From Tokens to Policy: Causal and Interpretable Heterogeneous Treatment Effects Identification
English summary
The paper introduces NEXIS, a method for identifying heterogeneous treatment effects (HTEs) in controlled experiments by re-framing the problem as Markov-blanket discovery on sufficient, aligned multi-modal pre-treatment representations. NEXIS iteratively selects latent interactors with provably consistent selection, avoiding spurious causal characterizations that arise from unmeasured effect modifiers. The approach is deployed on two anti-poverty programs in Africa, augmenting each with satellite imagery to capture previously unmeasured environmental modifiers. The results produce novel, interpretable prescriptive guidelines for optimizing the programs' next iterations.
Chinese summary
本文提出NEXIS方法，通过将异质处理效应（HTE）识别重构为在充分对齐的预训练多模态表示上进行马尔可夫毯发现，从而避免未观测效应修饰因子导致的虚假因果特征。NEXIS以可证明的筛选一致性迭代选择潜在交互因子。该方法被应用于非洲两个反贫困项目，结合卫星图像捕捉此前未测量的环境修饰因子，生成了可解释的优化指南以改进项目下一轮实施。
Key points
NEXIS re-frames HTE identification as a Markov-blanket discovery problem on aligned multi-modal pre-treatment representations.
NEXIS将对齐的多模态预治疗表示上的HTE识别重构为马尔可夫毯发现问题。
The method provides provable selection consistency, preventing spurious causal characterization from unmeasured heterogeneity drivers.
该方法提供可证明的筛选一致性，避免未测量异质性驱动因素导致的虚假因果特征化。
Applied to African anti-poverty programs with satellite imagery, it uncovered novel environmental effect modifiers and generated prescriptive policy guidelines.
结合卫星图像应用于非洲反贫困项目，揭示了新的环境效应修饰因子并生成可操作的策略指南。