New Models Improve Policy Impact and Public Service Efficiency
Interventions implemented at different points in time can have varying effects over time. The ability to accurately estimate such effects gives policymakers and researchers insights into a policy’s impact. But efforts to estimate time-varying treatment effects from observational data have largely focused on average outcomes, which may overlook significant differences in the distributions of counterfactual outcomes and lead to concerns over safety, fairness, and ethics. In a new study, researchers propose a conditional generative modeling approach that simulates all possible counterfactual outcomes under a time-varying treatment. The approach addresses the distribution mismatch in the observed data and the targeted counterfactual distribution via a marginal structural model, and it outperforms state-of-the-art baselines on both synthetic and real data. Flexible in nature, the approach is well suited to issues in health care and public policymaking.
The article, Counterfactual Generative Models for Time-Varying Treatments, is authored by Wu, S (Carnegie Mellon University), Zhou, W (Carnegie Mellon University), Chen, M (Princeton University), and Zhu, S (Carnegie Mellon University). Copyright 2023. All rights reserved. The article won first place in the Best Poster Competition for the 2023 YinzOR Student Conference, awarded to Wenbin Zhou, a new Ph.D. student at Carnegie Mellon University.
Service systems, such as emergency medical services and police departments, provide vital emergency aid to communities, but frequently face limited resources, including constrained budgets and personnel shortages. As a result, it is essential that the efficiency of these systems be evaluated and improved. In a new study, researchers present a generalized hypercube queue model, built on the original model by Larson, with a focus on its application to overlapping service regions such as police beats. The model takes into account heavy workloads, staff shortages, and the boundary effects of crime, and can be used to evaluate a service system’s general performance metrics. Researchers demonstrate the model’s versatility by applying it to overlapping police patrols and validate its accuracy using simulated examples. Using their model, they say, will make it possible to develop and optimize police dispatching policies that take into account queuing dynamics for more realistic systems.
The article, Generalized Hypercube Queuing Models with Overlapping Service Regions, is authored by Zhu, S and Xie, Y (Carnegie Mellon University). Copyright 2023. All rights reserved. It will be presented at the Institute for Operations Research and the Management Sciences (INFORMS) 2023. It was recently presented at a Scheduling and QUeueing At LLunch (SQUALL) seminar at Carnegie Mellon University.
About Heinz College of Information Systems and Public Policy
The Heinz College of Information Systems and Public Policy is home to two internationally recognized graduate-level institutions at Carnegie Mellon University: the School of Information Systems and Management and the School of Public Policy and Management. This unique colocation combined with its expertise in analytics set Heinz College apart in the areas of cybersecurity, health care, the future of work, smart cities, and arts & entertainment. INFORMS named Heinz College the #1 academic program for Analytics Education. For more information, please visit www.heinz.cmu.edu.