WHY MATHEMATICS?

Why Mathematics? | Varying-Coefficient Quantile Regression: Polynomial Splines, Knot Selection, Tail-Specific Curves and Rao-Score Tests

How a predictor slope can change smoothly with age, time or exposure, why spline choices matter, and how to compare a varying curve with a simpler linear model.

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

Define the conditional quantile, predictor and modifier clearly
Represent changing coefficients with polynomial spline bases
Select knots and assess a simpler null without hiding search
Inspect boundaries, support, uncertainty and transfer across tau

Chapter map

1. Model form
2. Modifier
3. Predictor
4. Coefficient curve
5. Intercept
6. Predictions
7. Observed outcomes
8. Median loss
9. Constant slope
10. Rival loss
11. Spline basis
12. Knot selection
13. Score test
14. Boundary
15. Quantile grid
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 form

This is where modelling choices begin to affect evidence. Write Q_tau=alpha_tau(z)+beta_tau(z)x.

A concrete situation

Use tau=.5.

What the mathematics does not license

The modifier z is part of the estimand. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Define variables. 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. Modifier

The central mathematical idea is concrete. Choose z before looking at curves.

A concrete situation

Use z=(0,1,2).

What the mathematics does not license

Data-driven modifiers multiply searches. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Predeclare. 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. Predictor

Here the geometry becomes operational. Set x=2 for the teaching rows.

A concrete situation

Hold x constant.

What the mathematics does not license

Real designs need x variation too. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Inspect support. 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. Coefficient curve

This step deserves a careful pause. Use beta(z)=1+.5z.

A concrete situation

Values are 1,1.5,2.

What the mathematics does not license

This curve is constructed. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Label example. 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. Intercept

The useful question is not only how to calculate, but what the calculation assumes. Use alpha(z)=3.

A concrete situation

Keep it constant here.

What the mathematics does not license

Intercept may also vary. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Fit declared structure. 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. Predictions

A small example makes the mechanism visible. Calculate 3+beta(z)2.

A concrete situation

Get 5,6,7.

What the mathematics does not license

Units follow y, not the basis scale. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Restore 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.

7. Observed outcomes

This is where modelling choices begin to affect evidence. Use y=(4,7,8).

A concrete situation

Residuals are -1,1,1.

What the mathematics does not license

Three cases are not evidence of smoothness. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Use real sample. 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. Median loss

The central mathematical idea is concrete. Sum half absolute residuals.

A concrete situation

Loss is 1.5.

What the mathematics does not license

Training fit favours flexibility. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Penalise or validate. 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. Constant slope

Here the geometry becomes operational. Set beta=1.

A concrete situation

Predictions are 5,5,5.

What the mathematics does not license

A deliberately simple rival may underfit. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Use fair alternatives. 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. Rival loss

This step deserves a careful pause. Residuals are -1,2,3.

A concrete situation

Loss is 3.

What the mathematics does not license

Two losses do not provide uncertainty. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Bootstrap appropriately. 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. Spline basis

The useful question is not only how to calculate, but what the calculation assumes. Represent beta(z) with polynomial splines.

A concrete situation

Estimate basis weights by quantile regression.

What the mathematics does not license

Basis coefficients are not direct effects. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Plot beta. 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. Knot selection

A small example makes the mechanism visible. Search a declared candidate set.

A concrete situation

Use stepwise selection as specified.

What the mathematics does not license

Selection adds uncertainty. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record 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.

13. Score test

This is where modelling choices begin to affect evidence. Assess a simpler linear null.

A concrete situation

Use Rao-score-type statistic.

What the mathematics does not license

A small p-value does not measure importance. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Report effect curves. 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. Boundary

The central mathematical idea is concrete. Inspect ends of the z range.

A concrete situation

Mark sparse regions.

What the mathematics does not license

Spline edges can be unstable. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show density. 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. Quantile grid

Here the geometry becomes operational. Repeat at lower and upper tau.

A concrete situation

Compare curve shapes.

What the mathematics does not license

Crossing and multiple testing matter. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Use simultaneous view. 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 tau, basis, knots and support.

A concrete situation

Teaching losses are 1.5 and 3.

What the mathematics does not license

Do not call a selected knot a natural threshold. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Bound claim. 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

Varying-coefficient quantile regression lets a coefficient change smoothly with an index such as age, time or baseline exposure, and it can change differently at each quantile. Kim proposed polynomial-spline estimation, stepwise knot selection and a Rao-score-type assessment against a simpler linear specification. The method asks a focused question: does the conditional percentile respond to x with one constant slope, or does that slope itself vary with z?

Consider Q_.5(Y|X,Z)=3+(1+.5Z)X. With X=2 and Z=(0,1,2), fitted medians are (5,6,7). If observed outcomes are (4,7,8), residuals are (-1,1,1) and total median loss is 1.5. A constant coefficient of 1 gives predictions (5,5,5), residuals (-1,2,3) and loss 3. The varying coefficient fits this teaching sample better, but three cases cannot choose spline degree, knot count or prove genuine smooth variation.

A spline basis turns the varying coefficient into ordinary unknown basis weights, but flexibility must be controlled. Too few knots hide curvature; too many follow noise and create unstable tail curves. A selected knot is not automatically a scientific threshold. The Rao-score-type test assesses a stated null under its assumptions, while repeated data-driven model searching requires honest accounting. Sparse support in z can make an attractive curve mostly extrapolation.

