Bayesian Causal Methods for Environmental Accountability Studies with Heterogeneous Effects

Dafne Zorzetto, Emma Landry, Michele Guindani, Donatello Telesca, Valeria Edefonti, Roberta De Vito, Falco J. Bargagli-Stoffi · 2026

Read Bayesian Causal Methods for Environmental Accountability Studies with Heterogeneous Effects by Dafne Zorzetto, Emma Landry, Michele Guindani, Donatello Telesca, Valeria Edefonti, Roberta De Vito, Falco J. Bargagli-Stoffi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13585896.txt.

Introduction

Dafne Zorzetto and co-authors review Bayesian tools for judging whether green rules deliver. Accountability studies come in two kinds. Some infer a policy's worth from pollution harm data, while others grade the policy head on. In both, the effect varies across groups, and finding who gains most matters.

Flexible Bayesian models suit that hunt, since they can sort units into like clusters. The review picks two families built on regression trees and on linked random shares. Fake data trials compare the two, showing where each shines. The authors give adoption advice for teams that grade air rules and other health linked policies.

Catalog record

CREATORDafne Zorzetto, Emma Landry, Michele Guindani, Donatello Telesca, Valeria Edefonti, Roberta De Vito, Falco J. Bargagli-Stoffi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE129 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13585896

