Professional Documents
Culture Documents
STATISTICS, NEUROSCIENCE
Correction for “A study of problems encountered in Granger
causality analysis from a neuroscience perspective,” by Patrick A.
Stokes and Patrick L. Purdon, which was first published August
4, 2017; 10.1073/pnas.1704663114 (Proc Natl Acad Sci USA 114:
E7063–E7072).
The authors wish to note the following: “In the Discussion
section of our paper, we misunderstood the method described in
Barnett and Seth (25) to compute the conditional Granger cau-
sality. We realize now that the spectral factorization method they
describe can be used to obtain the conditional Granger causality
with a single model fit, which would avoid the computational
problems associated with separate model fits. We apologize for
this error.”
25. Barnett L, Seth AK (2014) The MVGC multivariate Granger causality toolbox: A new
approach to Granger-causal inference. J Neurosci Methods 223:50–68.
www.pnas.org/cgi/doi/10.1073/pnas.1809324115
Edited by Apostolos P. Georgopoulos, University of Minnesota Medical School, Minneapolis, MN, and accepted by Editorial Board Member Thomas D.
Albright June 26, 2017 (received for review March 29, 2017)
Granger causality methods were developed to analyze the flow tional causality measure (22) suggested a means to assess direct
of information between time series. These methods have become influences within a larger network. Hence, the Granger–Geweke
more widely applied in neuroscience. Frequency-domain causal- approach seemed to offer neuroscientists precisely what they
ity measures, such as those of Geweke, as well as multivariate wanted—an assessment of direct, frequency-dependent, func-
methods, have particular appeal in neuroscience due to the preva- tional influence between time series—in a straightforward exten-
lence of oscillatory phenomena and highly multivariate experi- sion of standard time series techniques. Current, widely applied
mental recordings. Despite its widespread application in many tools for analyzing neuroscience data are based on this Granger–
fields, there are ongoing concerns regarding the applicability of Geweke (GG) paradigm (23–25).
Granger causality methods in neuroscience. When are these meth- Many limitations of the GG approach are well known, includ-
ods appropriate? How reliably do they recover the system struc- ing the requirements that the generative system be approxi-
ture underlying the observed data? What do frequency-domain mately linear, stationary, and time invariant. Analyses of the
causality measures tell us about the functional properties of oscil- GG approach applied to more general processes, such as neural
latory neural systems? In this paper, we analyze fundamental spiking data (2, 3), continuous-time processes (26), and systems
properties of Granger–Geweke (GG) causality, both computational with exogenous inputs and latent variables (4, 27), have been
and conceptual. Specifically, we show that (i) GG causality esti- shown to produce results inconsistent with the known functional
mates can be either severely biased or of high variance, both lead- structure of the underlying system. These examples illustrate the
ing to spurious results; (ii) even if estimated correctly, GG causal- perils of applying GG causality in situations where the genera-
ity estimates alone are not interpretable without examining the tive system is poorly approximated by the vector autoregressive
component behaviors of the system model; and (iii) GG causal- (VAR) model class. Other problems, such as negative causality
ity ignores critical components of a system’s dynamics. Based on estimates (28), have been observed even when the generative
this analysis, we find that the notion of causality quantified is system belongs to the finite-order VAR model class. Together,
incompatible with the objectives of many neuroscience investiga- these problems raise several important questions for neuroscien-
tions, leading to highly counterintuitive and potentially mislead- tists interested in using GG methods: Under what circumstances
ing results. Through the analysis of these problems, we provide are these methods appropriate? How reliably do these methods
important conceptual clarification of GG causality, with implica- recover the functional structure underlying the observed data?
tions for other related causality approaches and for the role of And what do frequency-domain causality measures tell us about
causality analyses in neuroscience as a whole. the functional properties of oscillatory neural systems?
