'Prior Exposure' Bayesian data analysis (BDA) workshop 2: Doing BDA in the social sciences

Description:

Workshop 2: Doing Bayesian data analysis in the social sciences

(Note that this is part of a series of four workshops that can be attended as stand-olone sessions or in sequence. Workshop 1 runs the previous day in the same venue - and meets the prerequisites for this workshop. Workshops 3 and 4 will run later this year. We plan to run similar workshops in 2016 and 2017).

Bursary information A limited number of ESRC funded bursaries (to cover travel and subsistance costs) are available to UK postgraduate social science students.

Prerequisites The prerequisites for this workshop are fulfilled by Workshop 1. They include a general under- standing of the core concepts of Bayesian statistical inference, and the general distinction between classical and Bayesian statistical methods.

Content This workshop aims to provide a solid theoretical and practical foundation for real-world BDA in psychology and social sciences. It will focus primarily on the linear statistical models and so-called conjugate prior distributions. The reason for this focus is twofold. First, linear models – which include t tests, ANOVA, and linear regression models – are the core of the standard repertoire of statistical models with which our audience will be familiar. Studying the Bayesian counterparts of these approaches will therefore be a natural transition. Second, Bayesian inference in linear models with conjugate priors is analytically tractable, and this entails, amongst other things, that we can use relatively simple formulae to calculate the posterior distribu- tion over the parameters and to make predictive inferences. This allows us to illustrate the general nature of Bayesian inference quickly and easily, postponing the computational and practical complications that arise as a consequence of performing Monte Carlo based numerical approaches to inference. In practical terms, this workshop will involve the use of the R statistical computing environment both to calculate posterior distribu- tions in linear models and to graphically illustrate them. Indeed, graphically illustrating, for example, how the posterior distribution is a weighted average of the prior and likelihood functions and how the contribution of the likelihood function grows rapidly with increasing data, provides compelling intuitive insight into the nature of Bayesian inference.

Learning outcomes On completion of this workshop, we expect attendees to be able to confidently perform and understand the Bayesian counterparts to many of the models with which they would be already famil- iar. They will also become familiar with new concepts – conjugate priors, posterior predictive distributions, marginalized likelihood functions – that arise only in the context of BDA.

Indicative reading

Lee, P. M. (2004). Bayesian Statistics: An Introduction. London, UK: Hodder Arnold.
Gelman, A., Carlin, J. B., Stern, H. S., Rubin, D. B. (2003). Bayesian data analysis (2nd ed.). Chapman & Hall.

Cost:

£10 (for postgraduate students)
£20 (for others)

Website and registration:

http://onlinestore.ntu.ac.uk/browse/extra_info.asp?compid=1&modid=2&deptid=39&catid=22&prodid=60

Region:

East Midlands

Keywords:

Quantitative Data Handling and Data Analysis, Statistical Theory and Methods of Inference, Bayesian methods, R, Bayesian data analysis , social sciences

Related publications and presentations:

Quantitative Data Handling and Data Analysis
Statistical Theory and Methods of Inference
Bayesian methods
R

Back to archive...