Sponsors:
Wendesday, March 17, 2010
Full-day courses:
-
M1 (morning): Objective Bayesian Model Choice
(by Susie Bayarri)
Half-day courses:
ABSTRACTS
A1:: Computational Bayesian Statistics (by Christian Robert)
In this short course, we will cover some notions of Monte Carlo techniques for Bayesian inference and in particular model choice. This will be backed by appropriate chapters of "Bayesian Core" Springer-Verlag, New York, 2007, ISBN 0-387-38979-2, namely, 3. Linear models and regression 4. Generalised linear models 5. Capture-recapture experiments with the support of specific datasets for each chapter. The goal is to introduce the principles of Bayesian computation through standard models with a minimum amount of theoretical discussion and an emphasis on practical implementation. Due to the time constraint, the illustration via R programming will necessarily be omitted, but all R programmes are available on http://www.ceremade.dauphine.fr/~xian/BCS and can be downloaded in advance (as well as the slides). Depending on publication schedule, the course will also borrow from "Introducing Monte Carlo Methods with R" to appear (tentatively) in January in the Use R! series of Springer Verlag.
F1: Bayesian Adaptive Design in Clinical Trials (by Brad Carlin and Peter Müller)
Overview: Thanks in large part to the rapid development of Markov chain Monte Carlo (MCMC)methods and software for their implementation, Bayesian methods have become ubiquitous in modern biostatistical analysis. In submissions to the U.S. FDA Center for Devices and Radiological Health, where data on new devices are often scanty but researchers typically have access to large historical databases, Bayesian methods have been in common use for over a decade and in fact were the subject of a recently-released FDA guidance document. Statisticians in earlier phases (especially Phase I oncology trials) have long appreciated Bayes' ability to get good answers quickly. Moreover, an increasing desire for adaptability in clinical trials (to react to trial knowledge as it accumulates) has also led to heightened interest in Bayesian methods.
This full-day course (4 consecutive workshops) introduces Bayesian methods, computing, and software, and then goes on to elucidate their use in Phase I, II, and III trials. We include descriptions of how the methods can be implemented in WinBUGS, R, and BRugs, the version of the BUGS package callable from within R. In particular, we will illustrate the different ways a Bayesian might think about power when designing a trial, and how a Bayesian procedure may e calibrated to guarantee good long-run frequentist performance (i.e., low Type I and II error rates), a subject of keen interest to the FDA.
Workshop 1: Introduction to Hierarchical Bayes Methods and Computing
Workshop 2: Bayesian design and analysis for Phase I studies
Workshop 3: Bayesian design and analysis for Phase II studies
Workshop 4: Bayesian design and analysis for Phase III studies
analysis, arm droppingStudents are invited to bring their own laptop computers to the session, and to have the latest versions of WinBUGS and R already installed on these computers. Both of these programs are freely available from http://www.mrc-bsu.cam.ac.uk/bugs/winbugs/contents.shtml and http://www.r-project.org/ respectively. The presentation will assume familiarity with basic Bayesian methods and MCMC algorithms, at the level of, say, Chapters 2 and 3 of Carlin and Louis (2009) or Chapters 2, 3, 5 and 11 of Gelman et al. (2004). The workshop's goal is to make these methods come alive in the software through real data examples that the students try for themselves during the presentation.
F2: Bayesian Modeling for Spatial and Spatio-Temporal Data with Applications to Environmental Sciences and Public Health
(by Sudipto Bannerjee)Proliferation of spatially referenced and spatiotemporal datasets and need for analysis is especially common in the broad fields of environmental sciences and public health. Here, spatially and temporally indexed data, consisting of one or several outcome variables, and associated predictors, are used to model physical characteristics, presence/absence, counts, or change. The focus of inference is on model parameters and/or subsequent prediction. Rarely is it safe, or even desirable, to assume independent model residuals. This assumptions is often violated because these data exhibit spatial, temporal and/or hierarchical structure. This course details hierarchical generalized linear models that accommodate spatial-temporal associations. In lecture, careful attention is paid to theoretical foundations of model specification, identifiability of parameters, and inference. Emphasis is laid on exploration and visualization of spatial-temporal data and model implementation. Upon course completion, participants can fit a diverse class of spatial-temporal models using the spBayes package in R (www.r-project.org) and the WinBUGS/OpenBUGS package (http://mathstat.helsinki.fi/openbugs). A familiarity with classical linear models and multiple regression is helpful, but not required. A laptop with a current versions of spBayes and WinBUGS/OpenBUGS
installed, while not required, may be useful.The following is a list of topics covered in the course. Each topic will include theory, examples, and data analysis along with live interactive computing demonstrations.
[a] Introduction to hierarchical linear models;
[b] Ingredients for modeling point-referenced spatial data;
[c] Bayesian Linear models for univariate point-referenced spatial data(kriging);
[d] Basics of the R statistical computing environment;
[e] Generalized linear models with spatial and temporal random effects;
[f] Hierarchical models for areally referenced datasets;
[g] Hierarchical models for spatial-temporal data;
[h] Ingredients for modeling multivariate spatially-referenced data;
[i] "Big-N problem": predictive process models for large datasets in space and/or time;
[j] Space varying coefficient models for spatial non-stationarity;
[k] Case studies from forestry, ecology and public health.
M1:
