WHY MATHEMATICS?

Why Mathematics? | GAMLSS Distributional Regression: Location, Scale, Shape, Link Functions and Conditional Centiles

How predictors can change the centre, spread, skewness and tail weight of a full conditional distribution, and why equal means can hide very different risks.

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

Choose a response family and define every distribution parameter
Model location, scale and shape through valid link functions
Fit parametric, smooth and random-effect terms with penalised likelihood
Validate residuals, centiles, subgroups and family sensitivity

Chapter map

1. Response family
2. Location
3. Scale
4. Shape
5. Predictors
6. Smooth terms
7. Quantile constant
8. Setting A
9. Setting B
10. Lower A
11. Lower B
12. Likelihood
13. Residuals
14. Family selection
15. Calibration
16. Claim

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

1. Response family

Here the geometry becomes operational. Choose a plausible distribution for y.

A concrete situation

Begin with Normal for teaching.

What the mathematics does not license

A familiar family can still fit badly. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Compare 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.

2. Location

This step deserves a careful pause. Model mu through a link.

A concrete situation

Set mu=10 in both settings.

What the mathematics does not license

Same location does not imply same risk. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep 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.

3. Scale

The useful question is not only how to calculate, but what the calculation assumes. Model positive sigma through a link.

A concrete situation

Use sigma=2 and 5.

What the mathematics does not license

Identity links can permit invalid negative scale. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Prefer valid link. 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. Shape

A small example makes the mechanism visible. Add skewness or kurtosis parameters when the family supports them.

A concrete situation

Keep them fixed in teaching case.

What the mathematics does not license

More parameters can overfit. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use 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.

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. Predictors

This is where modelling choices begin to affect evidence. Assign terms to each parameter equation.

A concrete situation

A variable may affect scale but not location.

What the mathematics does not license

Automatic selection can be unstable. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Publish formulas. 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. Smooth terms

The central mathematical idea is concrete. Use penalised splines where curvature is supported.

A concrete situation

Tune smoothness.

What the mathematics does not license

Too many effective degrees of freedom chase noise. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Check edf. 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. Quantile constant

Here the geometry becomes operational. Use z_.90=1.2816 for Standard Normal.

A concrete situation

Keep four decimals.

What the mathematics does not license

This constant is family-specific. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

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

8. Setting A

This step deserves a careful pause. Compute 10+1.2816 times 2.

A concrete situation

Q_.90=12.5632.

What the mathematics does not license

Rounding belongs at the end. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show arithmetic. 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. Setting B

The useful question is not only how to calculate, but what the calculation assumes. Compute 10+1.2816 times 5.

A concrete situation

Q_.90=16.408.

What the mathematics does not license

Equal means hide tail difference. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Compare plainly. 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. Lower A

A small example makes the mechanism visible. Compute 10-1.2816 times 2.

A concrete situation

Q_.10=7.4368.

What the mathematics does not license

Symmetry comes from Normal family. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Do not generalise. 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. Lower B

This is where modelling choices begin to affect evidence. Compute 10-1.2816 times 5.

A concrete situation

Q_.10=3.592.

What the mathematics does not license

Larger spread moves both tails. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep 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.

12. Likelihood

The central mathematical idea is concrete. Maximise penalised likelihood for all parameter equations.

A concrete situation

Track convergence.

What the mathematics does not license

A converged algorithm can fit the wrong family. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Inspect 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.

13. Residuals

Here the geometry becomes operational. Use distribution-aware residual diagnostics.

A concrete situation

Check centiles and tails.

What the mathematics does not license

Mean residual plots are insufficient. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use worm or QQ plots. 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. Family selection

This step deserves a careful pause. Compare fit, diagnostics and substantive plausibility.

A concrete situation

Use information criteria carefully.

What the mathematics does not license

Lowest score is not automatic truth. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Validate externally. 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. Calibration

The useful question is not only how to calculate, but what the calculation assumes. Check observed proportions below fitted centiles by subgroup.

A concrete situation

Seek coherent ordered centiles.

What the mathematics does not license

Global fit can hide subgroup failure. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Audit groups. 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

A small example makes the mechanism visible. State family, links, equations and support.

A concrete situation

Teaching .90 quantiles are 12.5632 and 16.408.

What the mathematics does not license

Do not call fitted distributions destiny. The habit of checking assumptions is one of the practical benefits of learning mathematics.

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

Generalised additive models for location, scale and shape, or GAMLSS, model more than a conditional mean. Rigby and Stasinopoulos allow several parameters of a chosen response distribution to depend on explanatory variables through link functions, linear terms, smooth terms and random effects. A predictor can therefore move the centre, change spread, alter skewness or affect tail weight, producing an estimated conditional distribution rather than one average line.

Take two settings with Normal conditional distributions. Both have location mu=10, while sigma is 2 in setting A and 5 in setting B. Using the standard-Normal .90 quantile 1.2816, the conditional .90 quantiles are 12.5632 and 16.408. The means match, yet the upper-tail decisions differ because scale changes. Their .10 quantiles are 7.4368 and 3.592, so larger spread moves both tails away from the centre.

The Normal example has only location and scale, while a broader GAMLSS family may include skewness and kurtosis parameters. Distribution choice is consequential: an excellent fit at the centre can still miss the tails, and flexible smoothers can overfit sparse regions. Link functions keep parameters in valid ranges, but validity of a parameter value does not guarantee adequacy of the distributional family.

Report the response family, parameter meanings, link functions, terms in each parameter equation, smoothness penalties, estimation algorithm, convergence, residual diagnostics, centile calibration and distribution-selection procedure. A fitted centile curve is conditional on the model and data support. It should not be sold as an individual destiny, a causal effect or a reason to ignore measurement error and subgroup fit.

