Please note: This course will be taught online only

Ryan Bakker is Reader in Comparative Politics at the University of Essex and currently on leave at the University of Sharjah in the UAE. His research focuses on Bayesian statistics as well as elections, political behaviour, and party systems, with an emphasis on elite–mass representation and democratic responsiveness. He is Principal Investigator of the Chapel Hill Expert Survey (CHES), and his work has appeared in the American Journal of Political Science, Political Analysis, Journal of Politics, European Journal of Political Research, and Party Politics.  

Course Description

This course takes you beyond traditional statistics into a world where uncertainty becomes insight, and complexity becomes clarity. You’ll learn how to harness Bayesian models to tackle data with hierarchical and latent structure, combining the best of two worlds: multilevel models (MLMs) that capture variation across levels and contexts, and measurement models (including factor analysis, SEM, and IRT) that uncover the invisible forces shaping what we observe.

Through an mix of lectures and hands-on labs, you’ll discover how Bayesian approaches allow you to:

  • Seamlessly handle missing or unbalanced data
  • Make fewer and more transparent assumptions
  • Quantify uncertainty in a principled way
  • Visualize, communicate, and compare complex models with confidence

By session’s end, you’ll be fluent in specifying, estimating, and interpreting advanced hierarchical and latent-variable models, all within a unified Bayesian framework.

Topics include:
• Fixed and random effects modeling
• Varying intercepts and slopes
• Multilevel models for linear and non-linear outcomes
• Bayesian factor analysis and latent variable modeling
• Structural equation modeling (SEM) and multilevel SEM
• Item response theory (IRT) for survey and test data
• Model comparison, posterior predictive checks, and visualization

You’ll work directly with R, JAGS, and Stan, gaining practical experience and reproducible templates you can immediately adapt to your own research. Whether you study politics, psychology, education, or economics, this course will show you how Bayesian methods reveal structure, nuance, and meaning that classical techniques often miss.

Background knowledge required

Maths
Calculus – Elementary
Linear Regression – Moderate

Statistics
OLS – Strong
Maximum Likelihood – Moderate

Computer Background
R – Moderate

This course introduces participants to Bayesian models for data with hierarchical and latent structure. Multilevel models (MLMs) allow researchers to model variation across groups and levels, while measurement models—including factor analysis, structural equation models (SEMs), and item response theory (IRT)—allow us to model the latent traits that underlie observed indicators.  The Bayesian approach to statistical modeling is particularly well-suited for such models in terms of:

–handling missing data

–having fewer necessary assumptions

–presenting post-estimation quantities of interest and their associated uncertainties

–much more…

The course will begin with a review of classical and Bayesian approaches to multilevel modeling before extending these ideas to the measurement domain. By the end of the week, participants will be able to specify, estimate, and interpret a range of hierarchical and latent-variable models within a unified Bayesian framework.

Topics covered include:

  • Fixed and random effects models
  • Varying intercepts and slopes
  • Multilevel models for linear and non-linear outcomes
  • Bayesian factor analysis and latent variable modeling
  • Structural equation modeling (SEM) and multilevel SEM
  • Item response theory (IRT) models for survey and test data
  • Model comparison, posterior predictive checking, and visualization

We will emphasize hands-on implementation using R, JAGS, and Stan, with a focus on model specification, diagnostics, and presentation of results. Participants will leave with a suite of reproducible code examples applicable to their own research.

Software and Resources:

R – https://cran.r-project.org/
JAGS – http://mcmc-jags.sourceforge.net/
Stan – https://mc-stan.org/users/

Daily Schedule (Tentative):

Day 1:  Introduction to Bayesian inference and MCMC methods
               Estimating simple Bayesian regression models inJAGS

Day 2:  Introduction to multilevel models
              – Fixed and random effects for varying intercepts/slopes, classical and Bayesian.
              – Bayesian MLMs for continuous outcomes
              – Including level-2 covariates

Day 3:  Multilevel logistic regression—classical and Bayesian approaches

Day 4: Multilevel models for GLMs

             -ordered/unordered choice models

             -count models

Day 5: Post-estimation and model fit for MLMs

Day 6: Introduction to Bayesian measurement models
             – Exploratory and confirmatory factor analysis
Day 7:  Bayesian IRT models

Day 8:  Bayesian SEMs—using latent variables in predictive models

               -Error in variables: latent left-hand side variables

Day 9: Bayesian SEMs continued

             -MIMIC (multiple indicator/multiple causes) models

Day 10: Wrapping up—walkthrough a project from start to finish