WHY MATHEMATICS?

Why Mathematics? | Regression Depth: Deepest Lines, Tilting to Vertical, Geometric Robustness and Leverage

How regression depth ranks whole lines by the observations they must cross before becoming vertical, why deepest fits resist leverage, and where ties and approximation matter.

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 predictor positions and record residual signs
Count crossings for both tilt directions at every cut
Take the minimum crossing count as line depth
Maximise depth and validate prediction magnitudes separately

Chapter map

1. Dataset
2. Candidate
3. Signs
4. Vertical cut
5. First tilt
6. Opposite tilt
7. Cut score
8. Depth
9. Shallow line
10. Deepest fit
11. Existence
12. Leverage
13. Exactness
14. Uncertainty
15. Transfer
16. Claim

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

1. Dataset

The useful question is not only how to calculate, but what the calculation assumes. Order x and retain signed response residuals.

A concrete situation

Use seven alternating residuals.

What the mathematics does not license

The teaching pattern is constructed. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Label illustration. 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. Candidate

A small example makes the mechanism visible. Choose line y=x.

A concrete situation

Compute residuals y−x.

What the mathematics does not license

Depth ranks a line, not individual points. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Keep distinction. 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. Signs

This is where modelling choices begin to affect evidence. Record +,−,+,−,+,−,+.

A concrete situation

No residual equals zero.

What the mathematics does not license

Magnitude is temporarily suppressed. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Retain magnitudes elsewhere. 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. Vertical cut

The central mathematical idea is concrete. Place a cut before, between or after ordered x-values.

A concrete situation

There are eight cut positions.

What the mathematics does not license

Duplicate x needs exact conventions. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

List cuts. 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. First tilt

Here the geometry becomes operational. Count above-left plus below-right.

A concrete situation

Counts alternate 3 and 4.

What the mathematics does not license

This count belongs to one rotation direction. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

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

6. Opposite tilt

This step deserves a careful pause. Count below-left plus above-right.

A concrete situation

Counts alternate 4 and 3.

What the mathematics does not license

Using only one direction overstates depth. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

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

7. Cut score

The useful question is not only how to calculate, but what the calculation assumes. Take the smaller directional count.

A concrete situation

Every cut gives 3.

What the mathematics does not license

Ties on the line need special handling. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

State no-ties case. 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. Depth

A small example makes the mechanism visible. Take the minimum across cuts.

A concrete situation

Candidate depth=3.

What the mathematics does not license

This simple sign formula is not a general multivariate algorithm. That distinction matters because a correct formula can still answer the wrong operational question.

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.

9. Shallow line

This is where modelling choices begin to affect evidence. Shift to y=x+2.

A concrete situation

All residuals are negative.

What the mathematics does not license

An extreme cut yields zero. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Compare geometry. 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. Deepest fit

The central mathematical idea is concrete. Maximise depth over candidate lines.

A concrete situation

Search lines through point pairs in simple regression.

What the mathematics does not license

Several fits may tie. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Report all ties. 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. Existence

Here the geometry becomes operational. Use the n/3 lower-bound result carefully.

A concrete situation

For n=7 a line of depth at least 3 exists.

What the mathematics does not license

This is not a half-breakdown guarantee. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Distinguish properties. 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

This step deserves a careful pause. Inspect how remote x-values affect tilting geometry.

A concrete situation

Depth attends to design position.

What the mathematics does not license

Prediction error still needs magnitude checks. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use both 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. Exactness

The useful question is not only how to calculate, but what the calculation assumes. State exact simple-regression or approximate multivariate computation.

A concrete situation

Record candidates and iterations.

What the mathematics does not license

Approximation can miss the maximum. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

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

14. Uncertainty

A small example makes the mechanism visible. Use depth regions or compatible bootstrap methods.

A concrete situation

Repeat the full deepest-fit search.

What the mathematics does not license

OLS covariance is mismatched. 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. Transfer

This is where modelling choices begin to affect evidence. Connect geometry to computer vision, calibration and robust trends.

A concrete situation

Explain why embedded fits resist tilting.

What the mathematics does not license

Depth does not prove causality. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use original units. 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 maximum depth, ties and prediction limits.

A concrete situation

Teaching candidate has depth 3.

What the mathematics does not license

Do not generalise from constructed signs. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Bound conclusion. 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

Regression depth ranks an entire fitted line by asking how difficult it is to make that line a nonfit. Geometrically, imagine tilting the candidate line until it becomes vertical. The depth is the smallest number of observations the line must cross during such a tilt. A shallow line can escape without crossing many points; a deep line is embedded in the data geometry.

Consider x=−3,−2,−1,0,1,2,3 and y=x+(1,−1,1,−1,1,−1,1). For candidate y=x, the residual signs alternate +,−,+,−,+,−,+. In the no-ties simple-regression sign calculation, every vertical cut gives tilt-direction crossing counts of 3 and 4, or 4 and 3. The smaller count is always 3, so this candidate has depth 3. For y=x+2 all residuals are negative, and an extreme cut gives a zero crossing count, so its depth is 0.

Rousseeuw and Hubert proved that every bivariate dataset has a line with depth at least n/3 and introduced the deepest regression method. The deepest line maximises depth, rather than minimising a residual magnitude. This makes it naturally attentive to leverage geometry and allows skewed or heteroscedastic errors in the framework. It does not say that every deepest line is unique or that residual magnitudes are irrelevant for prediction.

