Choose your reading route
eduKateSG · Why Mathematics? · Birge Ratios, Weighted Scatter and Uncertainty Consistency
Start with the question, audit the worked example, or jump to the assumptions and practice. Every chapter stays open for uninterrupted reading.
Understand the question
Define the quantity and follow the mechanism.Route 2 · Chapters 4–7
Reproduce the example
Audit the numbers one step at a time.Route 3 · Chapters 8–13
Check assumptions
See where design and calibration matter.Route 4 · Chapters 14–18
Explore applications
Connect the method to a concrete investigation.Route 5 · Chapters 19–22
Study and verify
Practise, inspect failure cases and read further.
Full chapter index · Practice and checking · Mathematics Learning Hub
When results come with uncertainties, a weighted mean may look precise even though the results disagree more than those uncertainties predict. The Birge ratio compares the observed weighted scatter with the degrees of freedom. It asks whether the uncertainty model and the data are statistically consistent as a whole.
This is one reason mathematics is important in measurement: a mean is not enough. We also need to check whether the pattern around the mean fits the assumptions used to calculate it.
Full chapter index
Chapters 1–3 · Understand the question
Chapters 4–7 · Reproduce the example
Chapters 8–13 · Check assumptions
Chapters 14–18 · Explore applications
Chapters 19–22 · Study and verify
Suppose n independent results x_i estimate one common measurand and have standard uncertainties u_i. The inverse-variance weighted mean is x_w = sum(x_i/u_i squared) divided by sum(1/u_i squared).
Calculate Q = sum[(x_i – x_w) squared/u_i squared]. Under the simple consistent normal model with correctly assessed independent uncertainties, Q is compared with a chi-square distribution with n – 1 degrees of freedom.
The Birge ratio is R_B = square root of Q/(n – 1). It is the square root of reduced chi-square.
A value near one means observed weighted scatter is on the scale expected from the stated uncertainties under the model. A value above one means more scatter; below one means less. Formal consistency requires the protocol's probability reference, not a universal rule that every value above one fails.
Take four results 98, 99, 101 and 102 units, each with standard uncertainty 1 unit. Equal uncertainties give equal weights, so the weighted mean is the arithmetic mean 100.
The standardised residuals are -2, -1, 1 and 2. Their squares sum to Q = 4 + 1 + 1 + 4 = 10.
| Result | Residual from 100 | Squared standardised residual |
|---|---|---|
| 98 | -2 | 4 |
| 99 | -1 | 1 |
| 101 | 1 | 1 |
| 102 | 2 | 4 |
There are n – 1 = 3 degrees of freedom. R_B = square root of 10/3, approximately square root of 3.3333, which is 1.8257.
The data scatter is about 1.83 times the scale represented by the simple uncertainty model after degrees-of-freedom adjustment. A decision still requires the relevant chi-square probability or protocol threshold.
| Quantity | Calculation | Result |
|---|---|---|
| Q | 4+1+1+4 | 10 |
| Degrees of freedom | 4−1 | 3 |
| R_B | √(10/3) | 1.8257 |
The weighted mean is estimated from the same four results, so one linear piece of information is used to locate the centre. Only three independent residual contrasts remain.
Dividing by n rather than n – 1 would ignore that fitted parameter. Degrees of freedom connect geometry, variance and probability.
With different u_i values, results with smaller uncertainty receive larger weight. The weighted mean may move toward them, and their deviations contribute more strongly to Q.
An understated small uncertainty can therefore inflate the Birge ratio and distort the mean. Review both values and uncertainty budgets.
The simple formula assumes independent results. Laboratories may share reference standards, transfer artefacts or common corrections. Correlation changes the appropriate quadratic form and degrees-of-freedom analysis.
Do not treat separate laboratory names as proof of independence. Read the comparison protocol.
A high ratio may reflect an unrecognised laboratory effect, drift, non-comparable measurands, an outlier, underestimated uncertainties or ignored correlation. It does not identify which cause is present.
The BIPM key-comparison report for gauge blocks used the Birge ratio to assess overall statistical consistency and found cases where large ratios indicated that a simple weighted mean was unreliable without modification.
A low ratio can occur by chance, but a persistently very low value can prompt checks for overestimated uncertainties, correlation, shared data or hidden averaging.
Small is not automatically better. The goal is a realistic uncertainty model.
Some analyses multiply uncertainties by a Birge ratio when scatter is high. That may be appropriate under a specified model, but it cannot explain non-comparable measurands or correct a biased result.
Any scaling rule must come from the governing procedure and be reported transparently.
Because R_B squared = Q/(n – 1), a formal test can be expressed directly through Q and the chi-square distribution. Critical ranges depend on n and probability level.
The ratio is a readable effect-scale summary, while Q supplies the distributional comparison.
The Birge ratio diagnoses extra scatter around a weighted mean. Mandel–Paule estimates a non-negative between-laboratory variance and recalculates weights and consensus.
One asks whether the simple model fits; the other builds a random-effects extension. Keep their purposes distinct.
Normalised error compares an individual result with an assigned or reference value using combined uncertainty. The Birge ratio looks at overall consistency across all participants.
The recent normalised-error guide develops the individual comparison.
Give four groups the same safe measurement task and ask each to report a result and an evidence-based standard uncertainty. Calculate the weighted mean, standardised residuals, Q and R_B.
