Calculate skewness, kurtosis, and shape metrics from data using Fisher, Pearson, Bowley, and Kelly methods. Includes skewness gauge, method comparison, and significance testing.
The skewness calculator measures the asymmetry of a data distribution using several common formulas, including Fisher, Pearson, Bowley, and Kelly skewness. It also reports kurtosis so you can review tail heaviness alongside left-right imbalance.
Skewness helps answer whether the data has a longer tail on one side and whether the mean is being pulled away from the median. Positive skew suggests a longer right tail, negative skew suggests a longer left tail, and values near zero suggest a roughly symmetric distribution.
Because different skewness formulas respond differently to outliers and sample size, this page puts them side by side instead of pretending one coefficient tells the whole story.
Skewness matters whenever symmetry is an assumption or when unusually long tails can change how you summarize a dataset. Looking at more than one skewness measure helps separate a genuinely asymmetric distribution from one that only looks skewed because of a small sample or a few extreme points.
With kurtosis, significance testing, and several asymmetry measures in one view, the calculator is more useful than a single skewness coefficient pasted into a report without context.
Fisher skewness g₁ = m₃/s³ where m₃ = (1/n)Σ(xᵢ−x̄)³ and s = √(m₂). Adjusted sample skewness G₁ = [n/((n−1)(n−2))]Σ((xᵢ−x̄)/s)³. Pearson's 2nd: Sk₂ = 3(mean−median)/s. Bowley: Sk_B = (Q₁+Q₃−2×Median)/(Q₃−Q₁). Standard error SES = √(6n(n−1)/((n−2)(n+1)(n+3))).
Result: G₁ = −0.0916 (approximately symmetric)
With n=20 exam scores, the adjusted sample skewness G₁ ≈ −0.09 indicates near-symmetry. The Z-score = −0.18 is well within ±1.96, so the skewness is not statistically significant. Pearson's 2nd coefficient (−0.11) agrees. The distribution of these scores is roughly symmetric.
No single skewness coefficient captures all aspects of asymmetry. Fisher's g₁ is the most common (used by Excel, R, Python), but it's sensitive to outliers. Pearson's formula is intuitive but assumes unimodality. Bowley and Kelly use quantiles and are robust — but they only see the middle of the distribution. Comparing multiple methods helps you understand which aspects of asymmetry are real versus outlier-driven.
Many statistical tests (t-tests, ANOVA, regression) assume normally distributed data. Significant skewness violates this assumption. Solutions include: log-transforming right-skewed data, using the Box-Cox transformation, applying non-parametric alternatives, or using robust methods. As a rule of thumb, |skewness| > 1 is a red flag for methods assuming normality.
While skewness measures left-right asymmetry, kurtosis measures tail heaviness. The normal distribution has kurtosis = 3 (excess = 0). Financial returns typically show excess kurtosis of 5-10, meaning extreme values happen far more often than a normal model predicts. Always report skewness and kurtosis together for a complete shape story.
Last updated:
Positive (right) skewness means the right tail of the distribution is longer or fatter than the left. Most data points cluster on the left, with some extreme high values pulling the mean above the median. Common examples: income, house prices, and wait times.
Negative (left) skewness means the left tail is longer. Most values cluster on the right, with some extreme low values pulling the mean below the median. Common examples: scores on an easy exam, age at retirement, and failure times of well-made products.
Fisher skewness (g₁/G₁) uses the third standardized moment — it considers exact deviations of each data point. Pearson's second coefficient uses 3(mean−median)/s — a simpler approximation. They usually agree in sign but can differ in magnitude. Fisher's is more precise; Pearson's is more intuitive.
Bowley (quartile) skewness = (Q₁+Q₃−2×Median)/(Q₃−Q₁) measures asymmetry using quartiles. It's bounded between −1 and +1, is resistant to outliers, and captures skewness in the middle 50% of the data. It ignores the tails entirely, which makes it robust but less sensitive than moment-based measures.
Kurtosis measures the "tailedness" of a distribution. Excess kurtosis = kurtosis − 3 (the normal distribution has kurtosis = 3). High excess kurtosis (leptokurtic) means heavier tails and more outliers. Together, skewness and kurtosis fully characterize the "shape" of a distribution beyond its center and spread.
Divide skewness by its standard error (SES) to get a Z-score. If |Z| > 1.96, the skewness is significant at the 5% level. SES depends on sample size: for small samples (n < 30), even moderate skewness may not be significant. For large samples (n > 300), even tiny skewness becomes "significant" but may not be practically meaningful.