Empirical Dynamic Modeling for Causal Analysis • fastEDM

The fastEDM R package implements a series of Empirical Dynamic Modeling tools that can be used for causal analysis of time series data.

Key features of the package:

powered by a fast multi-threaded C++ backend,
able to process panel data, a.k.a. multispatial EDM,
able to handle missing data using new dt algorithms or by dropping points.

Installation

You can install the development version of fastEDM from GitHub with:

# install.packages("devtools")
devtools::install_github("EDM-Developers/fastEDM-r")

Example: Chicago crime levels and temperature

This example, looking at the causal links between Chicago’s temperature and crime rates, is described in full in our paper:

library(fastEDM)

df <- read.csv(url(
  "https://github.com/EDM-Developers/fastEDM-r/raw/main/vignettes/chicago.csv"
))

crime_causes_temp <- easy_edm("Crime", "Temperature", data=df, verbosity=0)
#> ✖ No evidence of CCM causation from Crime to Temperature found.

temp_causes_crime <- easy_edm("Temperature", "Crime", data=df, verbosity=0)
#> ✔ Strong evidence of CCM causation from Temperature to Crime found.

Stata & Python Packages

This package is an R port of our EDM Stata package. Similarly, we are creating a fastEDM Python package. As the packages share the same underlying C++ code, their behaviour will be identical. If you plan to adjust some of the various low-level EDM parameters, check out the documentation of the Stata package for more details on their options and behaviours.