Then discuss whether the uncertainty estimates are comparable. Label the exercise as a model rather than a formal calibration comparison.
Use columns for x_i, u_i, weight 1/u_i squared, weighted value, residual and squared standardised residual. Sum weights and weighted values for x_w, then sum the final column for Q.
Calculate degrees of freedom and R_B in separately labelled cells. Keep unrounded numbers until display.
Validate positive uncertainties and consistent units. Compute the weighted mean with stable numerical summation, then Q and R_B. Return the result table so each contribution is visible.
For covariance, replace the scalar-weight formula with the documented matrix calculation. Test the equal-uncertainty example exactly.
Students can connect weighted averages, residuals, squares, degrees of freedom and probability. Ask why a result two uncertainty units from the mean contributes four to Q.
Parents can encourage careful language: a high ratio means the simple consistency model deserves review, not that one participant is automatically wrong.
Overall consistency checks appear in national metrology comparisons, calibration, physics constants and interlaboratory studies. They develop transferable skills in uncertainty, weighting and model checking.
Explore the broader pathway in the Mathematics Learning Hub and How Mathematics Works.
Replace 102 with 101 and recompute the mean, Q and R_B. Double all results and uncertainties and verify that R_B is unchanged. Halve only the uncertainties and explain why R_B doubles. Give the first result uncertainty 0.5 and recalculate the weighted centre.
State which assumptions each calculation uses.
### Is a Birge ratio above one always unacceptable? No. Sampling variability and the chosen probability criterion matter. Use the protocol's chi-square reference.
### Does R_B identify an outlier? No. It summarises overall weighted scatter. Inspect individual residuals and technical records separately.
### Is R_B unitless? Yes. Each residual is divided by an uncertainty in the same unit before squaring.
### Can I ignore correlations? No. Shared uncertainty components can change the consistency calculation materially.
Use the BIPM CCL-K4 final report for a real key-comparison discussion of Birge ratios and weighted means, and the current CIPM MRA measurement-comparisons guidance for comparison context. The four results here are original teaching data.
The Birge ratio asks a generous but demanding question: do the results and uncertainties tell one coherent story? Its value lies in checking the model before trusting the polished weighted mean.
Extended practice laboratory
Practice case 1: Move an outer result inward
Setup: Use the article’s Birge-ratio example and change one declared input for case 1. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 2: Move an outer result outward
Setup: Use the article’s Birge-ratio example and change one declared input for case 2. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 3: Double all values and uncertainties
Setup: Use the article’s Birge-ratio example and change one declared input for case 3. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 4: Halve only uncertainties
Setup: Use the article’s Birge-ratio example and change one declared input for case 4. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 5: Give one result smaller uncertainty
Setup: Use the article’s Birge-ratio example and change one declared input for case 5. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 6: Give one result larger uncertainty
Setup: Use the article’s Birge-ratio example and change one declared input for case 6. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 7: Recompute the weighted mean
Setup: Use the article’s Birge-ratio example and change one declared input for case 7. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 8: Audit Q contributions
Setup: Use the article’s Birge-ratio example and change one declared input for case 8. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 9: Use n instead of n minus one
Setup: Use the article’s Birge-ratio example and change one declared input for case 9. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 10: Add a fifth centred result
Setup: Use the article’s Birge-ratio example and change one declared input for case 10. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 11: Add a fifth extreme result
Setup: Use the article’s Birge-ratio example and change one declared input for case 11. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 12: Check a low ratio
Setup: Use the article’s Birge-ratio example and change one declared input for case 12. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 13: Check a high ratio
Setup: Use the article’s Birge-ratio example and change one declared input for case 13. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 14: Compare chi-square
Setup: Use the article’s Birge-ratio example and change one declared input for case 14. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 15: Introduce correlation
Setup: Use the article’s Birge-ratio example and change one declared input for case 15. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 16: Inspect standardised residuals
Setup: Use the article’s Birge-ratio example and change one declared input for case 16. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 17: Audit uncertainty budgets
Setup: Use the article’s Birge-ratio example and change one declared input for case 17. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 18: Avoid automatic scaling
Setup: Use the article’s Birge-ratio example and change one declared input for case 18. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 19: Compare normalised error
Setup: Use the article’s Birge-ratio example and change one declared input for case 19. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 20: Compare Mandel-Paule
Setup: Use the article’s Birge-ratio example and change one declared input for case 20. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 21: Round near a threshold
Setup: Use the article’s Birge-ratio example and change one declared input for case 21. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 22: Use covariance matrix form
Setup: Use the article’s Birge-ratio example and change one declared input for case 22. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 23: Record the protocol rule
Setup: Use the article’s Birge-ratio example and change one declared input for case 23. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.
Practice case 24: Write a model-check conclusion
Setup: Use the article’s Birge-ratio example and change one declared input for case 24. Task: Recalculate from raw values, preserve units and state which assumption or design condition changed.
Check: Check the relevant invariant, separate the numerical result from the claim it supports, and name the next diagnostic or source of evidence. Rebuild the relevant counts before interpreting the coefficient, preserve the stated design, and write one sentence separating the numerical result from the claim it can support. Then vary one defensible assumption, calculate again, and explain whether the conclusion changes because of new data, a new reference model, or a new decision purpose. Finish by naming the unit, tolerance, and evidence needed for transfer. Record one graph or table that would expose failure hidden by the summary.