A skewed histogram is one where the data bunches up on one side and stretches out with a longer tail on the other. Unlike a symmetrical, bell-shaped histogram where both sides mirror each other, a skewed histogram is lopsided. The direction of that longer tail tells you the type of skew and has real consequences for how you should interpret the data.
Right-Skewed vs. Left-Skewed
Skewness comes in two directions, and the naming convention trips people up at first. The skew is named for the direction of the tail, not the direction of the peak.
A right-skewed (positively skewed) histogram has most of its data clustered on the left side, with a long tail stretching to the right. Think of household income: most people earn moderate amounts, but a small number of very high earners pull the tail far to the right. Housing prices follow the same pattern. The bulk of homes sell in a typical range, but luxury properties create that rightward tail.
A left-skewed (negatively skewed) histogram is the opposite. The tail extends to the left while most data clusters on the right. Age at death in developed countries is a classic example: most people live into their 70s and 80s, but a smaller number of early deaths pull the tail leftward. Scores on an easy exam also tend to be left-skewed, since most students score high and only a few score very low.
The quickest way to identify skew when looking at a histogram is to find the tail. If the tail points right, it’s right-skewed. If the tail points left, it’s left-skewed.
How Skewness Shifts the Mean, Median, and Mode
In a perfectly symmetrical distribution, the mean and median sit at the same point. Skewness pulls them apart, and understanding how they separate is one of the most practical things to know about skewed data.
In a right-skewed distribution, the mean is larger than the median, and the median is larger than the mode. Those extreme high values in the tail drag the mean upward. For example, in one right-skewed dataset, the mode was 7, the median 7.5, and the mean 7.7. The mean got pulled toward the tail.
In a left-skewed distribution, the pattern reverses. The mean is the smallest of the three, the median falls in the middle, and the mode is the largest. In a left-skewed example, the mean was 6.3, the median 6.5, and the mode 7. Here, the low values in the left tail dragged the mean downward.
A simple rule: the mean always gets pulled toward the tail. That single fact explains most of what matters about skewed data in practice.
Why the Mean Can Be Misleading
When data is skewed, the mean often paints a distorted picture because it’s sensitive to extreme values. This is why news reports about “average” income can feel disconnected from most people’s reality. A few billionaires in the dataset pull the mean income well above what a typical person earns.
The median is more resistant to this kind of distortion. It sits at the exact middle of the dataset, so no matter how extreme the values in the tail are, the median barely moves. If you replaced the highest income in a dataset with a number ten times larger, the mean would jump dramatically, but the median would stay almost exactly where it was.
This is why median household income is generally more informative than mean household income, and why median home prices are the standard in real estate reporting. Whenever you’re working with skewed data, the median typically gives a more accurate sense of what’s “normal” in the dataset.
Recognizing Skew in Your Own Data
Beyond visual inspection of a histogram, there are a few clues that suggest skewness. If the mean and median of your dataset are noticeably different, that gap itself signals skew. The direction of the gap tells you which way: mean greater than median points to right skew, mean less than median points to left skew.
Data that has a natural floor but no ceiling tends to be right-skewed. Income can’t go below zero, but there’s no upper limit, so it skews right. Reaction times work the same way: there’s a minimum speed for human response, but slow outliers can stretch far to the right. Wait times, medical costs, and insurance claims all follow this pattern.
Data with a natural ceiling but no hard floor tends to be left-skewed. Test scores capped at 100, customer satisfaction ratings with a maximum of 5 stars, and age at retirement in a stable profession all tend to cluster near the upper limit with a leftward tail.
What To Do With Skewed Data
If you’re summarizing skewed data, report the median instead of (or alongside) the mean. This gives your audience a more honest picture of typical values. Including both numbers, plus noting the direction of skew, is even better since it tells the full story.
If you’re running statistical analyses that assume normally distributed data, skewness can be a problem. Many common tests, like t-tests and linear regression, work best with roughly symmetrical data. When your histogram is visibly skewed, you have a few options.
A log transformation is the most common fix for right-skewed data. It compresses the long tail by converting each value to its logarithm, which pulls extreme values closer to the rest of the data. Square root transformations serve a similar purpose but are less aggressive. For more flexibility, the Box-Cox transformation automatically finds the best power transformation to make your data more symmetrical.
These transformations don’t change what the data represents. They rescale it so that statistical methods designed for bell-shaped distributions can work properly. You can always convert results back to the original scale for interpretation.
Not every analysis requires normally distributed data, though. Non-parametric tests, which make no assumptions about the shape of your distribution, are an alternative when transformation feels inappropriate or when the skew is a meaningful feature of the data you don’t want to erase.