Conditional (multivariate) Granger causality

Computes conditional Granger causality from source to target given a set of conditioning channels, using the log ratio of the target's residual variance in a VAR fitted without the source (target + conditioning) versus with it (target + source + conditioning); both VARs share the same order (Geweke; Barnett & Seth 2014). Unlike the pairwise eegGrangerCausality(), conditioning removes indirect influences, so for a chain the conditional GC along a purely indirect path is near zero.

Usage

eegConditionalGC(
  x,
  target,
  source,
  conditioning = NULL,
  order = NULL,
  method = "ols",
  assay_name = NULL
)

Arguments

x

A PhysioExperiment object with 2D EEG data (time x channels).

target

Target channel (label or index) - the effect.

source

Source channel (label or index) - the putative cause.

conditioning

Channels to condition on (labels or indices). Defaults to all remaining channels.

order

MVAR model order, or NULL to select automatically on the full channel set.

method

MVAR estimator passed to PhysioCore::mvarFit() (default: "ols").

assay_name

Input assay name (default: the default assay).

Value

A list with the conditional GC value (in nats, non-negative in theory), the target, source, and conditioning labels, and the VAR order used.

References

Barnett, L., & Seth, A. K. (2014). The MVGC multivariate Granger causality toolbox. Journal of Neuroscience Methods, 223, 50-68.

See also

eegGrangerCausality(), eegDTF(), eegPDC()

Examples

if (FALSE) { # \dontrun{
pe <- make_eeg(n_time = 4000, n_channels = 3, sr = 250)
eegConditionalGC(pe, target = 3, source = 1, conditioning = 2)
} # }