Comments and ControversiesWiener–Granger Causality: A well established methodology
Introduction
For most of its history, neuroscience has primarily been concerned with examining the physiological correlates of experimentally delivered stimuli and overt behavioral responses. More recently, there has been growing interest in studying the effect that one part of the nervous system has on another, either in the absence of identifiable behavioral events or in the context of task performance. Such effects are typically examined by stimulating or lesioning the first part and investigating the outcome in the second. In peripheral and spinal pathways, the interventional techniques of stimulation and ablation have proven to be powerful methods for inferring causal influences from one neuron or neuronal population to another. For the study of causal relations within the brain, interventional techniques also have utility, although that utility is diminished by the high levels of convergence and divergence in brain pathways, as well as the highly nested reciprocity of projections. This article deals with a different approach to the problem of causal influence in the brain, called Time Series Inference (TSI). This approach, although relatively recent in neuroscience, is showing promise as a valuable adjunct to more traditional interventional approaches.
TSI, unlike stimulation or ablation, does not require intervention in the nervous system. It is based on temporal relations existing between time series recordings of neural activity, which may be obtained noninvasively as electroencephalographic (EEG), magnetoencephalographic (MEG), or functional Magnetic Resonance (fMRI) data. Of course, the time series may also be obtained from invasive single-unit, multi-unit, local field potential, or electrocorticographic recordings. TSI depends on the statistical predictability of one time series by another time series. If the two time series represent neural activity from different neurons or neuronal populations, then inference about causal relations between those neurons or neuronal populations is possible. Another advantage of TSI methods is that they naturally accommodate stochastic processes, and thus are well suited to the ubiquitous variability that is found in neural time series data. Furthermore, under proper conditions, TSI methods can be effectively used to relate neural activity to cognitive function.
In this article, we will describe Wiener–Granger Causality (WGC) as the type of TSI most commonly employed in neuroscientific studies. TSI methods, in one form or another, are well established and widely used in many different fields of study. The use of TSI methods in neuroscience is relatively new, but already has appeared in different implementations. For example, the popular Dynamic Causal Modeling (DCM) (Friston et al., 2003) approach is based on TSI, although this fact is not commonly appreciated. We will focus here on WGC techniques, which, as described below, are based on a relatively small set of straightforward assumptions. DCM, which we will not discuss at length, involves TSI with additional and more complicated assumptions (see accompanying articles in this issue).
In summary, our purpose here is to detail the implementation of WGC via AutoRegressive (AR) modelling, discuss the assumptions required to apply WGC to neural time series data, describe some of the limitations of its use, and provide some examples of applications to neural time series data. Finally, we will describe some potentially important extensions of currently employed techniques.
Section snippets
Time-domain WGC
The problem of defining ‘causality’ is non-trivial for complex systems, where, unlike simple systems observed in the every-day world, an intuitive understanding of cause and effect is lacking. In 1956 Norbert Wiener introduced the notion that one variable (or time series) could be called ‘causal’ to another if the ability to predict the second variable is improved by incorporating information about the first (Wiener, 1956). Wiener however lacked a practical implementation of his idea. Such an
Assumptions and challenges
In this section we identify some key challenges for WGC analysis, both in general and with regard to the commonly employed neuroimaging methods of fMRI, EEG, and MEG. In each case we outline recent work that may address, at least in part, these challenges.
Causal inference in neural systems
Any discussion of causal inference should be prefaced by one fundamental question: What do we expect from a measure of causality? One useful perspective is that causal measures in neuroscience should reflect effective connectivity, namely the directed influences that neuronal populations in one brain area exert on those in another (Friston, 1994). As applied to neuroimaging data (e.g., by way of DCM), effective connectivity analysis aims at identifying, from recorded neural time series data,
Acknowledgments
AKS is supported by EPSRC Leadership Fellowship EP/G007543/1 and by a donation from the Dr. Mortimer and Theresa Sackler Foundation. The authors thank Ms. Wei Tang, Dr. Lionel Barnett, Dr. Adam Barrett, Dr. Alard Roebroeck, Dr. Abraham Snyder, and our reviewers for helpful comments.
