Mathematics · Statistics · Descriptive Statistics
Skewness and Kurtosis Calculator
Calculates the skewness and excess kurtosis of a dataset to measure the asymmetry and tail heaviness of its distribution.
Calculator
Formula
g₁ is the sample skewness, G₂ is the sample excess kurtosis (kurtosis minus 3). x̄ is the sample mean, s is the sample standard deviation, n is the number of data points, and xᵢ are the individual values. Both formulas apply the bias-correction factors used by Excel, SPSS, and most statistical software. Excess kurtosis of 0 indicates a normal distribution; positive values indicate heavier tails (leptokurtic); negative values indicate lighter tails (platykurtic).
Source: Joanes, D.N. & Gill, C.A. (1998). Comparing measures of sample skewness and kurtosis. The Statistician, 47(1), 183–189.
How it works
Skewness and kurtosis are higher-order descriptive statistics that go beyond the mean and standard deviation to characterize the shape of a distribution. The mean tells you the central tendency, and the standard deviation tells you the spread, but neither reveals whether the distribution is symmetric or whether extreme values are unusually common. Skewness and kurtosis answer those questions directly, making them indispensable in exploratory data analysis, quality control, finance, and any domain where distributional assumptions matter.
The sample skewness (g₁) is computed using the bias-corrected formula adopted by Excel, SPSS, and most statistical packages. It involves summing the cubed standardized deviations — each data point minus the mean, divided by the sample standard deviation — then applying the correction factor n / [(n−1)(n−2)]. A positive skewness value indicates a right-skewed (positively skewed) distribution with a longer right tail, while a negative value indicates a left-skewed distribution. A value near zero suggests approximate symmetry. The sample excess kurtosis (G₂) uses the analogous bias-corrected fourth-moment formula. It measures how much the tails differ from a normal distribution: a value of 0 matches a normal distribution (mesokurtic), a positive value signals heavier-than-normal tails (leptokurtic), and a negative value signals lighter tails (platykurtic). Both formulas require at least 3 data points for skewness and at least 4 for kurtosis.
Practical applications span many fields. In finance, skewness and kurtosis of asset returns inform Value at Risk (VaR) models and option pricing — most financial returns exhibit negative skewness and excess kurtosis, meaning crashes are more likely than a normal model predicts. In manufacturing and quality control, engineers use these statistics to detect whether a process output distribution has drifted from normality before applying control charts. In biomedical research, kurtosis tests help determine whether parametric tests like t-tests and ANOVA are appropriate. In machine learning, feature skewness guides data transformation decisions such as log transforms or Box-Cox normalization.
Worked example
Consider the dataset: 2, 4, 4, 4, 5, 5, 7, 9 (n = 8).
Step 1 — Compute the mean: x̄ = (2 + 4 + 4 + 4 + 5 + 5 + 7 + 9) / 8 = 40 / 8 = 5.0
Step 2 — Compute the sample standard deviation: The sum of squared deviations is (2−5)² + (4−5)² + (4−5)² + (4−5)² + (5−5)² + (5−5)² + (7−5)² + (9−5)² = 9 + 1 + 1 + 1 + 0 + 0 + 4 + 16 = 32. So s = √(32 / 7) ≈ 2.1381.
Step 3 — Compute skewness: Compute each standardized value zᵢ = (xᵢ − 5) / 2.1381, then cube each. The sum of cubed standardized values is approximately −0.5338 + (−0.1040) × 3 + 0 × 2 + 0.4356 + 1.7424 ≈ 1.5242. Applying the correction: g₁ = (8 / (7 × 6)) × 1.5242 ≈ 0.2902. This positive value confirms a slight right skew.
Step 4 — Compute excess kurtosis: The sum of fourth-power standardized values is approximately 0.2851 + (0.0108) × 3 + 0 × 2 + 0.1898 + 3.0375 ≈ 3.5749. Applying the correction factor (8 × 9) / (7 × 6 × 5) × 3.5749 − 3 × 49 / (6 × 5) ≈ −0.5600. This slightly negative excess kurtosis indicates tails marginally lighter than a normal distribution.
Limitations & notes
Both statistics require a minimum sample size — at least 3 data points for skewness and 4 for kurtosis — but are highly unreliable for small samples (n < 30). For small n, standard errors of skewness and kurtosis are large, so apparent non-normality may be sampling noise rather than a true distributional feature. The formulas implemented here are the bias-corrected sample estimators (the same as Excel's SKEW and KURT functions and SPSS defaults); they differ from the simpler population moment formulas that some textbooks present. This calculator assumes all input values are numeric and independent; it does not handle time-series autocorrelation or weighted data. Extreme outliers have a disproportionate effect on both statistics because they involve third and fourth powers of deviations. For heavy-tailed distributions such as Cauchy or Pareto, the theoretical skewness and kurtosis may not even be defined (infinite), so sample estimates can be arbitrarily large and unstable. Always interpret skewness and kurtosis alongside histograms, Q-Q plots, and formal normality tests such as Shapiro-Wilk or Jarque-Bera for a complete picture.
Frequently asked questions
What is a good skewness value for a normal distribution?
A perfectly normal distribution has a skewness of exactly 0. In practice, values between −0.5 and +0.5 are generally considered approximately symmetric. Values between ±0.5 and ±1.0 indicate moderate skewness, and values beyond ±1.0 indicate high skewness that may warrant data transformation before applying parametric statistical methods.
What does excess kurtosis of 0 mean?
An excess kurtosis of 0 means the distribution has the same tail weight as a normal distribution (mesokurtic). This calculator reports excess kurtosis, which is the raw fourth-moment kurtosis minus 3. Some sources report raw kurtosis without subtracting 3, so a normal distribution would show 3 rather than 0 — always check which convention a source uses.
What is the difference between leptokurtic and platykurtic?
A leptokurtic distribution (excess kurtosis > 0) has heavier tails and a sharper central peak than a normal distribution, meaning extreme values occur more frequently. A platykurtic distribution (excess kurtosis < 0) has lighter tails and a flatter peak, with fewer extreme observations. Financial return series are typically leptokurtic, which is why fat-tail risk models are preferred over simple normal models.
Why does skewness matter in finance and risk management?
Negative skewness in asset returns means large losses are more probable than large gains of equal magnitude — a critical concern for portfolio risk management. Standard deviation-based risk measures like variance ignore this asymmetry. Incorporating skewness and kurtosis into models (e.g., Cornish-Fisher VaR adjustments) produces more realistic tail risk estimates and helps prevent dangerous underestimation of extreme loss events.
How many data points do I need for reliable skewness and kurtosis estimates?
As a practical guideline, at least 30 data points are needed before skewness estimates become reasonably stable, and at least 100 data points are recommended for kurtosis, which is sensitive to outliers due to its reliance on fourth powers of deviations. For formal normality testing, the Jarque-Bera test uses both statistics jointly and is best applied with n ≥ 50 for reliable p-values.
Last updated: 2025-01-15 · Formula verified against primary sources.