What Is Skewness and Kurtosis
Skewness measures the asymmetry of a distribution: positive skew means a long right-hand tail, negative skew a long left-hand tail. Kurtosis measures the weight of the tails — how much of the data sits far from the centre. A normal distribution has a skewness of 0 and an excess kurtosis of 0, so the two numbers together show how your data differ from a normal curve.
If you have ever looked at a histogram and wondered whether it leans to the left or the right, or whether its tails are fatter than a bell curve’s, you were already thinking about skewness and kurtosis. These two shape statistics appear in almost every normality check, and the Skewness and Kurtosis Calculator reports them alongside the mean, median and standard deviation. This guide explains what each one means, shows the formulas behind them, and works through three complete examples so you can read the numbers with confidence.
What Is Skewness?
Skewness is the third standardised moment of a distribution, and it measures asymmetry — how lopsided the data are around the centre. A distribution with a long tail stretching to the right is positively skewed (right-skewed); one with a long tail stretching to the left is negatively skewed (left-skewed). The direction is named after the tail, not the bulk of the data: in a right-skewed distribution most values sit on the left, and a few large values pull the tail out to the right.
Skewness is closely tied to the ordering of the mean, median and mode. In a right-skewed distribution the mean is pulled towards the long tail, so mean > median > mode. In a left-skewed distribution the order reverses: mean < median < mode. In a perfectly symmetric distribution all three coincide. That is why comparing the mean vs median vs mode is often the fastest way to spot skew — if the mean sits well above the median, the data are probably right-skewed.
Skewness is a pure number with no units, because each observation is first standardised by dividing by the standard deviation. Its sign gives the direction of the skew and its size gives the strength. A value of 0 means the distribution is symmetric. But skewness says nothing about the spread of the data and nothing about the weight of the tails — that is the job of kurtosis. It is also sensitive to sample size: with only a handful of observations, a single extreme value can move the coefficient a long way, so always check the standard error of skewness too; it is roughly the square root of 6 divided by the sample size.
What Is Kurtosis?
Kurtosis is the fourth standardised moment, and it measures the weight of a distribution’s tails — how much of the data lies far from the centre, relative to a normal curve. It is a statement about outliers and extreme values, not about how tall or flat the middle of the curve looks. The normal distribution is the reference point: its raw (Pearson) kurtosis is 3, so statisticians usually report excess kurtosis, which is raw kurtosis minus 3. A normal distribution therefore has an excess kurtosis of 0, just as the normal distribution has a skewness of 0.
Three words describe the three cases. A mesokurtic distribution has an excess kurtosis near 0 and tails like the normal curve’s. A platykurtic distribution has negative excess kurtosis and light tails, with fewer extreme values than a normal distribution. A leptokurtic distribution has positive excess kurtosis and heavy tails, with more extreme values than a normal distribution. The formal definitions are set out in the NIST/SEMATECH e-Handbook of Statistical Methods.
The peakedness myth. Many guides still describe kurtosis as “peakedness” — the idea that high kurtosis means a tall, sharp peak. That is wrong, and it is worth correcting explicitly: kurtosis measures the tails, not the peak. The reason is the fourth power. Kurtosis raises each standardised deviation to the fourth power, so an observation three standard deviations from the mean contributes 81 times as much as one a single standard deviation away. Distant values dominate the calculation, while the height of the centre barely matters. A distribution can have a flat top and still have high kurtosis if it has heavy tails, and a sharply peaked distribution can have low kurtosis if its tails are light. Remember this: kurtosis is a measure of tailedness and outlier character, not of peak height.
What Is the Difference Between Skewness and Kurtosis?
