About

Michael R. Elliott, PhD, is a Professor of Biostatistics at the University of Michigan School of Public Health and a Research Professor at the Institute for Social Research. He received his PhD in biostatistics from the University of Michigan in 1999. Prior to joining the University of Michigan in 2005, he held an appointment as an Assistant Professor at the Department of Biostatistics and Epidemiology at the University of Pennsylvania School of Medicine. Dr. Elliott's statistical research interests focus on the broad topic of missing data, including the design and analysis of sample surveys, causal and counterfactual inference, and latent variable models. He has worked closely with collaborators in injury research, pediatrics, women's health, social determinants of health, and smoking cessation research. His work involves developing model-based Bayesian approaches, methods to assess surrogate markers, and techniques to improve the generalizability of clinical trials. He has served as an editor for prominent statistical journals and has contributed significantly to research projects such as the Environmental Influences on Child Health Outcomes (ECHO) and studies related to women's health, environmental effects on health, driver safety, and smoking behavior.

Research topics

• Internal medicine
• Medicine
• Chemistry
• Pediatrics
• Demography
• Environmental health
• Cancer research
• Immunology

Selected publications

• The Impact of Layering Tobacco 21 Laws and Smoke-free Laws on US Adolescent Smoking Behaviors

  American Journal of Preventive Medicine · 2026-04-01

  article
  PublisherDOI

• Joint modeling of multiple longitudinal biomarkers and survival outcomes via threshold regression: variability as a predictor

  Biometrics · 2026-04-09

  articleOpen accessSenior author

  Longitudinal biomarker data and health outcomes are routinely collected in many studies to assess how biomarker trajectories predict health outcomes. Existing methods primarily focus on mean biomarker profiles, treating variability as a nuisance. However, excess variability may indicate system dysregulations that may be associated with poor outcomes. In this paper, we address the long-standing problem of using variability information of multiple longitudinal biomarkers in time-to-event analyses by formulating and studying a Bayesian joint model. We first model multiple longitudinal biomarkers, some of which are subject to limit-of-detection censoring. We then model the survival times by incorporating random effects and variances from the longitudinal component as predictors through threshold regression that admits nonproportional hazards. We demonstrate the operating characteristics of the proposed joint model through simulations and apply it to data from the Study of Women's Health Across the Nation to investigate the impact of the mean and variability of follicle-stimulating hormone and anti-Müllerian hormone on age at the final menstrual period.

  PublisherOA PDFDOI

• A contemporary examination of the effect of driver training for reducing crash risk in novice adolescent drivers: protocol for the DRIVER study, a random assignment trial

  Injury Epidemiology · 2025-12-01 · 1 citations

  articleOpen access

  BACKGROUND: Motor vehicle crashes and resultant fatalities remain disproportionately high among young drivers, with crash risk peaking immediately after licensure. Although graduated driver licensing laws (GDL) for young novice drivers have reduced such fatalities, driver error remains a leading cause; thus, prevention efforts that target improving skills in novice teen drivers before licensure are a strong candidate for reducing crash risk early in licensure. States with more comprehensive driver licensing laws that include mandated driver training before licensure in addition to GDL show lower crash rates post-licensure, but these effects were not determined through rigorous controlled studies of driver training. This paper describes the DRIVER study, a phase III randomized trial that tests the effectiveness of two different driver training programs in reducing young driver crash risk early in licensure in Pennsylvania, a state like many others that does not require formal training for young drivers. METHODS: Learner drivers age 16 and 17 years will be recruited and followed through the GDL learner phase and for six months post-licensure. Participants will be randomly assigned to one of three interventions: professional behind-the-wheel training (n = 333), online hazard training (n = 333), or an active control online vehicle and driver safety course, unrelated to hazard skills training (n = 333). The primary outcomes are on-road crash risk post-licensure (via kinematic hard braking events tracked through a smartphone-based app) and state license examination performance. Secondary outcomes include change in simulated driving performance from baseline to the time of license examination, self-reported and kinematically tracked risky driving behavior (e.g. cell phone use, speeding) and self-reported crashes. Participants will complete baseline surveys and cognitive assessments to determine potential moderating effects of cognitive maturation and risk-taking tendencies. DISCUSSION: Findings from the DRIVER study will provide insights into training effectiveness generally, and an evidence base for recommendations to policy makers, while also revealing for whom these interventions are less effective. TRIAL REGISTRATION: The study was registered on ClinicalTrials.gov Registry (NCT06413927) in May 2024, https://clinicaltrials.gov/study/NCT06413927 and last updated on August 11th, 2025. This protocol was developed per the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) Checklist.

