Must its mean equal its median?

If the mean exists it equals the center, which is a median; existence and uniqueness need qualification.

AI Glossary / Statistics and probability / Intermediate

Symmetric distribution

Definition

A symmetric distribution has the same probability pattern on either side of a central value. Formally, a random variable X is symmetric about c when X-c and c-X have the same distribution. Symmetry describes shape; it does not require a normal distribution, a single peak, or even a finite mean.

What it means

Values equally far above and below the center have matching probability behavior. For a continuous density, this can be expressed as equal density heights at c+d and c-d. For discrete outcomes, matching reflected values have equal probability masses.
The center is not automatically a well-defined population mean. If the mean exists, symmetry places it at the center. Some symmetric distributions have tails so heavy that the mean does not exist. A center can also be a median without requiring a unique median in every discrete distribution.

How it works

Specify the variable and the scale on which symmetry is considered. A distribution can be symmetric after a transformation and asymmetric on its original scale. Logarithmic and other nonlinear transformations do not generally preserve symmetry.
Identify a candidate center and compare reflected values or intervals. The probability assigned to an interval on one side should match the probability of its reflected interval on the other side for exact symmetry.
For a continuous distribution with density f, check whether f(c+d) equals f(c-d) for relevant d. Individual exact points have probability zero in a continuous model, so density equality should not be confused with nonzero point probabilities.
For a discrete distribution, compare its probability mass at reflected support points. Discrete symmetry can include substantial mass exactly at the center, and the cumulative distribution has jumps that need careful endpoint handling.
When the first absolute moment is finite, the mean equals the symmetry center. If a finite variance exists, it measures spread around that center. Symmetry alone guarantees neither moment; the Cauchy distribution is a standard counterexample.
A symmetric shape need not be bell-shaped. Uniform distributions are flat, and mixtures can be symmetric with two or more peaks. Different symmetric distributions can have sharply different tail probabilities and therefore different operational risks.
In observed data, assess approximate symmetry using plots and numerical summaries, while recognizing sampling variability. A small sample from a symmetric population need not look perfectly balanced. Conversely, a roughly symmetric histogram does not prove a specific probability model.
Zero skewness is not a general proof of symmetry. Where the relevant moments exist, symmetry implies zero third central moment, but the reverse implication can fail. Do not replace a shape assessment with one number alone.
When using a method that assumes symmetric errors, consider the conditional residual distribution relevant to that method. Symmetry of the raw outcome distribution does not automatically establish symmetry of regression errors or validate every inferential assumption.

Three practical examples

Measurement systems: assess whether positive and negative deviations around a calibration target have similar distributions.
Forecast evaluation: inspect whether overpredictions and underpredictions have balanced magnitudes on the chosen scale.
Simulation: choose a symmetric noise model while separately specifying its tail behavior and variance.

Think of it this way

Think of folding a probability picture along a vertical line through its center. Exact symmetry means the two halves match. The matching shape could be flat, sharply peaked, or have two hills; folding successfully does not make every picture a bell curve.

Worked example

Consider an original discrete distribution with values -2, -1, 0, 1, and 2. Assign probabilities 0.10, 0.20, 0.40, 0.20, and 0.10 respectively. The probabilities sum to one and match at reflected values, so the distribution is symmetric about zero.
Its mean is (-2)(0.10) + (-1)(0.20) + 0(0.40) + 1(0.20) + 2(0.10) = 0. Its variance is 4(0.10) + 1(0.20) + 0 + 1(0.20) + 4(0.10) = 1.20.
The probability of a value below zero is 0.30, and the probability above zero is also 0.30. They are not each 0.50 because the center itself has probability 0.40. This is why statements about equal halves need care for discrete distributions.
Adding ten to every possible value preserves the symmetric shape but moves the center and mean to ten. The variance remains 1.20. The example is discrete and is not a normal distribution despite its symmetry.

Illustrative case study

Fictional case: Maple Forecasting reviews errors from an AI demand model. A plot looks roughly balanced around zero, so a manager proposes assuming normal errors. Analysts compare tail behavior and find more extreme errors than a normal model would suggest. They retain the distinction between approximate symmetry and normality, then assess uncertainty with a method suited to the observed pattern. The review also separates errors by season, because an overall balanced plot can conceal different conditional behaviors.

Common misconceptions and mistakes

Do not equate symmetry with normality or a single peak. Equal mean and median do not prove symmetry. Symmetry does not guarantee a finite mean or variance. A symmetric distribution can be centered far from zero. Perfectly balanced counts are not required in every sample from a symmetric population. A transformation can change whether the distribution is symmetric.

Business use

Use symmetry as one part of assessing measurement error, forecast behavior, and distributional assumptions. Pair it with information about location, spread, tails, and subgroup patterns. For decisions affected by extreme outcomes, tail probabilities can matter more than whether the middle of the distribution appears balanced.

Check your understanding

What is the symmetry center in the example?

Answer: Zero.

Why is the probability below the center only 0.30?

Answer: There is probability mass of 0.40 at the center.

Does adding ten change the variance?

Answer: No; a shift changes location but not spread.

Related terms

Median
Mean
Skewness
Normal distribution
Distribution shape
Cauchy distribution
heavy tails
location parameter

Learn next

Compare normal, uniform, Cauchy, and symmetric mixture distributions. Then examine how transformations and conditioning can change distribution shape without changing the underlying observations.

Sources

Harrell, Glossary of Statistical Terms: https://hbiostat.org/glossary/ NIST/SEMATECH, Cauchy Distribution: https://www.itl.nist.gov/div898/handbook/eda/section3/eda3663.htm University of Vermont glossary: https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html The discrete probability example is original.

Previous: Symbolic regression Next: Symmetric matrix