Winsorized Mean Calculator
Find the Winsorized mean of any dataset by capping a chosen percentage of the smallest and largest values instead of deleting them — a robust average that keeps your full sample size intact. Paste your numbers, pick a Winsorizing level, and see the full step-by-step working alongside the trimmed mean, ordinary mean, and median.
Textbook-accurate method · Reviewed July 2026 · Uses the standard symmetric Winsorized-mean definition
Enter Your Data
Works with commas, spaces, tabs, or line breaks. Decimals and negatives are fine. Values are sorted automatically.
This is the percentage capped at each end (the standard symmetric convention). A 10% Winsorizing replaces the lowest 10% and highest 10% of values with the nearest value still inside that range — your sample size never shrinks.
Your Result
Winsorized Mean vs. Other Averages
What Gets Pulled In
Grey points are the originals; teal points show where they land after capping
- Kept as-is
- Original position
- Winsorized mean
What Is a Winsorized Mean?
A Winsorized mean is an average found by capping — not removing — a set percentage of the smallest and largest values in a dataset, then averaging everything as usual. Instead of throwing the extremes away, each one is reset to the nearest value still inside the “kept” range. The name comes from Charles P. Winsor, a biostatistician whose approach to handling stray measurements was picked up and popularized by the statistician John Tukey. Because every original data point is still represented in some form, the Winsorized mean keeps your full sample size, which sets it apart from methods that discard data outright.
Think of it as turning the volume down on your loudest values rather than muting them completely. A single freakishly high reading no longer gets to speak at full volume, but it still gets a seat at the table — just nudged in to match its more moderate neighbors.
How to Calculate a Winsorized Mean
The calculator above works through every step automatically, but here’s the logic behind it:
- Sort the data from smallest to largest.
- Choose a Winsorizing percentage for each end — 5%, 10%, and 20% are the most common choices.
- Work out how many values to cap on each side: multiply the count by the percentage and round to the nearest whole number. For a 10% cap on 10 values, that’s 10 × 0.10 = 1 value per side.
- Replace the capped values — set every value below the cutoff equal to the smallest value still inside the range, and every value above the cutoff equal to the largest value still inside the range.
- Average the full, adjusted dataset — that sum still gets divided by the original count, n, because no values were removed.
The Winsorized mean formula
If k is the number of values capped on each end and the sorted data is x₍₁₎, x₍₂₎, …, x₍ₙ₎, the Winsorized mean is:
Winsorized mean = ( k·x₍ₖ₊₁₎ + x₍ₖ₊₁₎ + … + x₍ₙ₋ₖ₎ + k·x₍ₙ₋ₖ₎ ) / n
In plain terms: the lowest k values are each replaced by x₍ₖ₊₁₎, the highest k values are each replaced by x₍ₙ₋ₖ₎, and the resulting n numbers are averaged in the usual way.
A worked example
Take the dataset 26, 29, 32, 33, 35, 39, 40, 41, 44, 90 — the 90 is a clear outlier pulling the average upward. The ordinary mean is 40.9. For a 10% Winsorized mean, we cap one value on each side: the lowest value, 26, is reset to 29 (the next value in), and the highest value, 90, is reset to 44 (the next value in). The adjusted dataset becomes 29, 29, 32, 33, 35, 39, 40, 41, 44, 44 — still ten numbers. Averaging those gives a Winsorized mean of 36.6, a far more representative figure than the outlier-skewed ordinary mean, while the sample size stays exactly where it started.
Winsorized Mean vs. Mean vs. Median
These three measures sit along a spectrum of how much protection they offer against extreme values:
| Measure | How it handles outliers | Best when |
|---|---|---|
| Mean | Fully exposed — every value counts as-is | Data is clean and roughly symmetric |
| Winsorized mean | Robust — extremes capped, sample size unchanged | Outliers exist but you want to keep every data point |
| Median | Most robust — only the middle value(s) matter | Heavily skewed data or many outliers |
For the full spread of a dataset in one pass, including the mean, median, and quartiles, try our Descriptive Statistics Calculator.