Skewness and kurtosis are two different shape statistics that answer two different questions. Skewness asks whether a distribution leans to one side: it is the third standardised moment, it has a direction (positive or negative), and it is driven by the asymmetry of the tails. Kurtosis asks how heavy the tails are: it is the fourth standardised moment, it has no direction, and it is driven by how many extreme values the data contain. A distribution can be strongly skewed with normal-looking tails, or perfectly symmetric with very heavy tails, which is why the two numbers are reported together. Both use the normal distribution as their reference: a normal curve has a skewness of 0 and an excess kurtosis of 0.
| Feature | Skewness | Kurtosis |
|---|---|---|
| Measures | Asymmetry (left-right balance) | Tail weight (how many extreme values) |
| Moment | Third standardised moment | Fourth standardised moment |
| Direction | Positive (right tail) or negative (left tail) | None — compares tails with the normal curve |
| Normal value | 0 | Excess kurtosis 0 (raw kurtosis 3) |
Is Kurtosis About Peakedness or Tails?
Tails. Kurtosis is a measure of tail weight, not of peak height, and the “peakedness” reading is a long-standing mistake. Because kurtosis raises deviations to the fourth power, observations far from the mean dominate the result, while the centre of the distribution contributes comparatively little. Two distributions can have the same peak and very different kurtosis if one has heavier tails, and two distributions with the same kurtosis can look very different in the middle. When you see a high kurtosis, read it as “this data set produces more extreme values than a normal distribution would” — never as “this curve is pointier”.
Skewness is read from the longest tail; kurtosis from how much of the data sits far from the centre — not from the height of the peak.
The Formulas for Skewness and Kurtosis
All skewness and kurtosis measures start from the central moments of the data. The second central moment, m2, is the variance; the third, m3, captures asymmetry; the fourth, m4, captures tail behaviour. Dividing by powers of the standard deviation makes each measure unit-free, so you can compare shapes across data sets measured in different units.
Moment skewness, g1. The population (or “moment”) coefficient is g1 = m3 / m23/2 — the third standardised moment, the average of the cubed standardised deviations. For a perfectly symmetric distribution m3 is 0, so g1 is 0. It is the textbook formula and the one the worked examples below report first.
Adjusted Fisher–Pearson skewness, G1. The version most software reports for a sample is G1 = g1 × √[n(n − 1)] / (n − 2). The correction removes some of the bias of the moment formula in small samples, and it is what Excel’s SKEW function and our statistics calculator report. For large samples G1 and g1 are nearly identical; for small samples G1 is always larger in size, which is why the examples below show both.
Pearson’s second coefficient. Pearson’s quick measure needs only the mean, median and standard deviation: 3 × (mean − median) / SD. It is handy when you do not have the raw data, but it uses only two summary numbers, so it can disagree with g1 for unusual distributions. It is also unreliable when the mean and median are nearly equal, because the small difference it divides is then mostly noise.
Bowley’s quartile skewness. Bowley’s measure uses the quartiles instead of the mean: (Q3 + Q1 − 2 × median) / (Q3 − Q1). It always lies between −1 and 1, it is unaffected by extreme values, and it is a useful cross-check when outliers distort the moment measures.
Kurtosis. Pearson’s kurtosis is b2 = m4 / m22, the fourth standardised moment; the normal distribution has b2 = 3. Excess kurtosis is b2 − 3, so the normal distribution has an excess kurtosis of 0.
Excel versus moment kurtosis. Excel’s KURT function reports sample excess kurtosis (often written G2), which applies a small-sample correction like the one in G1. The moment excess kurtosis reported by the site’s calculator is the plain b2 − 3. The two numbers describe the same data but are computed differently, so they are not equal: with 10 observations, the symmetric example below gives an Excel excess kurtosis of −1.200 and a moment excess kurtosis of −1.224. Use the Excel figure when you are matching Excel output, and the moment figure when you are matching a textbook formula.
