Welcome to the Bayesian Statistical Science Section

The Bayesian Statistical Science Section of the ASA provides a forum for statisticians and people who have interest in the Bayesian paradigm. The broad objectives of the Section are: to encourage research on theory and methods of statistical inference and decisionmaking associated with Bayes' theorem and to encourage the application and proper use of Bayesian procedures in the behavioral, biological, managerial, engineering, environmental, legal, medical, pharmaceutical, physical, and social sciences.

Current Affairs 

Continuing Education Courses at JSM 2014

The Section on Bayesian Statistical Science (SBSS) is pleased to sponsor the following Continuing Education Courses in the Joint Statistical Meetings to be held in Boston, MA, from August 02-07, 2014. Please note that the room numbers indicated below are subject to change. Participants are always encouraged to confirm these locations upon arrival at the conference front desk.

1. Sat, August, 02, 2014, 8:30 AM - 5:00 PM. Room CC-161. ​
Hierarchical Bayesian Modeling and Analysis for Spatial Data.
Instructor(s): Sudipto Banerjee, University of Minnesota, Bradley P. Carlin,
University of Minnesota, Alan Gelfand, Duke University.

Abstract: In this course we will describe hierarchical modeling and related Markov chain Monte Carlo (MCMC) methods for spatial statistics. We will begin by outlining and providing illustrative examples of the three types of spatial data: point-level (geostatistical), areal (lattice), and spatial point patterns. We then describe both exploratory data analysis tools and traditional modeling approaches for point-referenced data. Since our approach is fully model-based through the use of Gaussian processes, we develop the basics of spatial Gaussian process models. Approaches from traditional geostatistics (variogram fitting, kriging, etc.) will be briefly covered here. We then turn to areal data models, again starting with choropleth maps and other displays and progressing towards more formal model specifications, e.g., Markov random fields that underlie the conditional, intrinsic, and simultaneous autoregressive (CAR, IAR, and SAR) models widely used in areal data settings. The remainder of our presentation will cover hierarchical modeling for both univariate and multivariate spatial response data, including Bayesian kriging and lattice modeling, as well as more advanced issues such as anisotropy and nonstationarity. We also include a discussion of spatial point process models, and modern computational approaches for very large data sets (the so-called ``big N problem"). Short course participants should have an M.S. understanding of mathematical statistics at, say, the Hogg/Craig/Tanis or Casella/Berger levels, as well as basic familiarity with Bayesian modeling and computing at the Carlin/Louis or Gelman et al. levels. We will not assume any significant previous exposure to spatial or spatiotemporal methods.

2. Sun, August, 03, 2014, 8:30 AM - 5:00 PM. Room CC-161.
Bayesian Dynamic Models: Time Series Analysis & Forecasting.
Instructor(s): Raquel Prado, University of California at Santa Cruz, Mike West, Duke University

Abstract: This short-course covers basic principles and methods of Bayesian dynamic modeling in time series analysis and forecasting, with methodological details of central model classes explored in a range of examples. A main focus is on dynamic linear models and related methods of inference and forecasting, including multivariate time series analysis. Links between time and frequency domain, and stationary time series models, will be covered, as well as selected developments in nonlinear and non-Gaussian dynamic models and associated Monte Carlo Markov chain simulation methods for analysis. The course will conclude by contacting some some recent modeling and applied developments in multivariate time series and forecasting. The course draws on a range of examples and case studies from business, finance, signal processing and the biomedical sciences. The course material will be accessible to advanced students, academics and/or professionals with strong statistical modelling backgrounds and prior exposure to essentials of Bayesian analysis. Familiarity with- and working facility in- multivariate distribution theory and statistical inference are prerequisites. Prior exposure to some areas of time series analysis will be useful though is not necessary.

3. Mon, August, 04, 2014, 8:30 AM - 5:00 PM. Room CC-160B.
Nonparametric Bayesian Data Analysis.
Instructor(s): Peter Mueller, University of Texas at Austin, Fernando Quintana, Pontificia Univ Catolica de Chile

Abstract: In this short course we will discuss the use of nonparametric Bayesian inference (BNP) for a number of common statistical inference problems, including density estimation, regression, mixed effects models, classification and clustering. In the course of this discussion we review some of the popular BNP models, including Dirichlet process (DP) models, Polya tree models, DP mixtures and dependent DP (DDP) models. We will review some of the general modeling principles, including species sampling models, stick breaking priors, product partition models for random partition and normalized random measures with indpendent increments. We will briefly discuss some of the main computational algorithms and available software. Prerequisites: working knowledge of Bayesian data anlysis (at the level of a first graduate class in Bayesian inference), and basic knowledge of Markov chain Monte Carlo posterior simulation.

