Teaching Bayesian Statistics
By Diogo Ferrari, University of California, Riverside
The questions discussed in this article are: What challenges emerge when teaching introductory-level Bayesian statistics to non-statistics graduate students who want to become well-informed users of Bayesian methods, and what content-selection strategies can instructors adopt to overcome those challenges? Before addressing these questions, we may ask why this consideration is important and if there are any distinctive challenges to teaching Bayesian statistics that differ from teaching non-Bayesian statistics.
Bayesian statistics regained popularity among statisticians and applied researchers in various disciplines in the 1990s. By the end of the decade, scholars were debating whether the Bayesian approach to inference should be taught at the undergraduate level. The third edition of the 1997 issue of The American Statistician dedicated its Teachers’ Corner section to the topic. It included articles by Moore (Reference Moore, Panchapakesan and Balakrishnan), Berry (Reference Berry), and Albert (Reference Albert) and discussions by well-known statisticians: Jeffrey Witmer, Thomas Short, Dennis Lindley, David Freedman, and Richard Scheaffer. Moore (Reference Moore, Panchapakesan and Balakrishnan) advocated for not teaching Bayesian statistics at the elementary level because Bayesian methods were not widely used yet, conditional probabilities are difficult to understand, there were no suitable textbooks, and—unlike the frequentist approach—there was no standard set of prescriptive procedures to teach students. Berry (Reference Berry) and Albert (Reference Albert) countered these points, arguing that Bayesian inference is more intuitive and easier to grasp than frequentist approaches.