Day 2 Recap

What is the workflow for running a Bayesian analysis?

• Find a suitable likelihood.
• Identify the parameters in the model.
• Find appropriate priors (using, e.g., prior predictive checks).
• Fit the model.
• Check convergence.
• Compare posterior predictive distributions to the true data.
• If the model is off: Tweak it and begin again. Else: Report!

How do we fit a Bayesian model using brms?

• Actually much like we would with lme4-based analyses.
• And with a little more patience.

How do we check that it’s a good model?

• To check convergence, we use diagnostics like trace plots (🐛) and Rhat values.
• To see whether the model adequately captures the generative process behind the data, we use prior predictive cvhecks to compare data that the model generates to the true data we observed.

How would we report the results in a publication?

• Parameter estimates can be reported similarly to how you would report a frequentist model, bearing in mind that we are not rejecting any nulls, nor are we finding evidence for any effects.