Informs Annual Meeting 2017

SA51

INFORMS Houston – 2017

SA49

2 - Spatio-temporal Analysis of Urban Mobility of Transit Users from Smart Card Transactions Data Laiyun Wu, SUNY-Buffalo, 326 Bell Hall, University at Buffalo, Buffalo, NY, 14226, United States, laiyunwu@buffalo.edu, Jee Eun Kang, Samiul Hasan Characterizing individual trajectories is critical to understand the dynamics of human mobility and design reliable mobility models. The recent availability of Automatic Fare Collection (AFC) data has diversified the ways to understand human mobility, especially for the patterns of different type of transit users (e.g., regular workers, students, seniors and disabled persons). In this paper, we measure travel distance, travel time and stay time at activity locations for various user groups and present the statistical distributions of these characteristics. 3 - Bayesian Optimization to Fuse Large Scale Transportation Data Sets into Simulation Models Laura Schultz, George Mason University, Fairfax, VA, 22032, United States, lschult2@gmu.edu, Vadim Sokolov, Joshu Auld, Dominik Karbowski, Aymeric Rousseau In this paper we present Bayesian optimization techniques for calibrating large scale transportation models. Specifically, we address the problem of updating integrated demand-network models using dynamic traffic flow data, collected on freeways. We present several Bayesian optimization techniques and a distributed data flow framework, that allows scalable execution on highly parallel architectures. We demonstrate our approach using empirical results from the calibration of Chicago’s metropolitan area transportation model, with 10 million travelers and 28 million daily trips. 361D The Structural Modeling and Machine Learning Applications for Issues in Information Systems Sponsored: Information Systems Sponsored Session Chair: Lin Hao, University of Notre Dame, Notre Dame, IN, 46556, United States, lhao@nd.edu 1 - How Much is an Image Worth? Airbnb Property Demand Estimation Leveraging Large Scale Image Analytics Shunyuan Zhang, Carnegie Mellon University, 5562 Hobart Street, Apartment 614, Pittsburgh, PA, 15217, United States, shunyuaz@andrew.cmu.edu, Dokyun Lee, Param Vir Singh, Kannan Srinivasan We investigate the economic impact of images and lower-level image factors that influence demand in Airbnb. Using DiD analyses on a nine-month Airbnb panel data of 8211 properties, we find units with verified photos generate approximately 7% more demand. Leveraging computer vision techniques to classify image quality, we show 52.5% of this effect comes from high image quality. Next, we identify 12 human-interpretable image attributes and evaluate their economic impacts. Results suggest direct impacts of the attributes on demand. The study contributes to and bridges literature which traditionally ignored either demand side (photography) or systematic image characterization (marketing). SA51 This paper studies how online learning users dynamically adjust their learning plan every day given their preset deadlines, learning history, and peer effect. We develop a dynamic structural model in which users maximize utility considering the intertemporal trade-off between current and future learning amount. Theoretical properties enable us to distinguish forward-looking from myopic users, and we empirically test for forward-looking consumption patterns. We use a two-stage method to estimate structural parameters, and conduct counterfactual simulations to see how to help users learn better. 3 - Reward-based Crowdfunding Platform Design Shahryar Doosti, University of Washington, Mackenzie Hall, Seattle, WA, 98195, United States, shahryar@uw.edu, Yong Tan, Shengsheng Xiao With the emergence of reward-based crowdfunding platforms, there are a growing number of projects benefiting from crowd participation in their fund- raising process. With increasing competition, failed campaigns have been also increased. At the same time, there are multiple choices for entrepreneurs to design their campaigns in terms of design, reward scheme, and goal which can affect the success of the project. In this study, we use different methods of data mining to shed light on the important design questions in front of fund-seekers. 2 - A Dynamic Structural Model of Online Learning: Heterogeneous Studying Plan and the Peer Effect Ruoxin Zhou, Peking University, Beijing, China, zhourx@pku.edu.cn, Fei Ren

361B Behavioral Aspects of Judgment and Forecasts Invited: InvitedBehavioral Aspects of OR Invited Session Chair: Dilek Onkal, Bilkent University, Ankara, 06800, Turkey, onkal@bilkent.edu.tr 1 - The Effects of the Availability Bias on Facebook Users’ Perception of Privacy Risks Over Time Kathleen M. Whitcomb, University of South Carolina, Moore School of Business, Management Science Dept, Columbia, SC, 29208, United States, whitcomb@moore.sc.edu Past studies have shown that a short-term increase in risk perception for an activity or event can be achieved by exposing individuals to a vivid image known to induce negative affect. This effect has been demonstrated even when the image is unrelated to the activity. It has been hypothesized that the increase in risk perception generated by a vivid image inducing negative affect persists longer when the image is relevant to the target activity. We report the results of an experiment which demonstrates that a relevant negative image results in both a more enduring and greater increase in risk perception than does an irrelevant negative image. 2 - Cognitive Debiasing Training in Judgmental Frecasting: An Experimental Study Jorge Alvarado Valencia, PhD, Colombia, Pontificia Universidad Javeriana, Bogota, Colombia, jorge.alvarado@javeriana.edu.co, Jaime Vargas, Lina Merchan In order to reduce well-known cognitive biases in judgmental time series forecasting (noise modelling, small unjustified adjustments and overoptimism) a short cognitive debiasing training was tested against forecasting techniques training and no training at all with students in a judgmental adjustment task . Results show that cognitive debiasing training reduced small unjustified adjustments, but did not reduce noise modelling and was no better than absence of training. Forecasting techniques training led to a reduction in accuracy, mainly due to unjustified adjustments. 3 - Effects of Scenario Tone on Judgmental Forecast Adjustments Dilek Onkal, Bilkent University, Faculty of Business Administration, Ankara, 06800, Turkey, onkal@bilkent.edu.tr, Paul Goodwin, M. Sinan Gonul When presented with forecasts, users typically respond by judgmentally adjusting these predictions to incorporate their experience, intuition, and informational asymmetries. However, such adjustments may not always improve forecast accuracy; hence the need for support tools to improve communication and information sharing between forecasters and decision-makers. This research investigates the effects of providing differently-toned scenarios as forecast advice on individual and group-based judgmental predictions. Findings suggest important directions for designing and implementing effective forecast management systems to support both the providers and users of forecasts. 361C Large-scale Data Analytics in Urban Transportation Modeling I Sponsored: TSL, Urban Transportation Sponsored Session Chair: Shanjiang Zhu, George Mason University, Fairfax, VA, 22030, United States, szhu3@gmu.edu Co-Chair: Zhen Qian, Carnegie Mellon University, Pittsburgh, PA, 15213, United States, seanqian@cmu.edu 1 - Assessing Travel Behavior Responses to Washington Metro SafeTrack Project using Smartphone App Data Shanjiang Zhu, George Mason University, Nguyen Engineering Building, Suite 1300, 4400 University Drive Ms 6c1, Fairfax, VA, 22030, United States, szhu3@gmu.edu, Zhuo Yang, Lei Zhang This study investigates the behavioral reactions to the Washington Metro SafeTrack Project, a series of accelerated safety projects that required either single- track operations, or complete closure of a line segment. Longitudinal travel trajectories data were collected using a smartphone app among transit riders who used to use metro before the service disruptions. Algorithms were developed to infer the travel choices before, during, and after the network disruptions, which are crucial for assessing the impact of the unprecedented maintenance work at the Washington Metro. Findings from this study could help to inform agencies who are struggling with the aging infrastructure in the US. SA50

19