Prior exposure: Training in Bayesian data analysis to complement and enhance traditional statistical methods for psychology and the social sciences

点击翻译按钮获取中文摘要

Bayesian methods are now part of the standard repertoire of statistical methods in science and are becoming increasingly important in psychology and the social sciences in field as diverse as election forecasting, brain imaging and clinical psychology. However, while top international journals in these fields regularly publish quantitative work using sophisticated Bayesian methods, opportunities for psychologists and social psychologists in the UK to acquire skills in Bayesian data analysis are few and far between. Postgraduate social science training in the UK does not typically introduce these methods, while undergraduate and postgraduate training in traditional statistical methods will not cover some fundamental philosophical and statistical concepts necessary for advanced training in Bayesian methods.Offering training in Bayesian data analysis for psychologists and social scientists therefore involves overcoming a number of barriers. These include the differences in prior training (e.g., those with a basic quantitative social science grounding and those with experience of more advanced methods in a non-Bayesian context) and differences in technical competence (ranging from students familiar with spreadsheets and basic statistical packages to those with experience writing their own computer code for data analysis). However, perhaps the largest barrier is the difference in philosophy between Bayesian and traditional statistical methods. Although the different philosophies have historically been a source of conflict, recent developments in statistical science emphasise the practical advantages of drawing on both sets of methods. We propose a training package that aims to present Bayesian tools as a way to complement and enhance traditional statistical methods.Our training package comprises a series of four one-day workshops that begin with an introduction to the philosophy and practice of Bayesian data analysis and end with sophisticated Bayesian tools that allow researchers to handle messy real world data that are difficult to model using traditional approaches. Participants may attend one or more individual workshops (e.g., to understand basic Bayesian methods enough to review a paper or improve their collaboration with a more expert colleague) or attend the entire sequence. In addition to the workshops a final symposium will offer an opportunity for research students and early career researchers to present work involving Bayesian data analysis alongside papers by experts in the field. The overarching theme for the symposium is the teaching of Bayesian data analysis. On completion of the project the training resources will be made available to researchers under an open source license in order to facilitate the development of new training and to support the integration of Bayesian data analysis into the curriculum.