Chapter map

1. 1. Model
2. 2. Loss
3. 3. Calibration
4. 4. Scale equation
5. 5. Standardize
6. 6. Loss values
7. 7. Candidate B
8. 8. Minimise scale
9. 9. Breakdown
10. 10. Consistency
11. 11. Efficiency
12. 12. Leverage
13. 13. Algorithm
14. 14. Uncertainty
15. 15. Sensitivity
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. Model

This is where modelling choices begin to affect evidence. Specify regression coefficients and vertical residuals.

A concrete situation

Compare lines A and B.

What the mathematics does not license

Changing response direction changes the fit. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

State model. 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. Loss

The central mathematical idea is concrete. Choose a bounded nondecreasing rho of |u|.

A concrete situation

Use the stated Tukey-shaped rho.

What the mathematics does not license

Different rho functions define different estimators. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Write 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.

---

3. Calibration

Here the geometry becomes operational. Choose b consistently with rho and desired properties.

A concrete situation

Use b=.5 only for teaching.

What the mathematics does not license

b is not a generic probability cutoff. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Label 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.

---

4. Scale equation

This step deserves a careful pause. Solve n⁻¹Σrho(r_i/s)=b for positive s.

A concrete situation

A gives s≈.28813.

What the mathematics does not license

Zero-residual degeneracy needs safeguards. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

State root rule. 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. Standardize

The useful question is not only how to calculate, but what the calculation assumes. Compute u_i=r_i/s.

A concrete situation

The remote residual is far outside |u|=1.

What the mathematics does not license

Scale has response units. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Check dimensions. 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. Loss values

A small example makes the mechanism visible. Evaluate every bounded contribution.

A concrete situation

A averages to .5.

What the mathematics does not license

Rounding can move the root. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep 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.

---

7. Candidate B

This is where modelling choices begin to affect evidence. Repeat the same equation without retuning.

A concrete situation

B gives s≈2.97798.

What the mathematics does not license

Changing rho or b destroys comparability. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Freeze settings. 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. Minimise scale

The central mathematical idea is concrete. Search coefficients for the smallest robust scale.

A concrete situation

A wins the two-line comparison.

What the mathematics does not license

The coefficient search is non-convex. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use multiple starts. 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. Breakdown

Here the geometry becomes operational. Connect bounded rho and b to finite-sample resistance.

A concrete situation

High breakdown is achievable.

What the mathematics does not license

Exact value depends on design and tuning. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

State assumptions. 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. Consistency

This step deserves a careful pause. Apply the model-specific scale correction.

A concrete situation

Calibrate under the reference error law.

What the mathematics does not license

Uncorrected teaching scale is not universal. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Name correction. 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. Efficiency

The useful question is not only how to calculate, but what the calculation assumes. Calculate or cite efficiency for the exact tuning.

A concrete situation

Expect a trade-off with high breakdown.

What the mathematics does not license

Do not transfer numbers across rho choices. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report source. 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. Leverage

A small example makes the mechanism visible. Inspect hat geometry and robust distances.

A concrete situation

A small residual can coexist with high leverage.

What the mathematics does not license

Residual scale alone does not expose all leverage. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Add design diagnostics. 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. Algorithm

This is where modelling choices begin to affect evidence. Record start, objective, scale solver and tolerance.

A concrete situation

Track candidate scales.

What the mathematics does not license

A converged local point need not be global. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Use multi-start. 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. Uncertainty

The central mathematical idea is concrete. Use a sandwich or resampling method valid for the estimator.

A concrete situation

Preserve scale estimation in every replicate.

What the mathematics does not license

OLS covariance is mismatched. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Mirror fit. 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. Sensitivity

Here the geometry becomes operational. Vary rho, b and start within justified ranges.

A concrete situation

Check whether the selected line changes.

What the mathematics does not license

Tuning after seeing results is selective. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Predeclare primary fit. 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

This step deserves a careful pause. State bounded loss, b and scale-minimising result.

A concrete situation

A has lower teaching S-scale.

What the mathematics does not license

Do not call remote points errors. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Preserve context. 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

An S-estimator chooses regression coefficients that make a robust scale of the residuals as small as possible. For each candidate line, solve a scale equation such as n⁻¹Σρ(r_i/s)=b using a bounded loss ρ, then select the coefficients with the smallest positive scale s. The letter S points to scale: the regression problem is converted into a contest between residual spread estimates.

For a transparent teaching calculation use the bounded Tukey-shaped loss ρ(u)=1−(1−u²)³ for |u|<1 and ρ(u)=1 beyond, with b=.5. For line A, residuals 0,.1,−.1,.2,15 give s≈.28813. Their losses are approximately 0,.31957,.31957,.86085,1 and average .5. For line B, residuals −2,−.9,.1,1.2,17 give s≈2.97798, with losses about .83457,.24974,.00338,.41231,1 and average .5. The smaller robust residual scale favours A in this candidate comparison.

The particular rho function and b value above are teaching choices, not universal defaults. In theory, bounded rho and its calibration determine breakdown, consistency and efficiency properties. The scale equation may have edge cases if many residuals are exactly zero or if the candidate fit interpolates too many points. Software must define the positive solution, finite-sample correction and numerical safeguards.

S-estimation protects a residual scale against substantial contamination, yet its normal-model efficiency can be modest. It is often used as the high-breakdown foundation for a later MM-estimation stage. Students should separate the regression coefficients, residuals, scale equation, rho average, optimisation and final uncertainty. A compact “robust” label hides all six decisions.

---

Frequently asked questions

What does an S-estimator minimise?

A robust scale of the regression residuals.

How is scale defined?

Often by a bounded-loss equation whose average equals a calibrated constant b.

