Abstract: Many policy and practice questions involve wanting to know about the causal effects of interventions or risk factors in populations. However, while rigorous designs such as randomized controlled trials provide unbiased effect estimates for the sample at hand, they do not necessarily inform about effects in a target population of interest — they provide internal validity but not necessarily external validity. While there has been increasing discussion of this limitation of traditional trials, and increasing evidence that the participants in trials often differ from those in target populations, relatively little statistical work has been done developing methods to assess or enhance the ability of randomized trials to inform decision making in specific populations. This talk will provide a framework for thinking about internal and external validity in the context of population treatment effects, and will provide an overview of study designs and statistical methods for estimating population effects. Methods discussed will include interrupted time series approaches, methods for generalizing trial results to populations, and propensity score methods in large-scale data sources. The talk will also discuss the role of statisticians in developing and applying these methods, and open methodological questions that remain. Examples will come from public health and pubic policy, including topics such as gun policy, suicide prevention, and opioids.
Irene Felicetti: ilf@umich.eduRod Little Lectureship Award - Learning what works in populations for public health and public policy: The role of statistics (and statisticians)
Elizabeth Stuart, PhD. Associate Dean for Education Professor of Mental Health, Biostatistics, and Health Policy and Management Johns Hopkins University
October 8, 2020
3:30 pm - 5:00 pm
1755 SPH I
1415 Washington Heights
Ann Arbor, MI 48109-2029
Contact Information: Irene Felicetti: ilf@umich.edu
Abstract: Many policy and practice questions involve wanting to know about the causal effects of interventions or risk factors in populations. However, while rigorous designs such as randomized controlled trials provide unbiased effect estimates for the sample at hand, they do not necessarily inform about effects in a target population of interest — they provide internal validity but not necessarily external validity. While there has been increasing discussion of this limitation of traditional trials, and increasing evidence that the participants in trials often differ from those in target populations, relatively little statistical work has been done developing methods to assess or enhance the ability of randomized trials to inform decision making in specific populations. This talk will provide a framework for thinking about internal and external validity in the context of population treatment effects, and will provide an overview of study designs and statistical methods for estimating population effects. Methods discussed will include interrupted time series approaches, methods for generalizing trial results to populations, and propensity score methods in large-scale data sources. The talk will also discuss the role of statisticians in developing and applying these methods, and open methodological questions that remain. Examples will come from public health and pubic policy, including topics such as gun policy, suicide prevention, and opioids.