  PublisherDOI

• Using subject-level variability to predict time-varying outcomes: investigating the association between hormone variability and BMD trajectories over the menopausal transition

  Journal of Applied Statistics · 2025-11-26

  articleOpen accessSenior author

  Women are at increased risk of bone loss during the menopausal transition; in fact, nearly 50% of women's lifetime bone loss occurs during this time. In addition to level and rate of change in estradiol (E2) and follicle-stimulating hormone (FSH), the variability of these hormones may be a key predictor of bone loss; however, this remains unexplored in the existing literature. We introduce a joint model that explicitly characterizes the uncertainty in the means and variances of the hormone trajectories. Our method estimates both the individual mean marker trajectories and the individual residual variances, and links these variances to bone trajectories. In our application, for the first time, higher FSH variance interacted with time was associated with declines in bone mineral density (BMD) across the menopausal transition. At the final menstrual period, higher individual FSH variance predicted an average 0.26% decline in BMD, but this effect is moderated over time. Our results suggest that the mean and variance of FSH, rather than E2, may be a stronger predictor of menopausal bone health. In a variety of simulation studies, our method achieves >90% interval coverage, whereas naive two-stage alternatives often fail to propagate uncertainty in the individual-level variance estimates.

  PublisherDOI

• Methodology for a COVID-19 Recovery Surveillance Study Conducted Through an Academic–State Partnership

  Public Health Reports · 2025-07-10 · 6 citations

  articleOpen access

  OBJECTIVE: We describe the methodologic approach to rapidly launching a population-based surveillance study in spring 2020 to examine the lasting physical, mental, and economic effect of COVID-19 among adults in Michigan. MATERIALS AND METHODS: We established a partnership between the University of Michigan School of Public Health and the Michigan Department of Health and Human Services to conduct this study. Using a sequential stratified sampling strategy, we randomly selected adults with polymerase chain reaction-confirmed SARS-CoV-2 infection in the Michigan Disease Surveillance System. From 2020 through 2022, respondents completed detailed surveys on the lasting effect of their COVID-19 illness online in English or by telephone in English, Spanish, or Arabic, reflecting the diverse population in Michigan. We created and used sampling weights to reduce survey nonresponse bias and tested the performance of the weights with a nonresponse bias analysis. RESULTS: Of all sampled people (n = 17 584), 5521 completed our baseline survey a median of 4.5 months after their COVID-19 onset, for a response rate of 32.1%. Most respondents completed the survey online in every region except Detroit, where 67.1% completed the survey by telephone, highlighting the importance of multimode surveys to increase accessibility and generalizability. Our findings suggest minimal nonresponse bias in the weighted baseline sample. PRACTICE IMPLICATIONS: This unique academic-state partnership resulted in timely and actionable findings related to the lasting effect of COVID-19 that were unavailable elsewhere. While this effort was successful, it was built out of necessity given the limited resources available to local and state health departments to conduct surveillance during the COVID-19 pandemic.

  PublisherOA PDFDOI

• Joint impact of cigarette taxes and smoke-free laws on youth cigarette smoking and related disparities in the USA

  Tobacco Control · 2025-04-16 · 3 citations

  articleOpen access

  OBJECTIVE: To examine the impact of cigarette taxes on youth smoking in counties with and without workplace and hospitality smoke-free laws. METHODS: Using a nationally representative sample of 8th, 10th and 12th graders from the 2001-2021 Monitoring the Future study, we investigated the interaction of taxes and smoke-free policies on cigarette smoking participation, initiation and intention, examining differences by sociodemographic factors (sex, race and ethnicity, parental education, college educational expectations). We stratified models by grade, estimating the average marginal effects (AMEs) using modified Poisson regression with a sandwich variance estimator. RESULTS: Among 12th graders, higher taxes were associated with lower past 30-day smoking, and the relationship was stronger in populations covered by either hospitality or workplace smoke-free laws compared with 12th graders not covered (workplace: AME=-0.009, 95% CI=-0.016 to -0.001; hospitality: AME=-0.010, 95% CI=-0.017 to -0.003). We also examined three-way interactions between taxes, smoke-free policies, and sociodemographic subgroups. We found interactions for taxes with hospitality smoke-free laws and parental education for daily smoking initiation, such that higher taxes were effective in areas with smoke-free laws among 8th graders regardless of parental education, but in areas without smoke-free laws, only among 8th graders whose parents had a college education or more. We found no other statistically significant interactions. CONCLUSION: We found some evidence that taxes and smoke-free laws may work jointly to reduce cigarette smoking in certain youth populations. Policymakers should consider the complex tobacco control landscape and its effects on subpopulations when introducing laws.

