WHY MATHEMATICS?

Why Mathematics? | Shapiro–Francia Test: Normal Order Statistics, Squared Correlation, Tail Sensitivity and Normality Evidence

How a straight line on a normal probability plot becomes an auditable statistic, what the statistic detects, and why model checking still needs graphs and design knowledge.

Why mathematics is important becomes easier to see when a real decision depends on it. This guide follows the mechanism from definitions and dimensions to a worked calculation, validation, practical learning and responsible use. It keeps factual evidence separate from interpretation and treats every illustrative scenario as an example rather than a promise of results.

Choose a reading route

Order the observations and centre the sample
Calculate expected normal-order scores
Square the correlation between data and scores
Calibrate the statistic and diagnose the visible departure

Chapter map

1. Question
2. Ordering
3. Plotting positions
4. Normal scores
5. Centre data
6. Numerator
7. Score norm
8. Data norm
9. Correlation
10. Statistic
11. Tail effect
12. Skewness effect
13. Calibration
14. Dependence
15. Decision
16. Claim

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

1. Question

The useful question is not only how to calculate, but what the calculation assumes. Name the model property being checked.

A concrete situation

A residual model assumes normal errors.

What the mathematics does not license

Testing raw outcomes may answer the wrong question. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Write the target. 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. Ordering

A small example makes the mechanism visible. Sort observations from smallest to largest.

A concrete situation

1,2,3,4,5.

What the mathematics does not license

The original labels still matter for diagnostics. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Keep row IDs. 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. Plotting positions

This is where modelling choices begin to affect evidence. Assign a declared probability to each order.

A concrete situation

Use (i-.375)/(n+.25).

What the mathematics does not license

Other conventions give slightly different scores. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

State the convention. 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. Normal scores

The central mathematical idea is concrete. Transform probabilities with the inverse normal CDF.

A concrete situation

Scores are about -1.180,-.497,0,.497,1.180.

What the mathematics does not license

Rounded scores change the hand result. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Retain precision. 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.

Did You Know? In this chapter, the most important number is useful only together with its construction rule. Changing scaling, sample membership or validation design can change the numerical answer without changing the arithmetic.

5. Centre data

Here the geometry becomes operational. Subtract the sample mean.

A concrete situation

Deviations are -2,-1,0,1,2.

What the mathematics does not license

A location shift should not change normality. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Verify invariance. 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. Numerator

This step deserves a careful pause. Multiply ordered deviations by scores and sum.

A concrete situation

Symmetric products are positive.

What the mathematics does not license

Reversing scores changes orientation. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Audit signs. 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. Score norm

The useful question is not only how to calculate, but what the calculation assumes. Calculate the sum of squared scores.

A concrete situation

Both tails contribute strongly.

What the mathematics does not license

Approximate quantiles change the norm. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show terms. 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. Data norm

A small example makes the mechanism visible. Calculate total squared deviation.

A concrete situation

The example total is 10.

What the mathematics does not license

Zero variance is undefined. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Check spread. 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. Correlation

This is where modelling choices begin to affect evidence. Divide cross-product by both norms.

A concrete situation

Correlation is about .996.

What the mathematics does not license

This is not Pearson correlation between two measured variables. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Name objects. 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. Statistic

The central mathematical idea is concrete. Square the correlation.

A concrete situation

W prime is about .993.

What the mathematics does not license

Near one is not a p-value. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Separate statistic and calibration. 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. Tail effect

Here the geometry becomes operational. Move the largest value from 5 to 9.

A concrete situation

The top point bends away from the line.

What the mathematics does not license

One outlier can dominate. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Repeat calculation. 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. Skewness effect

This step deserves a careful pause. Compress the lower tail and stretch the upper.

A concrete situation

The probability plot curves.

What the mathematics does not license

Different alternatives have different power. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Describe the curve. 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

The useful question is not only how to calculate, but what the calculation assumes. Use sample-size-specific null results or validated software.

A concrete situation

The hand example has n=5.

What the mathematics does not license

A generic normal cutoff is invalid. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record method. 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. Dependence

A small example makes the mechanism visible. Check clusters and time order.

A concrete situation

Repeated students are not independent.

What the mathematics does not license

The null distribution changes. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Model dependence. 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. Decision

This is where modelling choices begin to affect evidence. Connect the result to the planned analysis.

A concrete situation

Residual diagnostics guide an interval estimate.

What the mathematics does not license

Automatic transformation may hurt interpretation. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Compare options. 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

The central mathematical idea is concrete. State evidence about the fitted model.

A concrete situation

The sample aligns closely with normal order scores.

What the mathematics does not license

Normality is not proven. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep the plot. 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

Day 1: define the decision, observation unit, X variables and Y or reference condition.
Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
Day 3: build a transparent baseline and write what it assumes.
Day 4: fit the main method inside a training-only pipeline; record every parameter.
Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
Day 6: compare validation metrics with the practical workload created by errors.
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 Shapiro–Francia statistic asks a visual question in numerical form: how closely do the ordered observations follow the expected order of observations from a normal distribution? After centring the data, it takes a squared correlation with normal-order scores. Values near one mean the plotted points are nearly straight; smaller values signal departures such as skewness, heavy tails or outliers.

