Probability trees and the value of a single intervention
This work addresses the fundamental challenge in statistical causality for researchers, offering an incremental improvement by applying probability trees and Bayesian estimation to enhance causal induction efficiency.
The paper tackles the problem of determining causal relationships from limited data by quantifying the information gain from a single intervention, showing that both anticipated and expected gains have simple expressions, resulting in an active-learning method that selects interventions with the highest gain, as illustrated through examples.
The most fundamental problem in statistical causality is determining causal relationships from limited data. Probability trees, which combine prior causal structures with Bayesian updates, have been suggested as a possible solution. In this work, we quantify the information gain from a single intervention and show that both the anticipated information gain, prior to making an intervention, and the expected gain from an intervention have simple expressions. This results in an active-learning method that simply selects the intervention with the highest anticipated gain, which we illustrate through several examples. Our work demonstrates how probability trees, and Bayesian estimation of their parameters, offer a simple yet viable approach to fast causal induction.