We gratefully acknowledge the U.S. Army Research Institute (ARI), Basic Research Program, for funding our project “Use Machine Learning to Model Time and Causality”.

The principal investigator (PI) is Heng Xu, and the co-PI is Nan Zhang.

Project Summary

Understanding how causes and effects unfold over time is essential for studying human behavior in dynamic environments. However, traditional methods often require researchers to guess the timing of these effects, leading to unreliable results. This project introduces a novel approach that uses machine learning to identify both the direction and timing of causal relationships without relying on such assumptions. By linking the predictive error of machine learning to causal dynamics, our method has the potential to reveal whether effects are immediate or delayed, and whether they are short-lived or long-lasting, providing a model-free supplement to existing panel models.