WHY MATHEMATICS?

Why Mathematics? | Bayesian Quantile Regression: Asymmetric Laplace Likelihoods, Priors, Posterior Sampling and Calibration

How check loss becomes a working likelihood, why prior odds change posterior conclusions, and why convergence and uncertainty calibration must be checked.

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

Connect the asymmetric Laplace likelihood to check loss
Combine likelihood information with explicit prior assumptions
Sample and diagnose the posterior coefficient distribution
Validate likelihood fit, prior sensitivity and interval calibration

Chapter map

1. Target
2. ALD likelihood
3. Data
4. Candidate two
5. Candidate one
6. Likelihood ratio
7. Prior odds
8. Scale
9. Posterior
10. Sampling
11. Convergence
12. Prediction
13. Misspecification
14. Sensitivity
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. Target

This step deserves a careful pause. Choose tau and define the conditional quantile.

A concrete situation

Use tau=.5.

What the mathematics does not license

Changing tau changes the estimand. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Record level. 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. ALD likelihood

The useful question is not only how to calculate, but what the calculation assumes. Write density proportional to exp minus check loss.

A concrete situation

Use scale one.

What the mathematics does not license

Parameterisations differ across software. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

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

3. Data

A small example makes the mechanism visible. Use y=1,2,10.

A concrete situation

Ten creates a long upper gap.

What the mathematics does not license

Three observations are illustrative only. Keeping this boundary clear makes the result easier to audit and transfer.

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.

4. Candidate two

This is where modelling choices begin to affect evidence. Set mu=2.

A concrete situation

Losses .5,0,4 total 4.5.

What the mathematics does not license

Loss is not posterior probability. A trustworthy workflow records the choice before seeing the final test result.

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.

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. Candidate one

The central mathematical idea is concrete. Set mu=1.

A concrete situation

Losses 0,.5,4.5 total 5.

What the mathematics does not license

Two candidates do not describe full posterior. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Sample parameter space. 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. Likelihood ratio

Here the geometry becomes operational. Compute exp(-4.5)/exp(-5).

A concrete situation

Ratio is exp(.5)=1.6487.

What the mathematics does not license

Normalising terms cancel only with same tau and scale. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

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

7. Prior odds

This step deserves a careful pause. Multiply likelihood ratio by prior odds.

A concrete situation

Posterior odds depend on both.

What the mathematics does not license

A hidden prior obscures the conclusion. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Publish prior. 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. Scale

The useful question is not only how to calculate, but what the calculation assumes. Estimate or assign positive sigma.

A concrete situation

Scale controls likelihood concentration.

What the mathematics does not license

Fixing it can understate uncertainty. Evidence becomes stronger when the limitation is tested instead of hidden.

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.

9. Posterior

A small example makes the mechanism visible. Normalise likelihood times prior.

A concrete situation

Obtain coefficient distributions.

What the mathematics does not license

A posterior is model-conditional. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Avoid guarantees. 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. Sampling

This is where modelling choices begin to affect evidence. Use MCMC or another verified algorithm.

A concrete situation

Run several chains.

What the mathematics does not license

A trace plot alone is insufficient. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Report 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.

11. Convergence

The central mathematical idea is concrete. Inspect R-hat, effective sample size and divergences where applicable.

A concrete situation

Seek stable chains.

What the mathematics does not license

More draws do not fix a wrong model. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Check 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.

12. Prediction

Here the geometry becomes operational. Generate posterior predictive conditional summaries.

A concrete situation

Compare observed tails.

What the mathematics does not license

ALD working likelihood may miss distribution shape. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Use checks. 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. Misspecification

This step deserves a careful pause. Separate quantile target from literal ALD truth.

A concrete situation

Consider robust covariance or alternative likelihoods.

What the mathematics does not license

Naive credible intervals may miscover. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Calibrate. 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. Sensitivity

The useful question is not only how to calculate, but what the calculation assumes. Change priors and scale assumptions.

A concrete situation

Track coefficient and prediction changes.

What the mathematics does not license

One prior run hides dependence. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Show range. 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

A small example makes the mechanism visible. Explain posterior uncertainty in original units.

A concrete situation

Use probabilities conditional on assumptions.

What the mathematics does not license

Do not conflate with causal probability. Keeping this boundary clear makes the result easier to audit and transfer.

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.

16. Claim

This is where modelling choices begin to affect evidence. State tau, likelihood, prior and computation.

A concrete situation

Teaching likelihood ratio is 1.6487.

What the mathematics does not license

Do not call it posterior odds without the prior. A trustworthy workflow records the choice before seeing the final test result.

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

Bayesian quantile regression combines a conditional-quantile target with a prior distribution and a likelihood-based updating rule. Yu and Moyeed used the asymmetric Laplace distribution because its negative log likelihood contains the same asymmetric check loss that defines ordinary quantile regression. The posterior therefore connects a familiar optimisation target to probability distributions over coefficients, while the prior makes assumptions about plausible coefficient values explicit.

For y=1,2,10, tau=.5 and scale one, location mu=2 gives check losses .5,0,4, totaling 4.5. Location mu=1 gives 0,.5,4.5, totaling 5. With the same scale and tau, the asymmetric-Laplace likelihood ratio for mu=2 versus mu=1 is exp(0.5), about 1.6487, before multiplying by prior odds. A stronger prior favouring one location can reverse the posterior comparison, which is why the prior cannot be treated as decoration.

