Synthetic tabular data are typically evaluated by distributional similarity, privacy distance, or predictive performance in synthetic training and real testing, but these criteria do not guarantee validity for causal inference. We demonstrate that fully generative tabular synthesizers, including GAN- and LLM-based models, can maintain predictive utility while distorting average treatment effect (ATE) estimates. The failure is structural: ATE preservation requires both a realistic covariate law and an accurate treatment-effect contrast, whereas prediction loss penalizes treatment-effect error only through an overlap-weighted term.
Thus, under imbalance or limited overlap, a generator may reproduce dominant observed outcomes while underlearning intervention-relevant contrasts. We formalize this mismatch through sensitivity and loss-decomposition results. Motivated by this causal analysis and intuition, we propose a hybrid synthetic-data framework for causal inference that generates covariates while modeling treatment and outcome mechanisms separately.
We evaluate the framework in three settings: ATE preservation under fully generative versus hybrid synthesis, augmentation for practical positivity problems, and diagnostic simulation engines for comparing OR, IPW, AIPW, and TMLE before real-data analysis. We also stress-test the hybrid construction across settings that vary overlap, covariate dimension, seed sample size, and treatment-effect complexity, including a logistic outcome-model misspecification check.
Across controlled simulation experiments, hybrid synthesis improves causal fidelity relative to fully generative baselines; the ACTG application shows improved predictive fidelity and potential for finite-sample estimator benchmarking. LLM-based hybrid synthesis is often more faithful than CTGAN in settings where causal fidelity can be assessed.
Blogger's Review: This article delves into the potential pitfalls of generative synthetic data in causal inference, highlighting the critical issue of preserving causal effects during data generation. The proposed hybrid framework not only holds theoretical significance but also demonstrates the possibility of enhancing causal inference accuracy in practical applications. Through sensitivity analyses, this work offers new perspectives and directions for future synthetic data research.