4. Tue, August, 05, 2014, 8:30 AM - 5:00 PM. Room CC-161.
Bayesian Inference.
Instructor(s): Bruno Sanso, University of California Santa Cruz.

Abstract: Bayesian methods have become increasingly popular with the advent of fast computational algorithms for the exploration of high dimensional probability distributions. The Bayesian paradigm provides a coherent framework to build models of high complexity, incorporate quantitative and structural prior information and account for all uncertainties in a probabilistic way. This course reviews the bases of Bayesian inference. As such, it focuses on key general concepts rather than the technical detail of specific methods. The course will start by presenting the basic elements of statistical inference that uses likelihood functions. We will then consider the problem of specifying prior distributions, proceed by describing the tools for both pointwise and interval estimation and prediction and present the Bayesian theory of hypothesis testing and model comparison. Finally we will review the elements of modern computational methods used in the applications of Bayesian models. The course targets students or professionals with a good knowledge of statistics that want to learn or refresh their knowledge of basic Bayesian inference. The level of mathematical sophistication will be kept as low as possible. Calculus and basic probability theory are considered a pre-requisite.

Upcoming Web-Based Lectures

Title: Overview, Hurdles, and Future Work in Adaptive Design
Presenter: Christopher Coffey, Department of Biostatistics, School of Public Health, University of Iowa
Date and Time: Wednesday, August 20, 2014, 12:00 p.m. - 2:00 p.m. Eastern time
Sponsor: Biopharmaceutical Section

Registration Deadline: Monday, August 18, at 12:00 p.m. Eastern time

In recent years, there has been substantial interest in the use of adaptive or novel randomized trial designs. Adaptive clinical trial designs provide the flexibility to make adjustments to aspects of the design of a clinical trial based on data reviewed at interim stages. Although there are a large number of proposed adaptations, all generally share the common characteristic that they allow for some design modifications during an ongoing trial. Unfortunately, the rapid proliferation of research on adaptive designs, and inconsistent use of terminology, has created confusion about the similarities, and more importantly, the differences among the techniques. Furthermore, the implementation of adaptive designs to date does not seem consistent with the increasing attention provided to these designs in the statistical literature. This webinar will first provide some clarification on the topic and describe some of the more commonly proposed adaptive designs. It will focus on some specific barriers that impede the use of adaptive designs in the current environment. Finally, there will be a discussion on future work that is needed to ensure that investigators can achieve the promised benefits of adaptive designs.

Title: Modeling and Simulation Approach to Optimize the Assessment of the Potential for QT Prolongation
Presenter: Daniel Weiner, SrVP & GM, Certara
Date and Time: Thursday, September 18, 2014, 12:00 p.m. - 2:00 p.m. Eastern time
Sponsor: Biopharmaceutical Section

Registration Deadline: Tuesday, September 16, at 12:00 p.m. Eastern time

Regulatory authorities require that all drugs be assessed for the potential for drug-induced QT interval prolongation, which is a biomarker for Torsades de Pointes, which can sometimes result in sudden death. Since 1988, there have been at least 14 drugs withdrawn from market as a result of QT prolongation (Stockbridge et al. Drug Safety 2013;36: 167-182). There are currently two Regulatory Requirements regarding this safety assessment: ICH E14 and S7B, which cover clinical and nonclinical evaluations, respectively.

The main topics of this webinar are
1) Review of the E14 Guidance and the concept of a Thorough QT (TQT) Study, including a discussion of how such studies should be analyzed
2) Discussion of what data to collect, on an ongoing basis, to assess the potential for QT prolongation
3) Discuss methods to determine if there is an emerging signal regarding the relationship between drug exposure and QT prolongation, and how to use that information for dose selection in the TQT and other subsequent clinical trials

Concepts will be illustrated with case studies.

*** More information and registration at the webinar page.


From the SBSS Mixer at JSM 2013