The asymmetric Laplace distribution can be a working likelihood rather than a literal claim that every conditional response follows that distribution. Posterior point estimates may target useful quantile coefficients under broad settings, yet uncertainty intervals can be miscalibrated when the likelihood is misspecified. Robust or sandwich corrections, alternative Bayesian constructions and simulation-based calibration may be needed depending on the inferential goal.

Report tau, parameterisation, scale treatment, prior, sampler, warm-up, chains, convergence, effective sample size, posterior predictive checks and sensitivity to the likelihood. A credible interval is conditional on the model and prior; it is not automatically a repeated-sampling confidence interval or a guarantee that a future observation falls inside a band.

Frequently asked questions

Why use the asymmetric Laplace distribution?

Its negative log likelihood contains the quantile check loss.

What does the prior do?

It changes posterior odds and regularises coefficient values.

Is the ALD always literally true?

No. It is often a working likelihood whose uncertainty still needs validation.

What is a credible interval?

A posterior probability interval conditional on the model and prior.

How is computation checked?

With multiple chains, convergence, effective sample size and predictive diagnostics.

What must be reported?

Tau, parameterisation, scale, prior, sampler, convergence, sensitivity and calibration.

Primary source and fact-check note

Keming Yu and Rana A. Moyeed, “Bayesian Quantile Regression,” Statistics & Probability Letters 54(4), 437–447 (2001), DOI 10.1016/S0167-7152(01)00124-9. 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, predictor support, time order, response orientation, missingness and dependence.

Objective audit

Write likelihood, prior, forest weights, check losses, penalties or lag equations explicitly.

Tuning audit

Record scale, priors, sampler settings, tree settings, lambda grids, tau grids and lag order.

Computation audit

Check posterior convergence, weight normalisation, solution paths, tied optima and solver tolerances.

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.

Coherence audit

Inspect quantile order, support, stability, effective neighbours, hit sequences and extrapolation.

Uncertainty audit

Use model-compatible posterior, sandwich or bootstrap methods and include tuning or repair.

Validation audit

Use untouched or rolling data for check loss, coverage, width and subgroup calibration.

Final claim

Translate the conditional tail result without causal, posterior or guarantee overclaims.

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: Target

Begin with this mechanism: Choose tau and define the conditional quantile. Recreate the concrete case: Use tau=.5. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Changing tau changes the estimand. Complete the transfer step: Record level. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: ALD likelihood

Begin with this mechanism: Write density proportional to exp minus check loss. Recreate the concrete case: Use scale one. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Parameterisations differ across software. Complete the transfer step: State formula. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Data

Begin with this mechanism: Use y=1,2,10. Recreate the concrete case: Ten creates a long upper gap. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Three observations are illustrative only. Complete the transfer step: Label example. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Candidate two

Begin with this mechanism: Set mu=2. Recreate the concrete case: Losses .5,0,4 total 4.5. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Loss is not posterior probability. Complete the transfer step: Keep units. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Candidate one

Begin with this mechanism: Set mu=1. Recreate the concrete case: Losses 0,.5,4.5 total 5. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Two candidates do not describe full posterior. Complete the transfer step: Sample parameter space. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Likelihood ratio

Begin with this mechanism: Compute exp(-4.5)/exp(-5). Recreate the concrete case: Ratio is exp(.5)=1.6487. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Normalising terms cancel only with same tau and scale. Complete the transfer step: Check conditions. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Prior odds

Begin with this mechanism: Multiply likelihood ratio by prior odds. Recreate the concrete case: Posterior odds depend on both. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A hidden prior obscures the conclusion. Complete the transfer step: Publish prior. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Scale

Begin with this mechanism: Estimate or assign positive sigma. Recreate the concrete case: Scale controls likelihood concentration. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Fixing it can understate uncertainty. Complete the transfer step: Run sensitivity. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Posterior

Begin with this mechanism: Normalise likelihood times prior. Recreate the concrete case: Obtain coefficient distributions. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A posterior is model-conditional. Complete the transfer step: Avoid guarantees. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Sampling

Begin with this mechanism: Use MCMC or another verified algorithm. Recreate the concrete case: Run several chains. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A trace plot alone is insufficient. Complete the transfer step: Report diagnostics. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Convergence

Begin with this mechanism: Inspect R-hat, effective sample size and divergences where applicable. Recreate the concrete case: Seek stable chains. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: More draws do not fix a wrong model. Complete the transfer step: Check geometry. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Prediction

Begin with this mechanism: Generate posterior predictive conditional summaries. Recreate the concrete case: Compare observed tails. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: ALD working likelihood may miss distribution shape. Complete the transfer step: Use checks. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Misspecification

Begin with this mechanism: Separate quantile target from literal ALD truth. Recreate the concrete case: Consider robust covariance or alternative likelihoods. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Naive credible intervals may miscover. Complete the transfer step: Calibrate. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Sensitivity

Begin with this mechanism: Change priors and scale assumptions. Recreate the concrete case: Track coefficient and prediction changes. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One prior run hides dependence. Complete the transfer step: Show range. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: Explain posterior uncertainty in original units. Recreate the concrete case: Use probabilities conditional on assumptions. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not conflate with causal probability. Complete the transfer step: Keep scope. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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: State tau, likelihood, prior and computation. Recreate the concrete case: Teaching likelihood ratio is 1.6487. Write responses, quantile levels, locations, scales, check losses, likelihoods, prior odds, posterior odds, coefficients, draws and original measurement 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 it posterior odds without the prior. Complete the transfer step: Bound conclusion. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, 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, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, 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.