What is a Kurtosis Calculator?
A kurtosis calculator measures how heavy the tails of a distribution are, that is how prone the data are to extreme values. To calculate kurtosis, divide the fourth central moment by the squared variance (m₄ / m₂²), then subtract 3 to get excess kurtosis, which is 0 for a normal distribution. Positive excess kurtosis (leptokurtic) means heavier tails; negative (platykurtic) means lighter tails.
What the Kurtosis Calculator does
Paste your data to get the sample excess kurtosis G₂ (Excel’s KURT, SPSS, SAS) and the population excess kurtosis g₂ (SciPy’s default), plus the raw Pearson kurtosis, the standard error of kurtosis, a z statistic, a histogram and a table of fourth-power deviations. Calculating kurtosis by hand takes a column of fourth powers and four formulas; the step-by-step solution shows every one with your numbers, so you can check homework or software output line by line.
The sample skewness is shown beside the kurtosis, so the page also works as a skewness and kurtosis calculator for a quick shape check. Kurtosis tells you about outlier-proneness. High kurtosis warns that rare, extreme values are more common than a normal model predicts, which matters for risk models, control limits and any method that assumes normal errors. Despite a common textbook description, kurtosis says little about how “peaked” the center is; it is dominated by the tails.
How to use the Kurtosis Calculator
- Type or paste at least four numbers into Data values, separated by commas, spaces or line breaks. A column copied from Excel or Google Sheets works. Thirty or more values give a much more stable estimate.
- Read G₂ and g₂ at the top. Remember they are excess kurtosis: 0 is the normal benchmark.
- Check the verdict for leptokurtic or platykurtic wording and the z test against normal tails.
- Open the moment table and steps to see m₂, m₄, the raw kurtosis and the small-sample adjustment.
How to calculate kurtosis by hand
To calculate kurtosis, compare the average fourth-power deviation with the squared variance. Five steps take you from raw data to the sample excess kurtosis that Excel and SPSS report:
- Find the mean x̄ of the n values.
- Square each deviation (x − x̄) and raise it to the fourth power, then add each column.
- Divide both sums by n to get m₂ and m₄, the second and fourth central moments.
- Compute g₂ = m₄ / m₂² − 3. This is the population excess kurtosis.
- Apply the small-sample correction G₂ = [(n + 1)g₂ + 6](n − 1) / [(n − 2)(n − 3)] to get the sample excess kurtosis.
Kurtosis formula
| Symbol | Meaning |
|---|---|
| m₂, m₄ | second and fourth central moments (divide by n) |
| n | number of values |
| SES | standard error of skewness, √[6n(n − 1) / ((n − 2)(n + 1)(n + 3))] |
| 3 | raw kurtosis of every normal distribution |
Raw kurtosis = excess kurtosis + 3. If a source reports kurtosis near 3 for normal-looking data it is using raw kurtosis; if it reports values near 0 it is using excess kurtosis.
G₂ matches SciPy’s kurtosis(x, bias=False) and Excel’s KURT; g₂ matches SciPy’s default kurtosis(x).
Formulas follow the NIST/SEMATECH e-Handbook of Statistical Methods and Joanes and Gill (1998), “Comparing measures of sample skewness and kurtosis”. The tails-not-peaks reading follows Westfall (2014), “Kurtosis as Peakedness, 1905–2014. R.I.P.”.
Worked example
Worked example: machine cycle times
Problem. A machine’s cycle time is recorded 13 times: 52, 48, 53, 51, 64, 52, 50, 54, 52, 40, 53, 55, 51 seconds. Most cycles are tightly grouped, but two are unusual. Find the kurtosis.
- 01Mean: x̄ = 675 / 13 = 51.9231 seconds.
- 02Σ(x − x̄)² = 324.9231, so m₂ = 324.9231 / 13 = 24.9941.
- 03Σ(x − x̄)⁴ = 41,845.1525, so m₄ = 41,845.1525 / 13 = 3,218.8579.
- 04Raw kurtosis b₂ = 3,218.8579 / 24.9941² = 5.1526, so g₂ = 5.1526 − 3 = 2.1526.
- 05G₂ = (14 × 2.1526 + 6) × 12 / (11 × 10) = 3.9422. SEK = 1.1909, so z = 3.9422 / 1.1909 = 3.3103.
Answer. The sample excess kurtosis is G₂ ≈ 3.94: the cycle times are strongly leptokurtic, with far heavier tails than a normal distribution would have, and z ≈ 3.31 makes this statistically significant. The skewness is almost zero (0.0266), so the extremes occur on both sides: the 40- and 64-second cycles.
