Proximal balancing improves causal effect estimates with hidden confounders
Proximal Balancing for Causal Effect Estimation under Unmeasured Confounding
Machine Learning
Summary
Estimating cause-and-effect from data can be tricky when some influencing factors are hidden. The authors address this by using proxy variables—indirect clues about the hidden factors—but without needing strict roles or complex modeling. Their method, called proximal balancing, finds a simpler summary that balances groups so their hidden influences are accounted for. This approach works well for different types of data, including images, and helps make better judgments about cause and effect.
What this means in practice
- •For healthcare data teams: Improve treatment effect estimates from observational healthcare data when patient confounders are partially unmeasured using proxy data.
- •For social policy analysts: Estimate causal impacts of policies using observational data with imperfect measurement of confounders accessed via proxy indicators.
Authors
Yonghan Jung
Abstract
Estimating causal effects from observational data is central to science and policy, but the effects are not identified when confounders are unmeasured. Proximal causal inference addresses this problem with proxies of the unmeasured confounders. However, existing proxy-based approaches either designate proxy roles and solve an inverse problem, which is ill-posed and hard to estimate with high-dimensional proxies, or use a latent-variable model, which assumes that the learned latent variable matches the hidden confounder and leaves bias when it does not. To address these challenges, we introduce proximal balancing. It carries the classical idea of covariate balancing to confounders that are observed only through proxies: it learns a low-dimensional summary of the covariates and proxies that makes the treatment groups comparable, and then adjusts for this summary. It needs no designated proxy roles, inverse problem, or latent model. We give identification theory, finite-sample guarantees, and a practical algorithm, PROBE. We demonstrate the method on low-dimensional, high-dimensional, and image proxies and on real-world data.