1 Definition

Fisher's skewness is a numerical measure of asymmetry in a distribution or data set. It indicates whether observations are spread more heavily to the left or to the right of a central value, usually the mean. Because it is standardized by the spread of the data, it is dimensionless and can be compared across different variables or samples.

1.1 Probability distribution skewness

For a probability distribution, skewness describes the degree to which the distribution departs from symmetry. A perfectly symmetric distribution, such as the normal distribution, has skewness equal to zero. When one tail extends farther than the other, the distribution is said to be skewed in the direction of the longer tail.

1.2 Sample skewness

In sample analysis, Fisher's skewness is used as a statistic computed from observed values. It summarizes the asymmetry of the sample and serves as an estimate of the population skewness. Different formulas may be used depending on whether the goal is descriptive reporting or statistical inference.

1.3 Standardized third central moment

Fisher's skewness is commonly defined as the third central moment divided by the cube of the standard deviation. This standardization makes the measure unitless and ties it directly to the shape of the distribution rather than to its scale.

1.4 Notation and symbols

The statistic is often denoted by the Greek letter gamma with a subscript indicating skewness, or by related symbols in sample formulas. In applied work, notation varies across textbooks, software packages, and statistical reports, but the underlying concept remains the same.

2 Mathematical formulation

Fisher's skewness can be expressed in population form or estimated from a sample. The formulas rely on central moments, especially the third central moment, which captures directional asymmetry.

2.1 Population formula

For a random variable with mean μ and standard deviation σ, population skewness is defined as the expected value of the cubed deviation from the mean divided by σ cubed. This expresses how strongly values deviate to one side of the center relative to the overall scale.

2.2 Sample formula

A common sample version replaces the population moments with their empirical counterparts. Given observations, the mean is computed first, then each deviation from the mean is cubed, summed, and standardized by the sample standard deviation raised to the third power.

2.3 Bias correction

Sample skewness is generally biased in small samples, meaning its average value may differ from the true population skewness. To reduce this effect, bias-corrected formulas are sometimes used, especially in inferential settings.

2.3.1 Adjusted Fisher-Pearson coefficient

The adjusted Fisher-Pearson coefficient is a corrected form of the sample skewness statistic. It multiplies the raw coefficient by a factor involving the sample size, improving its behavior as an estimator of population skewness for finite samples.

2.4 Relationship to central moments

Skewness is directly related to the third central moment, while variance depends on the second central moment. Together, these moments describe dispersion and asymmetry, making skewness part of a broader moment-based description of distribution shape.

3 Interpretation

The sign and magnitude of Fisher's skewness help summarize the shape of a distribution. Interpretation is often qualitative, though the numeric value can also be compared across samples.

3.1 Positive skewness

Positive skewness indicates a longer or heavier right tail. In such distributions, most observations cluster on the lower side, while a few larger values pull the tail to the right.

3.2 Negative skewness

Negative skewness indicates a longer left tail. Most values may lie above the center, with some relatively small observations extending the distribution toward the left.

3.3 Zero skewness

A skewness value of zero indicates symmetry in the third-moment sense. This does not necessarily mean the distribution is perfectly symmetric in every detail, but it does mean that the asymmetry measured by the third central moment vanishes.

3.4 Magnitude and asymmetry

The absolute value of skewness reflects the degree of asymmetry. Larger magnitudes indicate stronger departure from symmetry, while values near zero suggest a more balanced shape.

4 Estimation and computation

Skewness is commonly calculated from raw observations, though approximations and algorithmic methods are also used for large or grouped data. Accurate computation depends on numerical stability and on the formula selected.

4.1 Calculation from raw data

The basic procedure begins with the sample mean and the sample standard deviation. Each data value is centered by subtracting the mean, then the centered values are raised to the third power, averaged, and standardized.

4.2 Grouped data estimation

When data are summarized into intervals or frequency classes, skewness may be estimated from class midpoints and counts. Such estimates are less precise than those computed from raw data because grouping reduces detail in the tails and around the center.

4.3 Computational algorithms

Efficient algorithms compute moments in a single pass or in numerically stable stages. These methods are useful for large data sets, streaming data, and software implementations that must avoid loss of precision.

4.4 Software implementations

Statistical software often provides built-in skewness functions, though conventions may differ. Some packages return the unadjusted sample coefficient, while others use a bias-corrected form or offer multiple options.

5 Properties

Fisher's skewness has several mathematical properties that make it useful in statistical analysis. These properties also explain some of its limitations.

5.1 Dimensionless scale

Because skewness is standardized by the standard deviation, it has no units. This allows direct comparison between variables measured on different scales.

