Chapter map

1. 1. Region graph
2. 2. Adjacency matrix W
3. 3. Degree matrix D
4. 4. Graph Laplacian
5. 5. Independent component
6. 6. Spatial component
7. 7. Precision matrix
8. 8. Conditional mean
9. 9. Conditional variance
10. 10. ρ endpoint zero
11. 11. ρ endpoint one
12. 12. Likelihood link
13. 13. Prior for ρ
14. 14. Posterior mapping
15. 15. Graph sensitivity
16. 16. Held-out prediction

Visit the How Mathematics Works hub for the wider series and its connections to education, reasoning, science, computing, design and careers.

---

1. Region graph

Here the geometry becomes operational. List areas and defensible neighbours.

A concrete situation

Three regions form chain 1–2–3.

What the mathematics does not license

Borders do not always mean interaction. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Justify each edge. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

2. Adjacency matrix W

This step deserves a careful pause. Set wij for connected pairs.

A concrete situation

W12=W21=W23=W32=1.

What the mathematics does not license

Asymmetry changes the model. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Check symmetry. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

3. Degree matrix D

The useful question is not only how to calculate, but what the calculation assumes. Place neighbour counts on the diagonal.

A concrete situation

D=diag(1,2,1).

What the mathematics does not license

Islands create zero degrees. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Define island handling. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

4. Graph Laplacian

A small example makes the mechanism visible. Compute D−W.

A concrete situation

The chain Laplacian has row sums zero.

What the mathematics does not license

It is singular in the intrinsic limit. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Inspect rank. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

5. Independent component

This is where modelling choices begin to affect evidence. Use (1−ρ)I.

A concrete situation

At ρ=.6, the identity weight is .4.

What the mathematics does not license

Parameter scaling conventions vary. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Match software documentation. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

6. Spatial component

The central mathematical idea is concrete. Use ρ(D−W).

A concrete situation

The Laplacian weight is .6.

What the mathematics does not license

One global ρ assumes stationarity. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Test boundaries. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

7. Precision matrix

Here the geometry becomes operational. Combine Q=τ[(1−ρ)I+ρ(D−W)].

A concrete situation

Positive τ scales all entries.

What the mathematics does not license

Near-one ρ can create numerical problems. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Check eigenvalues. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

8. Conditional mean

This step deserves a careful pause. Derive the neighbour-weighted centre.

A concrete situation

For u1=.2,u3=−.1, mean u2=.0375 at ρ=.6.

What the mathematics does not license

This is a prior conditional value. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Separate it from posterior risk. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

9. Conditional variance

The useful question is not only how to calculate, but what the calculation assumes. Use the diagonal precision.

A concrete situation

Var(u2|others)=1/[τ(1.6)].

What the mathematics does not license

High degree shrinks more strongly. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Explain degree effects. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

10. ρ endpoint zero

A small example makes the mechanism visible. Verify the independent model.

A concrete situation

Q=τI.

What the mathematics does not license

Spatial evidence may still enter through covariates. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Keep terms separate. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

11. ρ endpoint one

This is where modelling choices begin to affect evidence. Verify the ICAR limit.

A concrete situation

Q=τ(D−W).

What the mathematics does not license

The constant vector is unpenalized. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Apply a sum-to-zero constraint. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

12. Likelihood link

The central mathematical idea is concrete. Combine the effect with a count or continuous-data model.

A concrete situation

Poisson log risk includes ui.

What the mathematics does not license

A good prior cannot repair a poor likelihood. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Check dispersion. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

13. Prior for ρ

Here the geometry becomes operational. Assign a declared prior on dependence.

A concrete situation

A prior distributes mass over 0 to 1.

What the mathematics does not license

Uniform is not neutral on all scales. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Run prior predictive checks. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

14. Posterior mapping

This step deserves a careful pause. Summarize regional effects and risks.

A concrete situation

Intervals widen in low-information regions.

What the mathematics does not license

Means alone overstate certainty. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Map intervals. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

15. Graph sensitivity

The useful question is not only how to calculate, but what the calculation assumes. Compare plausible adjacency and travel graphs.

A concrete situation

One bridge edge changes smoothing.

What the mathematics does not license

Choosing the nicest map is circular. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Report all declared graphs. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

16. Held-out prediction

A small example makes the mechanism visible. Predict later periods or reserved regions.