  PublisherOA PDFDOI

• Associations Between County-Level Vape-Free Air Law Coverage and E-Cigarette Use Behaviors Among U.S. Adolescents in Monitoring the Future

  Journal of Adolescent Health · 2025-12-06

  articleOpen access

  PURPOSE: We examined whether workplace and hospitality vape-free air law (VAL) population coverage was associated with adolescent e-cigarette use and related disparities in the United States. METHODS: We analyzed associations between county-level workplace and hospitality VAL coverage (100% vs. <100%) and current e-cigarette use (2014-2022) and first e-cigarette initiation (2015-2022) among US 8th, 10th, and 12th graders using nationally representative, cross-sectional Monitoring the Future data. We implemented weighted, grade-stratified, modified Poisson regression models, adjusted for individual-, county-, and state-level confounding factors, examining disparities by sex, race and ethnicity, parental education, and college educational expectations through two-way interactions. RESULTS: Workplace and hospitality VAL coverage was not associated with adolescent e-cigarette use or initiation overall. However, we did find that VAL coverage was associated with lower current e-cigarette use in some sociodemographic subgroups. Full (100%) hospitality VAL coverage was linked to lower e-cigarette use among male 8th and 12th graders, 12th graders with parents without a bachelor's degree, Hispanic 8th and 12th graders, and 8th and 12th graders of other races and ethnicities than their peers living in partially covered (<100%) counties. Full hospitality VAL coverage was linked to higher e-cigarette use among non-Hispanic Black 8th graders and non-Hispanic White 12th graders than their peers living in partially covered counties. DISCUSSION: Though not linked to adolescent e-cigarette use behaviors overall, we found that male adolescents and adolescents from low socioeconomic status backgrounds were more responsive to VAL with associations by race and ethnicity depending on grade and subgroup.

  PublisherOA PDFDOI

• Obesity from Childhood to Mid-adulthood in the United States: A Synthetic Cohort Approach to Measuring Health Trajectories

  Epidemiology · 2025-11-25

  articleOpen access

  BACKGROUND: Obesity dynamics early in life are likely important for long-term health, but have only been described piecemeal, because nationally representative longitudinal datasets are few and have limited follow-up duration. METHODS: We created a synthetic cohort by combining two US nationally representative datasets, the Early Childhood Longitudinal Study, Kindergarten Class of 1998-1999 (ECLS98; N = 21,120; ages 4-16 years; birth cohort 1991-1994), and the National Longitudinal Survey of Youth 1997 (NLSY97; N = 8,984; ages 12-41 years; birth cohort 1980-1984). We used the older-age cohort to impute future weight trajectories of children in the younger-age cohort by matching based on subject-level body mass index trajectories estimated via linear mixed models. We projected trajectories to age 41 years in 2035 for children observed up to a mean age of 13.5 years in 2007. RESULTS: The synthetic cohort (N = 10,102) showed that obesity prevalence increases from 10.0% at age 4 years to 56.3% at age 41 years. Obesity incidence peaks at ages 8 years (4.00/100 person-years [PY] [3.29-4.73]), 26 years (4.48/100 PY [3.04-5.92]), and 38 years (3.60/100 PY [0.00-8.91]). CONCLUSIONS: This synthetic cohort approach can be used to characterize dynamics of obesity and other conditions by maximizing data from shorter "life segments." Findings suggest that today's young adults will continue to become heavier as they age. In addition to prevention before kindergarten entry, other periods for obesity prevention could be middle childhood, mid-twenties, and late thirties.

  PublisherDOI

• Converging crises and maternal and child health: colonialism, extreme weather, and COVID-19

  Reproductive Health · 2025-11-05

  articleOpen access

  BACKGROUND: Climate change is a growing threat to human health, particularly in regions facing overlapping environmental hazards and social inequities. Puerto Rico-a U.S. territory with a colonial history-offers a unique case for examining how multiple disasters, including Hurricane Maria, ongoing earthquakes, and the COVID-19 pandemic, interact with structural vulnerabilities to affect maternal and child health. Despite increasing attention to climate-related health outcomes, little is known about the reproductive health impacts of cumulative disaster exposure in colonial contexts. METHODS: We used U.S. National Vital Statistics System data (2017-2021) to assess associations between disaster exposure and six maternal and newborn outcomes: preterm birth, low birthweight, term low birthweight, gestational hypertension, gestational diabetes, and excessive weight gain. Disaster exposure was defined based on the timing of hurricanes and the pandemic, using a three-month lag period. We analyzed data from Puerto Rico and used Florida and Texas as comparison sites. Multivariable log-binomial regression models estimated adjusted prevalence ratios. Effect modification was tested for (1) region within Puerto Rico and (2) colonial status, comparing Puerto Rico (territory) to Florida and Texas (states). Simulations were conducted to account for potential live-birth bias. RESULTS: Across 104,560 births in Puerto Rico, disaster periods were consistently associated with worse maternal health outcomes. For example, during the late post-hurricane period, gestational diabetes increased (RR = 1.19, 95% CI: 1.08, 1.31), while term low birthweight surprisingly appeared to decline (RR = 0.90, 95% CI: 0.83, 0.98). Associations with newborn health were mixed and may have been underestimated due to sharp declines in live births after disasters. Simulations suggested stronger disaster-related risks than observed in primary analyses. Effect modification by region and colonial status showed inconsistent but notable differences, particularly elevated maternal health risks in certain regions of Puerto Rico and compared to U.S. states. CONCLUSIONS: Our findings suggest that multiple disasters negatively affect reproductive health in Puerto Rico and that structural factors, including colonialism, may exacerbate these impacts. Public health responses must account for cumulative disaster exposure and systemic inequities to better support maternal and child health in marginalized settings, especially as climate change continues to intensify.

