What is a Skewness Calculator?
A skewness calculator measures how lopsided a distribution is. Paste your data to get the sample skewness G₁ (Excel’s SKEW), the population skewness g₁, a histogram and a significance test. Positive skewness means a longer right tail, negative a longer left tail, and values between −0.5 and 0.5 mean roughly symmetric data.
What the Skewness Calculator does
Paste your data to get the sample skewness G₁ (the adjusted Fisher–Pearson coefficient reported by Excel’s SKEW, Google Sheets, SPSS and SAS) and the population skewness g₁ (Excel’s SKEW.P, SciPy’s default). The skew calculator also gives the standard error of skewness, a z statistic for testing symmetry, Pearson’s median coefficient of skewness as a cross-check, a skewness graph (histogram of your data) and a moment table with every step.
Skewness matters because many methods assume roughly symmetric data. Strong skew is a reason to report the median and IQR rather than the mean and SD, to consider a log or square-root transformation, or to use a nonparametric test. Income, house prices and waiting times are typically right-skewed; scores on an easy test are typically left-skewed.
How to use the Skewness Calculator
- Type or paste at least three numbers into Data values, separated by commas, spaces or new lines. A column copied from Excel or Google Sheets works. Twenty or more values give a much more stable result.
- Read G₁ and g₁ at the top, then the verdict, which names the skew (symmetric, moderate or high, left or right) and tests whether it differs from zero.
- Look at the histogram to see the shape behind the number: the longer tail shows the direction of the skew.
- Open the moment table and the step-by-step solution to follow m₂, m₃, g₁, G₁ and the standard error.
How to calculate skewness by hand
You need only the mean and two sums. The moment table in the results lays out the same columns.
- Find the mean x̄ of the n values.
- Subtract the mean from each value to get the deviations (x − x̄).
- Square each deviation and cube each deviation, then add each column. Divide both sums by n to get m₂ and m₃.
- Divide m₃ by m₂ raised to the power 1.5. That is the population skewness g₁.
- For a sample, multiply g₁ by √[n(n − 1)] / (n − 2) to get G₁, the value Excel and SPSS report.
Skewness formula
| Symbol | Meaning |
|---|---|
| xᵢ, x̄ | data values and their mean |
| n | number of values |
| m₂, m₃ | second and third central moments (divide by n) |
| s | sample standard deviation |
G₁ is the version most statistics packages report and it is what this calculator treats as the main result. For large samples G₁ and g₁ are almost equal; for n = 12 the adjustment factor is √(12 × 11) / 10 = 1.1489.
z = G₁ / SES is a quick large-sample test of symmetry. SciPy’s skewtest uses D’Agostino’s transformation instead, which is more accurate for small n, so its z will not match exactly.
Pearson’s mode coefficient is only useful when the data have one clear mode, so the calculator reports the median version, which always exists.
Formulas follow the NIST/SEMATECH e-Handbook of Statistical Methods and Joanes and Gill (1998), “Comparing measures of sample skewness and kurtosis”. Software conventions are from the Excel SKEW documentation and SciPy’s scipy.stats.skew.
Worked example
Worked example: restaurant waiting times
Problem. Twelve customers waited 4, 2, 7, 3, 5, 12, 4, 3, 6, 4, 9, 5 minutes for a table. How skewed are the waiting times?
- 01Mean: x̄ = 64 / 12 = 5.3333; median = 4.5.
- 02Sum of squared deviations = 88.6667, so m₂ = 88.6667 / 12 = 7.3889.
- 03Sum of cubed deviations = 280.8889, so m₃ = 280.8889 / 12 = 23.4074.
- 04g₁ = 23.4074 / 7.3889^1.5 = 23.4074 / 20.0849 = 1.1654.
- 05G₁ = 1.1654 × √(12 × 11) / 10 = 1.3390. SES = √(792 / 1950) = 0.6373, so z = 1.3390 / 0.6373 = 2.1010.
Answer. The sample skewness is G₁ ≈ 1.339: the waiting times are highly right-skewed, with a long tail of slow seatings (9 and 12 minutes). Since z ≈ 2.10 > 1.96, the skew is statistically significant at the 5% level. Pearson’s median skewness agrees in direction: 3(5.3333 − 4.5) / 2.8391 = 0.8806, and the mode version gives (5.3333 − 4) / 2.8391 = 0.4697. Load this dataset into the calculator above to check every step.