A concrete situation

The model forecasts counts with intervals.

What the mathematics does not license

Interpolation is easier than extrapolation. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report predictive coverage. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

A seven-day practice lab

1. Day 1: define the decision, observation unit, X variables and Y or reference condition.
2. Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
3. Day 3: build a transparent baseline and write what it assumes.
4. Day 4: fit the main method inside a training-only pipeline; record every parameter.
5. Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
6. Day 6: compare validation metrics with the practical workload created by errors.
7. Day 7: present one supported conclusion, one unresolved uncertainty and one safe next experiment.

Questions for a student, parent or mentor

• Which part of the model is a mathematical definition, and which part is a choice?
• What small calculation would catch a sign, scale or dimension error?
• What evidence comes from training, validation and the untouched test?
• What benign condition could look like a fault or prediction failure?
• How would the answer change for another response, asset, school project or time period?
• What should a human review before acting on the score?

A method-specific audit before deployment

The Leroux conditional autoregressive model gives one spatial random effect a tunable balance between independence and neighbour agreement. Its precision matrix is proportional to Q(ρ)=(1−ρ)I+ρ(D−W), where W records neighbourhood connections and D contains each area’s number of neighbours. The parameter ρ lies between zero and one and changes the geometry of the prior.

When ρ=0, Q is the identity and regional effects are independent. As ρ approaches one, the graph Laplacian D−W dominates and neighbouring effects are encouraged to be similar; at ρ=1 the intrinsic CAR limit requires an identification constraint. A separate precision τ controls the overall magnitude of the random effects. This single component avoids the two-effect decomposition of the original BYM form.

Consider a three-area chain 1–2–3 with u1=.2 and u3=−.1. At ρ=.6, the conditional mean of u2 is .6(u1+u3)/(.6×2+.4)=.06/1.6=.0375. With precision τ, its conditional variance is 1/[τ(1.6)]. The calculation shows how both neighbours and the independent term contribute; it is a prior conditional calculation, not a fitted disease-risk estimate.

A single ρ applies across the whole map, so the model can oversmooth a true boundary or undersmooth a broad region. The graph, islands, scale and priors matter, while ρ can be weakly identified in sparse data. Responsible work publishes W and Q, checks eigenvalues and constraints, uses prior and posterior predictive diagnostics, maps uncertainty and compares plausible graphs without choosing the prettiest result.

---

Frequently asked questions

What is a CAR model?

It defines each regional random effect conditionally from neighbouring effects and a precision rule.

What does ρ control?

It balances independent variation against graph-based spatial dependence.

What happens at ρ=0?

The spatial precision reduces to an independent-effect structure.

What happens at ρ=1?

The intrinsic CAR graph Laplacian dominates and an identification constraint is required.

What does τ control?

It scales the overall precision, and therefore the variance, of the spatial effect.

Can one ρ represent every boundary?

Often imperfectly; localized discontinuities and nonstationary dependence require sensitivity checks or richer models.

---

Primary source and fact-check note

Brian G. Leroux, Xingye Lei and Norman Breslow, “Estimation of disease rates in small areas: a new mixed model for spatial dependence,” in Statistical Models in Epidemiology, the Environment, and Clinical Trials, 179–192 (2000), DOI 10.1007/978-1-4612-1284-3_4. Definitions and core claims in this article were checked against this primary publication. Calculations shown here were independently recomputed. Examples are educational illustrations; they are not reported industrial performance results.

Useful next reading

• Besag–York–Mollié disease mapping
• Oden’s Ipop statistics
• How Mathematics Works

Keep the options open. Mathematics education supports clearer questions, transferable problem-solving skills and informed choices across science, engineering, computing, AI, statistics, design and many other careers. It does not guarantee intelligence, income, admission or employment; those outcomes depend on far more than one subject or model.

Extended laboratory: topic-specific validation and responsible use

Graph audit

Publish W, D, islands and edge rationale.

Matrix audit

Verify Q entries, symmetry, rank and eigenvalues.

Endpoint audit

Recover independence at ρ=0 and ICAR at ρ=1.

Prior audit

Check τ and ρ before observing outcomes.

Constraint audit

Separate intercept and spatial effect.

Sensitivity audit

Compare only plausible graphs and priors.

Predictive audit

Score held-out counts or regions.