References (73)
- et al.
The variability of human, BOLD hemodynamic responses
Neuroimage
(1998) - et al.
Topographic organization of nonlinear interdependence in multichannel human EEG
Neuroimage
(2002) - et al.
Frequency decomposition of conditional Granger causality and application to multivariate neural field potential data
J. Neurosci. Methods
(2006) - et al.
BSMART: a Matlab/C toolbox for analysis of multichannel neural time series
Neural Netw.
(2008) - et al.
Dynamic causal modelling: a critical review of the biophysical and statistical foundations
Neuroimage.
(2011) - et al.
Analyzing information flow in brain networks with nonparametric Granger causality
Neuroimage
(2008) - et al.
Testing non-linearity and directedness of interactions between neural groups in the macaque inferotemporal cortex
J. Neurosci. Methods
(1999) - et al.
Dynamic causal modeling
Neuroimage
(2003) Dynamic causal modeling and Granger causality Comments on: the identification of interacting networks in the brain using fMRI: Model selection, causality and deconvolution
Neuroimage.
(2011)- et al.
Lexical influences on speech perception: a Granger causality analysis of MEG and EEG source estimates
Neuroimage
(2008)
Spurious regressions in econometrics
J. Econometrics
Partial Granger causality: eliminating exogenous inputs and latent variables
J. Neurosci. Methods
The use of time-variant EEG Granger causality for inspecting directed interdependencies of neural assemblies
J. Neurosci. Methods
The elusive concept of brain connectivity
Neuroimage
Information transfer in continuous processes
Physica D
Trends in EEG source localization
Electroencephalogr. Clin. Neurophysiol.
Testing the null hypothesis of stationarity against the alternative of a unit root
J. Econometrics
Analysis of dynamic brain imaging data
Biophys. J.
Mapping directed inuence over the brain using Granger causality and fmri
Neuroimage
Combining TMS and fMRI: from ‘virtual lesions’ to functional-network accounts of cognition
Cortex
A MATLAB toolbox for Granger causal connectivity analysis
J. Neurosci. Methods
Testing for nonlinearity: the method of surrogate data
Physica D
Inferring neural activity from bold signals through nonlinear optimization
Neuroimage
Detecting directional influence in fMRI connectivity analysis using PCA based Granger causality
Brain Res.
A new look at the statistical model identification
IEEE Trans. Automatic Control
Radial basis function approaches to nonlinear Granger causality of time series
Phys. Rev. E
Comparison of different cortical connectivity estimators for high-resolution eeg recordings
Hum. Brain Mapp.
Partial directed coherence: a new concept in neural structure determination
Biol. Cybern.
Granger causality and transfer entropy are equivalent for Gaussian variables
Phys. Rev. Lett.
Cortical functional network organization from autoregressive modeling of local field potential oscillations
Stat. Med.
Topdown control of human visual cortex by frontal and parietal cortex in anticipatory visual spatial attention
J. Neurosci.
Beta oscillations in a large-scale sensorimotor cortical network: directional influences revealed by Granger causality
Proc. Natl. Acad. Sci. U. S. A.
Mapping and correction of vascular hemodynamic latency in the BOLD signal
Neuroimage
Identifying neural drivers with functional MRI: an electrophysiological validation
PLoS Biol.
Effect of hemodynamic variability on Granger causality analysis of fMRI
Neuroimage.
Cited by (649)
A recurrent Hopfield network for estimating meso-scale effective connectivity in MEG
2024, Neural NetworksAugmenting Granger Causality through continuous ordinal patterns
2024, Communications in Nonlinear Science and Numerical SimulationDiscovering causal relations and equations from data
2023, Physics Reports