Chapter map

1. 1. Vectors
2. 2. Dimension
3. 3. Mean vector
4. 4. Covariance
5. 5. Whitening
6. 6. Affine invariance
7. 7. Empirical CF
8. 8. Normal CF
9. 9. Difference
10. 10. Weight
11. 11. Integral
12. 12. Kernel form
13. 13. Calibration
14. 14. Diagnostics
15. 15. Robustness
16. 16. Claim

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

---

1. Vectors

Here the geometry becomes operational. Define one independent vector per sampling unit.

A concrete situation

Each student contributes height and reach.

What the mathematics does not license

Two rows per student create dependence. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Keep one unit table. 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. Dimension

This step deserves a careful pause. Count measured coordinates p.

A concrete situation

The example has p=2.

What the mathematics does not license

Dimension relative to n affects stability. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Report n and p. 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. Mean vector

The useful question is not only how to calculate, but what the calculation assumes. Average each coordinate.

A concrete situation

Subtract the two coordinate means.

What the mathematics does not license

Missingness can change the analysed cohort. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Freeze rows. 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. Covariance

A small example makes the mechanism visible. Calculate the sample covariance matrix.

A concrete situation

Diagonal and cross-products enter.

What the mathematics does not license

Collinearity makes inversion unstable. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Inspect 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.

---

5. Whitening

This is where modelling choices begin to affect evidence. Apply S inverse one-half to centred vectors.

A concrete situation

Units become standardized coordinates.

What the mathematics does not license

Different matrix roots are equivalent geometrically. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Verify distances. 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. Affine invariance

The central mathematical idea is concrete. Rotate and rescale the original axes.

A concrete situation

The standardized evidence should agree.

What the mathematics does not license

Nonlinear transforms change the target. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Document transforms. 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. Empirical CF

Here the geometry becomes operational. Average exp(i t transpose z_j).

A concrete situation

Every residual contributes a phase.

What the mathematics does not license

A grid approximation can miss frequencies. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use the closed form. 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. Normal CF

This step deserves a careful pause. Write exp(-||t|| squared/2).

A concrete situation

It is determined by standardized normality.

What the mathematics does not license

It is not fitted separately at each t. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep one benchmark. 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. Difference

The useful question is not only how to calculate, but what the calculation assumes. Subtract the two complex functions.

A concrete situation

Use squared modulus.

What the mathematics does not license

Real and imaginary parts both matter. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Check both. 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. Weight

A small example makes the mechanism visible. Apply the Gaussian weight.

A concrete situation

Nearby frequencies receive controlled emphasis.

What the mathematics does not license

Bandwidth changes sensitivity. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use the published choice. 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. Integral

This is where modelling choices begin to affect evidence. Integrate over all t in R^p.

A concrete situation

The result is nonnegative.

What the mathematics does not license

Numerical truncation can bias it. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Prefer verified formula. 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. Kernel form

The central mathematical idea is concrete. Express the integral with pairwise distances.

A concrete situation

Distances 1,4,1 yield different similarities.

What the mathematics does not license

The toy cloud is singular in 2D. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Treat it as a warning. 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. Calibration

Here the geometry becomes operational. Use the validated lognormal or simulation approximation.

A concrete situation

Record software and version.

What the mathematics does not license

Approximation quality varies with n and p. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Run a null simulation. 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. Diagnostics

This step deserves a careful pause. Inspect scatterplots and squared radii.

A concrete situation

A line-shaped cloud explains failure.

What the mathematics does not license

One omnibus number hides mechanism. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Trace contributors. 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. Robustness

The useful question is not only how to calculate, but what the calculation assumes. Check duplicates, clusters and influential rows.

A concrete situation

Remove nothing without a rule.

What the mathematics does not license

Post-hoc deletion invalidates inference. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Report sensitivity. 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. Claim

A small example makes the mechanism visible. State joint normal-model evidence.

A concrete situation

The vector cloud departs from the fitted ellipse.

What the mathematics does not license

No causal mechanism is established. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Investigate design. 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 Henze–Zirkler approach compares the empirical characteristic function of standardized multivariate residuals with the characteristic function of a multivariate normal distribution. The squared difference is integrated with a Gaussian weight. This turns a continuum of frequency-domain comparisons into one affine-invariant statistic.

Start with two-dimensional observations, subtract the sample mean and transform by the inverse square root of the sample covariance matrix. The resulting residual cloud has a common centre and standardized geometry. Pairwise squared distances between these residuals enter a closed-form version of the weighted integral, so the calculation can be audited without numerically integrating every frequency.

Imagine standardized residuals z1=(-1,0), z2=(0,0) and z3=(1,0). Their squared pairwise distances are 1,4 and 1 for the three distinct pairs. A Gaussian kernel converts each distance into a similarity. The cloud is visibly confined to a line, so in two dimensions its covariance is singular; that makes it an excellent warning example because the ordinary full-rank test is not defined until the design supplies enough non-collinear information.

A significant result says the joint distribution is inconsistent with the fitted non-degenerate multivariate normal model. It does not locate the cause by itself. Follow with marginal and pairwise plots, Mahalanobis-distance diagnostics, checks for duplicate rows and clusters, and subject-matter review of the observations that contribute unusual geometry.

