Why divide by N for the whole finite population?

Because the descriptive definition averages the squared deviations over all N equally weighted values.

Does a large variance mean a large mean?

No. Shifting every value changes the mean without changing variance.

Can two different distributions share the same variance?

Yes. Variance is one summary and does not uniquely determine the distribution.

AI Glossary / Statistics and probability / Intermediate

Population variance

Definition

Population variance measures the average squared distance of values from their population mean. For a finite population of N equally weighted values, it is the sum of squared deviations from the population mean divided by N. For a random variable with a finite second moment, it is E[(X-mu)^2], where mu is the mean. It is commonly denoted sigma squared. Variance describes spread in squared units; its square root, the population standard deviation, returns to the original measurement units.

What it means

Two populations can have the same mean but very different variability. Variance quantifies one aspect of that difference by giving greater weight to larger deviations. It does not describe every feature of a distribution: skewness, tail behavior and multiple modes can differ even when means and variances match. The population must be clearly defined. A complete finite collection of recorded transactions has one descriptive variance, while an underlying process generating future transactions may have a different, unknown variance that those records only help estimate.

How it works

Define the population or probability model and calculate its mean. Square each deviation from that mean and average with the appropriate population weights. For a discrete probability distribution, use probability-weighted squared deviations; for a continuous distribution, integrate them against the density. An equivalent identity is Var(X) = E[X^2] - E[X]^2, when the moments exist. When only a sample is observed, choose an estimator appropriate to the sampling design. The familiar denominator n minus one estimates a population variance under independent, identically distributed sampling; it is not a replacement for the definition of the variance of a fully enumerated equally weighted finite population.

Three practical examples

Operations: describe variability in the complete set of a day's recorded processing times, with that day's cases explicitly defined as the finite population.
Probability modeling: calculate the variance implied by a distribution used to simulate demand or measurement noise.
AI evaluation: distinguish variability of individual model errors from uncertainty in an estimated average error, since these are different quantities with different interpretations.

Think of it this way

Think of measuring the spread of dots around their center. Squaring the distances prevents positive and negative deviations from canceling and gives distant dots more influence. The resulting number is a spread summary, not an average signed error and not the uncertainty of every statistic calculated from those dots.

Worked example

Consider an original finite population with values 2, 4, 6 and 8. Its mean is 5. Squared deviations are 9, 1, 1 and 9, summing to 20. Population variance is therefore 20/4 = 5, and standard deviation is sqrt(5), approximately 2.2361. The alternative identity gives mean square 30 minus squared mean 25, again 5. If these same four observations were instead treated as an independent sample for estimating an unknown population variance, the usual unbiased sample estimator would be 20/3, approximately 6.6667. The arithmetic differs because the statistical purpose differs. Finally, transforming every value to Y = 3X + 10 changes the mean to 25 and variance to 3^2 times 5 = 45. Adding a constant does not change spread, while multiplying all distances by three multiplies variance by nine.

Illustrative case study

In a fictional case, Cedar Logistics calculates the variance of all completed delivery times from one week. The report initially calls it the known variance of all future deliveries. Analysts revise the wording: the number describes that week's recorded population, while future performance depends on traffic, routes and operating conditions. They use additional data and an explicit model to estimate process variability and keep that estimate distinct from the week's exact descriptive calculation.

Common misconceptions and mistakes

Variance is expressed in squared measurement units; standard deviation uses the original measurement units. It cannot be negative when defined mathematically; a tiny negative value from a numerical calculation indicates computational error. A population variance can be zero when all values are identical. Some probability distributions have infinite or undefined variance, so the existence of a sample variance does not guarantee a finite theoretical one. The variance of a sample mean is not generally equal to the variance of individual observations. Independence and the sampling design determine how averaging changes uncertainty.

Business use

Businesses should select the spread measure that answers the operational question and label its units. Variance is useful in modeling and decomposition, while standard deviation or quantiles may be easier for stakeholders to interpret directly. Check whether extreme values represent errors or important genuine events before changing the data. When comparing periods, use consistent populations and weighting. A lower variance does not necessarily indicate better performance if the mean has worsened or an important subgroup has been omitted. Present location and variability together to avoid an incomplete comparison.

Check your understanding

Every measurement is doubled and then increased by 7. What happens to its population variance?

Answer: It is multiplied by four. Doubling multiplies squared deviations by four, while adding 7 leaves deviations from the new mean unchanged.

Related terms

Sample variance
Standard deviation
Standard error
second central moment
population mean
law of total variance

Learn next

Learn how variance changes under linear transformations, then study covariance and the conditions under which the variance of an average decreases with sample size.

Sample size

Sources

Penn State, STAT 414, Mathematical Expectation: https://online.stat.psu.edu/stat414/Lesson08 University of Vermont, David Howell, Population variance and sample variance glossary entries: https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html

Previous: Population protocol Next: Pose