mediation analysis

This study addresses the inconsistency in causal effect estimates between observational studies and randomized controlled trials (RCTs) by proposing the first unified framework for decomposing causal effect heterogeneity. The framework systematically identifies and quantifies three sources of heterogeneity: differences in covariate distributions, variation in mediating pathways, and shifts in outcome-generating mechanisms. Methodologically, it formally defines effect decomposition across data types (observational vs. experimental), integrating causal inference, sensitivity analysis, and decomposition modeling, while enabling robust parameter estimation under multiple hypotheses. Evaluated through simulation studies and an empirical analysis of the “Moving to Opportunity” experiment, the framework demonstrates improved interpretability, robustness, and policy generalizability in synthesizing evidence from heterogeneous data sources.

Existing causal mediation analysis relies heavily on untestable multiple ignorability assumptions, undermining the robustness of mechanistic inference. This paper proposes a novel identification framework grounded in heterogeneous treatment effects, which integrates explicit and implicit mediation pathways via causal decomposition—enabling simultaneous identification of total, direct, and indirect effects without requiring multiple ignorability. The method combines flexible heterogeneity modeling, Monte Carlo simulation–based validation, and a dedicated software implementation. It is empirically validated in two real-world applications: public resource governance and voter information dissemination. Simulation studies demonstrate that, compared to prevailing approaches, the proposed method achieves substantially improved estimation accuracy and bias control. By relaxing strong untestable assumptions and offering practical implementation tools, this work advances causal mediation analysis with a more robust, interpretable, and accessible framework for uncovering underlying causal mechanisms.

This study addresses the problem of testing whether a treatment effect operates entirely through observed mediators and identifying causal mechanisms under control for covariates. The authors propose a statistical test based on double machine learning, extending— for the first time—the joint evaluation of full mediation and causal mechanism identification to non-randomized treatment settings. By integrating conditional independence testing, the method achieves root-n consistent and asymptotically normal inference even in the presence of high-dimensional covariates. Simulation studies demonstrate favorable finite-sample performance, and the approach is successfully applied to two randomized experiments examining maternal mental health and social norms.

This paper systematically identifies and distills twelve key open scientific challenges in causal inference, centering on the critical disconnect among theory, methodology, and practice in cross-disciplinary applications. Method: It integrates the potential outcomes framework, causal graphical models, statistical modeling, and machine learning—emphasizing domain-knowledge integration and co-development of computationally tractable tools. It innovatively proposes a “theory–method–software–collaboration” quadruple research paradigm to bridge methodological advances with real-world deployment. Contribution/Results: The work delivers the first unified problem atlas spanning biomedical science, social science, and computer science; advocates a collaboration model driven by deep engagement of domain scientists; and provides a systematic roadmap for enhancing interpretability, robustness, and scalability of causal modeling. Collectively, it significantly advances interdisciplinary dialogue and methodological translation.

This paper addresses causal inference in multi-path structures with two sequential mediators. We propose a rigorous decomposition framework for path-specific necessary and sufficient (PNS) causal probabilities. Methodologically, we first define path-specific PNS under dual mediation, establish its nonparametric identifiability theorem, and develop a consistent, asymptotically normal estimator by integrating counterfactual reasoning, nonparametric estimation, and finite-sample simulation. Our key contribution is breaking the single-mediator limitation: we are the first to decompose the total PNS into interpretable, attributable components corresponding to distinct causal pathways—e.g., “teaching investment → learning motivation → academic performance.” Empirically, applied to educational data, our method quantifies the independent causal contributions of each mediated pathway to college admission outcomes, providing both theoretical foundations and computational tools for multi-mechanism causal attribution.

This study addresses the identification of natural direct effects in the presence of unmeasured confounding, particularly baseline confounders affecting the exposure–mediator pathway—a setting where conventional approaches rely on the often untenable cross-world counterfactual independence assumption. The authors propose a set of weaker identification conditions that circumvent this assumption, thereby establishing, for the first time in non-randomized vaccine studies, the identifiability of natural direct effects. Leveraging causal mediation analysis and semiparametric efficiency theory, they develop a multiply robust estimator that avoids stringent modeling assumptions on nuisance functions. Applied to real-world vaccine cohort data, the method effectively quantifies the direct causal effect operating through immune-mediated protection, substantially enhancing both the reliability of mechanistic interpretation and estimation efficiency.

This study addresses the challenge of causal inference in N-of-1 behavioral health case studies, where unobserved confounding impedes valid estimation. The authors propose the Ω causal estimator, which achieves identification without measuring or adjusting for confounders by leveraging functional contrasts over the support set of the outcome variable, requiring only the positivity assumption. This approach pioneers a support-based—rather than distribution-based—framework for causal inference, integrating de Finetti’s subjective probability interpretation with a theory of intervention–observation support consistency. A recall-baseline substitution mechanism bridges support-level contrasts to mean-level causal effects. The method’s feasibility is demonstrated in a case study on cognitive behavioral therapy for anxiety, offering clinicians a practical and robust tool for individualized causal inference.

Existing methods for causal mediation analysis face limitations in handling continuous or multidimensional mediators, non-binary treatments, and intermediate confounding, and often lack usability. This work proposes the crumble framework, which, for the first time, provides a unified nonparametric approach to estimating diverse mediation effects—including natural direct and indirect effects and stochastic intervention effects—under intermediate confounding, while accommodating treatment variables of arbitrary type. Built upon semiparametric theory and leveraging modified treatment policies, crumble enables flexible modeling with high interpretability. The framework’s robustness, practicality, and broad applicability are demonstrated through two real-data applications involving both binary and non-binary treatments.

This study addresses causal inference under network interference, where existing methods typically rely on accurate knowledge of the interaction network—a requirement often unmet in real-world settings due to missing, incomplete, or noisy network data. To overcome this limitation, the authors propose a network-free causal message passing approach that leverages the temporal dynamics of outcome variables to estimate both total treatment effects and spillover effects. Evaluated on a large-scale real-world field experiment, the method is compared against a bipartite graph-based approach that requires network information. Results show that, even without any network data, the proposed method yields effect estimates directionally consistent with the network-aware baseline across all metrics and achieves statistically significant alignment on key decision-relevant outcomes. This work provides the first empirical validation in a real experiment that temporal dynamics alone can effectively substitute for observed network structure, thereby eliminating dependence on network information.

This study addresses the lack of a systematic causal inference framework for within-subject pre-treatment derived outcomes in biomedical research by proposing a semi-parametric, modular causal inference approach tailored to the “within-subject treatment–between-subject analysis” setting, with a specific focus on natural direct effects in mediation analysis. The method integrates semi-parametric modeling, flexible machine learning algorithms, and multiply robust estimation to construct estimators that are robust to misspecification in both within- and between-subject models. To enable high-dimensional inference, it further incorporates a step-down procedure that controls the exceedance rate of the false discovery proportion. Extensive simulations demonstrate its superior performance, and the approach is successfully applied to assess the impact of stimulant medication on functional brain connectivity in children with autism spectrum disorder.