# Workshop Guide: Bayesian Statistics

## Audience

Analysts, researchers, data scientists, and decision-makers using models under uncertainty.

## 60-Minute Agenda

1. 0-10 min: Choose a grouped or uncertain decision problem.
2. 10-20 min: Write the generative story.
3. 20-35 min: Draft candidate models and priors.
4. 35-50 min: Choose prior and posterior predictive checks.
5. 50-60 min: Share decision summary and strongest model critique.

## 90-Minute Agenda

1. 0-10 min: Review the worked model criticism example.
2. 10-25 min: Teams define decision and generative story.
3. 25-40 min: Teams choose complete-pooling, no-pooling, and partial-pooling alternatives.
4. 40-55 min: Priors and prior predictive expectations.
5. 55-70 min: Computation and posterior predictive checks.
6. 70-90 min: Decision summary and critique.

## Team Exercise

Each team produces a model criticism report with:

- decision
- generative story
- candidate models
- priors
- prior predictive checks
- computational diagnostics
- posterior predictive checks
- model comparison
- recommendation

## Discussion Prompts

- What data would the priors generate before fitting?
- Which posterior predictive check matters for the decision?
- Where might pooling borrow too much information?
- What rollout decision follows from uncertainty?

## Facilitator Notes

Make teams explain the model in words before equations. The model is a checked assumption set, not a machine for producing authority.

Common failure modes:

- vague priors with implausible implications
- ignoring divergences or R-hat
- generic posterior predictive checks unrelated to the decision
- reporting posterior means without action guidance

## Review Standard

Use `final-assessment.md` as the rubric. A strong report makes assumptions visible and connects posterior uncertainty to a decision.