In this paper, we perform an analysis of GG causality meth-
Granger causality | time series analysis | neural oscillations | ods to help address these questions. We show how, due to the
connectivity | system identification
Significance
STATISTICS
However, there are ongoing concerns regarding its applicabil-
investigated functional neural systems at scales of organization ity in neuroscience. When are these methods appropriate? How
from the cellular level (1–3) to whole-brain network activity (4), reliably do they recover the functional structure of the sys-
under a range of conditions, including sensory stimuli (5–7), tem? Also, what do they tell us about oscillations in neural
varying levels of consciousness (8–10), cognitive tasks (11), and systems? In this paper, we analyze fundamental properties of
pathological states (12, 13). In such analyses, the time series data Granger causality and illustrate statistical and conceptual prob-
are interpreted to reflect neural activity from a particular source,
lems that make Granger causality difficult to apply and inter-
and Granger causality is used to characterize the directionality,
pret in neuroscience studies. This work provides important con-
NEUROSCIENCE
STATISTICS
−r120 0 x1,t−2
#
+ 0.7136 −r22 0 x2,t−2
Hence, the conditional frequency-domain causality is 0 0.5 −r32 x3,t−2
| {z }
Sw r (i) ,w r (i) (λ) A2
j j
fxi −→xj |xk (λ) = ln 0
"
0 0
#"
x1,t−3
# "
w1,t
#
Gj ,j (λ) Σfj ,j Gj∗,j (λ) + −0.356 0 0 x2,t−3 + w2,t ,
0 −0.3098 0 x3,t−3 w3,t
NEUROSCIENCE
r (i)
Σj ,j | {z }
= ln . A3
Gj ,j (λ) Σfj ,j Gj∗,j (λ) where r1 = 0.9, r2 = 0.7, and r3 = 0.8; θi = fi ∆t2π for i = 1, 2, 3;
1
r (i) and the assumed sampling rate is ∆t = 120 Hz. See S5. Construc-
The last equality follows from the assumed whiteness of {wj ,t }
r (i)
tion of Simulated Examples for the selection of parameter values
with covariance Σj ,j . for this example.
This equation is similar in form to the expression for uncon- We generated 1,000 realizations of this system, each of length
ditional frequency-domain causality in Eq. 2, suggesting, at least N = 500 time points (4.17 s). These simulated data have oscil-
superficially, an analogous interpretation. However, the transfor- lations at the intended frequencies that are readily visible in the
mations detailed in Eq. 3 suggest a more nuanced interpretation. time domain, as shown in Fig. 1. The conditional frequency-
domain causality was computed for each realization, using sepa-
Overview of Analysis. In the next section, we explore several rate full- and reduced-model estimates using the true model order
issues that arise in the application of the conditional and uncon- P = 3, as well as model orders 6 and 20. The VAR parameters for
0
0 estimates.
−10 −10 The bias in the conditional GG causality distribution under the
0 50 0 1 2 true model order occurs because, whereas GG causality is the-
Frequency (Hz) Time (sec) oretically defined using infinite histories, practical computation
requires using a finite order. Unfortunately, the reduced models
Fig. 1. VAR(3) three-node series system of example 1. Top shows the net- are being used to represent a subset of a VAR process, and sub-
work diagram of the system, with plots showing the frequency response of sets of VAR processes are not generally VAR, but instead belong
the internal dynamics (univariate AR behavior) of each node and the fre-
to the broader class of vector autoregressive MA (VARMA) pro-
quency responses of the channels connecting the nodes. Bottom shows the
cesses. VAR representations of VARMA processes are generally
spectrum of each node and a corresponding time series of a single realiza-
tion from the system. The high signal-to-noise ratio of this system and these
infinite order. Thus, when a finite-order VAR model is used for
data is evident in the highly visible oscillations in the time series and in the
the reduced process, as required in conditional Geweke causality
spectral peaks. computation, some terms will be omitted.
This can be seen more clearly by rewriting the full VAR model,
xj ,t Aj ,j (p) Aj ,i (p) Aj ,k (p) xj ,t−p wj ,t
" # P
" #" # " #
X
each model order were well estimated from single realizations of xi,t = Ai,j (p) Ai,i (p) Ai,k (p) xi,t−p + wi,t .
the process (Table S1). We also performed model order selec- xk ,t p=1 Ak ,j (p) Ak ,i (p) Ak ,k (p) xk ,t−p wk ,t
tion using a variety of methods—including the Akaike informa-
tion, Hannan–Quinn, and Schwarz criteria—all of which correctly Focusing on the reduced components, we see the VARMA form,
identified, on average, the true model order to be P = 3 (Fig. S1). XP
Fig. 2 shows the distributions of causality estimates for differ- xj ,t Aj ,j (p) Aj ,k (p) xj ,t−p
ent model orders for the truly causal connections from node 1 x = A (p) A (p) x +
k ,t k ,j k ,k k ,t−p
p=1
to node 2 and from node 2 to node 3, as well as the truly non- | {z }
causal connection from node 3 to node 1. Results for the true Autoregressive Term
model order 3 are shown in Fig. 2, Column 1, whereas those for X P
the order 6 and 20 models are shown in Fig. 2, Columns 2 and 3, wj ,t Aj ,i (p)
respectively. Each plot shows the median (blue line) and 90% wk ,t + Ak ,i (p) xi,t−p .
