Previously, I had used a "high level" Granger causality toolbox for the inference of directed edges. While there were some early results, I don't like to leave things as a "black box" without understanding the mechanics. In particular, I had observed that the $p$-values for the Granger toolbox seemed to be incorrectly scaled (which causes the FDR method to break down) and I had some serious concerns about the ability of the MVAR model to capture neural processes. So, I decided to implement Granger causality analysis from first principles, which is VAR modeling of time series data.
Getting started with VAR modeling, I found the Mathwork's tutorial to be really helpful.
No comments:
Post a Comment