Why bounded rho?

It prevents a few enormous standardized residuals from dominating the scale equation.

Does S-estimation guarantee high efficiency?

No. High breakdown can come with lower clean-model efficiency.

Does it solve leverage automatically?

No. Predictor-space leverage still needs compatible diagnostics.

What must be reported?

rho, b, correction, start, scale solver, coefficient optimiser, leverage checks and covariance method.

---

Primary source and fact-check note

Peter J. Rousseeuw and Victor J. Yohai, “Robust Regression by Means of S-Estimators,” in Robust and Nonlinear Time Series Analysis, Lecture Notes in Statistics 26, 256–272 (1984), DOI 10.1007/978-1-4615-7821-5_15. 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

• MM-estimators for robust regression
• Tukey bisquare M-estimator
• Hampel redescending M-estimator
• How Robust Statistics Work

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 sampling units, response, predictors, intercept, dependence, missingness and the scientific target.

Residual audit

Show every prediction, signed residual, squared residual, order statistic, subset membership and original unit.

Robustness audit

Connect h, bounded loss, b, start and scale to a precise finite-sample or asymptotic claim.

Solver audit

Record starts, seed, subset path, C-steps, scale roots, IRLS updates, objective and stopping tolerance.

Leverage audit

Inspect predictor-space geometry separately from response residuals.

Uncertainty audit

Use covariance or resampling that repeats the estimator’s complete selection and scale pipeline.

Sensitivity audit

Change one justified h, loss, b, starting fit, scale correction, tuning value or cluster at a time.

Final claim

Describe the fitted majority pattern and limitations without turning robustness into a verdict about people or records.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Model

Begin with this mechanism: Specify regression coefficients and vertical residuals. Recreate the concrete case: Compare lines A and B. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Changing response direction changes the fit. Complete the transfer step: State model. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Loss

Begin with this mechanism: Choose a bounded nondecreasing rho of |u|. Recreate the concrete case: Use the stated Tukey-shaped rho. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Different rho functions define different estimators. Complete the transfer step: Write formula. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Calibration

Begin with this mechanism: Choose b consistently with rho and desired properties. Recreate the concrete case: Use b=.5 only for teaching. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: b is not a generic probability cutoff. Complete the transfer step: Label choice. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Scale equation

Begin with this mechanism: Solve n⁻¹Σrho(r_i/s)=b for positive s. Recreate the concrete case: A gives s≈.28813. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Zero-residual degeneracy needs safeguards. Complete the transfer step: State root rule. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 5: Standardize

Begin with this mechanism: Compute u_i=r_i/s. Recreate the concrete case: The remote residual is far outside |u|=1. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Scale has response units. Complete the transfer step: Check dimensions. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Loss values

Begin with this mechanism: Evaluate every bounded contribution. Recreate the concrete case: A averages to .5. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Rounding can move the root. Complete the transfer step: Keep precision. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Candidate B

Begin with this mechanism: Repeat the same equation without retuning. Recreate the concrete case: B gives s≈2.97798. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Changing rho or b destroys comparability. Complete the transfer step: Freeze settings. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Minimise scale

Begin with this mechanism: Search coefficients for the smallest robust scale. Recreate the concrete case: A wins the two-line comparison. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The coefficient search is non-convex. Complete the transfer step: Use multiple starts. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 9: Breakdown

Begin with this mechanism: Connect bounded rho and b to finite-sample resistance. Recreate the concrete case: High breakdown is achievable. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Exact value depends on design and tuning. Complete the transfer step: State assumptions. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Consistency

Begin with this mechanism: Apply the model-specific scale correction. Recreate the concrete case: Calibrate under the reference error law. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Uncorrected teaching scale is not universal. Complete the transfer step: Name correction. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Efficiency

Begin with this mechanism: Calculate or cite efficiency for the exact tuning. Recreate the concrete case: Expect a trade-off with high breakdown. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not transfer numbers across rho choices. Complete the transfer step: Report source. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Leverage

Begin with this mechanism: Inspect hat geometry and robust distances. Recreate the concrete case: A small residual can coexist with high leverage. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Residual scale alone does not expose all leverage. Complete the transfer step: Add design diagnostics. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 13: Algorithm

Begin with this mechanism: Record start, objective, scale solver and tolerance. Recreate the concrete case: Track candidate scales. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A converged local point need not be global. Complete the transfer step: Use multi-start. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Uncertainty

Begin with this mechanism: Use a sandwich or resampling method valid for the estimator. Recreate the concrete case: Preserve scale estimation in every replicate. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: OLS covariance is mismatched. Complete the transfer step: Mirror fit. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Sensitivity

Begin with this mechanism: Vary rho, b and start within justified ranges. Recreate the concrete case: Check whether the selected line changes. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Tuning after seeing results is selective. Complete the transfer step: Predeclare primary fit. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: State bounded loss, b and scale-minimising result. Recreate the concrete case: A has lower teaching S-scale. Write responses, predictors, residuals, robust scales, standardized residuals, rho losses, b constants, coefficients, leverage measures and original measurement units with units, signs, subset labels and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call remote points errors. Complete the transfer step: Preserve context. Change exactly one response, predictor, sampling unit, h value, rho function, b constant, robust scale, starting subset, starting coefficients, tuning constant, leverage value, solver tolerance, resampling rule or covariance formula. Explain which objective changes, which observations influence the next step, which solution is selected, which conclusion survives and which claim must weaken. Keep design, predictions, residuals, ordering, subset selection, scale, score, weight, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full calculation table and once to a parent with the decision in original units. Add a counterexample in which a similar robust line comes from rounding, censoring, dependence, mixture structure, high leverage or a genuine minority process. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.