Advanced Bayesian Methods

Presenter(s): Dr Gabriel Katz


decorative image with NCRM logo

This resource looks at modern Bayesian computation. Focusing on the two most widely used Bayesian algorithms, the Gibbs Sampler, and the Metropolis-Hastings. It reviews s criteria used to assess model convergence, and when running Bayesian models, how to identify that the model is ready to be used to draw inferences about parameters. It discusses the goodness of fit criteria used in the Bayesian world, which differs from those used in frequency statistics. It concludes by discussing methods to speed up conversion or speed up execution time. All videos in this resource uses the a single slide set, which can be accessed here.

 

Introduction

In this video, Gabriel Katz, Associate Professor of Politics and Quantitative Methods at the University of Exeter introduces this online resource which will explore fundamental aspects of modern Bayesian computation.



   Download transcript    |   [ 850 Views ]

The Basics of Bayesian Computation

In this video, Gabriel goes over the basics of Bayesian computation, working through a series of exercises.



   Download transcript    |   [ 868 Views ]

Metropolis-Hastings

In this video, Gabriel looks at the second main algorithm used in Bayesian computations, which is the Metropolis Hastings algorithm and can be used when sampling from the conditional distributions is not possible.



   Download transcript    |   [ 827 Views ]

Gibbs Sampling

In this video, Gabriel illustrates the Gibbs Sampling algorithm and how it can be used to break down a complicated problem into a series of issue conditional problems.



   Download transcript    |   [ 4573 Views ]

Speeding up Bayesian computations

In this video, Gabriel talks though how researchers are able to use Bayesian methods to speed up calculations and reduce the time taken, whether that is using c+++, cluster computing services or cloud computing.



   Download transcript    |   [ 171 Views ]

Assessing Convergence

In this video, Gabriel explores the issue of convergence and how we assess convergence when estimating models in a Bayesian fashion using modern Bayesian computational approaches. This includes using traceplots and chains starting from different initial values.



   Download transcript    |   [ 580 Views ]



About the author

Gabriel earned his PhD from the California Institute of Technology (Caltech) in June 2010 and aslo holds an MSc. from Caltech and a Licenciatura in Economics from the Universidad de la Republica, in Uruguay. Gabriel's work focuses on Latin American Politics, Political Behavior, Political Economy and Research Methods, and has been published in leading economics and political science journals such as the American Journal of Political Science, Comparative Political Studies, the Economic Journal and the Journal of Law, Economics & Organization, among others. Gabriel's research uses advanced quantitative methods - mainly Bayesian statistics, but also causal inference and experimental techniques - to address substantive questions in comparative politics, political behaviour and political economy.

Primary author profile page



BACK TO TOP