How to Read the Numbers
Use these rules of thumb as a first reading, not as a verdict. They are the most commonly quoted cut-offs, and they work well for screening data before a formal normality test.
| Skewness (g1) | Reading |
|---|---|
| −0.5 to 0.5 | Approximately symmetric |
| 0.5 to 1 (or −1 to −0.5) | Moderately skewed |
| Above 1 (or below −1) | Highly skewed |
For excess kurtosis, a value near 0 means normal-like tails; a value above about 1 means heavier tails and more outliers than the normal curve; a value below about −1 means lighter tails and fewer outliers. With small samples, treat both numbers with caution: the standard errors are wide, so a skewness of 0.8 from only 10 observations is weak evidence. And neither coefficient alone proves normality or non-normality — pair them with a histogram or a Q–Q plot before you conclude anything.
What Does a Skewness of 0 Mean?
A skewness of 0 means the distribution is symmetric in the sense the third moment measures: the left and right sides balance, so the mean, median and mode coincide in a unimodal distribution. It does not mean the data are normal. A symmetric distribution can still be bimodal, flat-topped or heavy-tailed — the symmetric worked example below has a skewness of exactly 0 and an excess kurtosis of −1.2. A skewness of 0 rules out asymmetry; it says nothing about the tails.
Worked Example 1: Symmetric Data
The data are [2, 4, 6, 8, 10, 12, 14, 16, 18, 20]: ten evenly spaced values, as balanced as a data set can be. Every measure agrees:
| Measure | Value |
|---|---|
| Mean | 11.000 |
| Median | 11.000 |
| Standard deviation | 6.055 |
| Moment skewness (g1) | 0.000 |
| Adjusted skewness (G1) | 0.000 |
| Pearson’s coefficient | 0.000 |
| Excel excess kurtosis | −1.200 |
| Moment excess kurtosis | −1.224 |
| Standard error of skewness | 0.687 |
| Verdict | Nearly symmetric |
All three skewness measures are exactly 0, and the mean equals the median — the signature of perfect symmetry. Note the kurtosis, though: the excess kurtosis is negative, about −1.2, because evenly spaced values produce light tails. A skewness of 0 does not make the data normal — this distribution is flatter than a bell curve. Try the symmetric data in the calculator to see every step of the working.
Worked Example 2: Right-Skewed Data
The data are [1, 2, 2, 3, 3, 3, 4, 5, 9]: nine values clustered at the low end, with one large value pulling the tail to the right.
| Measure | Value |
|---|---|
| Mean | 3.556 |
| Median | 3.000 |
| Standard deviation | 2.351 |
| Moment skewness (g1) | 1.426 |
| Adjusted skewness (G1) | 1.728 |
| Pearson’s coefficient | 0.709 |
| Excel excess kurtosis | 3.648 |
| Moment excess kurtosis | 1.315 |
| Standard error of skewness | 0.717 |
| Verdict | Highly right-skewed |
Every measure points the same way. The moment skewness is 1.426, above the “highly skewed” cut-off of 1, and the adjusted G1 is larger still at 1.728. The mean (3.556) sits above the median (3), exactly as the mean > median > mode ordering predicts, and the single value of 9 drives the excess kurtosis up to 3.648 in Excel — a heavy right tail. Run this right-skewed data through the calculator to see how the 9 changes each figure.
Worked Example 3: Left-Skewed Data
The data are [1, 5, 6, 7, 8, 8, 9, 9, 10]: nine values bunched at the high end, with one small value pulling the tail to the left.
| Measure | Value |
|---|---|
| Mean | 7.000 |
| Median | 8.000 |
| Standard deviation | 2.739 |
| Moment skewness (g1) | −1.162 |
| Adjusted skewness (G1) | −1.408 |
| Pearson’s coefficient | −1.095 |
| Excel excess kurtosis | 2.229 |
| Moment excess kurtosis | 0.570 |
| Standard error of skewness | 0.717 |
| Verdict | Highly left-skewed |
All three skewness measures are below −1, so the data are highly left-skewed, and the mean (7) sits below the median (8), as the ordering predicts. The excess kurtosis is positive (2.229 in Excel) but smaller than in the right-skewed example — the left tail is heavy, but less extreme. Check the left-skewed data in the calculator and compare the two tails.
