Causality in Time Series - ClopiNet
Causality in Time Series - ClopiNet
Causality in Time Series - ClopiNet
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Roebroeck Seth Valdes-Sosaat lower field strengths. The cost of this greater specificity and higher effective spatialresolution is that SE-BOLD has a lower <strong>in</strong>tr<strong>in</strong>sic SNR than GRE-BOLD. The balloonmodel equations above are specific to GRE-BOLD at 1.5T and 3T and have been extendedto reflect diffusion effects for higher field strengths (Uludag et al., 2009).In summary, fMRI is an <strong>in</strong>direct measure of neuronal and synaptic activity. Thephysiological quantities directly determ<strong>in</strong><strong>in</strong>g signal contrast <strong>in</strong> BOLD fMRI are hemodynamicquantities such as cerebral blood flow and volume and oxygen metabolism.fMRI can achieve a excellent spatial resolution (millimeters down to hundreds of micrometersat high field strength) with good temporal resolution (seconds down to hundredsof milliseconds). The potential to resolve neuronal population <strong>in</strong>teractions at ahigh spatial resolution is what drives attempts at causal time series model<strong>in</strong>g of fMRIdata. However, the significant aspects of fMRI that pose challenges for such attemptsare i) the enormous dimensionality of the data that conta<strong>in</strong>s hundreds of thousands ofchannels (voxels) ii) the temporal convolution of neuronal events by sluggish hemodynamicsthat can differ between remote parts of the bra<strong>in</strong> and iii) the relatively sparsetemporal sampl<strong>in</strong>g of the signal.3. <strong>Causality</strong> and state-space modelsThe <strong>in</strong>ference of causal <strong>in</strong>fluence relations from statistical analysis of observed data hastwo dom<strong>in</strong>ant approaches. The first approach is <strong>in</strong> the tradition of Granger causality orG-causality, which has its signature <strong>in</strong> improved predictability of one time series by another.The second approach is based on graphical models and the notion of <strong>in</strong>tervention(Glymour, 2003), which has been formalized us<strong>in</strong>g a Bayesian probabilistic frameworktermed causal calculus or do-calculus (Pearl, 2009). Interest<strong>in</strong>gly, recent work hascomb<strong>in</strong>ed of the two approaches <strong>in</strong> a third l<strong>in</strong>e of work, termed Dynamic StructuralSystems (White and Lu, 2010). The focus here will be on the first approach, <strong>in</strong>itiallyfirmly rooted <strong>in</strong> econometrics and time-series analysis. We will discuss this tradition <strong>in</strong>a very general form, acknowledg<strong>in</strong>g early contributions from Wiener, Akaike, Grangerand Schweder and will follow (Valdes-Sosa et al., <strong>in</strong> press) <strong>in</strong> refer<strong>in</strong>g to the crucialconcept as WAGS <strong>in</strong>fluence.3.1. Wiener-Akaike-Granger-Schweder (WAGS) <strong>in</strong>fluenceThe crucial premise of the WAGS statistical causal model<strong>in</strong>g tradition is that a causemust precede and <strong>in</strong>crease the predictability of its effect. In other words: a variableX 2 <strong>in</strong>fluences another variable X 1 if the prediction of X 1 improves when we use pastvalues of X 2 , given that all other relevant <strong>in</strong>formation (importantly: the past of X1itself) is taken <strong>in</strong>to account. This type of reason<strong>in</strong>g can be traced back at least toHume and is particularly popular <strong>in</strong> analyz<strong>in</strong>g dynamical data measured as time series.In a formal framework it was orig<strong>in</strong>ally proposed (<strong>in</strong> an abstract form) by Wiener(Wiener, 1956), and then <strong>in</strong>troduced <strong>in</strong>to practical data analysis and popularized byGranger (Granger, 1969). A po<strong>in</strong>t stressed by Granger is that <strong>in</strong>creased predictabilityis a necessary but not sufficient condition for a causal relation between time series.82