Causal Partial Identification with Data Augmentation

ICLR 2026 Conference Submission22450 Authors

Keywords: Causal Inference, Partial Identification, Invariance, Data Augmentation

TL;DR: We repurpose symmetry-based data augmentation as an interventional tool to provably sharpen the bounds on causal effects derived from partial identification.

Abstract: We provide a first analysis for using knowledge of symmetries in data generation via data augmentation (DA) transformations for sharpening bounds on causal effects derived from observational data. The causal effect of the treatment $X$ on outcome $Y$ is generally not identifiable from observational data alone if their common causes, also known as confounders, are unobserved. Partial identification (PI) entails estimating bounds on such treatment effects by solving a constrained optimization problem that encodes different assumptions imposed on data generation. PI has use in many application domains where such bounds are sufficient to inform policy decisions, even if the treatment effect itself is not identifiable. To this end, we propose that the cheap and ubuquitous tool of DA, which is otherwise used for mitigating estimation variance, can also be repurposed for sharpening bounds in PI. This is especially useful when the data is complex (i.e., continuous, high-dimensional), as imposing additional constraints becomes expensive compared to a simple pre-processing via DA.

Primary Area: causal reasoning

Submission Number: 22450