Post

Based on 6 simulations, we provide practical recommendations. For context, the main limiting factor on using BF is practical: they are hard to compute. Researchers mostly have to resort to numerical estimation. BS is flexible but computationally expensive; SDDR less flexible, but cheaper.

The upshot: all methods can give accurate estimates, but the cost-benefit trade-off shifts depending on (1) computational cost of MCMC sampling, (2) normality of the posterior, (3) number of model parameters, and (4) test value tail probability.

1. BS or SDDR? If MCMC sampling is cheap, BS is a good default. But we found an unfavorable cost–benefit trend: with more parameters, BS becomes computationally expensive while BF estimates become less accurate. In models with many parameters, SDDR is often cheaper and at least as accurate.

Thanks for sharing! Very interesting! I'm honestly surprised by the low(ish) variability of the SDDR! In {bayestestR} we currently use logspline - I'm interested in hearing any recommendations you might have to improve there 🙂

I think the documentation should note the importance of a sufficient number of MCMC samples and tail probability of the test value. An estimate of the tail probability (`plogspline()`) might be a useful "diagnostic."

Can you elaborate on the tail probability? How would you do that, and what would you expect to see / hope not to see?

I prefer multiple bridge sampling, also known as adaptive umbrella sampling in other fields of science 🙂

I‘d be interested to read more. Is there a paper you would recommend?