Skewness

Skewness measures how lopsided a distribution is. A perfectly symmetric distribution, like the textbook bell curve, has zero skewness — fold it down the middle and the two halves match. Real data rarely cooperates. When a distribution has a long tail trailing off to the right, it is positively (right) skewed, and the rare large values drag the mean above the median. When the tail stretches left, it is negatively (left) skewed, and the mean falls below the median. The single number that captures this is the skewness coefficient: zero means symmetric, positive means a right tail, negative means a left tail, and the bigger the magnitude, the more pronounced the lean. It tells you, at a glance, which direction the unusual values are hiding.

Skewness matters because it decides whether the average is honest. Customer spend, order values, time-on-page, lifetime value, and income are almost always right-skewed: most people spend a little, a handful spend enormously, and those whales yank the mean upward until it no longer describes a typical customer. Report the mean order value of such data and you can quote a figure that almost no one actually spends. The median — the middle value, unmoved by extremes — usually tells a fairer story for skewed data. Knowing the skew also flags when a model's assumptions are at risk, since many techniques quietly assume roughly symmetric, normal-ish data and start to mislead when the distribution leans hard one way.

Skewness and kurtosis are the two shape statistics that go beyond the mean and the spread, and they answer different questions. Skewness is about direction — which way does the distribution lean, left or right. Kurtosis is about the tails and peak — how heavy the tails are and how much of the data piles into extreme values versus the shoulders. A distribution can be perfectly symmetric (zero skewness) yet have very heavy tails (high kurtosis), meaning rare extremes on both sides. It can also be strongly skewed with ordinary tails. Confusing them leads to the wrong worry: skewness warns you the average is pulled to one side, while kurtosis warns you outliers are more frequent and severe than a bell curve predicts.

In practice you read them together to understand a distribution's true shape. For right-skewed marketing data — say revenue per account — high positive skewness tells you the mean overstates the typical account, and the natural fix is often to take a logarithm, which compresses the long right tail and pulls the distribution back toward symmetry. That same log transform is a common cure for the heteroscedasticity that skewed targets create in regression. Kurtosis, meanwhile, would tell you how much to brace for extreme values in either direction, which matters for risk and for outlier handling. Treating the two as one blurs the picture: one is lean, the other is tail weight, and good distribution analysis names both.

Use skewness as a first check before you trust any average. Whenever you summarize a distribution that someone will act on — order values, session durations, deal sizes, customer lifetime value — compute or eyeball the skewness, and when it is strongly positive, lead with the median rather than the mean so your headline number reflects a typical case, not the pull of a few giants. Show the distribution itself with a histogram when you can, because a single statistic hides the story that a shape reveals. When you feed skewed data into a model that assumes symmetry, consider a log or square-root transformation to tame the lean, which often improves both fit and the validity of confidence intervals.

The discipline is to let skewness govern which summary you report and how you present results to non-statisticians. Pair it with segmentation — the skew in "all customers" often dissolves once you split whales from the rest into cleaner groups. Pair it with kurtosis to read tail weight separately from lean. The failure is to quote the mean of obviously skewed data as if it described a typical customer, to ignore the long tail when forecasting, or to confuse the lean with heavy tails and reach for the wrong remedy. Skewness, read early, keeps your averages from lying.

Skewness, from "skew" meaning to set at a slant, names the statistic for a distribution's asymmetry — which way its tail leans.

Etymology: source.

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Areas of marketing where skewness is a core concern: