01/11/2023: 10:30 AM - 12:15 PM MST
Invited
Quasi-experimental designs, such as difference-in-differences (DiD) methods, are commonly used for evaluating the effect of policy interventions. DiD approaches, for example, provide unbiased effect estimates if the trend over time would have been the same for the intervention and comparison groups in the absence of the intervention. However, traditional causal inference approaches using quasi-experimental designs may not be immediately applicable in many real-world settings, possibly due to the violation of parallel trends assumption, spillover effects, and increased complexities in data structures.
In this session, the speakers will propose novel causal inference methods for evaluating the effect of interventions in non-traditional settings and discuss their implications for real-world problems. Topics will include new causal inference methods for evaluating the spillover effects of the beverage tax, reconsidering DiD in the post-COVID era, policy effect evaluation under counterfactual neighborhood treatment assignments, and a non-parametric Bayesian approach to estimate the effect of a targeted effort to remove firearms. The applications to recent policy interventions, as well as novel quantitative method development, fit very nicely with the theme of ICHPS: "upgrading the pipeline from health data to health policy".
The session will have four speakers from three different institutions in the fields of biostatistics, statistics, quantitative social science, and health policy. This session will be chaired by Dr. Nandita Mitra.
Causal inference
Quasi-experimental designs
Difference-in-differences
Spillover effects
Covid-19
Gun violence
Organizer
Youjin Lee
Chair
Nandita Mitra, University of Pennsylvania
Presentations
We develop a Bayesian semi-parametric model for the impact of dynamic treatment rules on survival among patients diagnosed with pediatric acute myeloid leukemia (AML). The data are from a phase III clinical trial in which patients move through a sequence of four treatment courses. At each course, they undergo backbone chemotherapy that may or may not include an anthracycline-based (ACT) agent. While ACT is thought to more aggressively suppress AML, it is also cardiotoxic. Thus, treating overzealously with either may reduce survival. Our task is to estimate the potential survival probability under hypothetical dynamic treatment strategies, but there are several impediments. First, since ACT is not randomized, its effect on survival is confounded over time. Second, subjects initiate the next course depending on when they recover from the previous course - thus, timing is variable and potentially informs subsequent treatment decisions and survival. Third, patients may die or drop out before ever completing the full sequence. We develop a generative Bayesian semi-parametric model based on Gamma Process priors to address these complexities. At each treatment course, the model captures subjects' transition to subsequent treatment or death in continuous time. A g-computation procedure is used to compute a posterior over potential survival probability that is adjusted for time-varying confounding. We estimate the efficacy of AML treatment rules that dynamically assign ACT based on evolving cardiac function.
Speaker
Arman Oganisian, Brown University
COVID-19 has induced global historic disruptions in health, politics, and the economy. This has resulted in large-scale shocks, often several standard deviations in magnitude, in common indicators used for observational research, including unemployment, mortality, and educational outcomes. In this paper, we explore the implications of these shocks for evaluating three types of non-experimental interventions: 1) those that began prior to 2020 with major shocks in the post-intervention period; 2) those implemented alongside or as part of pandemic response; and 3) those that will be implemented in the coming years for which 2020 would normally be included in the pre-intervention period. While the traditional "common shocks" assumption requires that comparison units on average move in parallel prior to and during a major shock, we present statistical and practical reasons why alternative approaches may be more appropriate. We characterize a flexible set of assumptions related to shock processes, including "removable shocks", "common after-shocks" and "common recovery", and propose a method for selecting an optimal estimator based on analysis of untreated units. We present case studies, retrospectively exploring unemployment and uninsurance trajectories following the 2008 financial crisis.
Speaker
Alyssa Bilinski, Brown University
Public policy interventions are commonly evaluated using the difference-in-differences (DiD) approach. However, this approach does not directly account for the effect of the policy spilling over to neighboring regions such as nearby cities. For example, the implementation of an excise tax on sweetened beverages in Philadelphia was shown to be associated with a substantial decrease in volume sales of taxed beverages in Philadelphia but also showed an increase in beverage sales in bordering counties which were not subject to the tax. The latter association could potentially be explained by cross-border shopping behaviors of Philadelphia residents. Because spillover effects can offset the total effect of such interventions, particularly for specific sub-populations, understanding the dynamics of such effects is essential to holistically evaluate public policies.
To address these concerns, we extend difference-in-differences methods to identify the causal effects of policy interventions under various spillover conditions. We propose doubly robust estimators for the average treatment effect on the treated and on the neighboring control that relax standard assumptions on interference and model specification. Throughout, we discuss practical considerations of such analyses, including possible positivity violations and time-varying effects, as they relate to the Philadelphia beverage tax study.
Speaker
Gary Hettinger, University of Pennsylvania
Gun violence is a critical public health and safety concern in the United States. There is considerable variability in policy proposals meant to curb gun violence, ranging from increasing gun availability to deter potential assailants (e.g. concealed carry laws or arming school teachers) to restricting access to firearms (e.g. universal background checks or banning assault weapons). Many studies use state-level variation in the enactment of these policies in order to quantify their effect on gun violence. In this talk, we discuss recent methodological proposals for evaluating the impact of these policies, connecting frequentist and Bayesian approaches, and show how bringing in multiple sources of public health and safety data can provide important insight into how policies affect gun violence.
Speaker
Eli Ben-Michael, Harvard University