A Complete Beginner’s Guide to Bayesian Analysis
“How probable is it that my hypothesis explains this data, compared to some other hypothesis?”. This is a reasonable question that many of us want to answer. Surprisingly, traditional statistical techniques that use p-values (which I will call NHST for brevity) cannot answer that question. The Bayesian framework can. The framework can also find evidence to support the null hypothesis of no effect, something that is not possible under NHST. Some researchers have argued the traditional approach should no longer be used, with some journals now expressly forbidding NHST analyses in their articles.
This course is a practical, hands-on guide to Bayesian analysis for those who are entirely new to the area. The course will provide an intuition for the Bayesian framework and concepts such as priors, posteriors, and Bayes factors. The practical element of the course will focus on employing Bayes factors with reference priors. You will learn how the Bayes factor can be used as evidence in favour of a particular hypothesis relative to another.
Unlike many other courses, this course will require no computer programming or scripting. We will not be using BUGS, R, or MATLAB. Instead, this course will use an open-source platform called JASP, available on Windows, Mac, and Linux-based OSs. The software has many similarities with SPSS, which may help those who have prior SPSS experience. Prior SPSS experience is not expected or required, but a foundational understanding of the general linear model (t-tests and ANOVAs) is required. An understanding of model comparison will help, but is not a requirement.
Course Outline (subject to change)
1. Traditional Statistics, NHST, and the Issues Associated with Them
2. A Conceptual Understanding of Bayesian Analysis: Priors, Likelihoods, and Posteriors
3. The Bayes Factor for Comparing Hypotheses: A Conceptual Overview
4. Practical Lab Session: Comparing Two Groups with a Bayesian Approach (t-test)
5. Building Models to Explain the World: A Conceptual Overview
6. Practical Lab Session: Comparing More than Two Groups with a Bayesian Approach (ANOVA)
7. Practical Lab Session: Building Explanatory Models with a Bayesian Approach (Regression)
8. Interpreting and Reporting the Bayes Factor
9. Bayes or NHST? A Critical Discussion of the Pros and Cons
Who should attend?
The course is designed for researchers and postgraduate students who use quantitative statistics. This course may be particularly useful for (i) those who wish to engage with the cutting edge of statistical approaches, (ii) those who wish to be able to predict on the side of the null, and/or (iii) those who have no computer programming experience.
The course is not intended for those who wish to advance their existing foundational understanding. The above outline should act as a guide, but you may contact Chris Street with informal enquiries. The course will not cover specification or choice of prior distributions, non-standard distributions, distribution estimation techniques such as MCMC, or hierarchical modelling. These are more advanced concepts that would require some computational programming experience and existing knowledge of the foundations of the Bayesian framework.
There are no prerequisites for taking this course except for a basic understanding of t-tests and ANOVA. Familiarity with regression and model comparison will help, but this is not a requirement and will be discussed on the course.
Researchers from economics, education, epidemiology, medical research, psychology, public health, social work, sociology, and similar disciplines are welcome.
You can book a place on the course at the University of Huddersfield’s online store.