central interval (purple shading) of the distribution of causal- p=1
ity estimates. In addition, we show estimates for a single real-
| {z }
Moving-Average Term
ization (red line) and the 95% significance level (black dashes)
estimated by permutation (32). The details for the significance For the reduced model, the MA terms, specifically those contain-
level computation are described in S9. Permutation Significance ing xi , need to be appropriately projected onto the typically infi-
for Noncausality. Fig. 2, Column 4 shows the mean causality val- nite histories of the remaining xj and xk components. Truncation
ues for each order. The causality estimates for all cause–effect of the reduced VAR to finite order eliminates nonzero, although
pairs for model orders 3, 6, and 20 are shown in Figs. S2, S3, and diminishing, terms. Thus, when using a finite order, especially the
S4, respectively. true model order for the full system, the reduced model com-
The differences in these causality distributions reflect the ponents can be poorly represented, leading to biased causality
trade-off in bias and variance that would be expected with estimates.
increasing model order. Because the GG causality measure is The stochastic nature of the separate fits can further con-
defined for infinite-order VAR models, high-order VAR mod- tribute to artifactual peaks and valleys and even negative causal-
els are approximately unbiased. The similarity in the shape of ity (28). This is because when the full and reduced models
are estimated separately, their frequency structure can be mis- examine a bivariate case, such that the unconditional causality
aligned, producing spurious fluctuations in the resulting causal- can be computed correctly from a single model using the true
ity estimate. As seen in Fig. 2, negative values are extensive for VAR parameters. By using the true VAR parameters, we also
the truly noncausal connection from node 3 to node 1 and even eliminate the uncertainties associated with VAR parameter esti-
appear at model order 20 for the truly causal connection from mation. Thus, in this example, we focus solely on the functional
node 2 to node 3. This sensitivity to variability in the model relationship between system structure and the causality measure.
parameter estimates can be quite dramatic, depending on the We compare the causality of a set of similar unidirectional
specific system, and compounds the increased variability due to two-node systems, all with the structure shown in Fig. 3, Left.
high model order discussed above. To isolate the influence of transmitter and receiver behaviors,
we fix the channel as all pass (i.e., all frequencies pass through
Causality Interpretation: How Are the Functional Properties of the the channel without influencing the signal amplitude). In these
System Reflected in GG Causality? Given a frequency-domain systems, node 1, {x1,t }, is driven by a white noise input, {w1,t };
causality spectrum over a range of frequencies, a face value inter- resonates at frequency f1 ; and is transmitted to node 2. Node 2,
pretation would be that peaks represent frequency bands that {x2,t }, is driven by both node 1 and a white noise input, {w2,t },
are more causal or effective, whereas valleys represent frequency and resonates at frequency f2 . The channel is fixed as all-pass
bands of lesser effect. At the same time, we know that causal- unit delay, but the transmitter and receiver resonance frequen-
ity measures are functions of the VAR system parameters and cies are varied. The set of transmitter resonance frequencies is
reflect a combination of dynamics from the different components f1 ∈ {10, 20, 30, 40, 50} Hz. The set of receiver resonance fre-
of the system. How then do causality values relate to the under- quencies is f2 ∈ {10, 30, 50} Hz. The VAR equation of these
STATISTICS
lying structure or dynamics of the generative system? systems is thus
In this subsection, we examine the relationship between sys-
tem behavior and causality values by systematically varying the x1,t 2r1 cos (θ1 ) 0 x1,t−1
x2,t = 1 2r2 cos (θ2 ) x2,t−1 +
parameters of different parts of the system and assessing the
dependence of the causality on the component behaviors. We
2
−r1 0 x1,t−2 w1,t
use the terms “transmitter” to denote putatively causal nodes, + w2,t ,
0 −r22 x2,t−2
“receiver” to denote putative effect nodes, and “channel” to
denote the linear relationship linking the two in a particular where ri = 0.65 and θi = fi ∆t2π for i = 1, 2, and the assumed
NEUROSCIENCE
1
direction of connection. We will show that GG causality reflects sampling rate is ∆t = 120 Hz. The driving inputs are inde-
only the dynamics of the transmitter node and channel, with no pendent white Gaussian noises with zero mean and unit vari-
dependence on the dynamics of the receiver node. This suggests ance. The frequency-domain causality is computed for each sys-
that, even in simple bivariate AR systems, GG causality may tem using the single-model unconditional approach with the
be prone to misinterpretation and may not reflect the intuitive true parameter values. Fig. 3, Right shows the true frequency-
notions of causality most often associated with these methods. domain causality values for the different systems. Each plot
Example 2: Transmitter–receiver pair with varying resonance in Fig. 3, Right represents a single receiver resonance fre-
alignments. In this simulation study, we investigate how GG quency, with the different transmitter frequencies. For instance,
causality values change as the system’s frequency structure varies, Fig. 3, Top Right shows the values for all five transmit-
to determine whether the changes in the causality values agree ter settings with the receiver resonance set at 10 Hz. Simi-
with the face value interpretation of causality described above. larly, Fig. 3, Middle Right and Bottom Right shows the values
We structure the analysis to avoid the model order and model with the receiver resonance held fixed at 30 Hz and 50 Hz,
subset issues described in the previous section. Specifically, we respectively.
