Marginal Structural Models Overview

Principles

Definition

Marginal structural models is a class of statistical models developed to estimate the causal effect of a time-varying exposure or treatment on an outcome in the presence of time-varying confounding, where confounders may be affected by prior exposure and also influence future exposure or the outcome. This methodology employs inverse probability weighting techniques to create a pseudo-population where the exposure is independent of measured confounders over time, thereby enabling consistent estimation of marginal, population-level causal effects from observational longitudinal data.

Ontological type

Core Methods

Key Assumptions

Econometric Applications

Key developments

Key Figures

Propensity and MSM Foundations era

Donald B. Rubin [1] held appointments at Harvard University [3] and the University of California, San Francisco [4] during the Propensity and MSM Foundations era (1983-2000). In 1983 Rubin published The central role of the propensity score in observational studies for causal effects [7] and Assessing Sensitivity to an Unobserved Binary Covariate in an Observational Study with Binary Outcome [8], work that established the propensity score as a scalar balancing tool for addressing confounding and sensitivity concerns, laying the groundwork for IPW and weighting-based causal inference in this era. Paul R. Rosenbaum [2] was affiliated with the University of Pennsylvania [5] and Princeton University [6] during this period, and his collaboration on the central propensity score paper [7] helped crystallize the design-based emphasis in causal inference. Rosenbaum's [2] emphasis on design-based inference and matching during this era helped shape later weighting-based causal methods that underpin marginal structural modeling.

MSM Maturation and Weighting era

Peter C. Austin [1] is a prominent statistician associated with The University of Texas Health Science Center at San Antonio [3] and the University of Toronto [4] during the MSM maturation and weighting era. His key contributions include Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity‐score matched samples [7] and The Performance of Two Data-Generation Processes for Data with Specified Marginal Treatment Odds Ratios [8], which helped refine weighting-based causal inference by enabling practical balance assessment and evaluation of modeling assumptions in this era. Harvey Goldstein [2] is a leading figure associated with University College London [5] and the University of Bristol [6] during this era. His key contributions in this era include Multilevel Statistical Models [9], which advanced hierarchical modeling for clustered data and informed weighting-based causal inference through refined understanding of variation and diagnostic capabilities.

Dynamic Robust Weighting era

Stephen Burgess [1] is a prominent figure in this era whose work has spanned causal inference in genetics and policy-relevant evaluation, with affiliations at the University of California, San Francisco [3] and the University of Pennsylvania [4]. His key contribution during this era includes the Mendelian Randomization Analysis With Multiple Genetic Variants Using Summarized Data [7], which demonstrates how combining multiple genetic instruments with summarized statistics enables robust causal estimates when individual-level data are inaccessible. Adam S. Butterworth [2] is a leading quantitative thinker whose work in this era has ties to Harvard University [5] and the Massachusetts Institute of Technology [6]. His contributions center on the Mendelian Randomization Analysis With Multiple Genetic Variants Using Summarized Data [7], illustrating how summarized-data MR with multiple variants supports scalable causal inference—an approach that aligns with overlap-aware and weight-based robustness themes central to dynamic weighting in longitudinal settings [7].