Can the mean be outside the observed range?

An ordinary arithmetic mean cannot; a mean with nonnegative weights also stays within that range.

Is a sample mean automatically representative of a population?

No. Sampling and missing-data mechanisms determine whether the broader inference is justified.

AI Glossary / Statistics and probability / Beginner

Mean

Definition

The mean, usually meaning the arithmetic mean, is the sum of a set of numerical values divided by the number of values. A population mean describes a population or probability distribution; a sample mean summarizes observed data and may estimate a population mean. Weighted means assign different contributions to observations according to explicitly defined weights.

What it means

The mean describes the amount each observation would have if the observed total were shared equally. That makes it useful for quantities such as total workload per request or total consumption per customer.
It does not necessarily describe the most common observation or a value that anyone actually experienced. A small number of unusually large values can move the mean substantially. Whether that is a weakness or a useful reflection of the total depends on the question being asked.

How it works

For n observations x[1] through x[n], calculate x-bar = sum x[i]/n. Use consistent units and define which observations belong in the denominator. Including an extra zero is not the same as recording a missing value or excluding a case.
The population version is an expected value. For a discrete distribution, it is the sum of each possible value multiplied by its probability, when that expectation exists. Some heavy-tailed distributions do not have a finite mean, even though every finite sample has an arithmetic average.
For nonnegative weights with a positive total, the weighted mean is sum w[i]*x[i] divided by sum w[i]. Explain what the weights represent: frequencies, survey design, exposure, or another purpose. Different weighting choices can answer different population questions.
When combining group means, weight them by the corresponding group sizes if the goal is the overall mean of all observations. Taking an unweighted average of group means instead gives each group equal influence, regardless of how many observations it contains.
The arithmetic mean has a useful balance property: deviations from it sum to zero. It also minimizes the sum of squared distances from a single constant. That property explains its role in least squares, but it does not imply it is always the most informative measure of a typical case.
Distinguish the mean from the median and mode. The median concerns the middle of an ordered distribution, while the mode concerns its most frequent or highest-density region. In a skewed or multimodal dataset, these summaries can differ substantially.
Consider the measurement scale. Averaging arbitrary numeric category codes is generally meaningless. For ordered ratings, interpreting an average assumes something about the spacing of the categories. For multiplicative growth or some rate problems, another type of average may better match the target.
A sample mean's uncertainty is different from the spread of individual observations. Under independent, identically distributed observations with finite variance, its standard deviation decreases with the square root of sample size. Dependence, sampling weights, or clustering changes the appropriate calculation.
Inspect the distribution, missingness, and influential observations before reporting a mean alone. A single summary can conceal subgroups, extreme tails, or changing composition. Pair it with the sample size and suitable measures of spread when those features matter for interpretation.

Three practical examples

An operations team divides total processing time by completed requests to estimate average time per request.
A survey analyst uses appropriate weights to estimate a population average from unequal-probability sampling.
An AI evaluation team reports average loss while also checking difficult task groups and unusually large errors.

Think of it this way

Think of pouring the quantities from several containers into one pool and redistributing them equally. The level after redistribution is the mean. The picture explains why the total is preserved, but it also shows why the redistributed level need not match the amount in any original container.

Worked example

An original example has two requests in group A, taking 8 and 12 minutes. Group A's mean is 10 minutes. Group B has eight requests, each taking 20 minutes, so its mean is 20 minutes.
The unweighted average of the two group means is (10+20)/2 = 15 minutes. But the mean across all ten requests is (2*10 + 8*20)/(2+8) = 180/10 = 18 minutes. The first calculation describes an equal-weight average of groups; the second describes an equal-weight average of requests.
As a separate check of the mean's squared-distance property, observations 2, 4, 6, and 8 have mean 5. Their squared distances from 5 sum to 9+1+1+9 = 20. Using 4 instead gives 4+0+4+16 = 24.
More generally, replacing the mean by a constant a adds n*(a-x-bar)^2 to the squared-distance total. Here that addition is 4*(4-5)^2 = 4, confirming the difference. This is an algebraic property, not evidence that squared error is the right business loss in every application.

Illustrative case study

In a fictional case, Ashford Support reports an average response time across three teams. Managers later discover that the dashboard averages team-level means even though team volumes differ greatly.
Analysts revise the calculation to use total response time divided by total requests and label a separate equal-team comparison explicitly. They also add a tail summary because a small set of very slow responses matters operationally. The case illustrates choosing a denominator that matches the question; no real performance improvement is claimed.

Common misconceptions and mistakes

Do not assume the mean is an observed value, the most frequent value, or a guarantee that most observations are close to it. Do not average percentages or group means without considering their denominators.
A mean can be computed from finite data without a normality assumption. Normality may matter for particular inferential procedures, but it is not required for the arithmetic definition itself.

Business use

The mean is useful when totals and per-unit resource requirements matter. It supports budgeting, workload planning, and average-loss comparisons. Keep the unit of analysis explicit: per request, per customer, per day, and per team can produce different answers. When exceptional outcomes drive costs or user experience, report their behavior alongside the mean.

Check your understanding

Why is the overall request mean 18 rather than 15?

Answer: Eight of the ten requests belong to the group with mean 20, so an equal-request average must give that group more weight.

Related terms

Expected value
Median
Mode
Standard error
Arithmetic mean
Weighted mean
Geometric mean
Sampling weight

Learn next

Next, compare means, medians, and distribution plots on skewed data. Then study weighted aggregation and how the sampling design affects uncertainty in a reported average.

Sampling design

Sources

Frank Harrell, Glossary of Statistical Terms, mean: https://hbiostat.org/glossary/ Australian Bureau of Statistics, Statistical terms and concepts glossary: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/statistical-terms-and-concepts-glossary NIST/SEMATECH e-Handbook, Measures of Location: https://www.itl.nist.gov/div898/handbook/eda/section3/eda351.htm

Previous: MCS algorithm Next: Mean absolute deviation (m.a.d.)