Winsorized Mean vs. Trimmed Mean
The Winsorized mean’s closest relative is the trimmed mean, and the two are easy to mix up. Trimming deletes the extreme values outright, shrinking the sample size — a 10% trim on 10 values leaves you with only 8 numbers to average. Winsorizing replaces those same extreme values with their nearest neighbor instead of deleting them, so all 10 values are still averaged. The practical difference shows up in the denominator: trimmed means divide by a smaller n, Winsorized means divide by the original n. The calculator above reports the trimmed mean alongside the Winsorized mean so you can see exactly how the two diverge on your own data. If you specifically need the trimmed version, our Trimmed Mean Calculator covers that method in detail.
A note on notation: γ-Winsorized vs. per-side percentage
Some textbooks describe Winsorizing using a single combined figure — for example, a “20% Winsorized mean” meaning 10% capped on each side. This calculator, in line with the more common convention used across statistics software, asks for the per-side percentage directly, so entering 10% caps 10% at the bottom and 10% at the top. If you’re cross-checking against a textbook or a statistical package, confirm whether its percentage refers to one side or both combined — that’s almost always the source of any mismatch.
Where the Winsorized Mean Is Used
- Finance: smoothing returns data and index construction so single extreme trading days don’t dominate the average.
- Survey methodology: capping extreme response weights without discarding respondents entirely.
- Quality control: stabilizing process averages when a sensor occasionally reports a spurious reading.
- Sports and performance analytics: reducing the influence of one exceptional or disastrous outing while keeping every game in the record.
- Scientific research: handling measurement noise in small samples where every data point is valuable and dropping any of them is undesirable.
Frequently Asked Questions
A Winsorized mean is an average calculated after capping a fixed percentage of the smallest and largest values in a dataset, replacing each with the nearest value still inside the kept range, then averaging the full, unchanged sample size. It’s named after biostatistician Charles P. Winsor and is used to reduce the influence of outliers without discarding any observations.
Sort the data, decide a Winsorizing percentage for each end, and multiply the number of values by that proportion to find how many to cap on each side (rounding to a whole number). Replace the lowest values with the smallest kept value and the highest values with the largest kept value, then take the ordinary average of the full, adjusted dataset — the count, n, never changes.
A trimmed mean deletes the extreme values entirely, which shrinks the sample size before averaging. A Winsorized mean instead replaces the extreme values with the nearest remaining value, keeping the sample size exactly the same. Trimming discards the tails; Winsorizing caps them in place. Both reduce the pull of outliers, but Winsorizing preserves every original data point in some form.
No. That’s the defining feature of Winsorizing versus trimming. Every value that gets capped is replaced rather than removed, so the total count, n, used in the final average is identical to your original dataset size.
Use a Winsorized mean when your data has one or more outliers that would otherwise distort the ordinary average, but you want to retain the full sample size rather than reducing it through trimming or falling back on the median. It’s especially useful in finance, survey weighting, and quality-control settings where every observation still carries some information.
5% and 10% per side are the most common defaults, with 20% used for noisier data that has more extreme values to manage. Lighter Winsorizing changes the result less but offers less protection from outliers, while heavier Winsorizing offers more protection at the cost of pulling more of the distribution toward the center. Keep enough uncapped values in the middle that the result still reflects the shape of your data.
Methodology & formulas used
This calculator uses the standard symmetric Winsorized-mean definition, reviewed July 2026. All computation runs locally in your browser; your data is never uploaded.
- Values capped per end: k = round(n × p), where p is the per-side Winsorizing proportion
- Winsorized mean: replace the lowest k values with x₍ₖ₊₁₎ and the highest k values with x₍ₙ₋ₖ₎, then take the arithmetic mean of all n adjusted values
- Trimmed mean: shown for comparison — the arithmetic mean of the n − 2k values remaining after removing k from each end
- Ordinary mean: Σxᵢ / n, shown for comparison
- Median: the middle value (or average of the two middle values), shown for comparison
The Winsorizing percentage here is applied to each end, matching the convention used across most statistical software. If 2k is greater than or equal to n, there is no unaffected middle left and the result is undefined; in that case, lower the percentage or add more data. Results are rounded for display but computed at full precision.
References
- Wikipedia. Winsorized Mean — definition and properties. en.wikipedia.org
- Wikipedia. Winsorizing — transformation of statistics to limit extreme values. en.wikipedia.org
- NIST/SEMATECH Dataplot Reference Manual. WINSORIZED MEAN subcommand. itl.nist.gov