Workshop: Bayesian methods in decision making research

[siteorigin_widget class=”SiteOrigin_Widget_Hero_Widget”][/siteorigin_widget]
[siteorigin_widget class=”SiteOrigin_Widget_Headline_Widget”][/siteorigin_widget]

This mini workshop aims to provide an introduction to using Bayesian methods for both data collection for decision making experiments, as well as for the analysis of that data. While a brief overview of different methods and software packages will be provided, the focus will naturally be on the methods that I know best, having developed them myself, and with my collaborator (Tom Rainforth, University of Oxford).

The workshop should be useful for anyone with a general interest in Bayesian methods, but we will also zoom into details of using the DARC Toolbox (Vincent & Rainforth, preprint) for data collection, and the hierarchical Bayesian methods outlined in Vincent (2016).

Session 1 – Adaptive methods in data collection

We will examine how to run efficient Bayesian adaptive experiments using the DARC Toolbox (Vincent & Rainforth, 2017). This toolbox is general, in that it can be used for many different delayed and risky choice tasks, but we shall focus primarily upon temporal discounting experiments. The plan is to:

  • run through an example adaptive experiment
  • discuss Bayesian adaptive methods
  • look at how to customise experiments
  • run through some of the more advanced features

 

Session 2 – Bayesian data analysis

In the second session, we will look at how to use Bayesian methods to analyse already collected data. I’ll give an overview of a few different alternative packages, but then focus upon the hierarchical analysis of delay discounting data, which was focussed on in Vincent (2016). Because running analyses can take some time, waiting for MCMC sampling to complete, this session will be a bit less interactive. Nevertheless, code examples will be given for a few examples.

Both toolboxes are currently implemented in Matlab. Time will be focussed on the current capabilities available in these Matlab versions, but I will briefly discuss progress towards Python implementations.

I will attempt to screen capture the event, but can’t make any promises that the AV quality will be good enough to post online.

[siteorigin_widget class=”SiteOrigin_Widget_Headline_Widget”][/siteorigin_widget]
[siteorigin_widget class=”SiteOrigin_Widget_Headline_Widget”][/siteorigin_widget]

This session should be suitable for those who just want to watch, but we’ll also be running through some code if people would like to bring their laptops. If you’d like to do this, then you’ll need a laptop with Matlab installed, ideally version 2016a or later. Efforts are being made to port to Python, but this will take some time.

Session 1 – Adaptive methods in data collection

Demo code 1

Demo code 2

Session 2 – Bayesian data analysis

Note that this toolbox is slowly being ported to Python, using PyMC3 as the core sampling engine.

[siteorigin_widget class=”SiteOrigin_Widget_Headline_Widget”][/siteorigin_widget]
[siteorigin_widget class=”SiteOrigin_Widget_Headline_Widget”][/siteorigin_widget]

Vincent, B. T. (2016). Hierarchical Bayesian estimation and hypothesis testing for delay discounting tasks., 48(4), 1608–1620. http://doi.org/10.3758/s13428-015-0672-2

Vincent, B. T. & Rainforth, T. (2017, October 20). The DARC Toolbox: automated, flexible, and efficient delayed and risky choice experiments using Bayesian adaptive design. Retrieved from psyarxiv.com/yehjb