Australasian Mathematical Psychology Conference 2019

Knowledge is prior: Using past model fits to develpp informative priors in model selection.

Caroline Kuhne
Cognitive Psychology, University of Newcastle
Scott Brown
University of Newcastle
Andrew Heathcote
University of Tasmania
Dora Matzke
University of Amsterdam
Han Tran
University of Amsterdam

Bayesian inference is a powerful tool in which likelihoods are assigned to events and evidence. An advantage of Bayesian inference is that it gives researchers the opportunity to use prior distributions to encapsulate their beliefs and knowledge. A prior is a probability distribution that represents our belief about the parameters of the model. When an informative prior is used, with a distribution of values more or less likely chosen based on previous research findings – analysis does not start from scratch, but instead the cumulative effects of all data, past and present, can be taken into account. When the prior selected is uninformative, analysis is starting from scratch. In certain situations, the choice of an uninformative versus informative prior does not matter. However when it comes to model selection, such as choosing one regression model over another, priors do matter. The issue is knowing what prior to choose. The current project seeks to collate numerous model analyses across cognitive data sets, to establish sensible priors for the linear ballistic accumulator decision making model.