -1272101
vPMC 0 -2759719 0 0


[1] "Path coefficient matrix with a lag of 1 (direction goes from row to column):"

R.IFG R.TPJ R.Hipp vPMC
R.IFG 0.15362166 -0.015547828 -0.10018336 -0.14864550
R.TPJ -0.01469110 -0.004125846 0.08608147 -0.07414137
R.Hipp -0.09649103 -0.041054186 0.06071350 0.05501898
vPMC -0.04854562 -0.024198545 0.18920342 0.23005780


Would you have any thoughts as to why I get such values? I can send you the data if necessary.

3) Would it be possible to compare the fit for different instantaneous effects matrices?


Thank you very much in advance,

giuseppe