Choose your reading route
eduKateSG · Why Mathematics? · Biweight Scale, Residual Cutoffs and Robust Dispersion
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 19–25
Study and verify
Practise, inspect failure cases and read further.
Full chapter index · Practice and checking · Mathematics Learning Hub
Why is mathematics important when a measure of spread should use most observations but prevent distant values from dominating? The biweight scale estimator starts from the median and median absolute deviation, converts residuals into scaled distances, and gives central observations smoothly decreasing influence while excluding residuals beyond a cutoff.
The method connects powers, weights, robust scale and declared tuning constants. It complements biweight location: one estimates a centre, while the other estimates dispersion.
Did you know? In the NIST biweight-scale rule, a sufficiently distant observation contributes nothing to the weighted numerator or correction sum, yet it remains visible in the original dataset for investigation.
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 19–25 · Study and verify
Standard deviation squares every residual from the mean, so a single extreme value can dominate. MAD is resistant but may be less efficient under some models.
Biweight scale seeks a robust, efficient dispersion estimate by letting influence fall with standardized distance.
NIST uses the raw-data median y-prime and MAD = median|y_i − y-prime|. It defines u_i = (y_i − y-prime)/(9×MAD) and restricts the sums to u_i squared at most 1.
Inside the cutoff, the numerator uses squared residuals multiplied by (1 − u_i squared) to the fourth power. A separate correction sum uses (1 − u_i squared)(1 − 5u_i squared).
| Data | Median | MAD | u for 30 |
|---|---|---|---|
| 9, 10, 11, 12, 30 | 11 | 1 | 19/9 |
Consider 9, 10, 11, 12 and 30. The median is 11. Absolute deviations are 2, 1, 0, 1 and 19, so MAD = 1.
The u values are −2/9, −1/9, 0, 1/9 and 19/9. The last value lies outside |u| ≤ 1 and is excluded from the biweight sums.
| B weighted residual sum | A correction sum | Scale squared | Scale |
|---|---|---|---|
| 5.16955 | 3.569273 | 2.818595 | 1.678867 |
For the four included values, the weighted squared-residual sum is approximately 5.169550. The correction sum A = sum[(1−u squared)(1−5u squared)] is approximately 3.569273.
Using the NIST finite-sample form, scale squared is nB/[A(A−1)] = 5×5.169550/[3.569273×2.569273] ≈ 2.818595.
The square root gives a biweight scale of approximately 1.678867 units for the stated convention.
Keep more precision through the calculation and round only the final display. Software variants can use related biweight midvariance conventions, so name the exact statistic.
The observation 30 has |u| greater than 1 and receives no contribution to the restricted sums. Changing it to 300 leaves the median, MAD and included central residuals unchanged in this example.
That resistance does not prove 30 is wrong. It says the robust scale is designed not to let it dominate.
The denominator 9×MAD sets the residual scale. A larger tuning constant admits more distant observations; a smaller one rejects them sooner.
Nine is part of NIST’s documented convention, not a universal natural constant. Other software must be checked before values are compared.
Central residuals have u near zero, so (1−u squared) to the fourth is near one. As |u| approaches one, the weight approaches zero smoothly.
The estimator therefore changes influence gradually before the hard cutoff.
Biweight location iteratively updates a weighted centre. The NIST scale formula here is built around the raw median and MAD with a different tuning factor and finite-sample correction.
The biweight-location guide explains the feedback calculation for centre.
MAD is straightforward: take the median absolute distance from the median. Biweight scale uses those distances to construct u, then uses central residual magnitudes more fully.
Both remain vulnerable to poor sampling, dependence and limited measurement resolution.
Standard deviation is valuable when variance and Gaussian likelihood match the question. Biweight scale is valuable when tail contamination should not control the first estimate of spread.
Report the statistic by name; the numbers are not interchangeable.
Adding a constant to every observation leaves residuals, MAD and scale unchanged. Multiplying every observation by a positive factor multiplies MAD and the final scale by that factor while leaving u values unchanged.
These invariants make excellent unit tests.
If MAD is zero, u is undefined. This can happen with many ties or coarse resolution.
Do not insert a tiny denominator silently. Report the condition and use a documented fallback.
The correction term depends on n and the effective weight structure. Small samples can make the result unstable or undefined if too few observations receive weight.
State sample size, cutoff convention and software implementation.
Students can compute median, MAD and u for the five values, then mark which observations enter the sums. A plot of residual against weight makes the influence rule visible.