Frequently asked questions

What does GAMLSS model?

Several parameters of a conditional response distribution, not only its mean.

Why use link functions?

They connect predictors to parameters while respecting constraints such as positive scale.

Can predictors affect scale but not location?

Yes. Separate parameter equations allow different effects.

Why does distribution choice matter?

Centiles and tail probabilities depend on the assumed family.

How are centiles checked?

With distribution-aware residuals, coverage and subgroup calibration.

What must be reported?

Family, parameter meanings, links, terms, penalties, convergence, diagnostics and sensitivity.

Primary source and fact-check note

R. A. Rigby and D. M. Stasinopoulos, “Generalized Additive Models for Location, Scale and Shape,” Journal of the Royal Statistical Society: Series C 54(3), 507–554 (2005), DOI 10.1111/j.1467-9876.2005.00510.x. 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 observational unit, treatment or predictors, panel IDs, time order, response orientation, tail direction, missingness and dependence.

Objective audit

Write structural moments, check losses, individual-effect penalties, tail normalisations or multi-parameter likelihoods explicitly.

Assumption audit

List exclusion, relevance, rank, panel, tail-equivalence, response-family and link assumptions that the calculation needs.

Tuning audit

Record tau, penalty, threshold, tail sequence, family, smoothness and solver settings before interpreting output.

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.

Computation audit

Check moment balance, identification rank, optimisation ties, convergence, effective degrees of freedom and numerical sensitivity.

Uncertainty audit

Use weak-identification, cluster, extreme-value, simulation or likelihood methods matched to the actual design.

Validation audit

Use untouched units or future observations for loss, centile coverage, tail calibration and subgroup checks.

Final claim

Translate the conditional result without turning an assumption, percentile, fitted distribution or association into a guarantee.

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: Response family

Begin with this mechanism: Choose a plausible distribution for y. Recreate the concrete case: Begin with Normal for teaching. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A familiar family can still fit badly. Complete the transfer step: Compare alternatives. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Location

Begin with this mechanism: Model mu through a link. Recreate the concrete case: Set mu=10 in both settings. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Same location does not imply same risk. Complete the transfer step: Keep units. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Scale

Begin with this mechanism: Model positive sigma through a link. Recreate the concrete case: Use sigma=2 and 5. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Identity links can permit invalid negative scale. Complete the transfer step: Prefer valid link. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Shape

Begin with this mechanism: Add skewness or kurtosis parameters when the family supports them. Recreate the concrete case: Keep them fixed in teaching case. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: More parameters can overfit. Complete the transfer step: Use diagnostics. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra 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: Predictors

Begin with this mechanism: Assign terms to each parameter equation. Recreate the concrete case: A variable may affect scale but not location. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Automatic selection can be unstable. Complete the transfer step: Publish formulas. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Smooth terms

Begin with this mechanism: Use penalised splines where curvature is supported. Recreate the concrete case: Tune smoothness. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Too many effective degrees of freedom chase noise. Complete the transfer step: Check edf. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Quantile constant

Begin with this mechanism: Use z_.90=1.2816 for Standard Normal. Recreate the concrete case: Keep four decimals. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This constant is family-specific. Complete the transfer step: State source. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Setting A

Begin with this mechanism: Compute 10+1.2816 times 2. Recreate the concrete case: Q_.90=12.5632. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Rounding belongs at the end. Complete the transfer step: Show arithmetic. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra 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: Setting B

Begin with this mechanism: Compute 10+1.2816 times 5. Recreate the concrete case: Q_.90=16.408. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Equal means hide tail difference. Complete the transfer step: Compare plainly. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Lower A

Begin with this mechanism: Compute 10-1.2816 times 2. Recreate the concrete case: Q_.10=7.4368. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Symmetry comes from Normal family. Complete the transfer step: Do not generalise. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Lower B

Begin with this mechanism: Compute 10-1.2816 times 5. Recreate the concrete case: Q_.10=3.592. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Larger spread moves both tails. Complete the transfer step: Keep original units. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Likelihood

Begin with this mechanism: Maximise penalised likelihood for all parameter equations. Recreate the concrete case: Track convergence. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A converged algorithm can fit the wrong family. Complete the transfer step: Inspect objective. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra 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: Residuals

Begin with this mechanism: Use distribution-aware residual diagnostics. Recreate the concrete case: Check centiles and tails. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Mean residual plots are insufficient. Complete the transfer step: Use worm or QQ plots. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Family selection

Begin with this mechanism: Compare fit, diagnostics and substantive plausibility. Recreate the concrete case: Use information criteria carefully. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Lowest score is not automatic truth. Complete the transfer step: Validate externally. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Calibration

Begin with this mechanism: Check observed proportions below fitted centiles by subgroup. Recreate the concrete case: Seek coherent ordered centiles. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Global fit can hide subgroup failure. Complete the transfer step: Audit groups. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: State family, links, equations and support. Recreate the concrete case: Teaching .90 quantiles are 12.5632 and 16.408. Write responses, distribution families, location, scale, shape parameters, link scales, smooth terms, likelihoods, residuals, centiles, coverage and original units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call fitted distributions destiny. Complete the transfer step: Bound conclusion. Change exactly one observation, treatment assignment, instrument, rank restriction, panel visit, individual effect, penalty, tau, sample size, threshold, tail index, family, link, scale, shape parameter, smoother, solver tolerance, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, assumptions, tuning, computation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving weak identification, informative dropout, sparse tails, family misspecification, subgroup miscalibration, measurement error or high leverage. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.