Symmetric vs Right-Skewed vs Left-Skewed: Side by Side
| Dataset | n | Mean | Median | SD | g1 | G1 | Pearson | Excel excess kurt. | Moment excess kurt. | Verdict |
|---|---|---|---|---|---|---|---|---|---|---|
| Symmetric | 10 | 11.000 | 11.000 | 6.055 | 0.000 | 0.000 | 0.000 | −1.200 | −1.224 | Nearly symmetric |
| Right-skewed | 9 | 3.556 | 3.000 | 2.351 | 1.426 | 1.728 | 0.709 | 3.648 | 1.315 | Highly right-skewed |
| Left-skewed | 9 | 7.000 | 8.000 | 2.739 | −1.162 | −1.408 | −1.095 | 2.229 | 0.570 | Highly left-skewed |
Two patterns stand out. First, the adjusted G1 is always further from 0 than the moment g1, and the gap grows as the sample gets smaller — with only nine observations the adjustment matters. Second, the Excel excess kurtosis is always further from 0 than the moment value, for the same reason: both are small-sample corrections, and they push the numbers outward. The verdicts agree across all three skewness measures in every example, which is what you want to see before trusting a reading.
When Skewness and Kurtosis Matter
Skewness and kurtosis earn their keep in four places. First, normality checks: t-tests, ANOVA and linear regression all assume roughly normal data or residuals, and a skewness above 1 or an excess kurtosis far from 0 is the first sign that the assumption is shaky. Second, choosing a transformation: a right-skewed variable often behaves better after a log or square-root transformation, and the skewness coefficient tells you whether the transformation worked. Third, outlier warnings: kurtosis is extremely sensitive to extreme values, because it raises deviations to the fourth power — a single wild observation can dominate the figure, so always look at the data behind a large kurtosis. And remember the limit of both numbers: neither one alone proves normality. Pair them with a histogram, a Q–Q plot or a formal test such as Shapiro–Wilk before drawing conclusions.
Common Mistakes to Avoid
- Calling kurtosis “peakedness”. Kurtosis measures tail weight, not peak height — a flat-topped distribution can have high kurtosis if its tails are heavy.
- Treating a skewness of 0 as proof of normality. Symmetric data can be bimodal, flat or heavy-tailed; the symmetric example above has a skewness of 0 and an excess kurtosis of −1.2.
- Confusing raw kurtosis 3 with excess kurtosis 0. The normal distribution has a raw kurtosis of 3 and an excess kurtosis of 0; check which figure your software reports.
- Judging skew from the mean alone. A mean far from the median suggests skew, but you need the median (or the coefficient) to be sure — a single outlier can move the mean without skewing most of the data.
- Using Pearson’s coefficient when the mean and median are nearly equal. The small difference it divides is then mostly noise, so a tiny change in the data can flip the sign of the result.
How to Calculate Skewness and Kurtosis in Excel and on a TI-84
In Excel, enter your data in a column and use =SKEW(range) for the adjusted Fisher–Pearson skewness (G1), =SKEW.P(range) for the moment skewness (g1), and =KURT(range) for the sample excess kurtosis (G2). Note that KURT gives you the Excel figure, not the moment excess kurtosis — the symmetric example gives −1.200 in Excel and −1.224 as a moment calculation. A TI-84 has no built-in skewness or kurtosis function, so the quick route is to enter the data in a list, compute the mean and standard deviation with 1-Var Stats, and then apply the moment formulas — or skip the arithmetic and use the site’s skewness and kurtosis calculator, which reports all three skewness measures and both kurtosis figures for any data set.
Frequently Asked Questions
Tap a question to see the answer.
What is skewness and kurtosis?
Skewness measures the asymmetry of a distribution, and kurtosis measures the weight of its tails. A positively skewed distribution has a long right-hand tail; a negatively skewed one tails off to the left. Kurtosis tells you how many extreme values the data contain compared with a normal curve. A normal distribution has a skewness of 0 and an excess kurtosis of 0, so the two numbers together show how your data differ from normal.
What is a good skewness value?
A skewness between −0.5 and 0.5 is usually read as approximately symmetric and is fine for most purposes. Values between 0.5 and 1 (or −1 and −0.5) are moderately skewed, and values beyond 1 or below −1 are highly skewed. Whether a value is “good” depends on the analysis: many statistical tests tolerate mild skew, but a highly skewed variable may need a transformation before you use it.
What does a kurtosis of 3 mean?
A raw (Pearson) kurtosis of 3 means the distribution has the same tail weight as a normal distribution — that is the normal reference value. Its excess kurtosis is therefore 0. If a report says “kurtosis is 3” without saying which kind, check whether it means raw or excess kurtosis: an excess kurtosis of 3 would mean heavy tails, while a raw kurtosis of 3 means normal-length tails.
Is positive kurtosis good or bad?
Neither on its own — it depends on the context. Positive excess kurtosis means heavier tails and more extreme values than a normal distribution. That is a problem for tests that assume normality, but for risk analysis it can be the interesting part of the data: financial returns, for example, often have positive kurtosis because large moves happen more often than a normal curve would suggest.
Can skewness be zero for non-normal data?
Yes. A skewness of 0 only rules out asymmetry; it says nothing about the tails or the shape of the centre. A symmetric distribution can be bimodal, flat-topped or heavy-tailed. The symmetric worked example in this guide has a skewness of exactly 0 and an excess kurtosis of −1.2 — clearly not a normal curve. Always check kurtosis, and a histogram, alongside skewness.
How do you calculate skewness by hand?
First work out the mean and the standard deviation. Then take each observation, subtract the mean, divide by the standard deviation, and cube the result. Average those cubed values and you have the moment skewness g1. For the adjusted figure G1, multiply g1 by the square root of n(n − 1) divided by (n − 2). With more than a few values the arithmetic is tedious, so most people use a spreadsheet or a calculator.
Why does Excel give a different kurtosis from the moment formula?
Because Excel reports sample excess kurtosis (G2), which applies a small-sample correction, while the moment formula reports the plain b2 − 3. The two numbers describe the same data, so both are “correct” — they are simply computed differently, and the gap shrinks as the sample grows. For the symmetric example in this guide, Excel gives −1.200 and the moment formula gives −1.224.
When should you worry about skewness and kurtosis?
Worry when the numbers are large and your analysis assumes normality. Skewness beyond 1 or below −1, or an excess kurtosis far from 0 (say beyond 1 or below −1), is a signal to inspect the data — look for outliers, consider a transformation, or choose a method that does not assume normality. With small samples, though, treat the numbers cautiously: their standard errors are wide, so a large value from ten observations is only weak evidence.
Conclusion
Skewness and kurtosis are the two numbers that tell you the shape of your data. Skewness measures asymmetry — the direction and length of the longest tail — and kurtosis measures the weight of the tails, the share of the data that sits far from the centre. A normal distribution has a skewness of 0 and an excess kurtosis of 0, so both numbers are read against that reference. Check them side by side, remember that kurtosis is about tails rather than peak height, and never let a skewness of 0 alone convince you the data are normal. The three worked examples show how the numbers behave in practice; run your own data through the calculator and you will be reading shapes, not guessing at them.
Check Tools
Skewness and Kurtosis Calculator
Get all three skewness measures and both kurtosis figures for any data set, with the verdict and standard errors.
Normal Distribution Calculator
Work out probabilities and percentiles for the normal curve that skewness and kurtosis are measured against.
Standard Deviation Calculator
Find the standard deviation that standardises each observation in the skewness and kurtosis formulas.
Explore All Statistical Tools
Browse the full set of calculators and visualisers on the site.