[FieldTrip] Granger causality results out of range

[Michael Boyd (PGR)]

Hi,

We are currently using the granger causality function of the toolbox as described in the connectivity tutorial. 
When we put LFP data through this, the results seem stable (or at least produce results between 0 and 1). However when we locally rereference the LFP data (e.g. to each channel's adjacent channel or current source density) the results are no longer in range (i.e. large negative numbers for granger causality). Our attempts to solve this so far have been in vain, so could anyone suggest where we might look to try to track down this instability?

Script taken from connectivity tutorial:
cfg         = [];
cfg.order   = modelOrder;
cfg.toolbox = 'bsmart';
mdata       = ft_mvaranalysis(cfg, data);

cfg        = [];
cfg.method = 'mvar';
mfreq      = ft_freqanalysis(cfg, mdata);

cfg           = [];
cfg.method    = 'granger';
granger       = ft_connectivityanalysis(cfg, mfreq);

Thanks,
Michael