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bayesian Causal Methods for Environmental Accountability Studies with Heterogeneous Effects Authors: Dafne Zorzetto, Emma Landry, Michele Guindani, Donatello Telesca, Valeria Edefonti, Roberta De Vito, Falco J. Bargagli-Stoffi Published: 2026 ABSTRACT Purpose of Review:Environmental accountability studies evaluate whether regulatory interventions achieve their intended benefits. These studies fall into two broad categories: indirect accountability, which estimates the effects of exposure to higher levels of air pollution and uses them to indirectly predict policy impacts, and direct accountability, which evaluates specific interventions directly. The purpose of this review is to evaluate recent advances in causal inference and flexible Bayesian statistical modeling to support environmental accountability studies.Recent Findings:Recent studies, in both direct and indirect environmental accountability, deeply rely on causal inference to produce robust inferences and provide relevant and actionable insights to policy-makers and practitioners. In this context, heterogeneous treatment effects are central to both approaches as it is critically important to understand which subpopulations are most vulnerable (or resilient), requiring methods that can characterize effect variation. Bayesian nonparametric (BNP) methods have been proposed for this task due to their flexibility and ability to cluster units into coherent subgroups.Summary:We present a selection of BNP methods designed or adapted for each type of accountability, focusing on complementary approaches based on Bayesian Additive Regression Trees (BART) and Dependent Dirichlet Processes (DDP). Through Monte Carlo simulations, we compare these methods, identify their respective advantages, and provide recommendations for adoption and implementation in environmental accountability research. Introduction: Causal Inference to Inform The Accountability Framework in Environmental Health The Health Effects Institute introduced the “environmental accountability” framework to systematically evaluate the health effect of exposure to air pollution and whether environmental regulations, aimed at reducing such exposure, can achieve their intended public health goals [1]. This framework recognizes that demonstrating health benefits from air quality improvements requires moving beyond associational studies toward rigorous causal evaluation of specific interventions [2, 3]. The accountability framework links regulatory actions to changes in emissions, ambient concentrations, exposures, and ultimately health outcomes where each link requires distinct methodological approaches [3]. Landmark cohort studies established that long-term fine particulate matter (PM\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_{2.5}$$\end{document}) exposure increases cardiovascular and respiratory mortality [4, 5], results subsequently confirmed using administrative data covering tens of millions of older Americans [6]. These foundational findings, together with evidence of health effects at concentrations below current regulatory standards [7], underscore the need for methods that can rigorously quantify both the magnitude of exposure-response relationships and evaluate the effectiveness of policies designed to reduce exposures. Zigler and Dominici [3], building on the framework laid out by the Health Effect Institute [1], distinguished two distinct approaches to environmental accountability: indirect and direct accountability. Indirect accountability answers: “What is the relationship between pollution exposure and health?” These studies estimate the effects of exposure to higher levels of air pollution (often dichotomized at a policy relevant threshold, as we will see below) and use them to predict policy impacts, assuming that the estimated relationship would hold under regulatory implementation. Direct accountability answers: “Did this specific environmental regulatory intervention actually impact health?” Under the causal inference framework, invoked by Zigler and Dominici [3], these studies have the potential to evaluate well-defined policies (e.g., nonattainment designations, scrubber installations) and whether they caused health changes. The distinction is fundamental: indirect studies inform what might happen under hypothetical interventions, whereas direct studies reveal what did happen following actual policy implementation. Evidence from both approaches is needed to provide a complementary and well-rounded assessment of exposure–health relationships and the effectiveness of actions to reduce exposures, ultimately guiding protection and mitigation strategies [3, 8]. Both approaches must address heterogeneous treatment effects (HTE), which are ubiquitous in environmental health due to differential susceptibility, exposure patterns, and policy implementation [2, 9–11]. Substantial evidence indicates that air pollution health effects vary by age, with the elderly showing heightened vulnerability [11, 12]; by socioeconomic status and race/ethnicity [6]; and by pre-existing health conditions including diabetes, heart failure, and chronic lung disease [13]. These susceptibility patterns have direct implications for environmental health equity: communities that bear disproportionate pollution burdens are often those with elevated vulnerability to its effects [14]. In indirect accountability, key heterogeneous effects questions include who is most vulnerable to pollution exposure and where interventions will have greatest impact. In direct accountability, the focus shifts to where the intervention worked (and, potentially, through what mechanisms). Traditional approaches based on pre-specified subgroups are limited: they may miss important heterogeneity along unanticipated dimensions and suffer from multiple testing issues [15, 16]. Data-adaptive methods that discover heterogeneous groups while maintaining principled inference are therefore needed. Bayesian nonparametric (BNP) methods offer a data-adaptive solution by placing priors over flexible function spaces, allowing the data to reveal heterogeneity patterns while borrowing strength across similar observations [17–19]. This review focuses on four BNP approaches that provide flexible, data-adaptive solutions to HTE estimation in environmental health: Bayesian Causal Forests with post-hoc clustering (BCF+CART) and the Confounder-Dependent Bayesian Mixture Model (CDBMM) for indirect accountability [20, 21]; the Confounders-Aware SHared atoms BAyesian mixture (CASBAH) and Bayesian Principal Causal Forests (BPCF) for direct accountability [22, 23]. We want to emphasize that among the landscape of BNP methods for causal inference, we focus on these four approaches because they represent, to our knowledge, the current set of BNP methods purpose-built at the intersection of four defining criteria: (i) BNP modeling of the outcome-exposure relationship, (ii) heterogeneous treatment effect estimation and subgroup discovery, (iii) seamless integration of the accountability framework for environmental health regulation evaluation, and (iv) code availability, ensuring reproducibility and facilitating adoption by the broader environmental research community. This distinguishes them from BNP methods applied to environmental health that target exposure-response curves or average causal effects [24–26] or methods that do not provide a fully reproducible code. Furthermore, we want to note that, given these four defining criteria, this work should be approached as selective review rather than a broad review of Bayesian causal methods in environmental health. The four selected methods further span two conceptually distinct dimensions that are central to environmental health policy evaluation: the accountability pathway (indirect versus direct) and the mechanism for heterogeneity discovery (tree-based partitioning versus mixture-model clustering), providing a comprehensive view of current methodological options within a coherent organizational structure. Furthermore, we note that, while alternative frequentist methods are also available [16, 27], the reviewed Bayesian causal inference methods share key features attractive for environmental health: they provide full posterior inference for uncertainty quantification, they provide a flexible modeling of the relationship between exposure to higher levels of air pollution and health outcomes, and enable heterogeneity discovery without requiring pre-specification [28, 29]. These methods are evaluated through extensive Monte Carlo simulations under controlled conditions for both indirect and direct accountability studies. In causal inference settings, researchers typically do not have access to the ground truth (that is, the causal effect for each unit in the study). Simulation studies are therefore essential for assessing how the methods respond to different definitions and structures of treatment effect heterogeneity under scenarios with controlled ground truth. In these settings, model performance can be compared vis-a-vis with the data-generating-process ground truth to provide an evaluation of the methods performance [30]. In both the indirect and direct accountability simulation

Read the complete text