CCMS Colloquium: Teal Witter (CMC)

Causal questions lie at the heart of scientific inquiry, from evaluating economic policies to determining medical treatments. Yet, observational data alone is often insufficient due to the fundamental problem of causal inference: we cannot observe the counterfactual world where a specific intervention did not occur. Structural Causal Models (SCMs) offer a powerful solution by explicitly modeling the underlying mechanisms of a system. By formalizing data generation, SCMs allow us to use the do-operator to rigorously simulate interventions, answering questions like, “If a patient were administered prednisone and made to stop smoking, what would be their expected pain level?”

However, characterizing a system through individual queries is computationally daunting. As the number of features d grows, the landscape of possible interventions scales exponentially (2^d). To extract interpretable insights from this combinatorial complexity, we utilize the do-Shapley value, a game-theoretic framework that attributes the complicated dynamics of an SCM to individual features.

In this talk, I will present a new algorithmic approach that makes computing these values more tractable. We show that the causal landscape is structured into “irreducible sets”, a building block where multiple interventions yield identical effects. By leveraging this structure, we introduce an algorithm that computes do-Shapley values exactly, with runtime that depends on the graph’s complexity rather than 2^d. We further propose an estimator that targets these sets directly, producing more accurate estimates than prior work by several orders of magnitude or more.

Joint work with Álvaro Parafita, Tomas Garriga, Maximilian Muschalik, Fabian Fumagalli, Axel Brando, and Lucas Rosenblatt.