Applied Bayesian Statistics

Political Science 435

Spring, 2003

 

 

Professor: Jeff Grynaviski

Office: Pick 528

Email: grynaviski@uchicago.edu

Webpage: http://home.uchicago.edu/~grynav

Phone: 773-702-2370                        

 

 

Course Description

 

This course is designed as an introduction to Bayesian data analysis. To be clear, this course does not aspire to provide students with a smorgasbord of fancy new statistical tools to play with. Instead, it is designed to introduce you to an entirely different mode of inference that incorporates your beliefs about the world and the data that you collect to further inform those beliefs into a coherent whole. Unfortunately, this means that for much of the quarter we will be teaching you a different (and in some respects harder) way of doing really simple things that you already know how to do. However, by starting with very simple models, we will find stony ground on which to build increasingly complex models, many of which could not be estimated without the set of computational tools and inferential framework developed by applied Bayesians. By the end of the term, you should be in a position to extend the Bayesian approach in a wide assortment of directions that you can tailor to your own interests.

 

 

Required Course Materials

 

Peter Congdon. 2001. Bayesian Statistical Modeling (available at the Coop once the publisher replenishes its stock).

 

WinBugs 1.4. Downloadable from http://www.mrc-bsu.cam.ac.uk/bugs/winbugs/contents.shtml

 

Other Useful Texts

 

Gelman, Carlin, Stern, and Rubin, Bayesian Data Analysis

 

Gilks, Richardson, Spiegelhalter (editors), Markov Chain Monte Carlo in Practice

 

Gill, Bayesian Methods: A Social and Behavioral Sciences Approach

 

Lee, Bayesian Statistics: An Introduction

 

 

Course Assignments

 

Course grades will be based in equal parts on the following three criteria:

 

Class participation and attendance.

 

The course is designed as an introduction to an applied methodology, not as an opportunity to read and discuss the great works.  Unlike many graduate classes, this means that you might learn something in class that you wouldn’t get from the readings. It also means that the bulk of class time will be spent in lecture.  This is not to say that discussion or questions are not encouraged.  I will try to reserve some time for discussion and you are always, always encouraged to ask questions.  The best questions usually takes the form of something like, “huh?” or “I don’t understand that?”  If you don’t understand, chances are others are in the same boat, so do not be timid about asking questions.

 

6 or 7 homework assignments.

 

In addition to the regular reading assignments and working through the WinBugs code that comes with Congdon’s text, you will be expected to do a number of (mostly) computer assignments designed to get you familiar with the tools for applied Bayesian work.

 

Independent research project.

 

Students will be expected to write a fully Bayesian research paper where they apply a technique that they learned in class to a research question that interests them. Miracles are not expected—a Bayesian multiple regression would be more than adequate, especially since you may run into computational problems beyond the purview of the class with more sophisticated applications.

 

To facilitate finishing the project, I strongly encourage each of you to pick a data set the first week of classes that you will use for the project. Then, as you work through the homework exercises, to apply the techniques from the homework to your data set. By the end of the term, all you will have to do to finish is write an introduction and conclusion and staple the results together.

 

In choosing your data set, I recommend that you pay attention to the following criteria. One, identify a data set that is standard in your field of study and which scholars have used for simple analyses (i.e. multiple regression; logistic regression; event count models). Two, there should be different populations within the data set whose behavior may not “pool” (e.g. normal people and folks from the South; developed and lesser developed countries). Three, there should be some dependent variable in the data set for which there are competing extant theories in the literature.

 

Note: I do not accept incompletes unless you have a personal crisis of some sort.

 


Tentative Course Schedule

 

 

Part I. Foundations

 

Introductory Lecture (Powerpoint Viewer) 

 

Lavine. 2000. “What is Bayesian statistics and why everything else is wrong”

 

Western, Bruce and Simon Jackman. 1994. “Bayesian Inference for Comparative Research.” American Political Science Review 88: 412-423.  JSTOR

            - This piece provides a nice, easy to read, application of Bayesian methods. It will give you a pretty good feel for what’s in store.

 

 

Review of probability theory and the Bayesian Setup (Powerpoint)

 

Congdon, Chapter 1.

 

Degroot, Probability and Statistics. Chapter 1 and 2.1 – 2.2. Reg. Reserve/Photocopy

 

Definitti. 1937. “Foresight: Its Logical Laws, Its Subjective Sources.” Photocopy

            - This is an optional paper for those of you interested in intellectual history. This paper provided the intellectual foundations for subjective probability theory.

 

 

Review of Integral Calculus and Univariate PDFs and CDFs

 

Kleppner. Quick Calculus. Chapter 3. Online Reserve.

 

Degroot, Probability and Statistics Chapter 3.1 – 3.3. PDFs and CDFs. Reg. Reserve.

 

Homework 1-Due Friday, April 18, 2003

 

 

Multivariate Probability Models

 

Degroot, Probability and Statistics Chapter 3.4 – 3.7. PDFs and CDFs. Reg. Reserve.

