Methods workshop: “Bayesian Measurement in R/JAGS” with Dr. Xavier Fernández-i-Marín
The department's internal workshop on measurements and concepts proved to be both challenging and rewarding.
Dr. Xavier Fernández-i-Marín from LMU Munich, a leading scholar in methodology in the social sciences, gave researchers at the Department of comparative politics an introduction to different methodological tools to achieve good measurement for concepts of interest. The course aimed to allow the development of strategies for transforming raw data into meaningful measures.
Developing good measures of the concepts to work with is the first step for assessing causal relationships and associations amongst multiple variables.
By the end of the course, the participants increased their knowledge about (a) ways of designing various systems to transform raw measures into meaningful variables; (b) creating clusters and typologies of observations; and (c) performing measurement of such concepts through time.
The course used a Bayesian framework. The Bayesian framework is especially well suited for dealing with latent variables and for performing measurement. First, under Bayesian estimation there are no nuisance parameters, and the inference is done for both latent traits and item parameters. Second, it allows to systematically incorporate prior information. Third, the MCMC methodology overcomes many computational issues. The course progressed from basic modelling under classical estimation and ended with Bayesian and MCMC estimation.
The workshop had a dozen participants from the Department of Comparative Politics (Sampol) ranging from PhD candidates and postdoctoral fellows all the way to full professors. The course is part of Sampol’s ongoing drive to improve methods training among its academic staff.
In addition to being a leading scholar in methodology, Dr. Fernández-i-Marín is interested in questions related to global governance, the diffusion of policies and institutions, and the development of regulatory agencies. He has publications in leading journals such as the Journal of Statistical Software, Comparative Political Studies, the Journal of Public Administration Research and Theory, Governance, Public Administration, Regulation & Governance, or Policy Sciences, among others.
This was his third visit to the Department of Comparative Politics in Bergen, after successful workshops in Bayesian Hierarchical Modelling in December 2014 and June 2017.
The course was organized by Michaël Tatham and Tor Midtbø from the department.
- We learnt a lot, from conceptual issues with different measurement choices, all the way to the intricacies of Item Response Theory and dynamic Bayesian latent variable modelling, says Tatham.
Post-doc Andrea Fumarola and PhD student Jana Birke Belschner at the department were also impressed by the workshop and the sophistication of the techniques;
- It was very useful, and also tough. It made us think about the dialogue between different elements such as methodology and theory, and how we develop concepts and develop measures for them, says Fumarola.
Belschner had the same impression;
- We use so much time on data collection, but the concepts that are derived may be simplified. The workshop taught us more sophisticated techniques, and was really intellectually stimulating.