Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WA39

3 - Assigning Priorities (or Not) in Service Systems with Nonlinear Waiting Costs Nilay Argon, PhD, University of North Carolina, Chapel Hill, NC, United States, Huiyin Ouyang, Serhan Ziya For a single-server queueing system with two types of customers and nonlinear waiting costs, we compare static queueing rules that use information on customers’ types and order of arrival. Our main theorem ranks the type-based priority policies and first-come-first-serve rule according to their long-run average waiting costs under general cost functions. We then apply this result to polynomial cost functions to generate insights into when static prioritization is advantageous and when it is beneficial to consider state-dependent priority policies. 4 - Optimal Control Policies for an M/M/1 Queue with a Removable Server and Dynamic Service Rates Pamela Badian-Pessot, Cornell University, Ithaca, NY, 14850, United States, Mark E. Lewis, Douglas Down We consider an M/M/1 queue with a removable server that dynamically chooses one of two service rates. If the server is off, the system must warm up for a random, exponential amount of time, before it can begin processing jobs. We show under the average cost criterion, that work conserving policies are optimal. We then demonstrate the optimal policy can be characterized by two thresholds, one for turning on the server and another for the service rate. Finally, we explore the implications of such policies for system design. n WA39 North Bldg 226A Queueing Models for Healthcare Systems Sponsored: Applied Probability Sponsored Session Chair: Ohad Perry, Northwestern University, Evanston, IL, 60208, United States 1 - Run-through Experiments for Queues Harsha Honnappa, Purdue, Anna Tatara We develop a simulation and theoretical framework for selecting a fitted traffic model for a queueing system. Given a dataset of inter-arrival times that displays time-of-day effects, it is often the case that a modeler wishes to fit a non- homogeneous Poisson process to this dataset. The main question we address in this talk is how to select appropriate time-buckets to aggregate the arrivals in order to fit the Poisson model, by using a simulation estimate of an appropriate performance metric as a measure of ‘goodness of fit.’ 2 - Efficiently Selecting Arrival Models Zeyu Zheng, Stanford University, Stanford, CA, 94305, United States In modeling non-stationary and time-inhomogeneous data, a standard approach is to break the time horizon into multiple intervals, and then to fit a piecewise model. We propose a principled maximum likelihood approach that includes efficient model fitting, optimal interval boundary selection, and selection of model degree. 3 - Optimal Scheduling in Presence of Proactive Care Yue Hu, Columbia University, New York, NY, 10027, United States, Carri Chan, Jing Dong Healthcare is a limited resource environment where capacity is often reserved for needy patients. A new trend in medicine is the use of preventative care models that treat patients early to eliminate the need for more expensive resource later. We study the optimal scheduling policy for a two-class multi-server queueing model to understand how patients should be prioritized for care when preventative care is an option. We study both the transient system dynamics and the steady-state performance. In the first case, we analyze the most cost effective way to bring the system back to equilibrium when it is started far from “normal operation”. In the second case, we minimize the long-run average holding cost. 4 - A Fluid Diffusion Hybrid Limit for Service Systems with Fast and Slow Customers Lun Yu, Northwester University, Evanston, IL, 60208, United States Motivated by emergency departments, we consider a queueing system with two customer classes; the first class (of “high acuity”) requires long service times and receives priority over the second class (“low acuity”), whose average service times are substantially shorter. Unfortunately, the dynamics of such a system are intractable, and existing heavy-traffic regimes cannot capture the fact that, in practice, a non-negligible proportion of the arrivals from either class must wait for service. We propose a Fluid-Diffusion Hybrid limit to approximate the two queues, and demonstrate how it can be employed to study the benefits of de- pooling (namely, of having a “fast-track”).

n WA40 North Bldg 226B Decision Analysis and Research Study Design Sponsored: Decision Analysis Sponsored Session Chair: Saurabh Bansal, Penn State University, State College, PA, 16801, United States 1 - Qualitative Not as ôLesser Than: Guidelines for Deciding Between a Qualitative and Quantitative Decision Approach Kara M. Morgan, Quant Policy Strategies, LLC, 8375 Nemain Loop, Dublin, OH, 43016, United States In practice, decision methodologies can drive decisions, and biases against using qualitative methods can limit the options considered by decision analysts. There is a common conception that qualitative research is more appropriate for exploratory research rather than for informing decision making. In fact, there are many general cases in which taking a quantitative approach can obscure the issues and therefore reduce the potential impact of decision analysis on the decision. This talk will identify conditions under which qualitative methods could lead to better decisions and provide pointers for utilizing rigorous qualitative methods for decision making support. 2 - Research Methodologies and the Field of Decision Analysis Jeffrey M. Keisler, University of Massachusetts-Boston, M/5 249, 100 Morrissey Boulevard, Boston, MA, 02125, United States This presentation considers three questions. (1) What is a research methodology as opposed to any old methodology and what is the role of research methodologies in scholarship? (2) In what ways do and in what ways can decision analysis scholars incorporate common research methodologies? (3) How and why might we think of decision analysis itself as a research methodology? 3 - Accelerating Learning using Expert Judgment about the Relationship between Model and Reality Eva D. Regnier, Naval Postgraduate School, 555 Dyer Road, Bldg. 330, Room 287, Monterey, CA, 93943, United States, Melissa A. Kenney, Michael Gerst Because of their power to influence decisions, it’s useful to consider how the modeling process can be improved by formalizing the role of expert judgments in building, validating, and interpreting models. Bayesian reasoning - asking what model output we ought to expect as a function of the possible states of the modeled system - is rarely applied in mechanistic (vs. statistical) models. By formalizing expert judgments regarding the joint relationship between reality and model output, our framework (i) suggests new questions that should be asked in the model building and validation process and (ii) guides how model outputs should be interpreted, resulting in higher-value models and inferences. 4 - Using Software to Help Engage with Clients Max Henrion, Lumina Decision Systems, Inc, Campbell, CA, United States We will share experience on how decision analysts can software tools to facilitate understanding and analyzing new problems, by using influence diagrams to frame and bound decision problems, an agile modeling process driven by interactive sensitivity analysis, and model exploration and visualization to help clients get a visceral understanding of how decisions affect the objectives they care about. n WA41 North Bldg 226C DA Arcade I Sponsored: Decision Analysis Sponsored Session Chair: William Nicholas Caballero, Air Force Institute of Technology, 6465 Hemingway Road, Dayton, OH, 45424, United States 1 - Valuation of Strategic Adaptation Byunghee Choi, Penn State University, University Park, PA, United States, Robert D. Weaver We introduce a continuous time economic model of multiple stage processes with fixed stage commitment. Stage-based commitment is made within the context dynamic evolution of uncertainty with respect to process performance. Within this context, we introduce a method for valuation of strategic adaption to change in uncertainty and information evolution. We illustrate our approach with simulation.

424