An agent-based model of Bayesian inference as an educational tool

Abstract

The scientific and practical importance of correct Bayesian reasoning can hardly be overrated. But what is the best method to teach it and to fight widespread and harmful misconceptions like the false positive paradox? We studied the existing literature on the question and experimented with math cubes to find more intuitive visualizations. These inspired us to start experimenting with the programmable modeling environment NetLogo. The result is an agent-based model of Bayesian inference as an educational tool that combines the most successful teaching methods found so far: the use of natural frequencies, relatable visualization, interactive visualization, and visualization that can easily be reproduced with physical props. The model can be modified, improved, extended, and used in many ways. While testing in controlled experiments is still outstanding, the result thus far leads us to conclude that agent-based modeling is a most promising way of teaching Bayesian inference and illustrating it for practical purposes.