Controllable Generative Sandbox for Causal Inference

2026-03-03

AI Summary

Existing causal inference simulators struggle to simultaneously preserve distributional fidelity and enable controllable causal mechanisms in mixed-type tabular data. To address this challenge, this work proposes CausalMix—a generative framework based on variational autoencoders that models complex data distributions through a Gaussian mixture latent prior and data-type-specific decoders, while explicitly parameterizing key causal factors such as overlap, unobserved confounding strength, and treatment effect heterogeneity. CausalMix is the first method to unify realistic mixed-data generation with factorized, interpretable control over causal mechanisms, allowing independent adjustment of each causal component during simulation design. Empirical evaluations demonstrate that CausalMix achieves state-of-the-art distributional fidelity across multiple benchmarks while providing stable and fine-grained causal control, and it has been successfully applied to estimator evaluation and power analysis in a comparative study of prostate cancer treatments.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Constraint Satisfaction and Optimization: Mixed Discrete/Continuous Optimization

Application Category

• Web Mining and Content Analysis: Web data generation and simulation
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Method validation and study design in causal inference rely on synthetic data with known counterfactuals. Existing simulators trade off distributional realism, the ability to capture mixed-type and multimodal tabular data, against causal controllability, including explicit control over overlap, unmeasured confounding, and treatment effect heterogeneity. We introduce CausalMix, a variational generative framework that closes this gap by coupling a mixture of Gaussian latent priors with data-type-specific decoders for continuous, binary, and categorical variables. The model incorporates explicit causal controls: an overlap regularizer shaping propensity-score distributions, alongside direct parameterizations of confounding strength and effect heterogeneity. This unified objective preserves fidelity to the observed data while enabling factorial manipulation of causal mechanisms, allowing overlap, confounding strength, and treatment effect heterogeneity to be varied independently at design time. Across benchmarks, CausalMix achieves state-of-the-art distributional metrics on mixed-type tables while providing stable, fine-grained causal control. We demonstrate practical utility in a comparative safety study of metastatic castration-resistant prostate cancer treatments, using CausalMix to compare estimators under calibrated data-generating processes, tune hyperparameters, and conduct simulation-based power analyses under targeted treatment effect heterogeneity scenarios.

Problem

• causal inference
• synthetic data
• counterfactuals
• treatment effect heterogeneity
• unmeasured confounding

Innovation

• Causal Inference
• Generative Modeling
• Synthetic Data
• Treatment Effect Heterogeneity
• Unmeasured Confounding