WEDNESDAY, JUNE 11 - 8:00 a.m. - 3:00 p.m.

4. Modeling: Cause & Effect in Health Services Research

Faculty: Bryan Dowd, University of Minnesota

The problems posed by endogenous explanatory variables generally are threats to causal inference. This workshop will begin with an introduction to causal inference with emphasis on problems encountered in non-experimental (observational) data. Discussion of specific problems will include omitted variable bias, reverse causality, and true simultaneity. Discussion of estimation methods will include propensity scores, instrumental variable models—including difference models and two and three stage least squares, and selectivity correction models.

Level: Intermediate. Basic understanding of mathematical statistics, including ordinary least squares estimation is necessary. Some understanding of maximum likelihood estimation also will be helpful, but simple examples will be included as part of the workshop.

5. Introduction to Bayesian Methods

Faculty: Christopher Hollenbeak, Pennsylvania State College of Medicine, and David Vanness, University of Wisconsin Medical School

Bayesian methods are an increasingly important method of conducting statistical inference. This two-part workshop will introduce basic principles of Bayesian inference and provide hands-on experience fitting standard models using WinBUGS software. The first half of the workshop will focus on notions of probability, the difference between the classical and Bayesian approach, the selection of prior distributions, and estimation via Markov chain Monte Carlo (MCMC) simulation. The second half of the course will be a tutorial on WinBUGS software where participants will learn how to navigate the software and work through four models. Participants will learn how to use the graphical interface of WinBUGS to program classical normal and generalized linear models, use built-in tools to assess model convergence, and conduct Bayesian hypothesis tests from the posterior samples.

Level: Beginner/Intermediate. No familiarity with Bayesian methods is assumed. Familiarity with basic univariate and multivariate statistical methods is required.

Requirement: You must bring a Windows-based laptop with a CD-ROM drive and WinBUGS v. 1.4 (installed and ready) to the class.