AM08
MAKING BETTER DECISIONS IN CAUSAL ANALYSIS: AN INTRODUCTION TO STRUCTURAL EQUATION MODELING AND BAYESIAN NETWORKS
Course Type: Half Day
Course Level: Beginner
Overview: Causal networks are emerging as powerful tools that can be used to investigate new classes of research problems. Two popular modeling approaches - structural equation modeling and Bayesian networks - each are based on the causal network framework. Structural equation models are useful for evaluating hypotheses about mechanisms within a causal system and describing latent constructs from observed data. Bayesian networks, on the other hand, are useful for dynamically predicting unobserved variables as data accrue and for performing sophisticated cost-utility analyses. We will give an introductory overview of these models, show examples, and provide avenues for further skill development.
Background: This short course will make advanced causal modeling techniques accessible to a wide audience of researchers across many disciplines. Advances in structural equation modeling (SEM) and Bayesian networks (BNs) have linked together conceptual models, path diagrams, mathematical models and decision-analytic models. These techniques allow for i) combination of continuous, categorical and latent and observed variables; ii) modeling of causal relationships including multiple direct and indirect effects in a single analysis; iii) cutting-edge techniques for model selection and comparison; iv) compact representation of cost-utility problems; v) dynamic updates to model predictions as new clinical measures are obtained; and vi) the ability to break down complex causal structures into many smaller, more manageable local models which can then be combined together. These advantages are particularly applicable to both theoretical and applied research problems in medical decision making.
Format Requirements: Learners will experience a mixture of lecture and discussion. We will introduce basic concepts and vocabulary of SEMs and BNs, give real-world examples and conduct sample analyses. No prior knowledge of these techniques is required, although those with a basic understanding of statistics will benefit most from this course. While the course does not include a laboratory session, learners may wish to bring their laptop to download software packages from the Internet.
Description and Objectives: We present a basic overview of principles, some common nomenclature, diagrams, a little algebra (with only a handful of Greek letters!), and some real world examples. In this course, you will:
- Fundamentally enrich your way of thinking about certain problems in medical decision making;
- Learn fundamental concepts which underpin SEM and BN models;
- Gain knowledge of important resources and techniques for causal modeling;
- Be introduced to software for implementing SEM and BN analyses, including visualization of causal models;
- Be able to interpret results of advanced causal modeling techniques; and
- Understand advantages of SEMs and BNs over traditional statistical models.
Whether you just want to know how to read or critique an article that uses these techniques, or you want to engage a few graphical modeling researchers in some feisty methods discussions, signup for this course...we'd love to visit with you.
Jarrod E. Dalton, PhD
Cleveland Clinic
Assistant Professor of Medicine
Quantitative Health Sciences and Outcomes Research
Adam T. Perzynski, PhD
Case Western Reserve University at MetroHealth
Assistant Professor of Medicine
Center for Health Care Research and Policy
Joseph J. Sudano, PhD
Case Western Reserve University at The MetroHealth System
Assistant Professor of Medicine, Epidemiology and Biostatistics
Center for Health Care Research and Policy