Optimizing Public Policy
By Violet Chen
There has been a rising interest in applying modelling and optimization tools to improve public policy design. The Wednesday morning session, “Optimizing Public Policy,” featured three interesting talks on this topic. Theodore Papalexopoulos, PhD student from the MIT Operations Research Center, talked about “Redistricting Liver Allocation: A Simulation-Optimization Approach.” The presented work is joint with Dr. Dimitris Bertsimas, Dr. Nikolaos Trichakis, Yuchen Wang, Dr. Ryutaro Hirose, and Dr. Parsia Vagefi. They introduced a simulation-based optimization approach for balancing fairness and efficiency in liver allocation. From the perspective of policymakers, fairness is associated with lower mortality rates from allocation, and efficiency represents goals including lower transportation cost and fewer organ discards due to transportation delay. Their simulation-optimization approach is proposed to decide a grouping of 58 donation service areas (DSAs) into broader districts, which enable optimal allocation in terms of a combined objective of fairness and efficiency. They explained all three stages of their approach: simulation generates training partitions, then machine learning (ML) models are trained to provide a mortality oracle and a distance oracle, and last the balanced objective is optimized over ML model output. Using this approach, they create trade-off plots between different efficiency and fairness measures as decision-support tools for policymakers.
Dr. Larry Wein, Professor of Management Science at Stanford University, took over the stage and shared results from an interesting recent project “Analysis of the Number of Samples to Test from a Sexual Assault Kit (SAK),” joint work with Zhengli Wang, Kevin Macmillan, and Mark Powell. Dr. Wein started with an overview of applying operations management techniques to criminology questions. He gave examples of problems of reducing type I errors and type II errors. This study on SAK samples aims to help reduce type II errors, that is, avoid failing to convict sexual assault offenders. One serious issue with the current testing of SAKs is the existence of huge backlogs: nearly 400,000 SAKs are collected but not tested in the United States. Through cost-efficiency analysis, they conclude that “testing a SAK is a favorable, high-risk, huge-payoff lottery.” To ameliorate the backlog issue, their research proposes a machine learning model for choosing samples to test in SAKs. The samples are chosen to maximize the number of DNA generations that can be entered into CODIS, FBI’s DNA databases. Compared with the current selection based on the expertise of forensic nurses, their approach of combining police reports and a machine learning model is more efficient and effective in providing DNA yields.
The last talk of this session was given by Dr. Ronghuo Zheng, Professor of Accounting at the University of Texas at Austin. He presented joint work with Dr. Tinglong Dai and Dr. Katia P. Sycara about organ donation. Titled “Jumping the Line, Charitably: Analysis and Remedy of Donor-priority Rule,” this project looks into the impact of enabling donor priority on social welfare from organ allocation. One question motivating this study is why some people are registered organ donors while others are not. Their work uses the donor-priority rule: organ donors are prioritized in organ allocation, as the primary incentivizing scheme. To analyze the impacts of such a rule, they design a queuing structure which captures three-way trade-off among an abundance of organ supply, exclusivity of priority, and cost of donation. When the model only assumes heterogeneity in donating cost, they conclude that donor-priority rule improves social welfare—defined as the aggregate utilities of all individuals. When heterogeneity in the likelihood of requiring an organ transplant and organ quality is also incorporated, they show that the donor-priority rule can lower social welfare due to unbalanced incentives across different types of individuals. To rebalance the incentives and mitigate undesirable social welfare decrease, they use a freeze-period remedy and show that desirable mitigation can be achieved.