BAYESIAN DATA ANALYSIS USING BUGS AND R
A two-day short course sponsored by the University of Maryland Statistics Consortium and the Joint Program in Survey Methodology

MAY 5 - 6, 2003 
Presented at the University of Maryland

ANDREW GELMAN
Department of Statistics, Columbia University

INSTRUCTOR
Andrew Gelman is a professor in the Departments of Statistics and Political Science and founding director of the Quantitative Methods in Social Sciences Program at Columbia University.  He is the author of the books "Bayesian Data Analysis" and "Teaching Statistics: A Bag of Tricks" and over 100 articles on statistical methods, theory, and applications.  His statistical research interests include Bayesian inference, multilevel modeling, and statistical graphics and model checking.  His applied research interests include public opinion and voting, decision analysis, and environmental health.  His research has received awards from the American Statistical Association, the American Political Science Association, and other research organizations.  

PREQUISITE COURSE
Basic statistics course including basic familiarity with linear regression.

TOPICS AND APPROACH 

Why learn about Bayesian data analysis?
Bayesian inference is an effective and flexible approach for analyzing large data sets and combining information from different sources. Bayesian methods allow inference without overfitting for models with large numbers of parameters.  Important examples for sample surveys include small-area estimation, longitudinal data analysis, and multilevel regression.

Why Bugs and R? 
Bugs is a freely-available computer program that can fit Bayesian statistical models without the need for programming by the user. Bugs's modeling language is open-ended, making it easy to generalize a model or expand it to account for additional data. 
R is an open-source language for statistical computing and graphics. We run Bugs from R, which gives us flexibility in data manipulation before the analysis and display of inferences after.

What topics are covered? 
The course begins with Bayesian analyses of simple classical models such as linear and logistic regression.  These examples introduce Bugs and R as well as the basic ideas of Bayesian data analysis.  We then discuss how to fit and understand multilevel (hierarchical) models, in which parameters are modeled in batches.  We discuss general issues in practical Bayesian data analysis, including how to check the fit of a model to data and how to incorporate design information (e.g., strata or clusters in sampling) into a Bayesian model.  We also give some background on the computational algorithms used by Bugs as well as programming tips.  We illustrate with several applications from our own research using survey and social science data.  

Here is an approximate syllabus for the course: 

1.   R and Bugs for classical inference
2.   Model checking, model expansion, and debugging
3.   Hierarchical linear regression
4.   Hierarchical logistic and Poisson regression
5.   Understanding the Gibbs sampler and Metropolis algorithm
6.   Expressing a single model in several different ways
7.   Prior distributions
8.   Troubleshooting:  what to do when your simulations run too slowly, or get stuck, or don't make sense
9.   Accounting for design of surveys and experiments
10. Concluding discussion

What are the concepts introduced in the course?  
Bayesian data analysis has three steps: model building, inference, and model checking.  We integrate these using R and Bugs.  "Fitting" a model is neither the beginning nor end of analysis.  Bayesian hierarchical modeling can be seen as a generalization of least squares or maximum likelihood to allow partial pooling of inferences from different data sources.  

TENTATIVE SCHEDULE
Please be on time.  There is a 30 minute period at the beginning of class in which each participant will have to download software before instruction begins.

LOCATION 
The course will be held at The University of Maryland in the OACS Lab-0225 Lefrak Hall. The University of Maryland is located in College Park, MD, with nearby access to the College Park Metro Stop. Located on US Route 1 and approximately 2 miles from I-495.  Information, directions, and times will be sent with your confirmation letter.

REGISTRATION 
Online registration is required. Confirmation of registration and instructions will be sent after the registration form has been processed. Registration is not firm until you receive a confirmation letter. Payment by credit card is required. Post registration payment may be done online using the student's registration number. Please note registration number. Registration and Payment  deadline is  April 18, 2003.  Payment is due by April 18, 2003 or the registration will be cancelled.  Please notify JPSM as soon as possible if you need to cancel your registration. Cancellation between April 21 –April 23, 2003 will require a $100 administrative fee, the remainder will be reimbursed. Cancellation on or after April 24, 2003 is subject o the full fee amount.  

INQUIRIES
For information on this short courses contact JPSM at  301-314-7911; fax 301-314-7912;  or e-mail shortcourse@survey.umd.edu  

Statistics Consortium Web Page:  http://www.jpsm.umd.edu/stat 
JPSM Home page:  http://www.jpsm.umd.edu/