2015 Informs Annual Meeting

TB29

INFORMS Philadelphia – 2015

TB27 27-Room 404, Marriott

questions about the job. Although the ordeal positively selected candidates, it was the information in the answers that mattered for match formation. Although the overall number of matches and speed to fill a vacancy was unchanged, employers engaged in less recruiting activities and formed higher quality matches. 2 - Experiments as Instruments: Heterogeneous Position Effects in Sponsored Search Auctions

Multiple Criteria Decision Aiding Sponsor: Multiple Criteria Decision Making Sponsored Session Chair: Roman Slowinski, Prof., Poznan University of Technology, Pl. Marii Sklodowskiej-Curie 5, Poznan, PL, 60-965, Poland, roman.slowinski@cs.put.poznan.pl 1 - FITradeoff: Flexible and Interactive Tradeoff Elicitation Procedure Adiel T. DeAlmeida, Professor, Universidade Federal de Pernambuco, Caixa Postal 7462, Recife, PE, 50630-971, Brazil, almeidaatd@gmail.com, Adiel De Almeida Filho, Jonatas Araujo De Almeida, Ana Paula Costa FITradeoff is a Flexible and Interactive Tradeoff elicitation procedure for multicriteria additive models in MAVT scope. The classical tradeoff procedure is one of the approaches with strongest theoretical foundation. However, behavioral studies have shown inconsistences of DM during elicitation. The FITradeoff reduces DM’s effort in the process, by using partial information, thereby contributing for reducing inconsistences. It is implemented in a DSS, which is illustrated by applications. 2 - An Enhanced “Election Machine” for the Finnish Parliamentary Elections: Theory and Implementation Jyrki Wallenius, Professor, Aalto University School of Business, Runeberginkatu 22-24, Helsinki, Finland, jyrki.wallenius@aalto.fi, Tommi Pajala, Akram Dehnokhalaji, Pekka Korhonen, Pekka Malo, Ankur Sinha Web-based questionnaires to match candidates’ and voters’ views play an important role in Finland. We have collaborated with Helsingin Sanomat, who runs the most influential of such questionnaires, to enhance and further develop it. Our algorithm was tested in last April’s Parliamentary Elections. We describe our algorithm and the feedback. 3 - Multicriteria and Multiobjective Models for Risk, Reliability and Maintenance Context Rodrigo J P Ferreira, Assistant Professor, Universidade Federal de The use of multiple criteria and multiobjective models in risk, reliability and maintenance research has increased in recent years. These models may affect the strategic results of any organization, as well as, human life and the environment. In such situations, optimal solutions for one objective function cannot be suitable. These issues are presented according to the reference Multicriteria and Multiobjective Models for Risk, Reliability and Maintenance Decision Analysis. 4 - Constructive Preference Learning in Value-driven Multiple Criteria Sorting Roman Slowinski, Prof., Poznan University of Technology, Pl. Marii Sklodowskiej-Curie 5, Poznan, PL, 60-965, Poland, roman.slowinski@cs.put.poznan.pl, Milosz Kadzinski, Krzysztof Ciomek We present an interactive preference learning technique for multiple criteria sorting driven by a set of additive value functions compatible with a rich preference information acquired from the user. This information may include: (1) imprecise assignment examples, (2) desired class cardinalities, and (3) assignment-based pairwise comparisons. The output results are necessary and possible assignments, and extreme class cardinalities. TB28 28-Room 405, Marriott Empirical Market Design Cluster: Auctions Invited Session Chair: Peng Shi, MIT Operations Research Center, 1 Amherst Street, E40-149, Cambridge, MA, 02139, United States of America, pengshi@mit.edu 1 - Market Congestion and Application Costs John Horton, Assistant Professor, NYU Stern School of Business, 44 West Fourth Street, Kaufman Management Center, New York, NY, 10012, United States of America, John.Horton@stern.nyu.edu, Ramesh Johari, Dana Chandler We report the results of an experimental intervention that increased the cost of applying to vacancies in an online labor market by requiring workers to answer Pernambuco, Av. Professor Morais Rego, 1235., Recife, PE, 50670-901, Brazil, rodjpf@gmail.com, Adiel T De Almeida, Cristiano A V Cavalcante, Marcelo H Alencar, Adiel De Almeida Filho, Thalles V Garcez

Justin Rao, Researcher, Microsoft Research, 641 Avenue of Americas, New York, NY, 10014, United States of America, Justin.Rao@microsoft.com, Matthew Goldman

The Generalized Second Price auction has been shown to achieve an efficient allocation and favorable revenue properties provided the causal impact of ad position on user click probabilities is a constant the scaling factor for all ads. We develop a novel method to re-purpose internal business experimentation at a major search engine and we strongly reject the conventional multiplicatively- separable model, instead finding substantial heterogeneity of the causal impact of position on CTR. 3 - Optimal Design of Two-sided Market Platforms: An Empirical Case Study of Ebay Brent Hickman, Assistant Professor Of Economics, University of Chicago, 1226 E 59th St, Chicago, IL, 60637, United States of America, hickmanbr@uchicago.edu, Joern Boehnke, Aaron Bodoh-creed We investigate design of platform markets that house many auctions over time. We combine a unique dataset with a model of bidding where the option value of re-entering the market creates incentive for buyers to shade bids below private valuations in the current period. We show the model is identified using the Bellman equation for a representative bidder. We estimate the model and investigate the degree to which eBay is able to reduce transaction costs and approach the efficient allocation. 4 - Stability of Demand Models Across Policy Reforms: An Empirical Study with Boston Public Schools Peng Shi, MIT Operations Research Center, 1 Amherst Street, E40-149, Cambridge, MA, 02139, United States of America, pengshi@mit.edu, Parag Pathak In counterfactual analysis using demand modelling, an important but seldom checked assumption is that the proposed reform does not affect the demand model. We validate this assumption across a major school choice reform in Boston in 2014. To control for post-analysis bias, we precommit to forecasts before the reform. We find that while our prediction of the number of applicants were off, the logit and mixed-logit demand models we fit were stable before and after the reform. Chair: Tarun Mohan Lal, Mayo Clinic, mohanlal.tarun@mayo.edu 1 - Combating Attrition through New Developments in Transaction Analytics and Customer Dialogue Gerald Fahner, Analytic Science Senior Director, FICO, 181 Metro Drive, San Jose, United States of America, geraldfahner@fico.com ”Silent” attrition remains a costly problem requiring fast detection and insight to create effective retention offers. Our credit card case study shows how ensemble models instrumented with low-latency transaction features rapidly detect card- level and merchant category-level attrition. We explain our models and relate performance to profitability. We show how to boost persuasiveness of offers by customer dialogues to learn their preferences. Using a simulation we illustrate the value of dialogue. 2 - How Bringing Decision Optimization to the Cloud Will Democratize Optimization Susara Van Den Heever, IBM France, 1681 Route des Dolines, France, svdheever@fr.ibm.com, Xavier Ceugniet, Alain Chabrier, Stéphane Michel Even though Decision Optimization has been used effectively across industries for decades, it remains under-utilized. Complexity and costs are often cited as barriers to wider adoption. The emergence of cloud computing, as well as the renewed emphasis on cognitive analytics platforms, breaks down these barriers to bring the benefits of optimization to a wider audience. We will demonstrate this vision through a case study involving IBM Decision Optimization on Cloud, and IBM Watson Analytics. TB29 29-Room 406, Marriott Applications of Analytics II Sponsor: Analytics Sponsored Session

293