Pre-Conference Workshop C

back to Pre-Conference Workshops

Causal Inference and Directed Acyclic Graphs


Shabbar I Ranapurwala, PhD, MPH
Research Assistant Professor, Epidemiology
Research Scientist, Injury Prevention Research Center
University of North Carolina, Chapel Hill
Adjunct Assistant Professor, Epidemiology
University of Iowa

Shabbar Ranapurwala, PhD is a Research Assistant Professor of Epidemiology at UNC, Chapel Hill, Adjunct Assistant Professor of Epidemiology at the University of Iowa, and a research scientist at UNC Injury Prevention Research Center. Dr. Ranapurwala earned his PhD in Epidemiology at UNC and received postdoctoral training at the UI IPRC. His research interests include conducting intervention studies and applying causal inference research methods to observational data in injury and violence prevention. He has 12 refereed publications, has taught causal inference and directed acyclic graphs in doctoral level epidemiology courses, and serves on SAVIR’s science and research committee.


Stephen W. Marshall, PhD, MPH
Professor, Epidemiology
Director, Injury Prevention Research Center
University of North Carolina, Chapel Hill

Stephen Marshall, PhD, is an injury epidemiologist. He is the Director of the University of North Carolina (UNC) Injury Prevention Research Center, a professor of epidemiology in UNC’s Gillings School of Global Public Health, and faculty in UNC’s Matthew Gfeller Sports-Related Traumatic Brain Injury Center. Dr. Marshall has 25 years of experience and 300 publications in the fields of epidemiology and injury control. His career is dedicated to the twin goals of developing better research methods and translating research findings into tangible injury prevention gains. He serves on several national committees, including the Executive Committee of the SafeStates Alliance.



Rebecca Naumann, PhDc, MSPH
Research Associate, Injury Prevention Research Center
University of North Carolina, Chapel Hill

Becky Naumann, MSPH., is a doctorate candidate in epidemiology and Royster fellow at the University of North Carolina at Chapel Hill. Ms. Naumann has a Bachelor of Science in Environmental Health degree (University of Georgia) and a Master of Science in Public Health degree (Emory University). Prior to beginning her doctoral training, she worked as an epidemiologist on the Transportation Safety Team at the Centers for Disease Control and Prevention. Ms. Naumann has engaged in injury prevention research for more than 10 years. She has published 24 articles in the field of injury prevention and co-authored 4 book chapters.

Attendee Prerequisites:

This workshop requires a basic understanding of epidemiological study designs and Bradford Hill’s guidance on causal inference in epidemiology.

Course Goal:

To discuss the concepts of causal inference including counterfactuals, exchangeability, consistency, positivity, and no-interference and demonstrate the use of directed acyclic graphs to identify confounders in epidemiologic studies of injury and violence prevention.

Learning Outcomes:

At the conclusion of the workshop, participants will be able to:

  • Discuss the importance of counterfactuals in injury and violence research
  • Understand confounding and interference in injury and violence research
  • Understand the lack of consistency and positivity in injury and violence research
  • Develop and utilize directed acyclic graphs to identify potential confounders in injury and violence research

Course Description:

As our society goes through many unsettling social and economic changes, injury and violence prevention is poised to become the next frontier of public health. The rapid emergence of the opioid epidemic and the sustained increase in suicide rates over the past decade are challenging the status quo of how we perceive, investigate, and find enduring solutions to these problems. While well-designed randomized trials could potentially answer most research questions accurately, such trials are not possible to conduct due to ethical concerns, and we have to rely on observational studies to find solutions to our public health problems. For example, the Centers for Disease Control and Prevention (CDC) released opioid prescribing guidelines in March 2016 to limit over-prescribing of opioid pain relievers. However, as noted in the CDC guidelines themselves, the evidence base for the guidelines depends on a relatively modest number of observational studies that have multiple limitations including confounding, measurement error, and lack of generalizability. Modern epidemiologic methods of causal inference provide a framework to identify these problems in research, interpreting observational data in light of limitations, and improving the rigor of research to overcome these limitations. It is important that injury and violence prevention researchers are trained in methods for causal inference if we are to improve the validity of our research and the impact of potential solutions.


In this workshop, we will begin by discussing the concepts of counterfactuals, association, and causation, followed by defining confounding and illustrating examples of confounding in injury prevention. Further, we will identify and illustrate three key properties for causal inference (both mathematically and with examples), namely, exchangeability, consistency, and positivity. Then, we will define the properties of a confounder and compare conventional methods for identification of confounders (such as change-in-estimate) with directed acyclic graphs (DAGs). Using injury and violence examples, we will demonstrate how to develop a DAG and identify a minimally sufficient set of confounders to control for all observed confounding. We will share DAGs from our previously published injury and violence research to explain their development and operationalization. As an exercise, we will engage the audience to develop DAGs (individually or in groups) for specific injury/ violence research questions based on their areas of expertise or interest. Throughout the workshop, we will also include real world examples and illustrations of causal inference properties specifically in injury and violence prevention. The course will utilize a mixed format: some of the time will be used for lecture-style presentations, and some of the time for small group hands-on exercises.

Society for the Advancement of Violence and Injury Research


Powered by Wild Apricot Membership Software