Abstract
Evaluating causal relationship between biological signals is useful to understand biological systems under consideration. The purpose of this study is to discuss Granger causality between the signals applied filtering, which is useful in biomedical engineering and related fields. Biological signals are quite complex, and hence filtering processes are commonly important to extract necessary information from such signals. Granger causality between signals is estimated by a parametric method based on a multivariate auto-regression model. However, several studies reported that filtering processes affected such parametric estimation of Granger causality and often led to wrong directionality on causal relationships. The method proposed by Dhamala is not based on any parametric models but on Fourier transform and matrix factorization (nonparametric method). This nonparametric method may not be affected by filtering processes. In this study we compared these two methods by applying them to an analytically solvable model system consisting of two variables. We also applied these methods to physiological data and demonstrated the effectiveness of the nonparametric method.