5.2 Sensitivity to outliers

The cubing of deviations gives extreme values substantial influence. As a result, a few outliers can noticeably alter the skewness value, especially in small samples.

5.3 Dependence on sample size

Sample skewness can vary substantially when the number of observations is small. As sample size increases, the statistic typically becomes more stable and more representative of the underlying distribution.

5.4 Invariance under linear transformations

Skewness is unchanged by adding a constant or multiplying by a positive constant. Reversing the sign of the data, however, changes the sign of skewness because the distribution is mirrored.

6 Comparison with other skewness measures

Several alternative skewness measures exist, each emphasizing a different aspect of asymmetry. Fisher's measure is moment-based, while others rely more on quartiles or positional summaries.

6.1 Pearson's skewness coefficients

Pearson's coefficients use the mean, median, and standard deviation to describe asymmetry. They are often easier to interpret in simple data sets, but they do not capture shape in the same moment-based way as Fisher's skewness.

6.2 Bowley's skewness

Bowley's skewness is based on quartiles and is therefore less affected by extreme values. It summarizes asymmetry using the middle portion of the distribution rather than the full set of moments.

6.3 Moment-based versus quantile-based measures

Moment-based measures such as Fisher's skewness use all observations and are sensitive to tail behavior. Quantile-based measures are more robust to outliers but may overlook details in the extremes. The choice between them depends on the data and the analysis goal.

7 Applications

Fisher's skewness is widely used in statistics, data analysis, and modeling. It helps characterize shape, guide transformations, and inform method selection.

7.1 Descriptive statistics

In descriptive summaries, skewness provides a compact indication of whether a variable is balanced or tail-heavy. It is often reported alongside the mean, variance, and kurtosis.

7.2 Data exploration

During exploratory data analysis, skewness can highlight departures from symmetry that may not be obvious from central tendency alone. It can suggest whether a log transformation or other adjustment might be useful.

7.3 Distribution fitting

Skewness can aid in selecting or evaluating candidate distributions. A fitted model should ideally reproduce the observed asymmetry of the data, especially when tail behavior is important.

7.4 Statistical modeling

In modeling, skewness is relevant when assessing assumptions about residuals or response variables. It may influence decisions about transformations, robust methods, or the choice of a non-symmetric distribution.

8 Limitations

Although useful, Fisher's skewness is not a complete description of shape. It should be interpreted in context and supplemented with graphical and other numerical summaries.

8.1 Effect of extreme values

Because of the third power in its formula, extreme observations can dominate the statistic. This makes skewness informative about tail behavior, but also vulnerable to distortion from unusual values.

8.2 Non-normal data considerations

For highly irregular or multimodal distributions, a single skewness value may not adequately summarize the shape. Two distributions can share the same skewness while differing greatly in overall form.

8.3 Small-sample instability

With few observations, skewness estimates can be noisy and may change substantially if one value is added or removed. Caution is needed when interpreting small-sample results.

Fisher's skewness belongs to the broader study of distribution shape and moment-based statistics. It is often considered together with measures of spread and peakedness.

9.1 Kurtosis

Kurtosis measures the heaviness of tails and the concentration of values near the center. It complements skewness by describing a different aspect of distribution shape.

9.2 Central moments

Central moments are averages of powers of deviations from the mean. They provide a general framework for describing variance, asymmetry, and higher-order shape features.

9.3 Asymmetry in distributions

Asymmetry is a general property of distributions that are not mirror images around a central point. Skewness is one formal way to quantify that asymmetry.

</INTERNAL_LINK_CANDIDATES> Probability distribution (a random-variable law whose shape can be analyzed for skewness) Sample skewness (the skewness computed from observed data) Third central moment (the moment used to quantify asymmetry) Mean (the central value used in centering deviations) Standard deviation (the scale factor used to standardize skewness) Adjusted Fisher-Pearson coefficient (the bias-corrected sample skewness) Population skewness (the theoretical skewness of a distribution) Bias correction (a method to reduce small-sample estimation bias) Central moments (moments about the mean) Pearson's skewness coefficients (alternative skewness measures based on mean, median, and spread) Bowley's skewness (a quartile-based skewness measure) Quantile-based measures (robust asymmetry measures based on positions in the data) Descriptive statistics (summary measures used to describe data) Data exploration (initial analysis used to inspect distribution shape) Distribution fitting (matching data to a candidate probability model) Statistical modeling (use of skewness in model assessment and transformation choice) Outliers (extreme observations that can affect skewness) Kurtosis (a related measure of distribution shape) Linear transformation (a scaling or shifting operation affecting skewness predictably) Median (a central measure used in Pearson-style skewness)