Report the response, x variables, modifier z, tau grid, spline degree, candidate knots, selection rule, boundary treatment, score test, uncertainty and validation. Plot data density beneath coefficient curves and show simpler alternatives. Repeat the analysis across plausible knot sets and quantiles. Describe a varying conditional association unless the design and domain evidence support a causal interpretation.

Frequently asked questions

What varies in the model?

A regression coefficient changes smoothly with a declared modifier z.

Why use splines?

They express a smooth coefficient curve through estimable basis weights.

What is a knot?

A location where the polynomial pieces connect under continuity conditions.

What does the score test assess?

Whether the stated simpler model is adequate against the varying-coefficient alternative.

Why show support?

Sparse z regions can turn an attractive curve into extrapolation.

What must be reported?

Tau, modifier, basis, knots, selection, test, boundaries, uncertainty and validation.

Primary source and fact-check note

Mi-Ok Kim, “Quantile Regression with Varying Coefficients,” The Annals of Statistics 35(1), 92–108 (2007), DOI 10.1214/009053606000000966. 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 response, predictors, modifier, units, tau grid and plausible coefficient shape before selecting knots.

Objective audit

Write the varying-coefficient quantile equation and spline basis so every plotted curve corresponds to declared parameters.

Tuning audit

Record polynomial degree, boundary knots, candidate internal knots, stepwise rule, score test and multiplicity handling.

Computation audit

Verify basis values, continuity, fitted coefficient curves, residual loss, knot path and solver 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.

Support audit

Plot modifier density, predictor variation within modifier regions and tail counts beneath every coefficient segment.

Uncertainty audit

Repeat knot selection and coefficient fitting together, and show sensitivity to simpler bases and boundary choices.

Validation audit

Use held-out pinball loss and coefficient stability across samples, tau values and scientifically plausible subgroups.

Final claim

Describe smooth conditional association; do not promote a selected knot into a natural threshold or causal turning point.

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: Model form

Begin with this mechanism: Write Q_tau=alpha_tau(z)+beta_tau(z)x. Recreate the concrete case: Use tau=.5. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The modifier z is part of the estimand. Complete the transfer step: Define variables. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Modifier

Begin with this mechanism: Choose z before looking at curves. Recreate the concrete case: Use z=(0,1,2). Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Data-driven modifiers multiply searches. Complete the transfer step: Predeclare. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Predictor

Begin with this mechanism: Set x=2 for the teaching rows. Recreate the concrete case: Hold x constant. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Real designs need x variation too. Complete the transfer step: Inspect support. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Coefficient curve

Begin with this mechanism: Use beta(z)=1+.5z. Recreate the concrete case: Values are 1,1.5,2. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This curve is constructed. Complete the transfer step: Label example. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that 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: Intercept

Begin with this mechanism: Use alpha(z)=3. Recreate the concrete case: Keep it constant here. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Intercept may also vary. Complete the transfer step: Fit declared structure. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Predictions

Begin with this mechanism: Calculate 3+beta(z)2. Recreate the concrete case: Get 5,6,7. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Units follow y, not the basis scale. Complete the transfer step: Restore units. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Observed outcomes

Begin with this mechanism: Use y=(4,7,8). Recreate the concrete case: Residuals are -1,1,1. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Three cases are not evidence of smoothness. Complete the transfer step: Use real sample. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Median loss

Begin with this mechanism: Sum half absolute residuals. Recreate the concrete case: Loss is 1.5. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Training fit favours flexibility. Complete the transfer step: Penalise or validate. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that 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: Constant slope

Begin with this mechanism: Set beta=1. Recreate the concrete case: Predictions are 5,5,5. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A deliberately simple rival may underfit. Complete the transfer step: Use fair alternatives. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Rival loss

Begin with this mechanism: Residuals are -1,2,3. Recreate the concrete case: Loss is 3. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Two losses do not provide uncertainty. Complete the transfer step: Bootstrap appropriately. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Spline basis

Begin with this mechanism: Represent beta(z) with polynomial splines. Recreate the concrete case: Estimate basis weights by quantile regression. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Basis coefficients are not direct effects. Complete the transfer step: Plot beta. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Knot selection

Begin with this mechanism: Search a declared candidate set. Recreate the concrete case: Use stepwise selection as specified. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Selection adds uncertainty. Complete the transfer step: Record path. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that 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: Score test

Begin with this mechanism: Assess a simpler linear null. Recreate the concrete case: Use Rao-score-type statistic. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A small p-value does not measure importance. Complete the transfer step: Report effect curves. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Boundary

Begin with this mechanism: Inspect ends of the z range. Recreate the concrete case: Mark sparse regions. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Spline edges can be unstable. Complete the transfer step: Show density. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Quantile grid

Begin with this mechanism: Repeat at lower and upper tau. Recreate the concrete case: Compare curve shapes. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Crossing and multiple testing matter. Complete the transfer step: Use simultaneous view. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that 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 tau, basis, knots and support. Recreate the concrete case: Teaching losses are 1.5 and 3. Write response units, predictor x, modifier z, tau, spline basis values, knots, coefficient values, residuals, score statistics and losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call a selected knot a natural threshold. Complete the transfer step: Bound claim. Change exactly one observation, target quantile, base learner, gradient sign, correction step, shrinkage value, stopping round, hidden activation, network weight, smoothing parameter, penalty, censoring rule, predictor, modifier, spline degree, knot, fixed effect, lagged outcome, instrument, missing wave or forecast horizon. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete calculation and once to a parent with the real decision and original units. Add one counterexample involving correlated predictors, tail sparsity, validation leakage, quantile crossing, local optima, spline boundary instability, knot-search uncertainty, weak instruments, attrition or serial dependence. State the extra observation or test that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.