Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TA13

4 - Does Higher Availability Lead to Higher Use? Understanding the Relationship Between Vaccine Stock Availability and Immunization Coverage in Nigeria. Prashant Yadav, PhD, Harvard Medical School, Seattle, WA, 98109, United States, Emily Gooding, Eirini Spiliotopoulou The impact of increases in availability on coverage has not been rigorously evaluated in literature. We use data from Nigeria and a linear mixed effects model to estimate the effect of vaccine availability on routine immunization coverage and to identify factors which affect this relationship. We find that vaccine stockouts significantly decrease the number of children immunized and that for most vaccines, the effectlasts for several months after a stockout. The magnitude of the impact varies by vaccine. 5 - Sourcing Uncertain Product Quality: Competition, Learning, and Volume Postponement Alexander Rothkopf, Massachusetts Institute of Technology, Cambridge, MA, United States, Felix Lauton, Richard Pibernik Global-health buyers seek to introduce competition by incenting new generics suppliers with uncertain performance/quality to enter the market. We investigate the benefits of postponing a portion of the award to learn entrant’s performance and how the dynamics of learning and competition depend on the size of the postponed volume. n TA13 North Bldg 126B Joint Session MSOM-Health/APS: Data, Learning, and Decision-Making Sponsored: Manufacturing & Service Oper Mgmt/Healthcare Operations Sponsored Session Chair: Mohsen Bayati, Stanford University, Stanford, CA, 94305, United States Co-Chair: Khashayar Khosravi, Stanford University 1 - Matrix Completion Methods for Causal Panel Data Models Khashayar Khosravi, Stanford University, Stanford, CA, United States, Susan Athey, Guido Imbens, Nikolay Doudchenko, Mohsen Bayati A central tool in empirical operations management is average treatment effect (ATE) estimation from observational data. In this setting, the presence of potential confounders leads to biased estimates of ATE. One way to reduce this bias is to use panel data models where a subset of units is exposed to a binary treatment during some time-periods and the goal is estimating the counterfactual outcomes for all treated units/time-period pairs. We study a class of estimators that minimize the distance between the estimated matrix and the original matrix, while favoring less complex models. We prove the consistency of our estimators by extending the existing results in the matrix completion literature. 2 - Optimized Prediction of Mortality (OPOM): A Novel Machine-learning Approach for Liver Transplant Allocation Yuchen Wang, Massachusetts Institute of Technology, Cambridge, MA, 02142, United States, Dimitris Bertsimas, Jerry L. Kung, Nikolaos Trichakis, Ryutaro Hirose, Vagefi Parsia Since 2002, the MELD score has been used to rank candidates for liver transplantation. However, despite numerous revisions, MELD allocation does not allow for fair access to all waitlisted candidates. We developed an Optimized Prediction of Mortality(OPOM) to allow for more equitable allocation of these scarce resources. OPOM uses state-of-the-art machine learning Optimal Classification Tree models that were trained to predict a candidate’s three-month waitlist mortality. OPOM exhibited the highest out-of-sample AUC and can decrease 405.92 deaths every year compared to the current rules. Improved survival was noted across all candidate demographics, diagnoses, and geographic regions. 3 - Question-design for Healthcare Plan Recommendation Jonathan Z. Amar, MIT, 503 Franklin St, Aprt 3, Cambridge, MA, United States, Nikos Trichakis, Chaitanya Bandi We develop a novel question-design mechanism in order to improve healthcare plan recommendations. Our algorithm is based on accurately estimating the differences of utilities associated with every plan. Rather than being agnostic to it, our formulation is driven by available assortment, and therefore achieves better performance for plan recommendation. We establish theoretical justification for our algorithm which outperformed state-of-the-art methods on real datasets.

n TA14 North Bldg 126C Managing Queues: Capacity, Information, and Pricing Sponsored: Manufacturing & Service Oper Mgmt/Service Operations Sponsored Session Chair: Philipp Afeche, University of Toronto, Toronto, ON, M5S 3E6, Canada Co-Chair: Luyi Yang, Johns Hopkins University, Baltimore, MD, 21202, United States 1 - Pricing and Capacity Decisions for Shared Service Systems Under Competition We consider service systems where customers’ utility depends on price as well as their service experience, which in turn depends on how crowded the service environment is and who the service environment is shared with. We investigate how two such systems under competition make pricing and capacity decisions under competition. 2 - Managing Two-sided Platforms with Self-scheduling Agents and Impatient Customers Rouba Ibrahim, University College London, MS&I Department, UCL, Gower Street, London, WC1E 6BT, United Kingdom We study the operational management of service platforms with self-scheduling agents and impatient customers. Since the customer impatience distribution plays an important role, we propose controlling it using delay announcements, and characterise the interaction between three controls at the manager’s disposal: the staffing level, the compensation offered to agents, and the announcements made to customers. 3 - The Economics of Line Sitting Luyi Yang, Johns Hopkins University, 100 International Drive, Baltimore, MD, 21202, United States, Shiliang Cui We study an emerging business model of line-sitting in which customers seeking service can hire others (line-sitters) to wait in line on behalf of them. We develop a queueing-game-theoretic model that captures the interaction among customers, the line-sitting firm, and the service provider to examine the impact of line-sitting on the service provider’s revenue and customer welfare. We also contrast line- sitting with the well-known priority purchasing scheme as both allow customers to pay extra to skip the wait. 4 - Pricing in a Two-sided Market with Time-sensitive Customers and Suppliers Philipp Afeche, University of Toronto, Rotman School of Management, 105 St. George Street, Toronto, ON, M5S 3E6, Canada, Mustafa Akan We consider a firm that matches stochastically arriving and time-sensitive customers and suppliers. We characterize and compare the structure and performance of the profit-maximizing and socially optimal pricing policies. n TA15 North Bldg 127A Revenue Management and Marketing for Online Retail Sponsored: Manufacturing & Service Oper Mgmt/Service Operations Sponsored Session Chair: David Simchi-Levi, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States Co- Chair: Clark C. Pixton, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States 1 - E-word of Mouth on Action: Analysis of Operational Decisions when Customers are Resentful Nesim K. Erkip, Bilkent University, Dept of Industrial Engineering, Ankara, 06800, Turkey, Bahar Cavdar In this talk, we explore the impact of electronic word-of- mouth (WoM) communication in an online shopping system where there are two types of customers, namely premium and regular customers. Our results reveal insights about the long-run behavior of customer demand and operational decisions under different WoM signals on the perceived quality of different service types and customer behaviors. Eda Kemahlioglu-Ziya, NC State University, Raleigh, NC, United States, Wei Gu, H. Sebastian Heese, Serhan Ziya

248