Change 30 to 300 and explain which quantities stay fixed.
Use columns for residual, u, inclusion flag, numerator contribution and correction contribution. Sum only rows meeting the declared cutoff.
Keep the raw data untouched and label the tuning constant.
Validate numeric input, compute median and MAD, stop or branch if MAD is zero, then calculate restricted sums and the finite-sample denominator.
Unit-test the example to obtain B≈5.169550, A≈3.569273 and scale≈1.678867.
Biweight scale does not identify causes, certify normality or justify deleting observations. A zero weight is an estimator rule, not a factual verdict.
It also cannot repair selection bias or mixed populations.
Robust scale matters in astronomy, sensor analysis, manufacturing, environmental monitoring and experimental science. It supports diagnostics when occasional exceptional readings are plausible.
The transferable skill is to inspect how a weight function converts residual size into influence.
The Sn guide uses nested medians, while the Qn guide selects an ordered pairwise distance. Biweight scale instead uses smoothly weighted residual powers.
Students should show u and inclusion decisions before applying the long formula. Understanding the cutoff is more valuable than memorising it.
Parents can ask whether raw values, tuning constants and software conventions remain available.
Recalculate after replacing 12 with 13. Translate all values by 100. Double every value and predict the scale. Construct a tied dataset with MAD zero.
Compare biweight scale, MAD and standard deviation for the original sample.
### Is biweight scale the same as biweight location? No. One estimates dispersion and the other estimates centre.
### Why is 30 excluded? Its scaled residual exceeds the documented cutoff |u| ≤ 1.
### Does exclusion mean the value is erroneous? No. It limits influence in this estimator only.
### Can different software give different values? Yes, if it uses a different correction or biweight variant. Report the convention.
Read the official NIST biweight-scale documentation, the NIST biweight transformation manual and the NIST MAD definition. The worked calculation was independently reproduced.
Explore the Mathematics Learning Hub and How Mathematics Works for related foundations.
Biweight scale makes robust dispersion a visible weighting process. Mathematics controls influence without erasing the original observation, leaving science to investigate why the unusual value occurred.
Extended practice laboratory
Practice case 1: Find the median
Setup: Use the article’s biweight-scale example and change one declared input for case 1. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Find MAD
Setup: Use the article’s biweight-scale example and change one declared input for case 2. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Calculate u
Setup: Use the article’s biweight-scale example and change one declared input for case 3. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Apply the cutoff
Setup: Use the article’s biweight-scale example and change one declared input for case 4. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Calculate numerator terms
Setup: Use the article’s biweight-scale example and change one declared input for case 5. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Calculate correction terms
Setup: Use the article’s biweight-scale example and change one declared input for case 6. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Take the square root
Setup: Use the article’s biweight-scale example and change one declared input for case 7. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Move the far value
Setup: Use the article’s biweight-scale example and change one declared input for case 8. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Translate all values
Setup: Use the article’s biweight-scale example and change one declared input for case 9. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Scale all values
Setup: Use the article’s biweight-scale example and change one declared input for case 10. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Change the cutoff
Setup: Use the article’s biweight-scale example and change one declared input for case 11. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Create zero MAD
Setup: Use the article’s biweight-scale example and change one declared input for case 12. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Create few included values
Setup: Use the article’s biweight-scale example and change one declared input for case 13. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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 MAD
Setup: Use the article’s biweight-scale example and change one declared input for case 14. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Compare standard deviation
Setup: Use the article’s biweight-scale example and change one declared input for case 15. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Compare Sn
Setup: Use the article’s biweight-scale example and change one declared input for case 16. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Compare Qn
Setup: Use the article’s biweight-scale example and change one declared input for case 17. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Audit finite correction
Setup: Use the article’s biweight-scale example and change one declared input for case 18. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Audit software variant
Setup: Use the article’s biweight-scale example and change one declared input for case 19. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Handle missing data
Setup: Use the article’s biweight-scale example and change one declared input for case 20. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Plot weights
Setup: Use the article’s biweight-scale example and change one declared input for case 21. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Preserve raw values
Setup: Use the article’s biweight-scale example and change one declared input for case 22. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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: Avoid error claims
Setup: Use the article’s biweight-scale example and change one declared input for case 23. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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 robust-scale conclusion
Setup: Use the article’s biweight-scale example and change one declared input for case 24. Task: Recalculate from raw values, preserve units and state which mathematical or design condition changed.
Check: Check the relevant invariant, separate the number from its claim, 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.