Informs Annual Meeting 2017

MA32

INFORMS Houston – 2017

MA32

We propose, optimize and validate a methodological framework for estimating the extent of crew-propagated delays and disruptions. We identify factors that influence propagation, and incorporate them into a robust crew scheduling model. We develop a fast heuristic for solving the inverse of this model to generate crew schedules similar to real-world scheduling samples. Extensive out- of-sample validation tests using four large real-world airline networks demonstrate that the crew schedules produced by our approach generate propagation patterns similar to those observed in the real world. Moreover, parameters calibrated for one network are found to perform reasonably well for other networks. 2 - It’s Not Just Luck: Evidence of Impact Operational Improvements and Buffers on Performance from the U.S. Airline Industry Milind Sohoni, Indian School of Business, Ac4 Level 1, Isb Campus Gachibowli Hyderabad 500032, Hyderabad, 500032, India, milind_sohoni@isb.edu, Vinayak V. Deshpande, Chandrasekhar Manchiraju Amongst others, airline on time performance (OTP) is impacted by three important factors: (i) Factors outside an airline’s control such as weather, (ii) passive actions taken by airlines to improve OTP such as schedule padding, and (iii) active actions taken by airlines to improve operations. We analyze how each of these three affect OTP and airline rankings. 3 - Estimating Occupancy Rates at Delta Sky Clubs Emmanuel Carrier, Delta Air Lines, 5155 Vernon Springs Trl NW, Atlanta, GA, 30327, United States, emmanuel.carrier@delta.com, Keith Becker, Liming Yao One of the most common complaint about airport lounges is over-crowding. However, many lounge operators do not have accurate estimates of actual occupancy as exits are typically not recorded. In this presentation, we explore how to estimate occupancy rates by leveraging existing data sources such as lounge entry and aircraft boarding scans. Our results show that a more precise estimate of occupancy rates can improve customer satisfaction in two ways: by managing demand across lounges at airports with multiple locations and by increasing revenues through dynamic pricing during off-peak periods. 4 - Competition Policy, Privatization, and Efficiency: Evidence From the Global Airline Industry Hao Su, University of Maryland-College Park, 3346 Van Munching Hall, College Park, MD, 20743, United States, haosu@rhsmith.umd.edu, Nicola Volta, Davide Scotti, Megersa Abate, Martin E. Dresner This paper will examine how competition policy and airline privatization affect the efficiency of global airlines. We apply stochastic frontier models to a panel dataset including the largest worldwide airlines in the period 2005-2015. Our results will reveal the impact of state ownership on airline efficiency, as well as how government competition policy, such as open skies, impacts efficiency. 351D Adversary-focused Military Applications Sponsored: Military Applications Sponsored Session Chair: Brian J. Lunday, Air Force Institute of Technology, Air Force Institute of Technology, WPAFB, OH, 45433, United States, brian.lunday@afit.edu 1 - A Bilevel Programming Model for Integrated Air & Missile Defense Location Planning Aaron M. Lessin, United States Air Force, 2950 Hobson Way, Department of Operational Sciences, WPAFB, OH, 45433, United States, aaron.lessin@afit.edu, Brian J. Lunday, Raymond R. Hill The U.S. military must understand antiaccess/area-denial strategies using air defense weapons systems. We propose a bilevel programming formulation to locate weapons systems to maximize the minimum exposure of the aerial penetration path. We compare the result to three alternative metric-specific penetration paths, and illustrate that the proposed asset location strategy is robust with respect to each intrusion metric. A sensitivity analysis is also conducted to examine the effect of several model parameters on solution quality and required computational effort. MA34

351B Expertise versus Skill in Human Prediction Invited: Forecasting and Prediction Invited Session

Chair: Pavel Atanasov, PhD, New York University, Pytho LLC, 866 President Street, Brooklyn, NY, 11215, United States, pavel@pytho.io 1 - On the Evaluation of Beliefs: A Method for Assessing Credibility in Subjective Probability Judgment Josh Baker, University of Pennssylvania, Philadelphia, PA, 19103, United States, jbak@sas.upenn.edu In the face of uncertainty, decision makers must base their choices on subjective beliefs. Consequently, decision making under uncertainty requires decision makers to infer the quality of the beliefs at their disposal. In the absence of an objective criterion, drawing such inferences can be challenging. The purpose of this research was to develop a method for assessing the quality of beliefs when normative benchmarks are unknown. We demonstrate that simple models of belief credibility (individuals vs. wisdom of the crowds) can be used to improve the accuracy of an individual’s beliefs ex ante.” 2 - The Robust Bayesian Truth Serum Jens Witkowski, ET.H.Zurich, Rotbuchstrasse 30, Zurich, 8037, Switzerland, jensw@inf.ethz.ch, David C.Parkes The Bayesian Truth Serum (BTS), due to Prelec (2004), allows the truthful elicitation of private signals (e.g., experiences, or opinions) in regard to a true world state when this ground truth is unobservable. However, BTS ensures proper incentives only in the limit, i.e. as the number of respondents n goes to infinity. We present the Robust Bayesian Truth Serum (RBTS), which is incentive compatible for every n ≥ 3, taking advantage of a particularity of the quadratic scoring rule. RBTS is the first peer prediction mechanism to provide strict incentive compatibility for every n ≥ 3 without relying on knowledge of the common prior. Moreover, RBTS is numerically robust and ex-post individually rational. 3 - Identifying Latent Structures in Human Performance Data using COREX KSM.Tozammel Hossain, USC-Information Sciences Institute, 4676 Admiralty Way #1001, Marina Del Rey, CA, 90292, United States, tozammel@isi.edu We present an approach to analyzing human performance data using correlation explanation. A key step in our method is to identify latent cognitive factors from users’ responses over various psychological, political, and skill tests. The proposed method harnesses these factors to design effective cognitive tests, forecast users’ problem-solving performance, and compose optimal team for tackling problems. We demonstrate the effectiveness of our method on forecasters’ performance data collected in a geopolitical forecasting tournament for Good Judgement Project. 4 - Small Steps to Prediction Accuracy Pavel Atanasov, New York University, New York, NY, 10007, United States, pavel@pytho.io Pavel Atanasov, Pytho, 866 President Street, Brooklyn, NY, 11215, United States, pavel@pytho.io, Jens Witkowski, Barbara Mellers, Lyle Ungar, Philip Tetlock Under-reaction and over-reaction to new information are the Scylla and Charybdis of belief updating. We show that highly accurate forecasters successfully navigate these hazards by making frequent, small belief updates of their probabilistic estimates. Updating frequency and magnitude were reliable, valid out-of-sample indicators of accuracy across four seasons of a geopolitical forecasting tournament. Frequent updaters tended to consume more news, and scored higher on cognitive reflection, numeracy and fluid IQ tests. Cognitive training caused smaller, more incremental belief updating. Update magnitude may indicate how forecasters process information from multiple sources. 351C Advanced Analytics Models in Aviation Sponsored: Aviation Applications Sponsored Session Chair: Shervin Beygi, Boeing, Seattle, WA, 98109-2155, United States, shervin@umich.edu 1 - Modeling Crew Itineraries and Delays in the National Air Transportation System MA33

Vikrant Vaze, Dartmouth College, 14 Engineering Drive, Murdough Center, Hanover, NH, 03755, United States, vikrant.s.vaze@dartmouth.edu, Keji Wei

144