2016 INFORMS Annual Meeting Program

MB34

INFORMS Nashville – 2016

MB32 203A-MCC

model includes an estimate of predictive variance. We focus on evaluating the reliability of this predictive error from these various packages. When fitting the same data and model, the run times of certain packages can also differ by orders of magnitude. The study takes a practitioners point of view, using each package with minimal tuning. 3 - Time Management Policies In A Queueing System Ji-Eun Kim, PhD Student, The Pennsylvania State University, Imperial Towers, University Park, PA, 16801, United States, jxk594@psu.edu, David A. Nembhard, Hyeong Suk Na Many job assignment problems are organized from a company’s perspective to meet the demands of a schedule or to maximize workers’ productivity, often ignoring the heterogeneity of pacing styles among workers. We show that if one considers the diversity in pacing styles, system productivity can be increased using one or more approaches. The purpose of this study is to test job assignment policies to be used in a queueing system considering servers’ diversity in deadline reactivity. Empirical course website data was used to test a range of job assignment policies. 4 - Coordinating Station And Network Capacity In Urban Rail Transit System Li Li, Southwest Jiaotong University, No.111, North Section 1, Second Ring Road, Chengdu, China, speciallili@home.swjtu.edu.cn, Haifeng Yan, Gongyuan Lu, Wu You The performance of urban rail transportation is impacted by fluctuated passenger demand due to both the capacity constraint of station and line. The feature of high accessibility and volume makes a well coordinated train line plan in urban rail network very hard to be achieved. This research will present a stochastic integer programming model to demonstrate the mutual influence between passenger demand and train line plan. This model is solved by a simulation based approach which is applied in a real-world case in Chongqing Rail Transit Company. 204-MCC Simulation and Stochastic Optimization Sponsored: Manufacturing & Service Oper Mgmt, Healthcare Operations Sponsored Session Chair: Douglas Morrice, University of Texas-Austin, 2110 Speedway Stop B6500, Austin, TX, 78712-1750, United States, douglas.morrice@mccombs.utexas.edu 1 - Multimodularity In The Stochastic Appointment Scheduling Problem With Discrete Arrival Epochs Christos Zacharias, Assistant Professor, University of Miami, Coral Gables, FL, United States, czacharias@miami.edu, Tallys Yunes We address the problem of designing appointment scheduling strategies that account for patients’ no-show behavior, non-punctuality, emergency walk-ins and random service times. We maintain the discrete nature of the appointment scheduling problem by considering arrival epochs with discrete supports. We demonstrate that the optimal scheduling strategy minimizes a multimodular function, and a local search algorithm terminates with a globally optimal solution. 2 - Appointment Scheduling With Multiple Providers And Stochastic Service Times Michele Samorani, Santa Clara University, Santa Clara, CA, United States, samorani@ualberta.ca, S Abolfazl Soltani, Bora Kolfal We consider a multi-server appointment scheduling problem in which patients may not show up, and those who show up require stochastic service times. We model this problem as a Markov Chain and solve it through complete enumeration. Then, we employ statistical learning techniques to detect patterns among optimal solutions. We develop an effective heuristic method which uses these patterns to build near-optimal solutions. Our numerical experiments show that our methods result in higher-quality schedules than those obtained by existing models. We also test our heuristic with a field experiment made in collaboration with a local legal counseling clinic afflicted by high service time variability. 3 - Coordinated Appointment Scheduling Of An Integrated Practice Unit Douglas Morrice, The University of Texas, Austin, douglas.morrice@mccombs.utexas.edu, Dongyang Ester Wang, Kumar Muthuraman, Jonathan F Bard, Luci Leykum, Susan Noorily In this research, we develop a coordinated approach to patient appointment scheduling that enables a patient to receive multiple services on a single visit. The approach is compared to heuristics used in practice. A case study in pre-operative care involving the integration of Anesthesiology and Internal Medicine is used to motivate and illustrate the results. MB34

Structural Estimation in Operations Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Gabriel Weintraub, Columbia University, New York, NY, United States, gweintraub@columbia.edu 1 - The Efficacy Of Incentives In Scaling Up Marketplaces Ashish Kabra, INSEAD, Boulevard de constance, Fontainebleau, 77305, France, ashish.kabra@insead.edu, Elena Belavina, Karan Girotra Marketplaces spend billions in incentives to achieve scale, which is key to the efficacy and survival of marketplaces. Using detailed transaction data from a leading transportation marketplace, we estimate and compare the effects of incentives given to the “buyer” side and “seller” side of the marketplace as well as the effect of threshold and linear incentives. 2 - Spatial Competition And Preemptive Entry In The Discount Retail Industry Fanyin Zheng, Columbia Business School, fz2225@gsb.columbia.edu This paper studies how discount retailers make store location decisions by estimating a dynamic game model. It extends the empirical models of dynamic oligopoly entry by allowing for spatially interdependent entry and introducing machine learning tools to infer market divisions from data. The results suggest that preemptive incentives are important in chain stores’ location decisions and that they lead to loss of production efficiency. 3 - Ergodicity And The Estimation Of Markov Decision Processes Robert Bray, Kellogg, r-bray@kellogg.northwestern.edu I create a class of dynamic discrete choice estimators that exploit Markov chain ergodicity. The empirical likelihood of a Markov decision process depends only on the differences in the value function. And whereas the value function converges with Bellman contractions at the rate of cash flow discounting, the value function differences converge at the rate of cash flow discounting times the rate of Markov chain mixing (the subdominant eigenvalue of the state transition matrix). With this strong convergence result, I make Rust’s (1987) nested fixed point (NFXP) estimator 200 times faster in problems with more than 2,000 states. 4 - When Demand Projections Are Too Optimistic: A Structural Model Of Product Line Decisions Andres I Musalem, U. de Chile / Complex Engineering Systems Institute, Beauchef 851, Santiago, 8370456, Chile, amusalem@duke.edu A methodology is proposed to estimate structural models of product line competition. Not accounting for this endogeneity leads to overoptimistic estimates of demand due to a sample selection bias, which may generate misleading managerial recommendations. The methodology is illustrated using simulated and real data. Chair: Li Li, Southwest Jiaotong University, No.111, North Section Second Ring Road, Chengdu, China, speciallili@home.swjtu.edu.cn 1 - Operationalizing Industry Cluster Strategies Tayo Fabusuyi, Numeritics & Carnegie Mellon University, 5520 Baywood Street, Floor #3, Pittsburgh, PA, 15206, United States, tfabusuyi@cmu.edu, Juergen Pfeffer Local economic development organizations are often tasked with promoting the health of the regional economy. However, the unique composition of each geographical area calls for a distinct approach that reflects the peculiarities of the local economy. We present an approach by which the information in input- output is modeled and enriched using network analysis. Using a simulated policy intervention, we show how the approach can provide insight on regional economies and provide an application to industry cluster analysis. 2 - A Comparison Of Gaussian Process Modeling Software Collin Erickson, Northwestern University, 2145 Sheridan Road, Room C210, Evanston, IL, 60208, United States, collinerickson@u.northwestern.edu, Bruce Ankenman, Susan M Sanchez We have found that different software packages can give different results when fitting the same Gaussian process model, often called kriging. We compare various packages on a variety of test problems, finding that the accuracy of predictions can differ significantly. An attractive feature of Gaussian process fitting is that the MB33 203B-MCC Simulation II Contributed Session

159