Final claim

Report spatial dependence with uncertainty, not a mechanism.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Region graph

Begin with this mechanism: List areas and defensible neighbours. Recreate the concrete case: Three regions form chain 1–2–3. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Borders do not always mean interaction. Complete the transfer step: Justify each edge. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 2: Adjacency matrix W

Begin with this mechanism: Set wij for connected pairs. Recreate the concrete case: W12=W21=W23=W32=1. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Asymmetry changes the model. Complete the transfer step: Check symmetry. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 3: Degree matrix D

Begin with this mechanism: Place neighbour counts on the diagonal. Recreate the concrete case: D=diag(1,2,1). Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Islands create zero degrees. Complete the transfer step: Define island handling. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 4: Graph Laplacian

Begin with this mechanism: Compute D−W. Recreate the concrete case: The chain Laplacian has row sums zero. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: It is singular in the intrinsic limit. Complete the transfer step: Inspect rank. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 5: Independent component

Begin with this mechanism: Use (1−ρ)I. Recreate the concrete case: At ρ=.6, the identity weight is .4. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Parameter scaling conventions vary. Complete the transfer step: Match software documentation. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 6: Spatial component

Begin with this mechanism: Use ρ(D−W). Recreate the concrete case: The Laplacian weight is .6. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: One global ρ assumes stationarity. Complete the transfer step: Test boundaries. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 7: Precision matrix

Begin with this mechanism: Combine Q=τ[(1−ρ)I+ρ(D−W)]. Recreate the concrete case: Positive τ scales all entries. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Near-one ρ can create numerical problems. Complete the transfer step: Check eigenvalues. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 8: Conditional mean

Begin with this mechanism: Derive the neighbour-weighted centre. Recreate the concrete case: For u1=.2,u3=−.1, mean u2=.0375 at ρ=.6. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: This is a prior conditional value. Complete the transfer step: Separate it from posterior risk. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 9: Conditional variance

Begin with this mechanism: Use the diagonal precision. Recreate the concrete case: Var(u2|others)=1/[τ(1.6)]. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: High degree shrinks more strongly. Complete the transfer step: Explain degree effects. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 10: ρ endpoint zero

Begin with this mechanism: Verify the independent model. Recreate the concrete case: Q=τI. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Spatial evidence may still enter through covariates. Complete the transfer step: Keep terms separate. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 11: ρ endpoint one

Begin with this mechanism: Verify the ICAR limit. Recreate the concrete case: Q=τ(D−W). Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: The constant vector is unpenalized. Complete the transfer step: Apply a sum-to-zero constraint. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 12: Likelihood link

Begin with this mechanism: Combine the effect with a count or continuous-data model. Recreate the concrete case: Poisson log risk includes ui. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: A good prior cannot repair a poor likelihood. Complete the transfer step: Check dispersion. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 13: Prior for ρ

Begin with this mechanism: Assign a declared prior on dependence. Recreate the concrete case: A prior distributes mass over 0 to 1. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Uniform is not neutral on all scales. Complete the transfer step: Run prior predictive checks. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 14: Posterior mapping

Begin with this mechanism: Summarize regional effects and risks. Recreate the concrete case: Intervals widen in low-information regions. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Means alone overstate certainty. Complete the transfer step: Map intervals. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 15: Graph sensitivity

Begin with this mechanism: Compare plausible adjacency and travel graphs. Recreate the concrete case: One bridge edge changes smoothing. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Choosing the nicest map is circular. Complete the transfer step: Report all declared graphs. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.

Workshop 16: Held-out prediction

Begin with this mechanism: Predict later periods or reserved regions. Recreate the concrete case: The model forecasts counts with intervals. Write region IDs, neighbour edges, W entries, D entries, Laplacian entries, rho values, tau values, precision entries, conditional means, conditional variances, eigenvalues, posterior effects and predictive counts with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Interpolation is easier than extrapolation. Complete the transfer step: Report predictive coverage. Change exactly one observed count, expected count, population standard, gamma parameter, reliability weight, neighbour edge, rho value, precision, constraint, time period, interaction type, coordinate, triangle, gradient threshold, connection rule, null model, replicate count, random seed, observation period or validation region. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, geometry, likelihood, prior or null model, computation, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete regions, periods or samples outside tuning when testing transfer.