  PublisherOA PDFDOI

• Associations between county‐level e‐cigarette‐inclusive Tobacco 21 law population coverage and e‐cigarette use behaviors among United States adolescents in Monitoring the Future

  Addiction · 2025-12-04

  articleOpen access

  BACKGROUND AND AIMS: In the United States (US), Tobacco 21 (T21) laws set the minimum legal sale age for all tobacco products to 21 years. This study aimed to examine whether e-cigarette-inclusive T21 laws were associated with e-cigarette use behaviors and related disparities among US adolescents. DESIGN: We used nationally representative, repeated cross-sectional Monitoring the Future data to compare self-reported current e-cigarette use (2014-2022) and first e-cigarette initiation (2015-2022) among adolescents in counties with 100% ('full') versus <100% ('partial or no') e-cigarette-inclusive T21 law population coverage using modified Poisson regression, examining differences by sex, race and ethnicity, parental educational attainment and college educational expectations through interactions. SETTING: United States. PARTICIPANTS: 8th, 10th and 12th graders. MEASUREMENTS: County-level e-cigarette-inclusive T21 law population coverage was determined using Tobacco 21 Population Coverage Database and US Census Bureau population data. Current e-cigarette use was defined as any past 30-day use among the entire sample. First e-cigarette initiation was defined as first use in the current grade among adolescents who had not initiated use prior to the current grade. FINDINGS: Compared with 8th, 10th and 12th graders in counties with partial or no e-cigarette-inclusive T21 law coverage, 8th [marginal effect (ME) = -1.8%, 95% confidence interval (CI) = -3.1% to -0.6%], 10th (ME = -2.6%, 95% CI = -4.6% to -0.6%) and 12th graders (ME = -2.7%, 95% CI = -5.2% to -0.1%) in counties with full coverage had a lower current e-cigarette use prevalence. For current e-cigarette use, we also observed statistically significant interactions by sociodemographic factors. Across all grades, full [8th: predicted prevalence (PP) = 5.9%, 95% CI = 4.7%-7.1%; 10th: PP = 11.8%, 95% CI = 10.2%-13.4%; 12th: 18.1%, 95% CI = 15.6%-20.6%] versus partial or no coverage (8th: PP = 7.5%, 95% CI = 6.2%-8.8%; 10th: PP = 16.3%, 95% CI = 15.0%-17.6%; 12th: 23.4%, 95% CI = 21.9%-24.8%) was associated with lower current e-cigarette use among males but not females. By race and ethnicity, associations were statistically significant across all grades, but the magnitude and direction of these associations varied by subgroup and grade. Among 12th graders, full (PP = 16.1%, 95% CI = 13.9%-18.3%) versus partial or no coverage (PP = 20.5%, 95% CI = 19.0%-22.1%) was associated with lower current e-cigarette use among those who said they 'probably will' graduate from a four-year college but not among those with other educational expectations. We did not find sufficient evidence to support an association between e-cigarette-inclusive T21 law coverage and first e-cigarette initiation overall or across sociodemographic subgroups. CONCLUSIONS: E-cigarette-inclusive Tobacco 21 laws appear to be associated with lower current e-cigarette use among US adolescents. However, we lacked sufficient evidence to support an association with first e-cigarette use initiation. We also observed sociodemographic differences in these associations for current e-cigarette use.

  PublisherOA PDFDOI

Recent grants

• NIH Grant R01CA129101

  NIH · $899k · 2013

• Variance as a Predictor of Health Outcomes (Resub)

  NIH · $375k · 2021–2023

• NIH Grant R03AG031980

  NIH · $127k · 2012

• NIH Grant R01HL068987

  NIH · $432k · 2007

Frequent coauthors

• Roderick J. A. Little

  University of Michigan–Ann Arbor

  60 shared

• Flaura K. Winston

  Children's Hospital of Philadelphia

  43 shared

• Qixuan Chen

  Columbia University

  43 shared

• Malay Ghosh

  36 shared

• J. Sedransk

  36 shared

• Ye Yang

  Second Affiliated Hospital of Xinjiang Medical University

  36 shared

• Mary E. Thompson

  University of Waterloo

  36 shared

• David Haziza

  University of Ottawa

  36 shared