Points exactly on the candidate, repeated x-values and general-position failures need the paper’s exact conventions rather than an improvised sign count. In simple regression, exact algorithms can consider lines through pairs of points; in multiple regression, practical methods may be approximate. Report whether depth was exact or approximate, the tie handling, candidate set, maximum depth, all tied deepest fits and the prediction diagnostics used afterward.

Frequently asked questions

What has depth?

A candidate regression line or hyperplane, rather than one observation.

What does depth count?

The smallest number of observations crossed while tilting the fit to a vertical nonfit.

What is the deepest line?

A fit with maximum regression depth.

What lower bound holds in bivariate data?

At least one line has depth at least n/3.

Is the deepest fit always unique?

No. Several candidates may share maximum depth.

What must be reported?

Exact or approximate algorithm, ties, maximum depth, leverage and prediction diagnostics.

Primary source and fact-check note

Peter J. Rousseeuw and Mia Hubert, “Regression Depth,” Journal of the American Statistical Association 94(446), 388–402 (1999), DOI 10.1080/01621459.1999.10474129. 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, predictor orientation, dependence, ties, duplicates and missingness.

Calculation audit

Retain pair IDs, local medians, residual signs, cut counts, scores, weights and coefficient updates.

Robustness audit

State the exact breakdown, depth or influence property and the assumptions supporting it.

Solver audit

Record direct algorithm or IRLS start, scale, tuning, candidate set, tolerance and stopping criterion.

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 vertical residual magnitude.

Uncertainty audit

Use rank intervals, depth regions, robust covariance or resampling matched to the full estimator.

Sensitivity audit

Change one tie rule, intercept convention, candidate line, scale, tuning constant or leverage point at a time.

Final claim

Translate the line into original units while preserving minority observations and causal uncertainty.

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: Dataset

Begin with this mechanism: Order x and retain signed response residuals. Recreate the concrete case: Use seven alternating residuals. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The teaching pattern is constructed. Complete the transfer step: Label illustration. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Candidate

Begin with this mechanism: Choose line y=x. Recreate the concrete case: Compute residuals y−x. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Depth ranks a line, not individual points. Complete the transfer step: Keep distinction. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Signs

Begin with this mechanism: Record +,−,+,−,+,−,+. Recreate the concrete case: No residual equals zero. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Magnitude is temporarily suppressed. Complete the transfer step: Retain magnitudes elsewhere. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Vertical cut

Begin with this mechanism: Place a cut before, between or after ordered x-values. Recreate the concrete case: There are eight cut positions. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Duplicate x needs exact conventions. Complete the transfer step: List cuts. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: First tilt

Begin with this mechanism: Count above-left plus below-right. Recreate the concrete case: Counts alternate 3 and 4. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This count belongs to one rotation direction. Complete the transfer step: Show table. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Opposite tilt

Begin with this mechanism: Count below-left plus above-right. Recreate the concrete case: Counts alternate 4 and 3. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Using only one direction overstates depth. Complete the transfer step: Take both. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Cut score

Begin with this mechanism: Take the smaller directional count. Recreate the concrete case: Every cut gives 3. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Ties on the line need special handling. Complete the transfer step: State no-ties case. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Depth

Begin with this mechanism: Take the minimum across cuts. Recreate the concrete case: Candidate depth=3. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This simple sign formula is not a general multivariate algorithm. Complete the transfer step: Keep scope. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Shallow line

Begin with this mechanism: Shift to y=x+2. Recreate the concrete case: All residuals are negative. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: An extreme cut yields zero. Complete the transfer step: Compare geometry. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Deepest fit

Begin with this mechanism: Maximise depth over candidate lines. Recreate the concrete case: Search lines through point pairs in simple regression. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Several fits may tie. Complete the transfer step: Report all ties. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Existence

Begin with this mechanism: Use the n/3 lower-bound result carefully. Recreate the concrete case: For n=7 a line of depth at least 3 exists. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is not a half-breakdown guarantee. Complete the transfer step: Distinguish properties. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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 how remote x-values affect tilting geometry. Recreate the concrete case: Depth attends to design position. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Prediction error still needs magnitude checks. Complete the transfer step: Use both diagnostics. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Exactness

Begin with this mechanism: State exact simple-regression or approximate multivariate computation. Recreate the concrete case: Record candidates and iterations. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Approximation can miss the maximum. Complete the transfer step: Run sensitivity. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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 depth regions or compatible bootstrap methods. Recreate the concrete case: Repeat the full deepest-fit search. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers 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 pipeline. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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: Transfer

Begin with this mechanism: Connect geometry to computer vision, calibration and robust trends. Recreate the concrete case: Explain why embedded fits resist tilting. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Depth does not prove causality. Complete the transfer step: Use original units. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. 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 maximum depth, ties and prediction limits. Recreate the concrete case: Teaching candidate has depth 3. Write predictors, responses, candidate lines, signed residuals, vertical cuts, directional crossing counts, depth values, coefficients and original measurement units with units, signs, pair or point identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not generalise from constructed signs. Complete the transfer step: Bound conclusion. Change exactly one observation, predictor, response, pair rule, median convention, intercept rule, tie handling, candidate line, residual sign, cut position, robust scale, tuning constant, starting coefficient, leverage value, tolerance, interval or covariance formula. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, calculation, geometry, robustness property, optimisation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full table and once to a parent with the decision in original units. Add a counterexample involving rounding, repeated x-values, dependence, a step change, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.