🤖 AI Summary
This study addresses the challenge that existing predictive explanation methods conflate predictive importance with drivers of heterogeneous causal effects, thereby failing to attribute outcomes to genuine intervention effects. To resolve this, the authors propose a causal Shapley attribution framework grounded in intervention coalition games, integrated with DoubleML causal estimation to construct a local-to-global Causal Attribution Score (CAS). This approach explicitly disentangles causal effect modification from predictive contribution without altering the Shapley value formulation, enabling interpretable analysis of heterogeneous treatment effects. Empirical results demonstrate strong performance: on synthetic data, the method achieves a mean absolute error as low as 0.107; on real-world datasets—401(k) and Pennsylvania reemployment—it identifies key causal effect modifiers via Feature-CAS that markedly differ from those highlighted by conventional SHAP, substantially outperforming current baselines.
📝 Abstract
Predictive explanation methods attribute a model output; they do not, by themselves, attribute an intervention effect on the real-world outcome. We introduce the Causal Attribution Score (CAS), a compact score architecture for causal explanation. CAS starts from an identified interventional coalition game, allocates the joint intervention contrast with causal Shapley contributions, and converts those raw outcome-scale effects into Local CAS, Signed Local CAS, and two complementary Global CAS summaries. The innovation is not a new Shapley formula, but a local-to-global causal reporting layer with an explicit intervention target. In the known-truth benchmark, eight repeated primary-interaction simulations (n = 2,200 each, three actions) gave mean Local CAS MAE of 0.107 for coalition-aware CAS, compared with 0.173 for one-at-a-time normalisation and 0.213 for a global normalised absolute ATE vector. The paired advantage over one-at-a-time normalisation increased from -0.003 under additivity to 0.091 under strong interactions. On both empirical DoubleML datasets, 401(k) eligibility/net financial assets (n = 9,915) and Pennsylvania reemployment bonus/unemployment duration (n = 5,099), predictive SHAP/TreeSHAP rankings differed materially from Feature-CAS rankings of treatment-effect modifiers. In Pennsylvania, dep1 (exactly one dependent) moved from predictive global rank 13 to Feature-CAS rank 2 and was the leading local Feature-CAS modifier. These results isolate the added value of separating what predicts the outcome from what explains heterogeneity in an estimated causal effect.