The full code for this notebook is available on GitHub. A good first amelioration of these problems is to accompany your Bayes Factors with posterior predictive checks for absolute model fit and sensitivity analyses for robustness to theoretically unimortant changes in model structure and priors.

Namely, some researchers go for parameter estimation rather than Bayes Factors because BFs are too relative: (1) Bayes factors just change credence between models without quantifying whether the better model is actually credible enough, and (2) they are relative to the prior, yet people often discuss Bayes Factors as if they are purely data-driven. This is just a demonstration of syntax which ignores some controversies surrounding Bayes Factors. As flexibility increases so do complexity, so find you own golden middle way. In each section, I start with the most limited methods and then progress towards the all-purpose solutions. Then I go on to demonstrate Bayes Factors for mixed models using the same packages, including a more thorough discussion of pros and cons. I start with a TL DR section showing off the syntax for the simplest of all models: the intercept-only model.

Here, I compare different ways of computing Bayes Factors in R.

Bayes test models