Key Takeaways
- The tail gives skewness its direction: Right-skewed data have a longer right tail, while left-skewed data have a longer left tail.
- Skewness is not the same as volatility: Standard deviation measures spread; skewness describes asymmetry.
- The mean can be pulled toward the long tail: In many skewed data sets, the median is a more stable description of a typical observation.
- One coefficient is not the whole story: Histograms, outliers, multiple modes, and the specific skewness formula also matter.
How to Read Right and Left Skew
Symmetric distributions look roughly similar on both sides of the center. Skew appears when one tail stretches farther or carries more extreme observations than the other.
| Shape | What it means | Common effect |
|---|---|---|
| Right-skewed | Longer tail toward higher values | A few large observations can pull the mean upward |
| Left-skewed | Longer tail toward lower values | A few small observations can pull the mean downward |
| Approximately symmetric | Tails are broadly balanced | Mean and median are often closer together |
Direction refers to the tail, not to where most observations sit. Income data, for example, can be right-skewed when many observations sit in the lower and middle ranges while a small number reach very high values.
The relationship between the median and mean is useful context, but it is not a universal diagnostic rule. Multimodal data, bounds, and unusual outliers can produce more complicated patterns.
How Skewness Is Calculated
One common measure is the Fisher-Pearson coefficient of skewness, which uses the third central moment scaled by the standard deviation. Cubing deviations preserves their sign, so observations far above the mean contribute positively while observations far below it contribute negatively.
Statistical software may report an adjusted Fisher-Pearson coefficient that corrects for sample size. Other definitions, including quartile-based measures, also exist. Comparing skewness values from different tools therefore requires checking which convention each tool uses.
Why Skewness Matters in Finance and Business
Average values can hide important asymmetry. Household income, insurance losses, startup outcomes, transaction sizes, and some investment returns may contain a long tail of unusually large or small observations.
Risk analysis is one common use. Two return series can have similar averages and similar standard deviations while differing in the frequency or size of extreme outcomes on one side. Skewness adds information about that shape, although it still does not tell you the probability of every possible loss.
Business forecasting also benefits from recognizing skew. Using a mean as the “typical” order size or claim amount can be misleading when a small number of extreme observations dominate the average.
A Simple Skewness Example
Replacing the $400 purchase with a $5 purchase would create the opposite pattern: the low outlier would extend the left tail and pull the mean downward.
What Skewness Cannot Tell You
One coefficient does not show where outliers occur, whether the distribution has multiple peaks, or whether a particular probability model fits the data. NIST specifically recommends graphical tools such as histograms for seeing shape directly.
Sample size also matters. Small samples can produce unstable skewness estimates because one observation may have a large effect on the third moment.
Interpretation should therefore combine the coefficient with a plot, summary statistics, and knowledge of how the data were generated.
Frequently Asked Questions (FAQs)
What does positive skewness mean?
Positive skewness means the distribution is skewed to the right, with a longer or heavier tail toward higher values.
Is negative skewness bad?
No. The sign describes shape, not quality. Whether left skew is desirable depends on what the data represent.
Does skewness of zero mean the data are normal?
Not necessarily. A symmetric distribution can have skewness near zero without being normally distributed, and other features such as tail thickness may differ substantially.