WHY MATHEMATICS?

Why Mathematics? | MM-Estimators for Robust Regression: High-Breakdown Starts, Robust Scale and High-Efficiency Refinement

How MM regression separates resistance, scale and efficiency into three stages, why the fixed scale preserves protection, and why multiple roots and leverage still require checks.

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

Start from a consistent high-breakdown regression fit
Estimate and freeze a compatible robust residual scale
Refine coefficients with an efficient redescending M-loss
Audit roots, leverage, tuning and full-pipeline uncertainty

Chapter map

1. Stage one
2. Stage two
3. Freeze scale
4. Final loss
5. Standardize
6. Weights
7. Weighted fit
8. New residuals
9. Convergence
10. Multiple starts
11. Breakdown
12. Efficiency
13. Leverage
14. Uncertainty
15. Sensitivity
16. Claim

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

1. Stage one

The useful question is not only how to calculate, but what the calculation assumes. Compute a consistent high-breakdown regression start.

A concrete situation

Use line A as the teaching start.

What the mathematics does not license

An ordinary least-squares start can lose protection. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Name starter. 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. Stage two

A small example makes the mechanism visible. Estimate residual scale robustly at the start.

A concrete situation

Use s≈.28813.

What the mathematics does not license

Scale must be positive and compatibly calibrated. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record equation. 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. Freeze scale

This is where modelling choices begin to affect evidence. Hold the robust scale during efficient refinement.

A concrete situation

Keep s fixed for the teaching step.

What the mathematics does not license

Silent rescaling can alter breakdown inheritance. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Separate stages. 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. Final loss

The central mathematical idea is concrete. Choose a redescending rho with a larger efficiency tuning.

A concrete situation

Use Tukey c≈4.68506.

What the mathematics does not license

A common constant is model-specific. 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.

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. Standardize

Here the geometry becomes operational. Compute residuals divided by fixed s.

A concrete situation

Values include about 52.059.

What the mathematics does not license

Units must cancel. That distinction matters because a correct formula can still answer the wrong operational question.

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. Weights

This step deserves a careful pause. Evaluate bisquare IRLS weights.

A concrete situation

The remote point gets zero in this iteration.

What the mathematics does not license

Zero weight is not proof of error. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Keep provenance. 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. Weighted fit

The useful question is not only how to calculate, but what the calculation assumes. Solve weighted least squares for new coefficients.

A concrete situation

Get about .99068 and 1.03906.

What the mathematics does not license

This is one update, not final convergence. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Label iteration. 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. New residuals

A small example makes the mechanism visible. Recompute signed residuals for all points.

A concrete situation

Central residuals remain small.

What the mathematics does not license

Weights must be updated from the new fit. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Continue correctly. 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. Convergence

This is where modelling choices begin to affect evidence. Monitor coefficients, objective and score equation.

A concrete situation

Use a declared tolerance.

What the mathematics does not license

Small coefficient change can hide a different root. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Check objective. 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. Multiple starts

The central mathematical idea is concrete. Compare several valid high-breakdown starts.

A concrete situation

Retain competing final solutions.

What the mathematics does not license

Non-convex refinement may select local minima. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Report path. 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. Breakdown

Here the geometry becomes operational. Trace robustness to the initial fit and fixed scale.

A concrete situation

The start supplies the protection.

What the mathematics does not license

A badly chosen final loss can undermine claims. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

State theorem conditions. 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. Efficiency

This step deserves a careful pause. Tie final tuning to the exact reference model.

A concrete situation

Higher normal efficiency is a design goal.

What the mathematics does not license

Efficiency is not accuracy on every dataset. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use cautious wording. 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. Leverage

The useful question is not only how to calculate, but what the calculation assumes. Inspect predictor-space influence separately.

A concrete situation

Plot robust distances and residuals.

What the mathematics does not license

A leverage point may retain a small residual. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Use compatible 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.

14. Uncertainty

A small example makes the mechanism visible. Use covariance or bootstrap calculations for the full MM estimator.

A concrete situation

Include start, scale and refinement.

What the mathematics does not license

Treating final weights as fixed is incomplete. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Mirror pipeline. 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

This is where modelling choices begin to affect evidence. Vary valid starts, scale tuning and final tuning.

A concrete situation

Compare slopes and decisions.

What the mathematics does not license

Post-hoc tuning can manufacture stability. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Declare primary 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.

16. Claim

The central mathematical idea is concrete. Report all three stages and limits.

A concrete situation

One update moves slope to about 1.03906.

What the mathematics does not license

Do not call one update the final estimator. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep scope. 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

MM-estimation deliberately separates three goals. First compute a consistent high-breakdown regression start. Second estimate residual scale robustly while holding that start. Third refine the regression coefficients with an efficient redescending M-objective while keeping the scale fixed. Yohai’s name signals the combination of an initial robust procedure with a final M-estimation stage, rather than a single mysterious formula.

Continue the five-point example with initial line y=1+x and fixed robust scale s≈.28813 from the teaching S-equation. With Tukey bisquare tuning c≈4.68506, the standardized residuals are about 0,.3471,−.3471,.6941,52.059. The IRLS weights are about 1,.98905,.98905,.95658,0. One weighted least-squares update gives intercept≈.99068 and slope≈1.03906. The new residuals are about .00932,.07026,−.16880,.09214,14.85308. This is one refinement step, not a claimed final optimum.

