We recently (24/08/2012) found that the highpass temporal filtering originally applied to Sim19 and Sim20 was more aggressive than intended. Whereas, for all other simulations, the simulated data was conservatively highpass filtered at a cutoff frequency of 1/200s (to simulate preprocessing of FMRI data), this cutoff had originally been set at a fixed fraction of Nyquist, and hence for Sim19/20 (with the lower TR=0.25s) was unrealistically aggressive, at 1/17s. We have now re-run Sim19/20, with the highpass cutoff now set to the intended 1/200s. We found no significant differences in the bottom-line results, except for somewhat improved performance (in directionality estimation) by Patel's tau and LiNGAM.
Note that the simple two-node tests of Granger described in the Section "Effect of HRF variability and low TR" (in our original NeuroImage paper) did not involve any highpass filtering, and so are unaffected by the above issue.
The original simulation datasets can still be
downloaded here; the datasets linked to
in the main NetSim page contain the corrected
versions of Sim19 and Sim20.