2015 Informs Annual Meeting

TD74

INFORMS Philadelphia – 2015

TD72 72-Room 203A, CC Design and Analysis of Data with Complex Structure Sponsor: Quality, Statistics and Reliability Sponsored Session Chair: Xinwei Deng, Assistant Professor, Department of Statistics, Virginia Tech, 211 Hutcheson Hall, Blacksburg, VA, United States of America, xdeng@vt.edu Co-Chair: Ran Jin, Virginia Tech., Grado Department of Industrial and, Systems Engineering, Blacksburg, VA, 24061, United States of America, jran5@vt.edu 1 - Markov Switching Autoregressive Models with Applications in Cell Biology Ying Hung, Rutgers University, Piscataway, NJ, United States of America, yhung@stat.rutgers.edu We will introduce a new framework based on Markov switching autoregressive models for the analysis of experiments in cell biology. 2 - Sparse Particle Filtering Yun Chen, University of South Florida, 4202 E. Fowler Ave. ENB118, Tampa, FL, United States of America, yunchen@mail.usf.edu, Hui Yang Wireless sensor network has emerged as a key technology for monitoring space- time dynamics of complex systems. Distributed sensing gives rise to spatially-temporally big data. Realizing the full potentials of distributed sensing calls upon the development of space-time modeling of measured signals in dynamically-evolving physical environment. This paper will present a new approach of sparse particle filtering to model spatiotemporal dynamics of big data in distributed sensor network. 3 - Graphical Modeling with Functional Variables Ran Jin, Virginia Tech., Grado Department of Industrial and, Systems Engineering, Blacksburg, VA, 24061, United States of America, jran5@vt.edu, Hongyue Sun, Shuai Huang Graphical models are widely used to model variable relationship. Traditional graphical models are mainly used to model scalar variables. In this paper, a graphical model with functional variables is proposed. Functional regression models, combined with sparsity-inducing norms, are applied for the graphical modeling. A case study and simulation will be used to evaluate the proposed method. 4 - Optimal Design of Experiments for Generalized Linear Models Abhyuday Mandal, University of Georgia, Department of Statistics, 101 Cedar Street, Athens, GA, 30602-7952, United States of America, amandal@stat.uga.edu, Liping Tong, Jie Yang, Dibyen Majumdar Generalized linear models have been used widely for modeling the mean response both for discrete and continuous random variables with an emphasis on categorical response. Here we find efficient designs in the context of several optimality criteria, namely D-optimality, EW-optimality and Bayesian optimality. Regular fractional factorials with uniform replications are often used in practice. We show that these popular designs are often not optimal for binomial, Poission and multinomial cases.

2 - Generating and Comparing Pareto Fronts of Experiment Designs to Account for Multiple Objectives

Byran Smucker, Assistant Professor, Miami University, 100 Bishop Circle, 311 Upham Hall, Oxford, OH, 45056, United States of America, smuckerb@miamiOH.edu, Yongtao Cao, Tim Robinson

In many design scenarios the experimenter entertains multiple, conflicting objectives. The Pareto approach to experiment design is to construct a set of designs while explicitly considering trade-offs between criteria. The true Pareto front is not known, which creates problems in assessing front quality, and existing algorithms are inefficient, ineffective, or both. Here, we introduce an improved measure of front assessment, and present a new algorithm to generate Pareto fronts of designs. 3 - Cost Constrained ALT with Exponentially Changing Stress Durations David Han, University of Texas, One UTSA Circle, San Antonio, TX, United States of America, David.Han@utsa.edu When designing ALT, several variables such as the allocation proportions and stress durations must be determined carefully because of constrained resources. This talk discusses the optimal decision variables based on the popular optimality criteria under the constraint that the total cost does not exceed a pre-specified budget. A general scale family of distributions is considered to accommodate different lifetime models for flexible modeling with exponentially decreasing stress durations. 4 - Integration of Computer and Physical Experiments for Improving Predictive Inference Arda Vanli, Associate Professor, Florida State University, Tallahassee, FL, avanli@fsu.edu, Spandan Mishra A Bayesian predictive approach is developed to combine data from designed experiments on physical process and computer predictions. Predictive distribution of a regression model is used for inference on the outcome variable and issues including predictive capability, model adequacy and sensitivity to prior specifications are discussed. Applications from structural loss prediction and quality control are presented for illustrations. TD74 74-Room 204A, CC Bayesian Applications in Industrial Statistics Sponsor: Quality, Statistics and Reliability Sponsored Session Chair: Refik Soyer, The George Washington University, 2201 G St NW, Washington, DC, United States of America, soyer@gwu.edu 1 - What do Coin Tosses, Vessel Traffic Risk Assessment and Return Time Uncertainty Have in Common? Johan Rene Van Dorp, Professor, The George Washington University, 800 22nd Street NW, Suite 2800, Washington, DC, United States of America, dorpjr@gwu.edu, Jason Merrick Via a coin toss argument we will advocate decision making under uncertainty in vessel traffic risk assessment to be informed by relative risk comparisons by highlighting the analogy of an accident potentially occurring in a traffic situation with the toss of a biased coin. That same analogy is next used to demonstrate the large uncertainty bands that result for average return times of accidents in this context. 2 - Integrating Expert Judgement and Bayesian Analysis Thomas A. Mazzuchi, Professor And Chairman, George Washington University, Washington DC, DC, 20052, United States of America, mazzu@gwu.edu There is a growing need for marrying the fields of expert judgement and Bayesian Analysis that is, using the Expert Judgement approach to define prior distributions and for understanding the effects of the elicitation, codification and combination on the prior distribution and subsequent posterior analysis. This paper presents an investigation of the above for a simple model using the Classical Model for Expert Judgment by Cooke (2001). 3 - An Augmented Simulation Approach for Bayesian Design of Life Tests Refik Soyer, The George Washington University, 2201 G St. NW, Washington, DC, United States of America, soyer@gwu.edu In this talk we consider a Bayesian decision theoretic setup for optimal design of life tests. More specifically, we consider use of augmented probability simulation with a conjugate class of utility functions for design of life tests. We illustrate the implementation of the approach in one and two stage designs.

TD73 73-Room 203B, CC Recent Advances in Analyzing Experiments Sponsor: Quality, Statistics and Reliability Sponsored Session

Chair: Rong Pan, Associate Professor, Arizona State University, P.O. Box 878809, Tempe, AZ, United States of America, rong.pan@asu.edu 1 - Performance of Standard Mixture Designs in Modeling Ordinal Responses Mickey Mancenido, Arizona State University, 699 S. Mill Ave., Tempe, AZ, United States of America, mmanceni@asu.edu, Douglas Montgomery, Rong Pan Mixture designs for the proportional odds model — the widely used model for ordinal data — are lacking in literature. A viable surrogate are the standard mixture designs for linear models with normal errors. We are interested in the performance of the simplex-lattice with axial runs, simplex-centroid, computer- generated I-optimal, and the uniform space-filling designs when used in a mixture study with an ordinal response.

371