- Advanced Bayesian Models
CDSS, University of Mannheim
This second part of the Bayesian workshop series introduces more advanced models. We will cover Bayesian versions of latent variable models, such as factor models and item response theory or ideal point models, as well as models for simultaneous outcomes, such as seemingly unrelated regression and multivariate probit models. Furthermore, we will discuss models to deal with the ubiquitous problem of missing data in a fully Bayesian context. Each lecture is followed by lab sessions, where we replicate examples from the lecture and discuss how to understand fitted models through predictions and graphical displays..
- Regression designs and their applications
BA course, University of Mannheim
Does incumbent performance affect chances of re-election? Do get-out-the-vote election campaigns actually increase turnout? Do citizen information programs increase political participation? And if so, by how much? These and related questions in political science might seem straightforward, but answering them properly is harder than you might think. In this course we will encounter a framework for thinking about questions like these from the perspective of causal inference. This means thinking about how we can best estimate the effects of causes (such as peace keeping missions) on outcomes. In a perfect world we could conduct experiments, but often we are left with using limited and noisy pre-existing data. We will discuss a number of regression models, which will provide us with a toolbox for tackling these kinds of causal questions. The course will consist of both theoretical and practical sessions. [Syllabus]
- Theory building and causal inference
CDSS, University of Mannheim
- Modeling heterogeneity in cross-sectional and panel data
Essex Summer School
July/August 2012, 2013, and 2014
The course introduces a range of recently developed methods that deal with heterogeneity. Heterogeneity refers to differences between groups, such as individuals and countries, which are due to omitted or unobserved factors. Commonly employed approaches try to correct for such factors. We will take a different approach and build explicit models for heterogeneity between individuals or groups (such as countries or organizations) allowing us to answer questions such as: Does the effect of income on redistribution preferences differ between different subgroups of individuals? Does the effect of parties’ policy on vote choice differ between countries? Do different subgroups of a population follow different patterns of attitude or preference change? Do unemployment and poverty dynamics differ between (unobserved) subgroups?
More information can be found at the Essex Summer School website here.
- Advanced Bayesian Models for the Social Sciences
ICPSR Summer Programm
August 2013 and 2014
The course covers the theoretical and applied foundations of Bayesian statistical analysis at a level that goes beyond the introductory course. Therefore, knowledge of basic Bayesian statistics (such as that obtained from the 'Introduction to Applied Bayesian Modeling for the Social Sciences' workshop) is assumed. The course will consist of four modules. First, we will discuss Bayesian stochastic simulation (Markov chain Monte Carlo) in depth with an orientation towards deriving important properties of the Gibbs sampler and the Metropolis Hastings algorithm. Extensions and hybrids will be discussed. Second, the course will cover model checking, model assessment, and model comparison, with an emphasis on computational approaches. The third module introduces Bayesian variants of "workhorse" political science models, such as linear models, models for binary and count outcomes, discrete choice models, and seemingly unrelated regression. The fourth week will focus on more advanced Bayesian models, such as hierarchical/multilevel models, models for panel and time-series cross-section data, latent factor and item response theory (IRT) models, as well as instrumental variable models. Throughout the workshop, we emphasize not only estimation with modern programming software (R, and JAGS), but also how to communicate results effectively.
More information can be found at the ICPSR Summer School website here.
- Bayesian Analysis of Latent Variable and Hierarchical Models
University of Mannheim
25 - 27 September 2013
- Bayesian Generalized Linear Models
University of Berne
17-21 September 2012
This course introduces and extends the classical 'workhorse' social science models - linear, logit, probit models and their multilevel extensions - from a Bayesian perspective. In this course I will introduce the basics of Bayesian inference, showing how its interpretation of probability differs from the classical approach and how it is actually closer to how social scientists think about their models. I then introduce generalized linear models and show how they can be easily fitted using modern software for Bayesian inference. I introduce Bayesian model diagnostics and fit measures, which allow straightforward model comparisons and examination of model misspecification. The focus of the course will be on how to compute interesting quantities from those models, like predicted values or first differences in expected values for a changing covariate. Using the Bayesian approach to inference, their calculation is straightforward and one can easily construct appealing graphical displays.
- Multilevel analysis (with Tom Snijders)
Spring School, University of Oxford
More information can be found here
- Quantitative Methods in Political Science
Graduate Lecture in Political Science, University of Mannheim
This course introduces graduate students to quantitative methods in political science. During the first half of the course, we will focus on linear regression models. The topics covered include discussions of the mathematical bases for such models, their estimation and interpretation, model assumptions and techniques for addressing violations of those assumptions, and topics related to model specification and functional forms. During the second half of the course, students will be introduced to likelihood as a theory of inference, including models for binary and count data. Statistical language used is R.
- Models for Categorical Data
Undergraduate Course in Political Science, University of Mannheim
This course introduces advanced undergraduate students to models for categorical data. Its main focus lies on discrete choice models and their application. The use of simulation methods in emphasized, in order to calculate quantities of interest that can be easily communicated. Practical lab session using R and Zelig focus on presentation and interpretation of results.
- Basic and Multilevel Models for Categorical Data
Graduate Course in Social Science, University of Nijmegen
This course introduces students in the social science research master to models for categorical data. We discuss basic nonlinear probability models and their hierarchical or multilevel extensions. Practical lab session using Stata focus on calculations of quantities of interest that allow for clear communication and interpretation of results.
- Introduction to Missing Data
University of Mannheim, 2010 and 2011
This workshop introduces the problem of missing data and modern solutions for it. I discuss different types of missing data and still predominant (but deficient) strategies for dealing with them. I show that multiple imputation is a solid method to deal with a wide range of missing data that is easily carried out using Stata and R. Besides the classical multivariate normal imputation methods, I demonstrate how chained regression imputation can be used to impute genuinely categorical data.