How to interpret kurtosis
Read the sign first, then the size, then the standard error. The bands below are the ones this calculator uses for its verdict. They are rules of thumb for G₂, not formal cut-offs.
| Sample excess kurtosis G₂ | Shape | What it means for your data |
|---|---|---|
| Below −1 | Platykurtic | Light tails; values spread evenly with few extremes, like a uniform distribution |
| −1 to −0.5 | Slightly platykurtic | Somewhat fewer extreme values than normal data |
| −0.5 to 0.5 | Close to mesokurtic | Tail weight similar to a normal distribution |
| 0.5 to 1 | Slightly leptokurtic | Somewhat more extreme values than normal data |
| Above 1 | Leptokurtic | Heavy tails; outliers are more likely than a normal model predicts |
Is the kurtosis significant?
A G₂ of 1.5 means little from 12 values and a lot from 500. Divide G₂ by its standard error, SEK: if |z| = |G₂ / SEK| is above 1.96, the tail weight differs from normal at about the 5% level. SEK shrinks roughly like √(24 / n), so it is about 1.2 at n = 13, 0.83 at n = 30 and 0.22 at n = 500. With small samples, look at which values drive Σ(x − x̄)⁴; the calculator lists them when a few values dominate.
Skewness vs kurtosis
Skewness and kurtosis are the third and fourth standardized moments, and they are usually reported together to describe the shape of a distribution. To calculate kurtosis and skewness from the same data, find the mean and the central moments once, then use m₃ / m₂^1.5 for skewness and m₄ / m₂² − 3 for excess kurtosis. This page shows G₁ beside G₂; the Skewness Calculator gives the full skewness working, and the Descriptive Statistics Calculator reports both with the other summary statistics.
| Skewness | Kurtosis | |
|---|---|---|
| What it measures | Asymmetry: is one tail longer? | Tail weight: how prone to extreme values? |
| Moment | Third, (x − x̄)³ | Fourth, (x − x̄)⁴ |
| Normal distribution | 0 | 0 (excess) or 3 (raw) |
| Sign | Positive = longer right tail | Positive = heavier tails than normal |
| Possible range | Any real number | Excess ≥ −2, no upper limit |
| Excel function | SKEW | KURT |
Kurtosis of common distributions
These benchmark values help you place a result. Excess kurtosis can never be below −2, but it has no upper limit.
| Distribution | Excess kurtosis | Raw kurtosis | Description |
|---|---|---|---|
| Two equally likely values (coin flip) | −2 | 1 | Lowest possible; no tails at all |
| Uniform | −1.2 | 1.8 | Platykurtic |
| Normal | 0 | 3 | Mesokurtic benchmark |
| Logistic | 1.2 | 4.2 | Mildly leptokurtic |
| Laplace (double exponential) | 3 | 6 | Leptokurtic |
| Exponential | 6 | 9 | Leptokurtic and skewed |
| Student’s t with 5 df | 6 | 9 | Heavy tails |
Kurtosis measures tails, not peakedness
Many textbooks describe leptokurtic distributions as “sharply peaked” and platykurtic ones as “flat-topped”. That is misleading. Because each deviation is raised to the fourth power, values near the mean contribute almost nothing to m₄; the result is driven by the most extreme observations. Distributions can have a flat top and high kurtosis, or a sharp peak and low kurtosis.
A practical reading is: high kurtosis means a larger share of the variance comes from rare, extreme values. In the worked example the two outlying cycles (40 and 64 seconds) account for 99% of Σ(x − x̄)⁴.
Common use cases
- Finance and risk: daily returns typically show positive excess kurtosis (fat tails), so normal-based risk models understate the chance of large losses.
- Checking normality: together with skewness, kurtosis shows how far a sample departs from the normal shape assumed by t-tests and regression.
- Process monitoring: heavy-tailed measurement errors mean more points beyond control limits than a normal model predicts.
- Data screening: an unexpectedly large kurtosis often points to outliers or a mixture of two populations.
Assumptions and limitations
- Kurtosis needs large samples. With n under about 30 the standard error is large, and one value can swing the estimate.
- It is extremely sensitive to outliers. Check flagged values with the Lower and Upper Fence Calculator before interpreting it.
- Different software reports raw kurtosis, population excess kurtosis or sample excess kurtosis. Compare like with like.
- The z test uses a normal approximation that is poor for small samples. For a formal normality test, use the Shapiro–Wilk test or D’Agostino’s omnibus test.
- The calculator works on raw values. For a grouped frequency table, enter each class midpoint repeated by its frequency; the result is an approximation because values within a class are treated as equal.
Frequently asked questions
How do I calculate kurtosis?
How do I calculate kurtosis and skewness?
Why do we calculate kurtosis?
What is the difference between kurtosis and excess kurtosis?
What does high kurtosis mean?
What is an acceptable kurtosis for normality?
Why does Excel’s KURT give a different value from SciPy?
Can kurtosis be negative?
What do leptokurtic, mesokurtic and platykurtic mean?
Cite this page
Statculator. (2026, November 14). Kurtosis Calculator. Retrieved November 14, 2026, from https://statculator.com/descriptive-statistics/kurtosis-calculator/