Abstract Number:
1802
Submission Type:
Topic-Contributed Paper Session
Participants:
Chen-Pin Wang (1), Chen-Pin Wang (1), Karen Bandeen-Roche (2), Sahar Zangeneh (3), Chen-Pin Wang (1), Razieh Nabi (4), Johann Gaebler (5)
Institutions:
(1) UT Health Science Center San Antonio, N/A, (2) Johns Hopkins University, N/A, (3) RTI International, N/A, (4) Emory University, Rollins School of Public Health, N/A, (5) Stanford University, N/A
Chair:
Session Organizer:
Speaker(s):
Razieh Nabi
Emory University, Rollins School of Public Health
Session Description:
Health equity is the state in which everyone has a fair opportunity to attain their highest level of health. Health equity is attainable by optimizing structural fairness and eliminating health disparities. Various fairness algorithms and disparity estimands/estimators have been proposed towards different research goals. Built on these prior work, the proposed session will feature five presentations to address specific methodology shortcomings spurred from real-world applications: 1) measurement issues, 2) nonignorable missingness, 3) identification of heterogeneous disparities, 4) delineation of the mechanisms of fairness, and 5) theoretic insights under the causal modeling framework. Technical aspects of the presentations are summarized below.
1) Bandeen-Roche (JHU) considered health disparity research where the underlying risk factors (e.g., structural racism) cannot be measured directly. An advanced latent variable modeling framework was proposed to accommodate networking between the domains of structural racism and contextual specificity by place and time. These new methods aim to improve measures, and to better elucidate and address health disparities.
2) Zangeneh (RTI) proposed a general class of models and likelihood-based estimatiors when certain individual level data is subject to missingness, but available to all at the aggregated level from external sources. The proposed method does not require the missing at random (MAR) assumption, and thus it provides more robust estimates under a weaker assumption compared to those requiring MAR in the absence of aggregated data.
3) Wang (UTHSA) considered methods to assess counterfactual disparity of an endpoint outcome conditioned on principal strata of an intermediate variable that is prone to measurement errors and subsequent misclassification of principal strata. The proposed method incorporated risk adjustments with Bolk-Croon-Hagenaars method to mitigate measurement bias, followed by 1-step and 3-step error-less ML estimators to derive 'principal disparity' under respective data-guided identification assumptions. Efficiency, consistency and utilities of the two estimators are evaluated.
4) Nabi (Emory) proposed methods to make optimal but fair policies which "break the cycle of injustice" by correcting for the unfair dependence of both policies and outcomes on protected attributes. The proposed methods were developed for causal inference and constrained optimization to learn optimal policies in a way that addresses multiple potential biases which afflict data analysis in sensitive contexts. This proposal is equipped with the theoretical guarantee that the chosen fair policy will induce a joint distribution for new instances that satisfies given fairness constraints.
5) Gaebler (Stanford) and colleagues considered two families of causal fairness algorithms: (i) constraining effects of policies on counterfactual disparities; and (ii) constraining effects of protected attributes on policies. The authors showed that analytically and empirically both fairness algorithms almost always result in strongly Pareto dominated policies, meaning that there is an alternative, unconstrained policy favored by every stakeholder with preferences drawn from a large, natural class. The results highlight formal limitations and potential adverse consequences of common mathematical notions of causal fairness.
Sponsors:
Health Policy Statistics Section 1
Justice Equity Diversity and Inclusion Outreach Group 2
Mental Health Statistics Section 3
Theme:
Statistics and Data Science: Informing Policy and Countering Misinformation
Yes
Applied
Yes
Estimated Audience Size
Small (<80)
I have read and understand that JSM participants must abide by the Participant Guidelines.
Yes
I understand and have communicated to my proposed speakers that JSM participants must register and pay the appropriate registration fee by June 1, 2024. The registration fee is nonrefundable.
I understand