Choose your reading route
eduKateSG · Why Mathematics? · Trimmed Means, Tail Removal and Robust Location
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 few tail values make an ordinary mean unrepresentative of the central pattern? A trimmed mean sorts the data, removes a declared percentage from each tail and averages the remaining observations.
The method links order statistics, percentages and averages. Its apparent simplicity requires discipline: the trim percentages, rounding rule and target population must be stated before the result is interpreted.
Did you know? A trimmed mean changes both the sum and the denominator. That is why it is not the same calculation as replacing tail values and keeping the original count.
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
An ordinary mean uses every magnitude and can move sharply with an extreme value. A median resists tails but may ignore useful spacing in the central region.
A trimmed mean averages a chosen central fraction, offering a tunable compromise.
NIST defines the trimmed mean by dropping declared percentages from the lower and upper tails. A 90% trimmed mean keeps the middle 90% after removing 5% at each end.
NIST allows separate lower and upper percentages through P1 and P2. The percentages must be reported.
| Original | Trim each tail | Retained |
|---|---|---|
| 1,2,3,4,5,6,7,8,9,100 | 10% = one value | 2,3,4,5,6,7,8,9 |
Consider 1, 2, 3, 4, 5, 6, 7, 8, 9 and 100. The ordinary mean is 145/10 = 14.5.
Trim 10% from each tail. With n = 10, remove one lowest and one highest observation: 1 and 100.
| Ordinary mean | Trimmed sum | Retained n | Trimmed mean |
|---|---|---|---|
| 14.5 | 44 | 8 | 5.5 |
The retained values are 2 through 9. Their sum is 44 and the retained count is 8.
The 10%-per-tail trimmed mean is 44/8 = 5.5.
Once 100 is removed by the prespecified tail rule, changing it to 1,000 does not change the trimmed mean.
The raw value must still be preserved. Robust calculation is not permission to discard evidence.
Trimming removes observations from the calculation and reduces the denominator. Winsorising replaces tail values with boundary values and keeps n positions.
The Winsorised-mean guide shows the replacement mechanism.
A 50% trimmed mean averages values between the lower and upper quartiles under the stated convention. A median uses only the middle position or middle pair.
The two can coincide in simple symmetric samples but are not identical estimators.
A small trim limits only the most extreme values. A large trim increases resistance while using less of the data’s spacing.
Choose the percentage from the study design, accepted method or sensitivity plan, not from which value gives the preferred answer.
Equal trimming suits a symmetric contamination concern. One-sided processes may justify different lower and upper percentages.
Asymmetric trimming changes the estimand and can shift location deliberately, so explain the design.
If n times the trim proportion is not an integer, software may floor, round or interpolate. Different conventions can retain different observations.
State the rule and software version, especially for small samples.
Several observations may equal the boundary. Removing some tied values but not others can look arbitrary even when it follows an index rule.
Show retained values or the exact cutoff algorithm.
Adding a constant to every value adds it to the trimmed mean. Multiplying every value by a positive factor multiplies the result by that factor.
These equivariance checks catch unit and formula errors.
Define the analysis set before trimming. Removing inconvenient records as missing and then trimming again creates hidden double selection.
Document exclusions, duplicates and missingness separately.
The trimmed mean estimates the centre of a trimmed population distribution, not automatically the ordinary population mean. Its standard error also needs a method suited to trimming.
Do not attach the ordinary mean’s formula by habit.
Students can place the ten values on cards, remove one card from each end and average the remainder. Then replace 100 with 1,000 and compare summaries.
Repeat with 20% per tail to see the resistance-information tradeoff.
Sort a copy of the data, calculate the lower and upper removal counts, mark retained rows and average only those rows.
Keep IDs beside values so the audit identifies what was trimmed.
Validate proportions, sort with stable identifiers, apply the documented count rule and return the trimmed mean, retained n and excluded IDs.
Unit-test the example to retain eight values and obtain 5.5.
Trimming does not prove tail observations are errors, create normality or repair a biased sample. It can hide important extremes if reported without raw context.
The method is robust to magnitude after removal, but sensitive to the chosen percentage.
Trimmed means appear in survey analysis, finance, laboratory summaries, sports scoring and robust inference. They are useful when tail influence must be controlled transparently.
The transferable skill is to declare a data-selection rule before calculating.
The Hodges–Lehmann guide builds a robust centre from pairwise averages. The biweight-location guide downweights residuals rather than removing fixed tail counts.
Students should list retained observations and the new denominator. This prevents “trimmed” from becoming a hidden calculator setting.
Parents can ask whether the percentage was chosen in advance and whether conclusions change under a reasonable sensitivity analysis.
Trim one value from each tail of 2, 3, 4, 5, 6 and 50. Compare with a Winsorised mean. Use unequal tail trims and describe the changed target.
Translate every value by 10 and predict the trimmed result.
### Are trimmed values deleted from the dataset? No. They are excluded from this calculation but should remain in the raw record.
### Is trimming always symmetric? No. Lower and upper percentages can differ if the design justifies it.
### Does a trimmed mean estimate the ordinary mean? Not automatically. It targets a trimmed location under the stated rule.
### Is it the same as Winsorising? No. Winsorising replaces tail values and keeps the original count.
Read the official NIST trimmed-mean documentation, the NIST measures-of-location handbook and the NIST Winsorized-mean documentation. The worked calculation is original.
Explore the Mathematics Learning Hub and How Mathematics Works for related foundations.
A trimmed mean makes resistance a visible selection rule. Its strength comes from a declared percentage, a correct denominator and an audit trail that keeps every original observation available.
Extended practice laboratory
Practice case 1: Sort the data
Setup: Use the article’s trimmed-mean 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: Compute trim counts
Setup: Use the article’s trimmed-mean 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: Remove lower tail
Setup: Use the article’s trimmed-mean 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: Remove upper tail
Setup: Use the article’s trimmed-mean 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: Recalculate denominator
Setup: Use the article’s trimmed-mean 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: Move the far value
Setup: Use the article’s trimmed-mean 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: Trim 20 percent
Setup: Use the article’s trimmed-mean 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: Use unequal tails
Setup: Use the article’s trimmed-mean 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: Compare Winsorising
Setup: Use the article’s trimmed-mean 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: Compare median
Setup: Use the article’s trimmed-mean 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: Compare ordinary mean
Setup: Use the article’s trimmed-mean 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: Handle fractional counts
Setup: Use the article’s trimmed-mean 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: Handle cutoff ties
Setup: Use the article’s trimmed-mean 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: Translate all data
Setup: Use the article’s trimmed-mean 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: Scale all data
Setup: Use the article’s trimmed-mean 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: Preserve IDs
Setup: Use the article’s trimmed-mean 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: Handle missing values
Setup: Use the article’s trimmed-mean 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 prespecification
Setup: Use the article’s trimmed-mean 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: Check small n
Setup: Use the article’s trimmed-mean 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: Estimate uncertainty carefully
Setup: Use the article’s trimmed-mean 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 the raw sample
Setup: Use the article’s trimmed-mean 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: Avoid deletion claims
Setup: Use the article’s trimmed-mean 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: Name the estimand
Setup: Use the article’s trimmed-mean 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 trimmed-location conclusion
Setup: Use the article’s trimmed-mean 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.