Skew (or skewness) is a measure of how lopsided a set of data is. When data is perfectly balanced, like a classic bell curve, the skewness value is zero. When the data bunches up on one side and stretches out with a long tail on the other, that’s skew. It shows up constantly in everyday data, from household incomes to test scores, and understanding it helps you pick the right way to summarize and analyze numbers.
How Skew Works
Picture a histogram of your data. If the shape is symmetrical, with roughly equal amounts of data on both sides of the center, the skewness is zero or very close to it. But most real-world data isn’t perfectly symmetrical. Skewness puts a number on that asymmetry.
A skewness value of zero means the data is balanced. Positive values mean the data is skewed to the right, and negative values mean it’s skewed to the left. The farther the value is from zero, the more lopsided the distribution. As a general benchmark, an absolute skewness value greater than 2 is considered a substantial departure from a normal, symmetric shape.
Right Skew vs. Left Skew
The direction of skew refers to where the long tail points, not where most of the data sits. This trips people up at first.
- Right skew (positive skew): The right tail is long relative to the left. Most values cluster on the lower end, but a few unusually high values stretch the distribution to the right. Housing prices are a classic example. Most homes in a city sell near the median price, but a handful of luxury properties pull the tail far to the right.
- Left skew (negative skew): The left tail is long relative to the right. Most values cluster on the higher end, with a few unusually low values dragging the tail left. Age of death from natural causes fits this pattern. Most people who die of heart disease or cancer do so at older ages, but some die younger, creating that leftward tail.
What Happens to the Mean and Median
In a perfectly symmetric distribution, the mean, median, and mode all land at the same spot. Skew pulls them apart, and the direction they separate tells you something useful.
The mean gets dragged toward the tail, because extreme values have an outsized effect on averages. If you’re looking at income data (right-skewed), a few millionaires pull the mean higher than the median. That’s why news reports about “average income” can be misleading. The median, which sits in the exact middle of the data, resists that pull and gives a better sense of what’s typical.
The rule is straightforward: if the mean is higher than the median, the data is right-skewed. If the mean is lower than the median, it’s left-skewed. Comparing these two numbers is a quick way to check for skew before doing any formal calculation.
Why Skew Matters in Practice
Many common statistical methods assume your data follows a roughly normal (symmetric) distribution. When data is heavily skewed, those methods can give misleading results. Averages become poor summaries, and statistical tests may overstate or understate significance.
Recognizing skew helps you choose the right summary statistic. For skewed data, the median is usually more informative than the mean. If someone asks what houses “typically” cost in your city, you’d want to give them the median price, not the mean, because a few mansion sales would inflate the average.
When skewed data needs to be analyzed with methods that assume normality, a common fix is the log transformation. Taking the logarithm of each value compresses the long tail and often pulls the data into a more symmetric shape. This works especially well when the original data follows a pattern where values multiply rather than add, which is common in biological and financial data. One practical note: log transformations only work on positive numbers, so if your data includes zeros, you’d add a small constant to each value first.
How to Interpret Skewness Values
When software calculates skewness for your dataset, here’s how to read the number:
- Between -0.5 and 0.5: The data is fairly symmetrical.
- Between -1 and -0.5, or 0.5 and 1: Moderately skewed.
- Beyond -1 or 1: Highly skewed.
- Beyond -2 or 2: Substantially non-normal. At this point, standard statistical methods that assume symmetry are likely unreliable without transformation.
For large datasets (more than 300 observations), you can rely on the raw skewness value along with a visual check of the histogram. For smaller samples, the skewness value is less stable and should be interpreted more cautiously.
Skewness vs. Kurtosis
Skewness and kurtosis are often reported together, but they measure different things. Skewness captures the left-right asymmetry of a distribution. Kurtosis captures how heavy the tails are, meaning how likely the data is to produce extreme outliers. A dataset can be perfectly symmetric (zero skew) but still have unusually thick tails (high kurtosis). Think of skewness as describing the shape’s balance and kurtosis as describing its tail weight.
Skew in Geometry
Outside of statistics, “skew” has a specific meaning in three-dimensional geometry. Two lines are called skew lines if they don’t intersect and aren’t parallel. This is only possible in three or more dimensions. In a flat, two-dimensional plane, non-parallel lines always cross eventually, but in 3D space, two lines can point in completely different directions and never meet. Think of a power line running north-south above a road running east-west beneath it. They don’t touch and they aren’t parallel, so they’re skew.