Seminars
Fall 2020
Building an Identifiable Dynamic Linear Model
Albert H. Lee III
Department of Statistical Sciences and Operations Research
Virginia Commonwealth University
Thursday, November 5, 2020
11:00 am - 12:00 pm
Zoom Link: https://vcu.zoom.us/j/95182867687
Abstract
Dynamic Linear Models (DLM) provide a flexible approach to modeling time series problems in which the underlying process changes through time. Unlike the frequentist methods developed by Holt-Winters or Box and Jenkins, the DLM utilizes Bayesian statistical principles and many works using DLM have incorporated some of these principles. Yet these works assume specified a priori variance parameters yielding a partial Bayesian model. The current work restricts the unknown observation variance parameter to be strictly greater than the unknown system variance parameters, thereby providing full model identifiability. A comparison is made between a Partial Bayesian DLM and a Fully Bayesian DLM. Three examples are provided illustrating the versatility available with both the Partial and Full Bayesian DLM. Sampling Importance Resampling and Metropolis Hastings Sampling are used to develop the Fully Bayesian Models. One step ahead forecasts and 95% Credible Intervals are illustrated within each model and then a holdout forecast is illustrated for each.
Speaker bio
Albert Lee is an Instructor in the Department of Statistical Sciences and Operations Research at VCU. He is currently pursuing his PhD. in Systems Modeling and Analysis from VCU. He earned his bachelor of science degree from the University of North Carolina Wilmington in 2005 and his master's degree in statistics from VCU in 2008.
A Comparison of Experimental Designs for Calibration
Christine M. McAnulty
Department of Statistical Sciences and Operations Research
Virginia Commonwealth University
Thursday, October 29, 2020
11:00 am - 12:00 pm
Zoom Link: https://vcu.zoom.us/j/93573874822
Abstract
The impact of experimental design choice on the performance of statistical calibration is largely unknown. Calibration is a technique that uses available experimental data to model the relationship between input and response variables to ultimately infer inputs based on newly observed response values. The purpose of this study is to investigate the performance of several experimental designs with regards to inverse prediction via a comprehensive simulation study. Specifically, we compare several design types including traditional response surface designs, algorithmically generated variance optimal designs, and space-filling designs. Results indicate that the choice of design has an impact on calibration performance and provides overall support for the use of I-optimal designs.
Speaker bio
Christine M. McAnulty is a PhD candidate in the Department of Statistical Sciences and Operations Research at VCU.
Robust MAVE through Nonconvex Penalized Regression
Jing Zhang
Department of Statistical Sciences and Operations Research
Virginia Commonwealth University
Thursday, October 22, 2020
11:00 am - 12:00 pm
Zoom Link: https://vcu.zoom.us/j/94370584757
Abstract
High dimensionality has been a significant feature in modern statistical modeling. Sufficient dimension reduction (SDR) approach is an efficient tool to explore the low dimensional projection subspace without losing full regression information between the response and the high dimensional predictors. Minimum average variance estimation (MAVE) is a popular method for dimension reduction. However, it is not robust to the outliers in the response due to the use of least squares. In this study, we proposed a new robust SDR method based on MAVE that introduces a mean-shift parameter as an indicator of the influence of each observation in the data. A penalty term on these mean shift parameters can help identify outliers and thus achieve robust estimation. Simulation studies show that our method has high prediction accuracy on estimating the effective dimension reduction directions in the presence of outliers in the response. We also show that our method is
less sensitive to the choice of the initial values.
Speaker bio
Jing Zhang is a PhD candidate in the Department of Statistical Sciences and Operations Research at VCU. She earned her Bachelor of Applied Science degree in materials science and engineering from Xi'an University of Architecture and Technology. Her Master's degree in materials science and engineering comes from SUNY Stony Brook. She has worked as a quality analyst at the Tangram Company, LLC.
Framework for Bayesian EWMA and CUSUM Control Charts using Loss Fuctions
Chelsea L. Jones
Department of Statistical Sciences and Operations Research
Virginia Commonwealth University
Thursday, October 8, 2020
11:00 am - 12:00 pm
Zoom Link: https://vcu.zoom.us/j/98172384591
Abstract
There has been a generous amount of research in profile monitoring constructing and implementing Bayesian control charts. In this work we will construct several Bayesian EWMA and CUSUM charts to observe possible drift shifts in a process. We define the framework of these charts using posterior and posterior predictive distributions informed by different loss functions. A simulation study is then performed, and the performances of the charts are evaluated via average run length (ARL), standard deviation of the run length (SDRL), average time to signal (ATS), and standard deviation of time to signal (SDTS). A sensitivity analysis is completed on the decision parameters, out-of-control shift size, and hyper-parameters of the distribution. We provide recommendations for use of the Bayesian EWMA and CUSUM charts based on these results.
Speaker bio
Chelsea Jones is a PhD candidate in the Department of Statistical Sciences and Operations Research at VCU. She earned her Bachelor of Science degree in actuarial science and her Master of Science degree in applied mathematics from Virginia State University.
SEIRD model for Qatar: A Case Study
Dr. Edward Boone
Department of Statistical Sciences and Operations Research
Virginia Commonwealth University
Thursday, October 1, 2020
11:00 am - 12:00 pm
Zoom Link: https://vcu.zoom.us/j/98539248986
Abstract
The Covid‐19 outbreak of 2020 has required many governments to develop mathematical-statistical models of the outbreak prevalence for policy and planning purposes. This work provides a tutorial on building a compartmental model using the Susceptibles, Exposed, Infected, Recovered and Deaths (SEIRD) model for the State of Qatar. A Bayesian framework is used to perform both parameter estimation and predictions. The use of interventions in the model attempts to quantify the impact of various government attempts to slow the spread of the virus. Predictions are also made to determine when the peak of Active Infections will occur. The talk uses the data from the Johns Hopkins Corona Virus Mapping project.
Speaker bio
Dr. Edward Boone is a Professor in the Department of Statistical Sciences and Operations Research at VCU. He is also the director of VCU's Statistics and Analytics Consulting Lab. Dr. Boone's research interests include the application of Baysian statistical methods to problems in the environment, healthcare, and national security. He earned his doctorate in statistics from Virginia Polytechnic and State University in 2003.
Spring 2020
The Uses of Statistics and Data-Analytics at a Federal Reserve Bank
Dr. Kartik B. Athreya
Federal Reserve Bank of Richmond
Friday, March 6, 2020
11:00 am - 12:00 pm
Grace Harris Hall, Room 4155
Abstract
Dr. Athreya will talk about the uses of statistics, and career pathways, at the Richmond Fed and in the Federal Reserve System. He also plans to spend some time speaking on the relation between statistics and (macro)economics. That latter part is not immediately about career opportunities, but rather to give some perspective on how the use of statistics is complicated by the (highly) non-experimental data of economics. That is to help show stats students how parts of statistics (like econometrics and macroeconomic models) have evolved to deal with very specific problems in social science settings.
Speaker bio
Dr. Kartik B. Athreya is the Executive Vice President and Director of Research at the Federal Reserve Bank of Richmond. Athreya's work has been published in a variety of academic journals, including the Journal of Monetary Economics, American Economic Journal: Macroeconomics, and International Economic Review. He is also an associate editor at the Journal of Economic Dynamics and Control. In recent years, Athreya has taught a doctoral course in macroeconomics at the University of Virginia and authored a book entitled Big Ideas in Macroeconomics (2013, MIT Press). He earned his doctorate from the University of Iowa in 2000.
Measuring Sensitivity to Nonignorable Incompleteness
Dr. Daniel F. Heitjan
Southern Methodist University and UT Southwestern Medical Center
Friday, February 21, 2020
12:00 pm - 12:50 pm
Academic Learning Commons (MCALC), Room 1104
Abstract
Statisticians have long recognized the potential biasing effects of nonignorable missing data mechanisms. For example, if larger observations are more likely to be missing or censored, then standard estimates such as the sample mean of the observed data or the Kaplan-Meier curve are invalid. Unfortunately, methods that attempt to estimate or test the degree of nonignorability are unsatisfactory. My idea is to embed the reference ignorable model in a nonignorable model in which I define a nonignorability parameter to represent the degree of departure from missing at random. I then conduct a sensitivity analysis to evaluate the dependence of the MLE of the parameter of interest varies with the degree of nonignorability. If it takes a large value of the nonignorability parameter to substantially affect the estimated parameter of interest, we judge the standard analysis to be insensitive. In this talk, I describe an approach to such a sensitivity analysis based on the index of local sensitivity to nonignorability (ISNI) statistic. An R package is available to conduct this analysis for the univariate GLM and a range of models for clustered or longitudinal data. I will demonstrate applications in live-data examples.
Speaker bio
Dr. Daniel F. Heitjan is Professor and Chair of Statistical Science at Southern Methodist University and Professor of Population & Data Sciences at UT Southwestern Medical Center. A native of Detroit, he earned a BSc in Mathematics (1981), an MSc in Statistics (1984), and a PhD in Statistics (1985) from the University of Chicago. He served on the faculties of UCLA (1985-1988), Penn State (1988-1995), Columbia University (1995-2002), and the University of Pennsylvania (2002-2014) before moving to Texas in 2014. Dr. Heitjan has over 200 publications in the literature of medicine and statistics, and is an elected Fellow of the American Statistical Association (1997), the Institute of Mathematical Statistics (2012), and the Society for Clinical Trials (2017). His research interests include clinical trial design, Bayesian methods, the theory of inference with incomplete data, and statistical methods in health economics, pharmacogenomics, and smoking cessation research.
Check out our past events.