1 Definition and interpretation

Skewness describes the lack of symmetry in a distribution or data set around its central value. When a distribution is symmetric, its left and right sides have similar shapes and tail lengths. When it is skewed, one side extends farther or contains more extreme values than the other. The concept is used to summarize shape in descriptive statistics and to compare how different data sets depart from symmetry.

1.1 Symmetry and asymmetry

A symmetric distribution has matching or nearly matching tails on both sides of the center. In such cases, the mean, median, and mode often lie near one another. Asymmetry appears when one tail is stretched out more than the other, causing the center of the distribution to be pulled toward the longer tail. Skewness is therefore not simply about the location of data values, but about their imbalance around the middle.

1.2 Positive skewness

Positive skewness, also called right skewness, occurs when the right tail is longer or more pronounced. Most observations cluster on the left, while a smaller number of large values extend the distribution to the right. In many right-skewed data sets, the mean exceeds the median because the larger values influence the average more strongly.

1.3 Negative skewness

Negative skewness, or left skewness, occurs when the left tail extends farther than the right. The bulk of the data lies toward higher values, while a smaller number of low values create the long left tail. In such distributions, the mean is often less than the median, since low extreme values pull the average downward.

1.4 Zero skewness

Zero skewness indicates perfect symmetry in an idealized sense, though real data may show only approximate symmetry. A distribution with zero skewness does not necessarily have the shape of a normal curve; it may still be flat, bimodal, or otherwise nonnormal. The term mainly indicates balance between the two sides rather than any particular form.

2 Measures of skewness

Skewness can be quantified in several ways, depending on whether one wants a moment-based summary, a robust statistic, or a measure based on quartiles and other quantiles. Different formulas emphasize different parts of the distribution and may behave differently in small samples or in the presence of outliers.

2.1 Moment-based measures

Moment-based measures use algebraic summaries of the distribution, especially deviations from the mean. They are closely tied to the mathematical moments used in probability theory. These measures are common in theoretical work and in many statistical software packages.

2.1.1 Third standardized moment

The third standardized moment is a classic measure of skewness. It is formed by taking the third central moment and dividing by the cube of the standard deviation. This standardization makes the measure dimensionless and comparable across scales. Positive values indicate right skewness, negative values indicate left skewness, and values near zero suggest symmetry.

2.1.2 Sample skewness

Sample skewness is the empirical version computed from observed data. Several formulas exist, varying in how they adjust for sample size and bias. Although the computed value estimates population skewness, it can be unstable for small samples or highly irregular data. Software implementations may differ slightly in definition, so results are not always identical across programs.

2.2 Pearson’s coefficients

Pearson’s coefficients express skewness through simple relationships among the mean, median, and mode. One common form compares the distance between the mean and mode with the standard deviation, while another uses the mean and median. These coefficients are easy to interpret and useful when only summary statistics are available, though they are less general than moment-based measures.

2.3 Bowley’s skewness

Bowley’s skewness is based on quartiles rather than the mean and variance. It compares the spread of the central 50 percent of the data on either side of the median. Because it depends on quantiles, it is less sensitive to extreme observations than moment-based formulas. This makes it useful for skewed data sets with outliers.

2.4 Quantile-based measures

Quantile-based measures assess asymmetry by comparing selected percentiles, such as the lower and upper tails or different parts of the interquartile range. These methods can capture skewness in a robust way and are often helpful when distributions are heavy-tailed or when outliers would distort moment estimates. Their interpretation is usually intuitive, since they refer directly to positions in the ordered data.

3 Graphical identification

Skewness can often be recognized visually before it is calculated. Graphical methods provide an immediate sense of shape and can reveal whether asymmetry is mild, pronounced, or driven by a few unusual values. They are especially useful for exploratory data analysis.

3.1 Histograms

Histograms display the frequency of observations in consecutive intervals. A skewed histogram typically shows a concentration of data on one side and a longer tail on the other. While bin width can affect appearance, the overall direction of skewness is usually visible. Histograms are often the first tool used to inspect distribution shape.

3.2 Box plots

Box plots summarize data using quartiles, the median, and whiskers. Skewness may appear as an off-center median line, unequal whisker lengths, or isolated outliers on one side. Because box plots compress the data into a few summary features, they are compact and effective for comparing multiple distributions.

3.3 Density plots

Density plots show a smoothed estimate of the distribution’s shape. They can make asymmetry easier to see than histograms because they reduce the effect of arbitrary bin choices. A longer tail on one side of the peak is a common visual sign of skewness. These plots are useful when the data set is large enough to support stable smoothing.

3.4 Cumulative distribution functions

Cumulative distribution functions, or CDFs, present the proportion of observations below each value. Skewness is not as visually direct here as in a histogram, but asymmetry may be inferred from how quickly the curve rises on one side and flattens on the other. In comparative settings, differences between CDFs can reveal whether one distribution is more asymmetric than another.

