Dissertations and Theses
Date of Degree
1-1-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy (Ph.D.)
Department
Epidemiology and Biostatistics
Advisor(s)
Zachary Shahn
Committee Members
Denis Nash
Mustafa Hussein
Oliver Dukes
Subject Categories
Econometrics | Epidemiology | Public Health
Keywords
G-estimation, g-methods, difference-in-differences, econometrics, structural nested mean models
Abstract
Estimating causal effects in the presence of time-varying treatments, confounders, and effect heterogeneity remains a central challenge across epidemiology, health policy, and economics. Structural nested mean models (SNMMs) provide a flexible and principled framework for addressing these challenges, yet they remain underutilized due to perceived complexity and limited intuitive guidance. This dissertation advances both the accessibility and the applied utility of SNMMs through an innovative tutorial and two substantive policy applications within difference-in-differences (DiD) frameworks.
Chapter 1 provides an accessible, intuition-driven introduction to g-estimation of SNMMs for applied researchers. Using a sequential multiple assignment randomized trial (SMART) for depression treatment as a motivating example, the chapter demystifies g-estimation by translating standard mathematical notation into plain language, introducing extensive visualizations beyond traditional directed acyclic graphs, and developing analogies to improve conceptual understanding. A fully worked example with code is presented to bridge theory and implementation.
Chapter 2 applies marginal SNMMs (mSNMMs) to estimate heterogeneous effects of Medicaid expansion under the Affordable Care Act on county-level uninsurance rates from 2008–2019. Effects are allowed to vary by expansion cohort, time since expansion, and time-varying county poverty. Identification relies on conditional parallel trends assumptions that may be more plausible compared to time-varying DiD assumptions. Medicaid expansion reduced uninsurance by 3–8 percentage points immediately following adoption, with effects growing to 7–10 percentage points five years post-expansion. Among early (2014) expansion states, effects were substantially larger in higher-poverty counties, a pattern that weakened in later cohorts. While marginal effects were broadly consistent with existing DiD estimators, only mSNMMs enabled joint estimation of cohort-, duration-, and poverty-specific heterogeneity.
Chapter 3 extends SNMMs to settings with continuous, repeated treatments, using state minimum wage increases from 2010–2018 as an illustrative example. SNMMs were estimated under time-varying conditional parallel trends assumptions using doubly robust estimating equations. Sustained effects of a $1.00 minimum wage increase were examined across prior wage levels ($7.25–$12.50) and up to three years post-increase. Estimated effects on unemployment and food insecurity were consistently small and statistically insignificant. For poverty, small statistically significant increases were observed in states with the lowest minimum wages in the prior year, but these findings were not robust to modest parallel trends assumption violations in sensitivity analyses.
This dissertation makes dual contributions by advancing methodological accessibility through pedagogical innovation and demonstrating policy-relevant applications of G-estimation of SNMMs. Furthermore, G-estimation of SNMMs identified under parallel trends expands the toolkit for evaluating staggered policy adoption and repeated and continuous exposures with greater flexibility than existing methods. Future work should prioritize software development, interactive educational resources, and applications to emerging policy questions.
Recommended Citation
Shamsunder, Meghana G., "To “Guess-timate” or G-Estimate: Understanding and Applying G-Estimation of Structural Nested Mean Models to Estimate Complex Treatment Effects in Public Health and Health Policy" (2026). CUNY Academic Works.
https://academicworks.cuny.edu/sph_etds/121