A strong start matters because a non-convex redescending M-stage can have several local solutions. A compatible fixed scale matters because recomputing an ordinary scale at every efficient update can surrender the protection inherited from the start. The final tuning constant controls efficiency and influence under a stated error model. Neither “50% breakdown” nor “95% efficiency” should be quoted without the exact estimator, dimensions, calibration and assumptions.

MM regression is valuable when a team wants both resistance to substantial contamination and good precision near a clean reference model. It still does not turn zero weights into proof of bad data, guarantee the same answer under clustered errors or automatically control predictor-space leverage. Report all three stages, loss functions, constants, scale corrections, starting algorithm, convergence path, leverage diagnostics and covariance method.

Frequently asked questions

What are the three MM stages?

High-breakdown regression start, robust scale estimate, then efficient redescending M-refinement.

Why freeze the scale?

It helps preserve the robustness inherited from the initial estimator while coefficients are refined.

What does the final tuning control?

Influence and asymptotic efficiency under a stated reference model.

Can the final stage have several solutions?

Yes. Redescending objectives are non-convex, so robust starts and path checks matter.

Does zero weight prove bad data?

No. It is an influence decision for one fitted procedure.

What must be reported?

Every stage, rho functions, constants, corrections, starts, convergence, leverage checks and covariance method.

Primary source and fact-check note

Victor J. Yohai, “High Breakdown-Point and High Efficiency Robust Estimates for Regression,” The Annals of Statistics 15(2), 642–656 (1987), DOI 10.1214/aos/1176350366. 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 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.

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.

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.

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: Stage one

Begin with this mechanism: Compute a consistent high-breakdown regression start. Recreate the concrete case: Use line A as the teaching start. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: An ordinary least-squares start can lose protection. Complete the transfer step: Name starter. 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: Stage two

Begin with this mechanism: Estimate residual scale robustly at the start. Recreate the concrete case: Use s≈.28813. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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 must be positive and compatibly calibrated. Complete the transfer step: Record equation. 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: Freeze scale

Begin with this mechanism: Hold the robust scale during efficient refinement. Recreate the concrete case: Keep s fixed for the teaching step. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Silent rescaling can alter breakdown inheritance. Complete the transfer step: Separate stages. 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: Final loss

Begin with this mechanism: Choose a redescending rho with a larger efficiency tuning. Recreate the concrete case: Use Tukey c≈4.68506. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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 common constant is model-specific. 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.

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: Standardize

Begin with this mechanism: Compute residuals divided by fixed s. Recreate the concrete case: Values include about 52.059. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Units must cancel. 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: Weights

Begin with this mechanism: Evaluate bisquare IRLS weights. Recreate the concrete case: The remote point gets zero in this iteration. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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 weight is not proof of error. Complete the transfer step: Keep provenance. 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: Weighted fit

Begin with this mechanism: Solve weighted least squares for new coefficients. Recreate the concrete case: Get about .99068 and 1.03906. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: This is one update, not final convergence. Complete the transfer step: Label iteration. 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: New residuals

Begin with this mechanism: Recompute signed residuals for all points. Recreate the concrete case: Central residuals remain small. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Weights must be updated from the new fit. Complete the transfer step: Continue correctly. 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.

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: Convergence

Begin with this mechanism: Monitor coefficients, objective and score equation. Recreate the concrete case: Use a declared tolerance. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Small coefficient change can hide a different root. Complete the transfer step: Check objective. 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: Multiple starts

Begin with this mechanism: Compare several valid high-breakdown starts. Recreate the concrete case: Retain competing final solutions. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Non-convex refinement may select local minima. Complete the transfer step: Report path. 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: Breakdown

Begin with this mechanism: Trace robustness to the initial fit and fixed scale. Recreate the concrete case: The start supplies the protection. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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 badly chosen final loss can undermine claims. Complete the transfer step: State theorem conditions. 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: Efficiency

Begin with this mechanism: Tie final tuning to the exact reference model. Recreate the concrete case: Higher normal efficiency is a design goal. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Efficiency is not accuracy on every dataset. Complete the transfer step: Use cautious wording. 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.

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: Leverage

Begin with this mechanism: Inspect predictor-space influence separately. Recreate the concrete case: Plot robust distances and residuals. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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 leverage point may retain a small residual. Complete the transfer step: Use compatible 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 14: Uncertainty

Begin with this mechanism: Use covariance or bootstrap calculations for the full MM estimator. Recreate the concrete case: Include start, scale and refinement. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Treating final weights as fixed is incomplete. Complete the transfer step: Mirror pipeline. 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 valid starts, scale tuning and final tuning. Recreate the concrete case: Compare slopes and decisions. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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: Post-hoc tuning can manufacture stability. Complete the transfer step: Declare primary 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 16: Claim

Begin with this mechanism: Report all three stages and limits. Recreate the concrete case: One update moves slope to about 1.03906. Write responses, predictors, high-breakdown starts, residual scales, standardized residuals, redescending weights, coefficient updates, objectives 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 one update the final estimator. Complete the transfer step: Keep scope. 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.