How to interpret a skewness value
A widely used rule of thumb (due to Bulmer) classifies the size of the skewness. The bands are conventions, not tests; always look at a histogram and consider the sample size. A skewness of 1.5, for example, is highly right-skewed: the mean sits well above the median and a few large values form a long right tail.
| Skewness | Description | Typical shape |
|---|---|---|
| Below −1 | Highly left-skewed | Long tail of low values, e.g. scores on an easy exam |
| −1 to −0.5 | Moderately left-skewed | Noticeable left tail |
| −0.5 to 0.5 | Approximately symmetric | Tails of similar length |
| 0.5 to 1 | Moderately right-skewed | Noticeable right tail |
| Above 1 | Highly right-skewed | Long tail of high values, e.g. incomes or waiting times |
How to spot skewness in a graph
Skewness in a graph shows up as one tail stretching further than the other. The skewness diagram the calculator draws is a histogram, but a box plot or dot plot shows the same clues. The direction of the skew is the side of the long tail, not the side of the peak.
- Right-skewed (positive): the peak is on the left and the bars trail off to the right. The order is usually mode < median < mean.
- Left-skewed (negative): the peak is on the right and the bars trail off to the left. The order is usually mean < median < mode.
- Symmetric (near 0): the two halves roughly mirror each other, and the mean, median and mode are close together.
- In a box plot: the median line sits off-center in the box and one whisker is clearly longer. Draw one with the Box Plot Calculator.
Skewness test: is the skew significant?
A sample from a perfectly symmetric population will still show some skew, so a nonzero G₁ is not proof of a skewed population. The skewness test divides G₁ by its standard error, SES, and compares the result with ±1.96. If |z| is larger, the skew is significant at about the 5% level.
The standard error shrinks as the sample grows, which is why small samples need large skewness before it counts as significant.
| Sample size n | Standard error of skewness | |G₁| needed for z > 1.96 |
|---|---|---|
| 10 | 0.687 | 1.35 |
| 20 | 0.512 | 1.00 |
| 30 | 0.427 | 0.84 |
| 50 | 0.337 | 0.66 |
| 100 | 0.241 | 0.47 |
| 200 | 0.172 | 0.34 |
Skewness of common distributions
Knowing the theoretical skewness helps you judge whether your sample value is plausible for a given model.
| Distribution | Skewness | Shape |
|---|---|---|
| Normal, uniform, t (df > 3) | 0 | Symmetric |
| Exponential | 2 | Strong right skew |
| Chi-square with k df | √(8 / k) | Right skew that fades as k grows |
| Poisson with mean λ | 1 / √λ | Right skew that fades as λ grows |
| Binomial (n, p) | (1 − 2p) / √(np(1 − p)) | Right skew if p < 0.5, left if p > 0.5 |
Why software reports different skewness values
There are three common sample versions, and they differ mainly for small n. Excel SKEW, Google Sheets SKEW, SPSS, SAS and pandas’ .skew() report G₁. Excel SKEW.P, SciPy’s skew() with its default bias=True and R’s moments package report g₁. Some R packages, including e1071 by default, report b₁ = g₁((n − 1)/n)^1.5. This page shows all three so you can match your software.
Common use cases
- Checking assumptions: decide whether data are symmetric enough for a t-test or whether a transformation or nonparametric test is safer.
- Choosing summary statistics: strong skew favors the median and IQR over the mean and SD.
- Grading: a left-skewed (negatively skewed) test means most students scored high and the test was easy; a right-skewed test means it was hard.
- Finance and risk: skewness of returns shows whether large losses or large gains are more likely than a normal model suggests.
- Quality control: a skewed process distribution can signal a physical limit, such as a dimension that cannot go below zero.
Assumptions and limitations
- Skewness is very sensitive to outliers because deviations are cubed; a single extreme value can dominate it. Check the data with the fence calculator.
- Sample skewness is unreliable for small samples. With n under about 20 the standard error is large (≈0.64 for n = 12).
- A skewness of 0 does not guarantee symmetry: some asymmetric distributions have zero skewness. Look at a histogram.
- The z test assumes the population is normal under the null hypothesis and uses a normal approximation; for a formal normality check use the Shapiro–Wilk test.
Frequently asked questions
How do I calculate skewness?
What does a skew of 1.5 mean?
What does positive skewness mean?
What is an acceptable skewness value?
Why is my skewness different from Excel or SciPy?
How do I test whether skewness is significant?
Is the mean always greater than the median in right-skewed data?
How do I calculate skewness and kurtosis together?
Cite this page
Statculator. (2026, November 14). Skewness Calculator. Retrieved November 14, 2026, from https://statculator.com/descriptive-statistics/skewness-calculator/