3
10, 20, 30, 40, 10 30 50 30
and 50 Hz 30 Hz
1 time-step
1 2 2
20 30 40 30
Spectrum 1 (dB) Spectrum 2 (dB) 30 30
15 12
10 8 1
4
0
0
−10 −4 0
0 20 40 60 0 20 40 60 0 20 40 60
3 Fig. 3. System and unconditional frequency-
10, 20, 30, 40, 10 50 50 50 domain causality values for different resonance
and 50 Hz 50 Hz alignments of the VAR(2) unidirectional trans-
1 time-step mitter–receiver pair of example 2. The channel is
1 2 2 all-pass unit delay. Three receiver resonances are
20 50 40 50 tested: 10 Hz, 30 Hz, and 50 Hz, shown in the
Spectrum 1 (dB) Spectrum 2 (dB) 30 50 Top, Middle, and Bottom rows, respectively. For
15 25 each setting of the receiver resonance, five trans-
10 20 mitter resonances are tested: 10 Hz, 20 Hz, 30 Hz,
1
10 40 Hz, and 50 Hz. Left column shows the spec-
0 tra. Right column shows the causality. As discussed
0 in the text, the causality reflects the spectrum of
−10 −10 the transmitter and channel, independent of the
0
0 20 40 60 0 20 40 60 0 20 40 60 receiver dynamics. Hence, all three plots of the
Right column are identical and similar to the node
Frequency (Hz) Frequency (Hz) Frequency (Hz) 1 spectrum plots of the Left column.
STATISTICS
PP p
det I − p=1 Ai,i (p) z are outside the unit circle—which is GG causality, in terms of both its statistical properties and its
conceptual interpretation. These problems have important prac-
the case for the systems in example 2. Similarly, in the conditional tical consequences for neuroscience investigations using GG
case (22), causality and related methods. Moreover, as we discuss below,
1
Z π these problems can be avoided in large part by paying more care-
fx −→xj |xk (λ)d λ ≤ Fxi −→xj |xk , [7] ful attention to the modeling and system identification steps that
2π −π i are fundamental to data analysis (Fig. 4).
NEUROSCIENCE
P
∞ p
with equality iff the roots of det p=1 G (p) z are inside Alternative Approaches for Estimating GG Causality. The potential
the unit circle, where G is the system function of Eq. 4. Hence, for negative causality estimates and spurious peaks in frequency-
the frequency-domain causality is a decomposition of the time- domain GG causality has been recognized previously (28). The
domain causality. problem has been framed primarily as one in which conditional
Because the time-domain conditional causality is indepen- causality estimates are sensitive to mismatches between the sepa-
dent of the receiver node, the relationship expressed in Eq. 7 rate full and reduced models; i.e., spurious values occur when the
would, at a minimum, impose strict constraints on the form spectral features of the full and reduced models do not “line up”
of any receiver dependence in the frequency-domain condi- properly. Our work identifies the bias–variance trade-off in the
tional causality. Example 1 from the previous subsection qual- VAR model order as a more fundamental difficulty. Model order
itatively agrees with this (Fig. 2, Column 3). We see that selection is problematic even in the context of single-model para-
receiver dynamics do not appear to be reflected in the con- metric spectral estimation. AR models have the property that
ditional frequency-domain causality. In the causality estimates as model order is increased, the number of potential oscillatory
from node 1 to node 2 (Fig. 2, Top), the 10-Hz receiver reso- modes increases, which can alter estimated peak locations. This
interpreting the form or dynamics of such connections. In such question (Fig. 4), including the processes of model formulation
cases, under a VAR model, the direct connection between nodes and model selection. Although these steps can be challenging and
is represented by the corresponding cross-components of the complex, they are essentially a quantitative formalization of the
VAR parameter matrices, which indicate the absence of a con- scientific method. Unfortunately, causality methods have been
nection if all of these parameters are equal to zero. Hypothesis used, in effect, to circumvent many of these painstaking steps (Fig.
