# Final Assessment: Bayesian Statistics

## Assignment

Write a Bayesian model criticism report for a grouped or uncertain decision problem.

Your report must define the decision, generative story, candidate models, priors, prior predictive checks, computation diagnostics, posterior predictive checks, model comparison, and decision recommendation.

## Required Artifact

Submit a model criticism report.

Minimum sections:

- decision
- generative story
- candidate models
- priors
- prior predictive expectations
- computational diagnostics
- posterior predictive checks
- model comparison
- decision summary
- strongest critique

## Rubric

| Criterion | Strong | Needs revision |
|---|---|---|
| Decision | Names the decision the model supports | Reports a model without a decision |
| Generative story | Explains how the data could arise | Starts with formulas only |
| Priors | Priors are interpretable and checked before fitting | Priors are vague defaults |
| Computation | R-hat, ESS, divergences, and mixing are checked before interpretation | Ignores sampler diagnostics |
| Posterior predictive checks | Tests decision-relevant summaries and model mismatch | Shows generic fit plots only |
| Comparison | Compares models with prediction, calibration, interpretability, and decision consequences | Chooses only by one score |
| Recommendation | Translates uncertainty into rollout, holdback, or no-action guidance | Reports posterior means only |
| Critique | Names where the model may be structurally wrong | Treats the model as true |

## Pass Criteria

Pass if the report makes model assumptions visible and connects posterior uncertainty to a decision.

## Submission Checklist

- [ ] The decision is explicit.
- [ ] The generative story comes before formulas.
- [ ] Priors are checked with prior predictive simulation.
- [ ] Diagnostics are checked before posterior summaries.
- [ ] Posterior predictive checks are decision-relevant.
- [ ] Model comparison includes practical consequences.
- [ ] Recommendation states uncertainty and downside risk.

## Certificate Language

Completed Bayesian Statistics in Practice by producing a model criticism report with prior predictive checks, diagnostics, posterior predictive checks, and decision recommendation.