---

Frequently asked questions

What does the Henze–Zirkler test assess?

It assesses whether an independent vector sample follows a non-degenerate multivariate normal distribution.

What is a characteristic function?

It is an expected complex exponential that uniquely represents a probability distribution.

Why standardize the residuals?

Standardization removes location, scale and linear coordinate choice from the target geometry.

What does Gaussian weighting do?

It makes the integrated squared discrepancy finite and controls which frequencies receive emphasis.

Can the test work with singular covariance?

The ordinary full-rank procedure requires an invertible covariance matrix.

Does rejection identify the cause?

No. Plots and component diagnostics are needed to locate skewness, tails, clusters or dependence.

---

Primary source and fact-check note

Norbert Henze and Bernhard Zirkler, “A Class of Invariant Consistent Tests for Multivariate Normality,” Communications in Statistics—Theory and Methods 19(10), 3595–3617 (1990), DOI 10.1080/03610929008830400. 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

• Mardia skewness and kurtosis
• Shapiro–Francia normality test
• 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

Design audit

Freeze independent multivariate rows and inclusion rules.

Geometry audit

Reproduce centring, covariance and transformations.

Statistic audit

Show every score, distance or component contribution.

Calibration audit

Match sample size, dimension and dependence treatment.

Graph audit

Preserve probability plots, scatterplots and radius diagnostics.

Magnitude audit

Explain the practical consequence in original units.

Sensitivity audit

Use prespecified alternatives and influence checks.

Final claim

Keep inference inside the model and sampling design.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Vectors

Begin with this mechanism: Define one independent vector per sampling unit. Recreate the concrete case: Each student contributes height and reach. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Two rows per student create dependence. Complete the transfer step: Keep one unit table. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 2: Dimension

Begin with this mechanism: Count measured coordinates p. Recreate the concrete case: The example has p=2. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Dimension relative to n affects stability. Complete the transfer step: Report n and p. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 3: Mean vector

Begin with this mechanism: Average each coordinate. Recreate the concrete case: Subtract the two coordinate means. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Missingness can change the analysed cohort. Complete the transfer step: Freeze rows. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 4: Covariance

Begin with this mechanism: Calculate the sample covariance matrix. Recreate the concrete case: Diagonal and cross-products enter. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Collinearity makes inversion unstable. Complete the transfer step: Inspect eigenvalues. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 5: Whitening

Begin with this mechanism: Apply S inverse one-half to centred vectors. Recreate the concrete case: Units become standardized coordinates. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Different matrix roots are equivalent geometrically. Complete the transfer step: Verify distances. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 6: Affine invariance

Begin with this mechanism: Rotate and rescale the original axes. Recreate the concrete case: The standardized evidence should agree. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Nonlinear transforms change the target. Complete the transfer step: Document transforms. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 7: Empirical CF

Begin with this mechanism: Average exp(i t transpose z_j). Recreate the concrete case: Every residual contributes a phase. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A grid approximation can miss frequencies. Complete the transfer step: Use the closed form. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 8: Normal CF

Begin with this mechanism: Write exp(-||t|| squared/2). Recreate the concrete case: It is determined by standardized normality. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: It is not fitted separately at each t. Complete the transfer step: Keep one benchmark. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 9: Difference

Begin with this mechanism: Subtract the two complex functions. Recreate the concrete case: Use squared modulus. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Real and imaginary parts both matter. Complete the transfer step: Check both. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 10: Weight

Begin with this mechanism: Apply the Gaussian weight. Recreate the concrete case: Nearby frequencies receive controlled emphasis. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Bandwidth changes sensitivity. Complete the transfer step: Use the published choice. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 11: Integral

Begin with this mechanism: Integrate over all t in R^p. Recreate the concrete case: The result is nonnegative. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Numerical truncation can bias it. Complete the transfer step: Prefer verified formula. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 12: Kernel form

Begin with this mechanism: Express the integral with pairwise distances. Recreate the concrete case: Distances 1,4,1 yield different similarities. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The toy cloud is singular in 2D. Complete the transfer step: Treat it as a warning. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 13: Calibration

Begin with this mechanism: Use the validated lognormal or simulation approximation. Recreate the concrete case: Record software and version. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Approximation quality varies with n and p. Complete the transfer step: Run a null simulation. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 14: Diagnostics

Begin with this mechanism: Inspect scatterplots and squared radii. Recreate the concrete case: A line-shaped cloud explains failure. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One omnibus number hides mechanism. Complete the transfer step: Trace contributors. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 15: Robustness

Begin with this mechanism: Check duplicates, clusters and influential rows. Recreate the concrete case: Remove nothing without a rule. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Post-hoc deletion invalidates inference. Complete the transfer step: Report sensitivity. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: State joint normal-model evidence. Recreate the concrete case: The vector cloud departs from the fitted ellipse. Write rows, dimensions, mean vectors, covariance matrices, standardized residuals, frequencies, characteristic functions, weights, pairwise distances and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: No causal mechanism is established. Complete the transfer step: Investigate design. Change exactly one observation, row, coordinate, tie, plotting position, normal score, covariance entry, distance, bandwidth, component statistic, dimension, dependence estimate, null approximation or simulation seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible plot or geometry, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from skewness, heavy tails, clusters, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.