Posterior predictive checks and sensitivity
Research context
research method · conditional · As of 2026-10-10
A posterior predictive distribution integrates p(y_new|θ) over p(θ|data), retaining parameter uncertainty under the declared model.
Key knowledge
A posterior predictive distribution integrates p(y_new|θ) over p(θ|data), retaining parameter uncertainty under the declared model. Compare replicated residual patterns, held-out standards, range dependence and alternative priors/models. Agreement on fitted points may conceal model discrepancy; extrapolation needs a separately justified uncertainty model.
Values and equations
p(y_new|data)=∫p(y_new|θ)p(θ|data)dθ
- y_new
- New response with observation units
- θ
- Model parameter vector with specified units
- p
- Predictive and parameter densities with corresponding reciprocal units
Conditions and scope
- Validation data must be appropriately independent of fitting/model selection; account for shared standards and systematic effects.
- Report prior sensitivity, posterior sampler convergence/effective sample size and predictive diagnostics. A credible interval is conditional on the model and need not have automatic frequentist coverage.
Sources · 2
References
Sources retain their recorded scope. A source link alone does not establish verification of every statement.
Evidence and review
Inspected §§8.5, 10.4–10.6, 11.4, 11.6–11.10, 12 and Annex B: shared effects, inverse calibration, time series and prior sensitivity; source check dated 2026-10-10. Scope is limited to the cited statements, equations and assumptions, not the whole publication. Inspected between-day excess scatter and hierarchical ratio-model account; project summary, not original experimental full text; source check dated 2026-10-10. Scope is limited to the cited statements, equations and assumptions, not the whole publication.
2026-10-10 · Author source inspection of the displayed claim, model assumptions and equations; independent exact-content and branch-inventory review is a separate gate. Contemporary status is a dated snapshot, not a promise of ongoing monitoring.
Entry details
Permanent link to this entry ↗- Record identity
- research-measurement-posterior-model-checks
- Revision
- research.1
- Structured reference record
- Postgraduate & research reference