For a five-value illustration, order 1,2,3,4,5 and use plotting positions (i-0.375)/(n+0.25). Their standard-normal quantiles are approximately -1.180,-0.497,0,0.497,1.180. The sample deviations are -2,-1,0,1,2. Their correlation is about 0.996, so W prime is about 0.993. This is a hand-worked geometric illustration, not a published small-sample p-value.

The statistic is scale and location invariant because centring and standardisation remove the measurement origin and unit. That useful invariance does not remove dependence, selection, censoring, rounding or mixtures. A time series with serial dependence can produce an attractive probability plot while violating the independent-sample null used for calibration.

Normality is a model property, not a badge of good data. A small p-value can identify a consequential mismatch, while a large p-value may simply reflect low power. Show the probability plot, explain the feature that bends it, and decide whether the planned analysis needs normal observations, normal residuals, or only a large-sample approximation.

Frequently asked questions

What does the Shapiro–Francia test assess?

It assesses whether ordered sample values align with expected normal-order statistics.

What is W prime?

It is a squared correlation, equivalently a normalized squared linear combination of ordered observations.

Why is a value near one expected?

Normal observations tend to follow a nearly straight normal probability plot.

Does a large p-value prove normality?

No. It means the sample has not supplied strong evidence against the fitted normal model.

Can it be used on residuals?

Yes, when residual normality is the actual modelling assumption and dependence is addressed.

What should accompany the test?

A probability plot, sample size, calibration method and description of the practical consequence.

Primary source and fact-check note

S. S. Shapiro and R. S. Francia, “An Approximate Analysis of Variance Test for Normality,” Journal of the American Statistical Association 67(337), 215–216 (1972), DOI 10.1080/01621459.1972.10481232. 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

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.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

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.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Question

Begin with this mechanism: Name the model property being checked. Recreate the concrete case: A residual model assumes normal errors. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Testing raw outcomes may answer the wrong question. Complete the transfer step: Write the target. 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: Ordering

Begin with this mechanism: Sort observations from smallest to largest. Recreate the concrete case: 1,2,3,4,5. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The original labels still matter for diagnostics. Complete the transfer step: Keep row IDs. 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: Plotting positions

Begin with this mechanism: Assign a declared probability to each order. Recreate the concrete case: Use (i-.375)/(n+.25). Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Other conventions give slightly different scores. Complete the transfer step: State the convention. 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: Normal scores

Begin with this mechanism: Transform probabilities with the inverse normal CDF. Recreate the concrete case: Scores are about -1.180,-.497,0,.497,1.180. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Rounded scores change the hand result. Complete the transfer step: Retain precision. 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.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 5: Centre data

Begin with this mechanism: Subtract the sample mean. Recreate the concrete case: Deviations are -2,-1,0,1,2. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A location shift should not change normality. Complete the transfer step: Verify invariance. 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: Numerator

Begin with this mechanism: Multiply ordered deviations by scores and sum. Recreate the concrete case: Symmetric products are positive. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Reversing scores changes orientation. Complete the transfer step: Audit signs. 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: Score norm

Begin with this mechanism: Calculate the sum of squared scores. Recreate the concrete case: Both tails contribute strongly. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Approximate quantiles change the norm. Complete the transfer step: Show terms. 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: Data norm

Begin with this mechanism: Calculate total squared deviation. Recreate the concrete case: The example total is 10. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Zero variance is undefined. Complete the transfer step: Check spread. 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.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 9: Correlation

Begin with this mechanism: Divide cross-product by both norms. Recreate the concrete case: Correlation is about .996. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is not Pearson correlation between two measured variables. Complete the transfer step: Name objects. 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: Statistic

Begin with this mechanism: Square the correlation. Recreate the concrete case: W prime is about .993. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Near one is not a p-value. Complete the transfer step: Separate statistic and calibration. 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: Tail effect

Begin with this mechanism: Move the largest value from 5 to 9. Recreate the concrete case: The top point bends away from the line. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One outlier can dominate. Complete the transfer step: Repeat calculation. 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: Skewness effect

Begin with this mechanism: Compress the lower tail and stretch the upper. Recreate the concrete case: The probability plot curves. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Different alternatives have different power. Complete the transfer step: Describe the curve. 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.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 13: Calibration

Begin with this mechanism: Use sample-size-specific null results or validated software. Recreate the concrete case: The hand example has n=5. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A generic normal cutoff is invalid. Complete the transfer step: Record method. 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: Dependence

Begin with this mechanism: Check clusters and time order. Recreate the concrete case: Repeated students are not independent. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The null distribution changes. Complete the transfer step: Model dependence. 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: Decision

Begin with this mechanism: Connect the result to the planned analysis. Recreate the concrete case: Residual diagnostics guide an interval estimate. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Automatic transformation may hurt interpretation. Complete the transfer step: Compare options. 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 evidence about the fitted model. Recreate the concrete case: The sample aligns closely with normal order scores. Write observations, plotting positions, normal quantiles, centred values, cross-products, squared norms, correlations, statistics and calibrated probabilities with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Normality is not proven. Complete the transfer step: Keep the plot. 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.