tests for such connections have been described in the time series 4, annotation 1). As we have seen in our analysis, this can mask
analysis literature (30, 46). This approach is analogous to that pro- many potential points of failure, including poor representation
posed by Kim et al. (1) for point process models of neural spik- of the system by the model, and computational and interpreta-
ing data. To our knowledge, the VAR versions (30, 46) of this tional problems with statistics computed from the model. Perhaps
approach have not been used in neuroscience applications, but more importantly, the use of causality methods supplants a clear
may offer potential advantages in computational and statistical statement of the scientific question of interest with an ambiguous
efficiency compared with more widely applied permutation-based statistic and a loaded concept that carries philosophical connota-
tests on causality measures, such as those outlined in ref. 25. tions, detracting from the question at hand.
A number of causality approaches have been developed that In many cases, the scientific question of interest is simply to
purportedly handle a wider range of generative dynamic systems, determine the presence or absence of a direct connection from
beyond those that might be well approximated by VAR models. one region to another. This question can often be answered
Transfer entropy methods (31, 47, 48) can be viewed as a gener- directly from the estimated model. In the case of VAR mod-
alization of the Granger approach and are applicable to nonlin- els, as discussed earlier, a hypothesis test could be performed
STATISTICS
ear systems. However, such methods require significant amounts on the off-diagonal VAR components (30). In the transfer func-
of data, are computationally intensive, and present their own tion case, this is given by the unnormalized PDC. In DCM, this is
model identification issues, such as selection of the embedding explicitly represented in the selected model’s connections (49).
order. These approaches are strictly time-domain measures, and In other cases, the scientific question centers on characterizing
their similarity to GG causality (31) suggests that they would face the mechanistic interactions between components of a system.
similar problems with interpretation such as independence of As we have argued, this can be ascertained by examining the
receiver dynamics. Dynamic causal modeling (DCM) uses neuro- component dynamics from the estimated model. For instance,
physiologically plausible dynamic models to characterize causal in the analysis of electrocorticographic oscillations in nonhu-
NEUROSCIENCE
relationships (49). The models, estimation methods, and notions man primate somatosensory cortex described in ref. 5, one could
of causality involved in DCM are different from the time series- examine the component node and channel dynamics to charac-
based causality approaches discussed here. The interpretation terize more specific interactions between system components.
and statistical properties of DCM are the subject of ongoing Yet in other cases, the scientific question may relate to pre-
analysis (50–53). dicting the effects that structural changes might have on system
dynamics. For instance, Epstein et al. (38) wanted to evaluate
Causality Analysis in the Framework of System Identification. The whether surgical resections could alleviate seizures. This question
problems in causality methods that we have discussed arise when could also be investigated from the estimated model, by removing
these methods are applied without clear reference to the underly- connections or nodes, simulating data using the model, and then
ing models, their estimation methods and statistical measures, or evaluating the influence of these interventions on the generation
the scientific question of interest—that is, without clear reference of seizure oscillations. The validity of such statements would be
to the process of system identification (54, 55). System identifi- incumbent on the applicability of the model and would require
cation comprises the set of steps undertaken to study a scientific experimental validation. However, by structuring the problem in
1. Kim S, Putrino D, Ghosh S, Brown EN (2011) A Granger causality measure for point 28. Chen Y, Bressler SL, Ding M (2006) Frequency decomposition of conditional Granger
process models of ensemble neural spiking activity. PLoS Comput Biol 7:e1001110. causality and application to multivariate neural field potential data. J Neurosci Meth-
2. Kispersky T, Gutierrez GJ, Marder E (2011) Functional connectivity in a rhythmic ods 150:228–237.
inhibitory circuit using Granger causality. Neural Syst Circuits 1:9. 29. Granger CWJ (1969) Investigation causal relations by econometric models and cross-
3. Gerhard F, et al. (2013) Successful reconstruction of a physiological circuit with known spectral methods. Econometrica 37:424–438.