4 Mathematical properties

Skewness has several important mathematical characteristics that affect how it is interpreted and computed. These properties help distinguish skewness from more familiar summaries such as the mean or variance.

4.1 Relationship to moments

Moment-based skewness depends on third-order departures from the mean. Because the third moment involves cubed deviations, values farther from the center contribute disproportionately. This is why skewness reflects asymmetry rather than mere spread. In contrast, the first moment gives location and the second moment measures dispersion.

4.2 Sensitivity to outliers

Skewness can be highly sensitive to extreme observations, especially when calculated from moments. A few large values can strongly affect the third moment and shift the result even if most data are fairly balanced. For this reason, robust alternatives based on medians or quantiles are often preferred in contaminated or heavy-tailed data.

4.3 Invariance under linear transformations

Skewness is unaffected by changes of measurement scale that use positive linear transformations. Multiplying all values by a positive constant or adding a constant does not alter the direction or degree of asymmetry. A reflection across the origin changes the sign of skewness, because it reverses the tail orientation.

4.4 Existence and undefined cases

Skewness may be undefined or not meaningful in some cases. If the variance is zero, standardized formulas cannot be computed because division by the standard deviation is impossible. For distributions with extremely heavy tails, the relevant moments may fail to exist, so a moment-based skewness measure is not defined in the usual mathematical sense.

5 Sampling and estimation

In practice, skewness is often estimated from finite samples rather than known populations. Estimation introduces variability, and different formulas can produce slightly different results. Care is needed when interpreting small samples or data with unusual tail behavior.

5.1 Estimators of population skewness

Estimators are formulas designed to approximate the skewness of an underlying population. Some are direct plug-in versions of the theoretical definition, while others adjust for sample size. The choice of estimator depends on whether the goal is theoretical analysis, descriptive reporting, or comparison across samples.

5.2 Bias and correction

Many raw sample skewness formulas are biased, especially in small samples. Bias correction attempts to improve the estimator’s average accuracy by adjusting for finite-sample effects. Such corrections are common in statistical software and can make reported values more comparable to population skewness under standard assumptions.

5.3 Standard error of skewness

The standard error of skewness measures how much an estimate may vary from sample to sample. It is useful for assessing whether an observed skewness value is meaningfully different from zero. In large samples, the standard error generally decreases, making the estimate more precise.

5.4 Small-sample behavior

With small data sets, skewness estimates can be noisy and unstable. A single unusual observation may dominate the result, and different formulas may disagree noticeably. For this reason, small-sample skewness should usually be interpreted alongside plots and other summary measures rather than in isolation.

6 Applications

Skewness is used across many branches of statistics and applied analysis. It helps summarize data shape, guides model choice, and supports diagnostic checks in practical settings.

6.1 Descriptive data analysis

In descriptive analysis, skewness provides a concise description of distribution shape. It helps analysts understand whether the data cluster evenly around the center or lean toward one tail. This information can guide the choice of summary statistics, such as whether the median may be more representative than the mean.

6.2 Normality assessment

Because a normal distribution is symmetric, skewness is one clue in evaluating departures from normality. Large skewness may indicate that transformation, nonparametric methods, or an alternative model is more appropriate. It is usually considered together with kurtosis, plots, and formal tests.

6.3 Financial data

Financial returns, income measures, and related variables often show skewed distributions. Positive or negative skewness can indicate that extreme gains or losses occur more often than a symmetric model would suggest. Analysts use skewness to summarize risk and to compare the shape of return distributions across assets or time periods.

6.4 Quality control and process monitoring

In quality control, skewness can signal changes in a manufacturing or measurement process. A shift in asymmetry may reflect wear, contamination, or other process changes that affect the distribution of outputs. Monitoring skewness alongside mean and variance can improve sensitivity to certain types of variation.

Skewness is closely connected to several other descriptive concepts that help characterize distributions. These ideas often appear together in statistical summaries and graphical analysis.

7.1 Kurtosis

Kurtosis describes the heaviness of tails and the sharpness of the central peak relative to a reference distribution. It is distinct from skewness, which measures asymmetry rather than tail weight. Both are fourth- and third-order shape summaries, respectively, and together they provide a broader picture of distribution form.

7.2 Mean, median, and mode

The mean, median, and mode are measures of central tendency that often shift relative to one another in skewed data. In right-skewed distributions, the mean tends to lie to the right of the median, while in left-skewed distributions it often lies to the left. The mode may remain near the highest point of the distribution and can help indicate the dominant cluster of values.

7.3 Tail behavior

Tail behavior refers to how quickly probabilities or frequencies decline away from the center. Long or heavy tails are often associated with pronounced skewness, though the two ideas are not identical. Tail shape is important in risk analysis, outlier detection, and the study of extreme events.

7.4 Central tendency and dispersion

Central tendency and dispersion are the broader categories that include location and spread. Skewness complements these measures by describing whether spread is balanced on both sides of the center. Together, they provide a compact statistical profile of a distribution.