2015 Informs Annual Meeting

SB65

INFORMS Philadelphia – 2015

SB63 63-Room 112B, CC

3 - Using Decision Analysis and Large Data Analytics to Enhance Decision Quality Mazen Skaf, Partner & Managing Director, Strategic Decisions Group, 745 Emerson St, Palo Alto, CA, 94301, United States of America, mskaf@sdg.com Increasingly, in multiple domains, the use of large data analytics is enhancing the way we apply DA from identifying a decision opportunity, to framing, generating new alternatives, and evaluating alternatives. We present cases from various industries to illustrate how using large data analytics contributes to decision quality and how, in some situations enables a totally new approach to alternatives generation and strategy development. SB65 65-Room 113B, CC Quantifying Uncertainty in Decision Analysis Practice Sponsor: Decision Analysis Sponsored Session Chair: Christopher Hadlock, PhD Student, The University of Texas at Austin, 1 University Station Austin, TX 78713, Brad Powley, Senior Consultant, Strategic Decisions Group, 745 Emerson Street, Palo Alto, CA, 94301, United States of America, bpowley@sdg.com On occasion, it is useful for a decision analyst to encode a decision maker’s probability distribution with one or more infinite tails. However, an infinite tail of one probability distribution might be heavier than that of another, adding a wrinkle to probability encoding. This talk defines heavier tails and introduces a theory of tail behavior tailored to help a decision analyst encode a probability distribution with one or more infinite tails. 2 - Assessment Error and Discrete Approximations to Continuous Distributions Robert Hammond, Decision Analyst, Chevron, 1400 Smith St, Decision analyses often use continuous probability distributions, elicited from subject matter experts, and discretized for use in decision trees. In practice, cognitive biases reduce the precision of the assessments, and discretization introduces approximation error. We compare several discretization methods when the assessments are imprecise, and show that assessment precision often has significantly more impact on model performance than accuracy tradeoffs between discretization methods. 3 - Decision Making under Incomplete Information: Sequential Probability Assessment Heuristics Tao Huang, The University of Texas at Austin, 204 E. Dean Keeton Street, Stop C2200, ME department, ETC II, Room 5.152, Austin, TX, 78712, United States of America, huangt55@gmail.com, Eric Bickel Expected utility of an alternative is not explicitly computable in cases where probability mass function is partially known. Previous research has tried to establish dominance to solve the problem. In practice, however, dominance is rarely established and further probability assessments are needed. In this paper, we propose a novel way called Sequential Probability Assessment Heuristic that iteratively selects a feasible assessment based on a method from machine learning to solve this problem. 4 - Assessment-adaptive Discretizations (AAD) for Decision Analysis Practice Christopher Hadlock, PhD Student, The University of Texas at Austin, 1 University Station Austin, TX 78713, United States of America, cchadlock@austin.utexas.edu, Eric Bickel It is common practice in decision analysis to discretize continuous uncertainties into several points, and then assign probabilities to these points. Shortcuts pre- specify percentiles to assess, along with the probabilities to assign to each of them. However, shortcuts optimized towards matching the mean often differ from those optimized to match the variance. We present assessment-adaptive discretization methods, which adapt the probability assignments based upon the observed assessments. Houston, TX, United States of America, rhammond@chevron.com, Eric Bickel United States of America, cchadlock@austin.utexas.edu 1 - A Theory of Tail Behavior for Decision Analysis

Doing Good with Good OR I Cluster: Doing Good with Good OR Invited Session Chair: Lisa Maillart, University of Pittsburgh, Pittsburgh, PA, lisa.maillart@engr.pitt.edu Co-Chair: Itai Ashlagi, MIT, 100 Main St, Cambridge, MA, 02139, United States of America, iashlagi@mit.edu 1 - Optimal Policy Design to Motivate Blood Donation: Evidence from a Randomized Field Experiment and a Structural Model Tianshu Sun, University of Maryland Smith School of Business, 3330 Van Munching Hall, College Park, MD, 20740-2840, United States of America, tianshusun@rhsmith.umd.edu Using a randomized field experiment involving 80,000 participants, we test the effect of different policies in driving donation. We find: blood banks can use group reward to motivate group formation to increase donation; such group reward is four times more cost effective than individual reward. We build a structural model and perform simulations to identify optimal incentive and targeting strategy. 2 - Improving Blood Collection Policies for Cryoprecipitate Chenxi Zeng, Georgia Tech, Atlanta, GA, United States of America, czeng8@gatech.edu, Turgay Ayer, Chelsea White Iii, Roshan Vengazhiyil, John Deshane Working closely with the American Red Cross (ARC), we have developed and analyzed a donor collection model for whole blood that is to be processed into cryoprecipitate (cryo), a critical blood product for controlling massive hemorrhaging. Our numerical results show that the proposed solution approaches may reduce the expected cryo collection cost by up-to 70%, compared with the current practice. Implementation of the model at the ARC supports our estimates. 3 - Ebola Treatment Facility Location Planning in Guinea Chu Qian, Georgia Tech, GA, United States of America, qianchu31@gmail.com, Charmaine Chan, Matt Daniels, Javiera Javieria, Caleb Mbuvi, Ivan Renaldi, Jonathan Sutomo, Kimberly Adelaar In the recent Ebola outbreak, treatment facilities were critical but beds were unavailable in some areas while unused in others. A spatial simulation was built to project the spread within Guinea, overlaid with heuristics on when and where to place treatment facilities. Results showed units set up quickly or in advance could have saved more than 2000 lives. SB64 64-Room 113A, CC Applications of Decision Analysis & Large-Scale Data Analytics Sponsor: Decision Analysis Sponsored Session Chair: Mazen Skaf, Partner & Managing Director, Strategic Decisions Group, 745 Emerson St, Palo Alto, Ca, 94301, United States of America, mskaf@sdg.com 1 - Expert Calibration and Elicitation for Large Scale Investment Decisions Saurabh Bansal, Assistant Professor, Penn State University, 405 Business Building, University Park, PA, 16802, United States of America, sub32@psu.edu, Genaro Gutierrez We discuss a new approach for the use of expert calibration and elicitation for estimating probability distributions. The approach has been in use since last two years at a Fortune 500 firm for making an annual large-scale investment decision. Practical insights for expert elicitation are also discussed. 2 - Reducing Risk and Improving Incentives in Funding Entrepreneurs Samuel E. Bodily, Professor, Darden Graduate Business School, Univ. of Virginia, 100 Darden Boulevard, Charlottesville, VA, 22903, United States of America, BodilyS@Darden.virginia.edu Backer financing mechanisms that encourage an entrepreneur are identified, risk analysis models are developed, and insights are obtained about how mechanisms (e.g. equity capital, incentive gifts, insurance, revenue contracts, and derivative swaps) can best reduce risk to the entrepreneur and give proper incentives, at given cost to the backers. Attention is given to avoiding problems of moral hazard and providing proper incentives. Certainty equivalents for financing alternatives are derived.

87