connectivity from spiking activity alone. PLoS Comput Biol 9:e1003138. 30. Lütkepohl H (2005) New Introduction to Multiple Time Series Analysis (Springer,
4. Eichler M (2005) A graphical approach for evaluating effective connectivity in neural Berlin).
systems. Philos Trans R Soc Lond B Biol Sci 360:953–967. 31. Barnett L, Barrett AB, Seth AK (2009) Granger causality and transfer entropy are
5. Brovelli A, et al. (2004) Beta oscillations in a large-scale sensorimotor cortical net- equivalent for Gaussian variables. Phys Rev Lett 103:238701.
work: Directional influences revealed by Granger causality. Proc Natl Acad Sci USA 32. Blair RC, Karniski W (1993) An alternative method for significance testing of wave-
101:9849–9854. form difference potentials. Psychophysiology 30:518–524.
6. Roebroeck A, Formisano E, Goebel R (2005) Mapping directed influence over the 33. Kay SM (1988) Modern spectral estimation: Theory and application (Prentice Hall,
brain using Granger causality and fMRI. Neuroimage 25:230–242. Englewood Cliffs, NJ).
7. Wang X, Chen Y, Bressler SL, Ding M (2007) Granger causality between multiple inter- 34. Stokes PA (2015) Fundamental problems in Granger causality analysis of neuroscience
dependent neurobiological time series: Blockwise versus pairwise methods. Int J Neu- data. PhD thesis (Massachusetts Institute of Technology, Cambridge, MA).
ral Syst 17:71–78. 35. Hu S, Liang H (2012) Causality analysis of neural connectivity: New tool and limitations
8. Baccalá LA, Sameshima K, Ballester G, do Valle AC, Timo-Iaria C (1998) Studying the of spectral Granger causality. Neurocomputing 76:44–47.
interaction between brain structures via directed coherence and Granger causality. 36. Hughes SW, Crunelli V (2005) Thalamic mechanisms of EEG alpha rhythms and their
Appl Signal Process 5:40–48. pathological implications. Neuroscientist 11:357–372.
9. Barrett AB, et al. (2012) Granger causality analysis of steady-state electroencephalo- 37. Wagner AD, Koutstaal W, Maril A, Schacter DL, Buckner RL (2000) Task-specific repe-
graphic signals during propofol-induced anaesthesia. PLoS One 7:e29072. tition priming in left inferior prefrontal cortex. Cereb Cortex 10:1176–1184.
10. Seth AK, Barrett AB, Barnett L (2011) Causal density and integrated information as 38. Epstein CM, Adhikari BM, Gross R, Willie J, Dhamala M (2014) Application of high-
measures of conscious level. Philos Trans R Math Phys Eng Sci 369:3748–3767. frequency Granger causality to analysis of epileptic seizures and surgical decision
11. Graham S, et al. (2009) Role of medial cortical, hippocampal and striatal interactions making. Epilepsia 55:2038–2047.
during cognitive set-shifting. Neuroimage 45:1359–1367. 39. Kamiński MJ, Blinowska KJ (1991) A new method of the description of the informa-
12. Cadotte AJ, et al. (2010) Granger causality relationships between local field poten- tion flow in the brain structures. Biol Cybern 65:203–210.
tials in an animal model of temporal lobe epilepsy. J Neurosci Methods 189:121– 40. Kamiński M, Ding M, Truccolo WA, Bressler SL (2001) Evaluating causal relations in
129. neural systems: Granger causality, directed transfer function and statistical assessment
13. Wang S, et al. (2007) Revealing the dynamic causal interdependence between neural of significance. Biol Cybern 85:145–157.
and muscular signals in Parkinsonian tremor. J Franklin Inst 344:180–195. 41. Baccalá LA, Sameshima K (2001) Partial directed coherence: A new concept in neural
14. Kopell NJ, Gritton HJ, Whittington MA, Kramer MA (2014) Beyond the connectome: structure determination. Biol Cybern 84:463–474.
