Causal Effect Estimation with Learned Instrument Representations

📅 2026-02-10
📈 Citations: 0
Influential: 0
📄 PDF

career value

248K/year
🤖 AI Summary
This work addresses the challenge of causal effect estimation in the presence of unobserved confounders and without explicit instrumental variables (IVs). It proposes ZNet, a novel model that integrates representation learning with IV methodology by leveraging a structural causal model–guided neural architecture to automatically learn latent instrumental representations from observed covariates. These learned representations satisfy the three core IV conditions without requiring pre-specified instruments, effectively disentangling confounding and instrumental components. The model is trained via empirical moment conditions to ensure theoretical consistency. Experiments demonstrate that ZNet not only accurately recovers true instrumental variables when they exist but also constructs effective latent instruments in their absence, serving as a plug-and-play module that significantly enhances the performance of various two-stage IV estimators.

Technology Category

Application Category

📝 Abstract
Instrumental variable (IV) methods mitigate bias from unobserved confounding in observational causal inference but rely on the availability of a valid instrument, which can often be difficult or infeasible to identify in practice. In this paper, we propose a representation learning approach that constructs instrumental representations from observed covariates, which enable IV-based estimation even in the absence of an explicit instrument. Our model (ZNet) achieves this through an architecture that mirrors the structural causal model of IVs; it decomposes the ambient feature space into confounding and instrumental components, and is trained by enforcing empirical moment conditions corresponding to the defining properties of valid instruments (i.e., relevance, exclusion restriction, and instrumental unconfoundedness). Importantly, ZNet is compatible with a wide range of downstream two-stage IV estimators of causal effects. Our experiments demonstrate that ZNet can (i) recover ground-truth instruments when they already exist in the ambient feature space and (ii) construct latent instruments in the embedding space when no explicit IVs are available. This suggests that ZNet can be used as a ``plug-and-play''module for causal inference in general observational settings, regardless of whether the (untestable) assumption of unconfoundedness is satisfied.
Problem

Research questions and friction points this paper is trying to address.

instrumental variable
unobserved confounding
causal effect estimation
observational study
Innovation

Methods, ideas, or system contributions that make the work stand out.

instrumental variable
representation learning
causal inference
unobserved confounding
ZNet
F
Frances Dean
University of California, Berkeley; University of California, San Francisco
J
Jenna Fields
University of California, Berkeley; University of California, San Francisco
R
Radhika Bhalerao
University of California, Berkeley; University of California, San Francisco
M
Marie Charpignon
University of California, Berkeley; University of California, San Francisco
A
Ahmed Alaa
University of California, Berkeley; University of California, San Francisco