INFORMS 2021 Program Book

INFORMS Anaheim 2021

SD08

2 - Evidence-Based Policy Learning Jann Spiess, Stanford Graduate School of Business, Stanford, CA, 94305-8526, United States, Vasilis Syrgkanis The past years have seen the development of machine-learning algorithms to estimate personalized treatment-assignment policies from randomized controlled trials. Yet such algorithms often do not take into account that treatment assignments are frequently subject to hypothesis testing. In this project, we explicitly take significance testing of the effect of treatment-assignment policies into account, and consider assignments that optimize the probability of finding subsets with a statistically significant positive treatment effect. We provide an efficient implementation using decision trees, and demonstrate its gain over selecting subsets based on positive (estimated) treatment effects. Compared to standard tree-based regression and classification tools, this approach tends to yield substantially higher power in detecting positive treatment effects. 3 - Discovering Causal Models with Optimization: Confounders, Cycles, and Feature Selection Nur Kaynar, University of California-Los Angeles, Los Angeles, CA, 90049-5554, United States, Auyon Siddiq, Frederick Eberhardt The recent advancements in graphical approaches to causality have opened new opportunities to learn the underlying causal relations systematically from observational data. In this work, we propose a new method for causal structure discovery that allows for both unmeasured confounders and feedback cycles. Our new representation of the inference as an integer optimization enables us to solve instances in minutes that are intractable for current state-of-the-art methods. We then demonstrate how our method can be used to shed light on the validity of a proposed instrument in a simple and intuitive manner. SD08 CC Room 303C In Person: Social Media and Platform Economy General Session Chair: Luna Zhang, University of Washington-Tacoma, Tacoma, WA, 98402-3100, United States 1 - Analysis of Orange County 311 Non-emergency Call System During the Covid-19 Pandemic Duygu Pamukcu, Virginia Tech, Blacksburg, VA, United States, Christopher W. Zobel Local governments are responsible for maintaining necessary services and quickly and timely informing citizens before, during, and after an emergency. To do this, governments implement smart information and communication technologies in public services. This study examines the 311 non-emergency call system reactions in the U.S. metropolitan areas to reflect the COVID-19 pandemic. We investigate if the 311 system can capture citizen needs and complaints about the pandemic. 311 system collects and reports a highly structured and location-based dataset similar to social media data in some ways. We provide a comparative analysis with geolocated tweets from the same region to examine if the 311 system is a valuable source of information and if there are additional advantages of using the 311 system over social media data to identify service needs during a crisis. 2 - Digital Platforms and Race-Related Classroom Curriculum: Evidence from Black Lives Matter Ananya Sen, Carnegie Mellon University, Pittsburgh, PA, United States We study whether digital platforms can be a force for equality in the context of systemic racism. We use requests made by teachers on DonorsChoose.org as a measure of demand for race-related conversations in the classroom. We use the precise timing of high-profile police brutality events to identify the effect on race- related requests. We find a significant increase in race-related requests with the effect being driven by the killing of George Floyd in 2020. These requests are related to books written by Black authors and those that have Black protagonists. There are significant spillovers related to other minority communities such as Asians and Hispanics. The impact is higher for schools that witnessed a protest in the city. We find no polarization on the platform with requests coming from both Republican and Democratic zip-codes. 3 - Revenue-sharing Designs for Platforms Luna Zhang, University of Washington-Tacoma, Tacoma, WA, 98402-3100, United States, Hemant K. Bhargava, Kitty Wang Platforms motivate value creators by sharing platform revenue with them. Major platforms today use a linear revenue-sharing scheme. We explore alternative designs for revenue-sharing between the platform and creators, and apply our insights to the tensions that have arisen between platforms and creator-partners.

SD09 CC Room 303D In Person: Design and Control of Queues and Applications in Healthcare Systems General Session Chair: Jing Dong, Columbia University, New York, NY, 10027-6945, United States 1 - A New Approach to Global Stability of Multiclass Queueing Networks Feiyang Zhao, The University of Texas at Austin, Austin, TX, 78731-2434, United States, John Hasenbein, Itai Gurvich The focus of this research is on the global stability of stochastic processing networks, under a wide class of control policies. A framework for global stability is developed, in which the resources are given freedom to decide their own priority policy, under some general constraints. We offer a new approach for determining sufficient conditions for global Skorohod Problem (SP) stability, which builds on, and makes connections to, suitable robust optimization problems, with the collection of priority policies as the uncertainty set. We also show how global SP stability of a family of policies is inherited from the stability of static-priority policies. 2 - Skills-based Routing under Demand Surges Jinsheng Chen, Columbia University, New York, NY, 10027-6714, United States, Jing Dong, Pengyi Shi Many service systems employ dedicated staffing with cross-training to provide partial flexibility. Servers primarily serve specific classes of customers, but may serve other classes if necessary, at the cost of inefficiency. For example, during a pandemic, nurses trained in other specializations may be reassigned to take care of patients who have contracted the infectious disease.We consider a multi-class multi-pool parallel server system with partial flexibility, under general time- varying arrival rates. We derive near-optimal scheduling policies that minimize the sum of holding and “overflow” costs. Our policy is simple, intuitive, and makes use of future arrival rate information. SD10 CC Room 304B In Person: Cutting-Edge Methods for Data-Driven Decision-Making General Session Chair: Fei Fang, Duke University, Durham, NC, 27705-4547, United States 1 - Online Learning via Offline Greedy Algorithms: Applications in Market Design and Optimization Rad Niazadeh, Chicago Booth School of Business, Chicago, IL, 94305-5008, United States, Negin Golrezaei, Fransisca Susan, Joshua Wang, Ashwinkumar Badanidiyuru Motivated by online decision-making in time-varying combinatorial environments, we study the problem of transforming offline algorithms to their online counterparts. We focus on offline combinatorial problems that are amenable to a constant factor approximation using a greedy algorithm that is robust to local errors. For such problems, we provide a general framework that efficiently transforms offline robust greedy algorithms to online ones using Blackwell approachability. Demonstrating the flexibility of our framework, we apply our offline-to-online transformation to several problems at the intersection of revenue management, market design, and online optimization. 2 - Decision Forest: A Nonparametric Approach to Modeling Irrational Choice Yi-Chun Chen, UCLA Anderson School of Management, Los Angeles, CA, United States, Velibor Misic We propose a new nonparametric choice model that can represent any customer choice model, including those that are inconsistent with weak rationality. In the proposed model, each customer type is associated with a binary decision tree, which represents a decision process for making a purchase based on checking for the existence of specific products in the assortment. We theoretically characterize the model complexity and propose two practical estimation methods. Using real- world transaction data, we show that the proposed model outperforms benchmark models in out-of-sample predictive ability. We also demonstrate how the proposed model can extract insights about substitution and complementarity effects and identify interesting customer behaviors within a specific product category.

34