The dynome. Neuron 83:1319–1328. 42. Korzeniewska A, Manczak M, Kamiński M, Blinowska KJ, Kasicki S (2003) Determi-
15. Cannon J, et al. (2014) Neurosystems: Brain rhythms and cognitive processing. Eur J nation of information flow direction among brain structures by a modified directed
Neurosci 39:705–719. transfer function (dDTF) method. J Neurosci Methods 125:195–207.
16. Buzsáki G, Wang XJ (2012) Mechanisms of gamma oscillations. Annu Rev Neurosci 43. Korzeniewska A, Crainiceanu CM, Kus R, Franaszczuk PJ, Crone NE (2008) Dynamics
35:203–225. of event-related causality in brain electrical activity. Hum Brain Mapp 29:1170–1192.
17. Buzsáki G, Watson BO (2012) Brain rhythms and neural syntax: Implications for effi- 44. Astolfi L, et al. (2007) Comparison of different cortical connectivity estimators for
cient coding of cognitive content and neuropsychiatric disease. Dialogues Clin Neu- high-resolution EEG recordings. Hum Brain Mapp 28:143–157.
rosci 14:345–367. 45. Florin E, Gross J, Pfeifer J, Fink GR, Timmermann L (2011) Reliability of multivariate
18. Olbrich E, Achermann P (2005) Analysis of oscillatory patterns in the human sleep EEG causality measures for neural data. J Neurosci Methods 198:344–358.
using a novel detection algorithm. J Sleep Res 14:337–346. 46. Dufour JM, Renault E (1998) Short run and long run causality in time series: Theory.
19. Anderson KL, Rajagovindan R, Ghacibeh GA, Meador KJ, Ding M (2010) Theta oscil- Econometrica 66:1099–1125.
lations mediate interaction between prefrontal cortex and medial temporal lobe in 47. Schreiber T (2000) Measuring information transfer. Phys Rev Lett 85:461–464.
human memory. Cereb Cortex 20:1604–1612. 48. Quinn CJ, Coleman TP, Kiyavash N, Hatsopoulos NG (2011) Estimating the directed
20. Bollimunta A, Mo J, Schroeder CE, Ding M (2011) Neuronal mechanisms and information to infer causal relationships in ensemble neural spike train recordings.
attentional modulation of corticothalamic alpha oscillations. J Neurosci 31:4935– J Comput Neurosci 30:17–44.
4943. 49. Friston KJ, Harrison L, Penny W (2003) Dynamic causal modelling. Neuroimage
21. Geweke J (1982) Measurement of linear dependence and feedback between multiple 19:1273–1302.
time series. J Am Stat Assoc 77:304–313. 50. Daunizeau J, David O, Stephan KE (2011) Dynamic causal modelling: A critical review
22. Geweke JF (1984) Measures of conditional linear dependence and feedback between of the biophysical and statistical foundations. Neuroimage 58:312–322.
time series. J Am Stat Assoc 79:907–915. 51. Lohmann G, Erfurth K, Müller K, Turner R (2012) Critical comments on dynamic causal
23. Bressler SL, Seth AK (2011) Wiener–Granger causality: A well established methodol- modelling. Neuroimage 59:2322–2329.
ogy. Neuroimage 58:323–329. 52. Friston K, Daunizeau J, Stephan KE (2013) Model selection and gobbledygook:
24. Seth AK (2010) A MATLAB toolbox for Granger causal connectivity analysis. J Neurosci Response to Lohmann et al. Neuroimage 75:275–278, discussion 279–281.
Methods 186:262–273. 53. Lohmann G, Müller K, Turner R (2013) Response to commentaries on our paper: Criti-
25. Barnett L, Seth AK (2014) The MVGC multivariate Granger causality toolbox: A new cal comments on dynamic causal modelling. Neuroimage 75:279–281.
approach to Granger-causal inference. J Neurosci Methods 223:50–68. 54. Box GEP, Jenkins GM, Reinsel GC (2008) Time Series Analysis: Forecasting and Control
26. McCrorie JR, Chambers MJ (2006) Granger causality and the sampling of economic (Wiley, Hoboken, NJ).
processes. J Econom 132:311–336. 55. Ljung L (1999) System Identification: Theory for The User (Prentice Hall PTR, Upper
27. Guo S, Seth AK, Kendrick KM, Zhou C, Feng J (2008) Partial Granger causality – elimi- Saddle River, NJ), 2nd Ed.
nating exogenous inputs and latent variables. J Neurosci Methods 172:79–93. 56. Priestley MB (1994) Spectral Analysis and Time Series (Academic, London).