[2603.03587] Controllable Generative Sandbox for Causal Inference

Statistics > Methodology arXiv:2603.03587 (stat) [Submitted on 3 Mar 2026] Title:Controllable Generative Sandbox for Causal Inference Authors:Qi Zhang, Harsh Parikh, Ashley Naimi, Razieh Nabi, Christopher Kim, Timothy Lash View a PDF of the paper titled Controllable Generative Sandbox for Causal Inference, by Qi Zhang and 5 other authors View PDF HTML (experimental) 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 providin...