Research map / Probabilistic and scientific modeling / Treatment effects

Causal Identification for Complex Functional Longitudinal Studies

ICLR 2025 · Probabilistic and scientific modeling › Treatment effects · approach: Proximal and deep causal effect estimation

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a nonparametric causal identification framework for functional longitudinal data with time-varying outcomes, mortality and censoring, generalizing g-computation, IPW, and doubly robust formulas, and checks g-computation with Gaussian-process Monte Carlo simulations.

Problem

Causal identification of marginal mean counterfactual outcomes when treatments, confounders and outcomes are continuous-time functional processes, creating uncountably infinite treatment-confounder feedback and infinite-dimensional data, with death and censoring.

Approach

Extends the coarsened-data framework and stochastic process/measure theory to continuous time: partitions of time with treatment intervened to a stochastic regime G converge in total variation to a target distribution, yielding g-computation, IPW and doubly robust identification formulas that are shown to impose no restrictions on observed data.

Key results

• Monte Carlo g-computation approximations converge toward the analytical target value 0 as sample size n (100 to 2500) increases, with reduced variance.
• Increasing grid size K (10, 50, 250) alone does not consistently improve accuracy or reduce variance, whereas increasing both n and K does.
• The identification is shown to be nonparametric: the set of identifiable observed-data distributions is dense in total variation norm.

Measured effects reported in the paper

Surprising findings

• Refining the time grid alone does not consistently improve approximation accuracy or variance in the simulations.

Limitations

• Only population-level identification; no estimation or inference, and no sample-level (sparse/irregular) data analysis.
• IPW and doubly robust formulas are not numerically evaluated since they cannot be directly approximated without estimation.
• Nonparametric claim is 'almost' nonparametric, relying on a piecewise continuous path-space regularity condition.

Experimental setup

Datasets: simulated Gaussian process data (glucose/insulin-style outcome and treatment)

Metrics: bias of g-computation approximation boxplot spread across 10,000 repeats

Builds on: coarsened data framework g-computation formula inverse probability weighting doubly robust formula

Closest papers by meaning

• Composable Causality: A Toolkit for Systematic Time-Series Causal Discovery and Treatment-Effect Benchmarking (NeurIPS 2026 · similarity 0.88)
• Temporal Causal Mediation through a Point Process: Direct and Indirect Effects of Healthcare Interventions (NeurIPS 2023 · similarity 0.88)
• Instrumental Variable Estimation for Causal Inference in Longitudinal Data with Time-Dependent Latent Confounders (AAAI 2024 · similarity 0.86)
• Data-Driven Covariate Selection for Nonparametric and Cycle-Agnostic Causal Effect Estimation (NeurIPS 2026 · similarity 0.86)
• Local Learning for Covariate Selection in Nonparametric Causal Effect Estimation with Latent Variables (NeurIPS 2025 · similarity 0.85)
• Causal Attribution Analysis for Continuous Outcomes (ICML 2025 · similarity 0.85)