Evaluation of Bayesian Model


New Member
How can I evaluate if a created bayesian model is reliable model that makes good predication?
To conclude if the created bayesian model produces a low error rate.
If you want to prove the certainty of your estimated regression coefficients, you usually use 0.05 and 0.95 quantiles of your posterior parameter samples.

If you want to evaluate the "overall correctness" of our model, one possibility is the technique of "posterior predictive model checking". The basic idea is: For each simulated parameter-set of your posterior sample, you use the corresponding regression model in combination with Monte-Carlo simulations in order to produce "artificial observations". These new virtual datasets you can compare in many respects with the original data, e.g., comparing residual patterns, in order to verify that your model works properly.