Why Mathematics? | Effective Sample Size for Bayesian Reliability Models: Autocorrelation, Bulk and Tail ESS, Quantiles and Precision

Begin with serial dependence. A Markov chain deliberately makes each draw depend on its predecessor. A slowly moving Weibull shape produces long runs of similar values. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. File size is not information size. Inspect dependence.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Record the estimand, chain structure, autocorrelation method and software version. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use saved draws are not independent as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Did You Know? Effective sample size belongs to a chosen posterior quantity; one chain can have high ESS for a mean and low ESS for a tail quantile. That is why a careful report includes the construction, not only the final scalar.

Contents · Next chapter

Now translate correlation into independent-equivalent information. Lag-k autocorrelation measures linear dependence between draws k steps apart. The example uses 0.60, 0.30 and 0.10. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Negative and nonlinear dependence need careful treatment. Plot several lags.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Report bulk, tail and local ESS for the quantities that drive decisions. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use autocorrelation by lag as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. Add one plus twice the relevant autocorrelation sum. The teaching sum is 1+2×1=3. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Arbitrary truncation can make the estimate unstable. Use validated sequence estimators.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Keep MCMC ESS distinct from observed sample size and weighted-sample ESS. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use integrated autocorrelation time as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Begin with serial dependence. Divide N by the integrated autocorrelation time. 4,000/3 is about 1,333. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. The number is quantity-specific and approximate. Report method and target.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Use effective draws per second when comparing computational efficiency. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use worked ess about 1,333 as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Primary research: https://doi.org/10.1214/ss/1177011137. Read the authors’ definition and conditions before carrying the method into a censored reliability model.

Contents · Previous chapter · Next chapter

Now translate correlation into independent-equivalent information. ESS divided by N describes information retained per draw. 1,333/4,000 is about 0.333. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. High computational cost per draw is not reflected. Compare time per effective draw.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Pair ESS with R-hat, MCSE, trace or rank plots and model diagnostics. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use efficiency ratio as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. At large lags fewer paired observations remain and estimates fluctuate. A tail of small sample correlations changes the raw sum. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Summing every lag can produce nonsense. Use initial positive sequences.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Record the estimand, chain structure, autocorrelation method and software version. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use why autocorrelations become noisy as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Did You Know? Effective sample size belongs to a chosen posterior quantity; one chain can have high ESS for a mean and low ESS for a tail quantile. That is why a careful report includes the construction, not only the final scalar.

Contents · Previous chapter · Next chapter

Begin with serial dependence. Pair autocovariances and truncate with positivity and monotonicity principles. Noisy late pairs are excluded. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Finite-sample estimates still have uncertainty. Keep chain length adequate.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Report bulk, tail and local ESS for the quantities that drive decisions. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use geyer sequence ideas as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Now translate correlation into independent-equivalent information. Modern estimators use within- and between-chain information with rank-based transformations. Four chains contribute to one bulk ESS. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Concatenating nonconverged chains can inflate confidence. Check R-hat first.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Keep MCMC ESS distinct from observed sample size and weighted-sample ESS. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use multiple-chain ess as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. Bulk ESS measures efficiency for central location and rank-normalised means. Median log lifetime is the target. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. A large bulk ESS does not guarantee precise extreme quantiles. Add tail ESS.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Use effective draws per second when comparing computational efficiency. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use bulk ess as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Begin with serial dependence. Tail ESS examines indicators for lower and upper quantile regions. Reliability at the 5th and 95th percentiles mixes slowly. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. One tail can be much worse than the centre. Report both tails.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Pair ESS with R-hat, MCSE, trace or rank plots and model diagnostics. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use tail ess as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Now translate correlation into independent-equivalent information. Local ESS can be calculated for probabilities near a chosen quantile. The 0.99 lifetime quantile drives a warranty limit. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Global ESS averages away local difficulty. Match ESS to the estimand.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Record the estimand, chain structure, autocorrelation method and software version. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use quantile-specific efficiency as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Did You Know? Effective sample size belongs to a chosen posterior quantity; one chain can have high ESS for a mean and low ESS for a tail quantile. That is why a careful report includes the construction, not only the final scalar.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. Derived survival probabilities require their own ESS. R(1,000 hours) is computed at every draw. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Parameter ESS does not automatically equal prediction ESS. Diagnose generated quantities.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Report bulk, tail and local ESS for the quantities that drive decisions. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use reliability probabilities as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Second authoritative source: https://doi.org/10.1214/20-BA1221. Compare its reference distribution, diagnostics and finite-simulation cautions with this worked example.

Contents · Previous chapter · Next chapter

Begin with serial dependence. Few observed failures can create strong posterior correlation between shape and scale. A ridge reduces ESS for median lifetime. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. More chains do not remove weak identification. Improve design or prior information.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Keep MCMC ESS distinct from observed sample size and weighted-sample ESS. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use censored-data effects as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Now translate correlation into independent-equivalent information. Population means and group deviations often mix at different rates. Batch variance has ESS 120 while mean life has ESS 2,000. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Reporting only the best value hides risk. Show the minimum relevant ESS.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Use effective draws per second when comparing computational efficiency. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use hierarchical parameters as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. Some algorithms can yield ESS estimates above the nominal number of draws. Alternating draws reduce variance of a mean. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Very large estimates can also signal estimator noise. Interpret with diagnostics.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Pair ESS with R-hat, MCSE, trace or rank plots and model diagnostics. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use negative autocorrelation as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Begin with serial dependence. Keeping every kth draw reduces stored autocorrelation but usually discards useful information. Saving every tenth draw shrinks the file. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Thinning rarely improves ESS per computation. Thin only for justified storage or rank needs.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Record the estimand, chain structure, autocorrelation method and software version. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use thinning as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Did You Know? Effective sample size belongs to a chosen posterior quantity; one chain can have high ESS for a mean and low ESS for a tail quantile. That is why a careful report includes the construction, not only the final scalar.

Contents · Previous chapter · Next chapter

Now translate correlation into independent-equivalent information. When chains mix adequately, longer runs usually increase ESS roughly with work. Doubling post-warmup length nearly doubles bulk ESS. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Poor geometry can prevent proportional gains. Measure ESS per second.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Report bulk, tail and local ESS for the quantities that drive decisions. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use more iterations as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. A better coordinate system can reduce posterior correlation and raise ESS. Non-centred batch effects move more freely. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Reparameterisation must preserve the intended prior. Validate the transformed model.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Keep MCMC ESS distinct from observed sample size and weighted-sample ESS. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use reparameterisation as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Begin with serial dependence. For a posterior mean, MCSE is approximately posterior SD divided by square root ESS. ESS converts dependence into precision. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. This formula needs an appropriate central limit approximation. Report both quantities.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Use effective draws per second when comparing computational efficiency. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use ess and monte carlo error as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Now translate correlation into independent-equivalent information. MCMC ESS concerns numerical simulation, not the number of tested components. Fifty failures analysed with ESS 2,000 still have fifty observed failures. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Computational precision cannot create scientific information. Keep both sample sizes distinct.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Pair ESS with R-hat, MCSE, trace or rank plots and model diagnostics. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use ess is not data sample size as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. Particle or importance-weight ESS uses weight concentration, while MCMC ESS uses serial dependence. Both share an independent-equivalent intuition. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. The formulas and failure modes differ. Name the estimator.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Record the estimand, chain structure, autocorrelation method and software version. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use ess is not importance ess as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Did You Know? Effective sample size belongs to a chosen posterior quantity; one chain can have high ESS for a mean and low ESS for a tail quantile. That is why a careful report includes the construction, not only the final scalar.

Contents · Previous chapter · Next chapter

Begin with serial dependence. Continue sampling until relevant ESS or MCSE targets are met after convergence checks. A warranty quantile needs more tail ESS. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. Stopping on a favourable estimate can bias reporting. Predefine targets.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Report bulk, tail and local ESS for the quantities that drive decisions. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use stopping by target precision as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Now translate correlation into independent-equivalent information. Record chain counts, draws, transformations and estimator version. Two packages may use different legacy formulas. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. A bare ESS number is hard to reproduce. Save diagnostic settings.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Keep MCMC ESS distinct from observed sample size and weighted-sample ESS. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use software and reproducibility as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Contents · Previous chapter · Next chapter

Precision depends on the particular posterior quantity. ESS connects sequences, correlations, sums, ratios and square roots. Verify tau=3 and ESS≈1,333. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. The main insight is that dependence reduces information. Compare two correlation patterns.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Use effective draws per second when comparing computational efficiency. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use parent and student learning route as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Useful next reading: eduKate Mathematics Learning System, approach to learning mathematics, Bukit Timah Mathematics curriculum overview, cross-subject learning approach, previous Why Mathematics article, Mathematics Learning Hub. Choose the destination that matches the next question rather than treating links as a checklist.

Contents · Previous chapter · Next chapter

Begin with serial dependence. ESS is not the number of observations, a universal property of a chain or a substitute for convergence. Bulk and tail values answer different precision questions. This step matters because autocorrelation, independent-equivalent information and reliability precision require the mathematical object, prediction target and reliability question to stay aligned. A large ESS cannot validate the scientific model. Read Geyer and the improved R-hat and ESS paper.

A reproducible audit should record variables, units, observation boundaries, censoring codes, model version and software settings. Pair ESS with R-hat, MCSE, trace or rank plots and model diagnostics. Rebuild the displayed number from saved inputs before interpreting it. A result that cannot be reproduced is not ready to guide reliability work.

Try a small learning experiment: sum autocorrelations 0.60, 0.30 and 0.10, form an autocorrelation time of three and divide 4,000 by three. Use common misconceptions and next reading as the focus. Predict the direction of change before using software, alter one defensible input, and finish with two sentences: what the calculation supports, and what remains uncertain. This habit turns mathematics education into transferable problem-solving rather than formula recall.

Quick answers

What must be fixed before checking a model or algorithm? Define the dataset unit, censoring mechanism, time scale, parameterisation, simulation design and diagnostic rule.

Does a satisfactory diagnostic prove the model is true? No. It supports a particular check under stated assumptions and simulation conditions.

Which mathematics matters? Probability defines the reference distribution, simulation makes it observable, proportions summarise tail events, variance describes uncertainty and graphs expose structured mismatch.

How should a student practise? Preserve intermediate quantities, predict directions before calculating, change one input at a time and explain the result in ordinary language.

What follows an internal check? Sensitivity analysis, independent computation, held-out or later-data prediction and a separate decision analysis when cost or safety matters.

Contents · Previous chapter · Mathematics Learning Hub