[Paper Review] Spatial causal inference in the presence of unmeasured confounding and interference

Georgia Papadogeorgou, Srijata Samanta|arXiv (Cornell University)|Mar 14, 2023

Spatial and Panel Data Analysis4 citations

TL;DR

This paper proposes a Bayesian spatial causal inference framework that jointly models interference and unmeasured spatial confounding in observational spatial data. By integrating spatial random effects and exposure models, it identifies causal effects even when confounders are unmeasured and spatial dependence exists, proving parameter identifiability and showing improved inference over standard methods in simulations and a U.S. county-level study on SO₂ emissions and cardiovascular mortality.

ABSTRACT

This manuscript unites causal inference and spatial statistics, presenting novel insights for causal inference in spatial data analysis, and drawing from tools in spatial statistics to estimate causal effects. We introduce spatial causal graphs to highlight that spatial confounding and interference can be entangled, in that investigating the presence of one can lead to wrongful conclusions in the presence of the other. Moreover, we show that spatial dependence in the exposure variable can render standard analyses invalid. To remedy these issues, we propose a Bayesian parametric approach based on tools commonly-used in spatial statistics. This approach simultaneously accounts for interference and mitigates bias from local and neighborhood unmeasured spatial confounding. From a Bayesian perspective, we show that incorporating an exposure model is necessary. Under a specific model formulation, we prove that all parameters are identifiable including the causal effects, even in the presence of unmeasured confounding. We illustrate the approach with a simulation study. We evaluate the effect of local and neighboring sulfur dioxide emissions from power plants on county-level cardiovascular mortality from observational spatial data in the United States, where unmeasured spatial confounding and interference might be present simultaneously.

Motivation & Objective

• To address the intertwined challenges of spatial interference and unmeasured spatial confounding in causal inference with spatial data.
• To demonstrate that ignoring one (e.g., interference) while accounting for the other (e.g., confounding) can lead to incorrect causal conclusions.
• To develop a Bayesian parametric model that simultaneously accounts for interference and mitigates bias from local and neighborhood unmeasured spatial confounders.
• To establish theoretical identifiability of all model parameters under unmeasured confounding, even in the presence of spatial dependence.
• To evaluate the method’s performance using simulation studies and a real-world case study on sulfur dioxide emissions and cardiovascular mortality in U.S. counties.

Proposed method

• Proposes a Bayesian hierarchical model that incorporates spatial random effects to capture unmeasured spatial confounding.
• Uses a spatial exposure model to account for neighborhood-level exposure effects, enabling estimation of interference effects.
• Introduces spatial causal graphs to visualize and analyze the entanglement between interference and confounding in spatial data.
• Employs a likelihood-based framework with proper priors to ensure posterior inference and parameter identifiability.
• Uses second-degree county adjacency matrices to define neighborhood exposure, allowing for spillover effects from neighbors of neighbors.
• Applies Markov Chain Monte Carlo (MCMC) methods for posterior computation and uncertainty quantification.

Experimental results

Research questions

• RQ1 Can spatial interference and unmeasured spatial confounding be entangled in a way that leads to incorrect causal conclusions if only one is modeled?
• RQ2 How does spatial dependence in the exposure variable affect the validity of standard causal estimators?
• RQ3 Can a Bayesian spatial model jointly account for interference and unmeasured spatial confounding while ensuring parameter identifiability?
• RQ4 What is the performance of the proposed method compared to standard OLS in estimating local and interference effects under spatial dependence and unmeasured confounding?
• RQ5 To what extent does the inclusion of weather variables affect the estimation of causal effects in real-world spatial data?

Key findings

• The proposed Bayesian model significantly reduces bias in estimating local and interference effects compared to OLS, especially under unmeasured spatial confounding and exposure dependence.
• The simulation study shows that standard OLS estimators have poor coverage of 95% credible intervals when interference or confounding is present, while the proposed method maintains appropriate coverage.
• In the case study, the model estimates a positive causal effect of sulfur dioxide emissions on cardiovascular mortality, with credible intervals that reflect uncertainty due to spatial dependence and unmeasured confounders.
• The inclusion of weather variables had minimal impact on OLS estimates, suggesting that demographic and emission variables dominate the outcome variation in the model.
• Theoretical analysis proves that all model parameters, including those related to interference and confounding, are identifiable even with unmeasured spatial confounders.
• Spatial causal graphs reveal that misattributing spatial structures to interference or confounding can lead to erroneous causal interpretations, highlighting the need for joint modeling.