 

 

Part II. Bayesian Analysis of Basic Models

 

Bayesian analysis of one-parameter models

 

Congdon, Chapter 2.5-2.6

 

Gill. 2002. Bayesian Methods. Chapter 3.

 

Homework 2-Due Friday, April 25, 2003

 

Slave Revolts Data

 

 

 

Bayesian analysis of the one-parameter normal model with WinBugs introduction

 

Congdon, Chapter 2.1-2.2

 

 

 

Three philosophies of prior selection applied to the two-parameter normal model—Lecture 1 classical Bayesians with Winbugs implementation

 

Three philosophies of prior selection applied to the two-parameter normal model—Lecture 2 modern parametric Bayesians and an introduction to mcmc

            - Excel implementation of MCMC

 

Three philosophies of prior selection applied to the two-parameter normal model—Lecture 3 subjective Bayesians

 

Congdon, Chapter 2.3-2.4

 

Gilks, Richardson, and Spiegelhalter. 1996. “Introducing Markov Chain Monte Carlo.” In Markov Chain Monte Carlo in Practice. Reserve / Photocopy

 

Gill, ND. A Primer on Markov Chain Monte Carlo.” (GS-view)

 

Jackman, Simon 2000. "Estimation and Inference via Bayesian Simulation: An Introduction to Markov Chain Monte Carlo" American Journal of Political Science 44: 375-404. JSTOR.

            - This paper is optional for now given that its purpose is to show how to estimate more complicated models than we are talking about in class, but it is still quite useful if given a close read.

 

Homework 3-Due Friday, May 2, 2003

 

Slave Revolts Data—revised for WinBugs

 

 

Hierarchical Models and Bayesian Shrinkage

 

Congdon, Chapter 5

 

Bernardo, Jose M. 1984. “Monitoring the 1982 Spanish Socialist Victory: A Bayesian Analysis.” Journal of the American Statistical Association 79: 510-515.  JSTOR

 

Homework 4-Due Friday, May 16, 2003

 

 

Part III. Bayesian Analysis of the General Linear Model

 

Bayesian Regression with improper and conjugate priors

 

Congdon, Chapter 4.

 

 

Bayesian Regression with conjugate and convenient priors

 

Western, Bruce. “Unionization and Labor Market Institutions in Advanced Capitalism, 1950-1985.” American Journal of Sociology 99:1314-1341.

 

 

 

MCMC Convergence Diagnostics

 

Gelman. 1996. “Inference and Monitoring Convergence.” In MCMC in Practice.

 

 

The Choice of Priors and Bayesian Hypothesis Testing

 

Berk, Richard A., Bruce Western, and Robert E. Weiss 1995. "Statistical Inference for Apparent Populations." Sociological Methodology 25:421-458.

 

Bollen. 1995. “Apparent and Nonapparent Significance Tests.” Sociological Methodology 25:459-268.

 

Firebaugh. 1995. “Will Bayesian Inference Help? A Skeptical View” Sociological Methodology 25:469-472.

 

Rubin. 1995. “Bayes, Neyman, and Calibration.” Sociological Methodology 25:473-479.

 

Berk, et al. “Reply” Sociological Methodology 25: 481-485.

 

JSTOR

 

 

Bayesian Model Assessment.

 

Congdon, Chapter 10.

 

Clark, 2001. “Testing Non-Nested Models of International Relations: Evaluating Realism.” American Journal of Political Science  45: 724-744. JSTOR

 

 

Hierarchical Linear Models, part I

 

Hierarchical Linear Models, part II

 

Congdon, Chapter 8

 

Spiegelhalter, et al. 1996. “Hepatitis B: a case study in MCMC methods.” In MCMC in Practice.

 

Western, Bruce. 1998. “Causal Heterogeneity in Comparative Research: A Bayesian Hierarchical Modeling Approach.” American Journal of Political Science 42: 1233-1259. JSTOR

 

Lindley and Smith. 1972. “Bayes Estimates for the Linear Model.” Journal of the Royal Statistical Society, Series B, 34: 1-41. (JSTOR Optional)

 

Homework 5. Due May 29, 2003.

 

Data for homework 5.

 

 

 

General Linear Models

 

Congdon, Chapter 8

 

Quinn, Martin, and Whitford. 1999. “Voter choice in multi-party democracies: A test of competing theories and models.” American Journal of Political Science 43: 1231-1247. JSTOR

 

Martin 2001. “Congressional Decision Making and the Separation of Powers.” American Political Science Review 2001. JSTOR

 

Example: WinBugs code and data for a Poisson regression model with example of slow convergence.

 

 

Models for Missing Data

 

King, et al. 2001. “Multiple Imputation of Missing Data.” American Political Science Review 95: 49-69. JSTOR

 

 

Latent Variable Models

 

Jackman. 2000. “Estimation and Inference are Missing Data Problems: Unifying Social Science Statistics via Bayesian Simulation.” Political Analysis 8:4, pp. 307-332.

 

 

Advanced Topics

 

ecological inference; latent variables