Is a sample mean a statistic?

Yes; it is calculated from the observations.

Is a statistic always a single number?

No. More general definitions include vector-valued or other data summaries, although introductory examples are often scalar.

Does every statistic estimate a parameter?

No. Some are used primarily for description, diagnostics, or hypothesis testing.

AI Glossary / Statistics and probability / Beginner

Statistic

Definition

A statistic is a quantity calculated from observed data using a specified rule. Examples include a sample mean, sample proportion, median, variance, or correlation. In statistical inference, a statistic is distinguished from a population parameter: the statistic is computed from the available observations, while the parameter describes the population or probability model. A statistic can summarize data, estimate a parameter, or help test a hypothesis. The same numerical result can serve different purposes depending on the question and design.

What it means

The word statistics can name the whole discipline or refer to multiple calculated summaries. A statistic is one such data-based quantity. Before a sample is observed, its value can vary across possible samples. After collection, the observed statistic is a particular number or other specified data summary.

A statistic is not automatically an accurate estimate of the quantity people care about. Its usefulness depends on the data source, the calculation rule, the target population, and how uncertainty and bias are handled. A clearly calculated number can still answer the wrong question.

How it works

Begin with a well-defined dataset and a purpose. Choose a rule appropriate to the measurement scale and target. To calculate a sample mean, add the observed values and divide by their number. To calculate a sample proportion, divide the count satisfying a stated condition by the eligible denominator.

Distinguish the statistic as a function of data from its realized value. An estimator is a statistic used to estimate a parameter; the observed result is an estimate. A test statistic is selected to compare observed data with what a hypothesis and its assumptions would predict.

Specify exclusions, missing-value handling, weights, and transformations. These decisions can change the statistic even when the underlying records are the same. For inference, study the sampling distribution or an appropriate model-based uncertainty procedure. The fact that a statistic is easy to compute does not establish that standard formulas for its uncertainty apply to clustered, selected, or dependent observations.

Three practical examples

1. A support manager calculates the median resolution time to summarize the center of a skewed collection of completed tickets.

2. A quality analyst calculates the proportion of sampled products that fail a specified inspection criterion.

3. A machine-learning team computes mean absolute prediction error on a held-out evaluation dataset, defining exactly which cases and weights enter the calculation.

Think of it this way

Think of a statistic as a recipe applied to ingredients. The recipe determines how the ingredients become a summary, but a correctly followed recipe cannot fix missing ingredients or make them represent a different population. Both the calculation and the source material matter.

Worked example

Consider an original dataset of five ticket-resolution times, measured in hours: 2, 3, 3, 7, and 10. Their sum is 25, so the sample mean is 25/5 = 5 hours. The median is the middle ordered value, 3 hours. Three of the five times are at most 3 hours, giving a sample proportion of 3/5 = 0.60.

The squared deviations from the mean are 9, 4, 4, 4, and 25, summing to 46. Using the usual n - 1 denominator, the sample variance is 46/4 = 11.5 square hours. Each of these quantities is a statistic, but each summarizes a different feature.

Now replace the final value 10 with 30. The mean becomes 45/5 = 9 hours, while the median remains 3 hours and the proportion at most 3 hours remains 0.60. This comparison shows that different statistics respond differently to the same change in data.

If these five records are merely convenient examples, their mean is still a valid descriptive statistic for those records. Claiming that it estimates the average for every customer ticket requires a defensible selection or modeling argument. The arithmetic alone does not supply that argument.

Illustrative case study

In a fictional service company, a dashboard labels the average resolution time simply as “performance.” A new analyst discovers that it excludes unresolved tickets and gives equal weight to tickets with very different business consequences.

The team clarifies what the statistic measures, adds the number of unresolved cases, and separates reporting of typical time from reporting of costly delays. It keeps the original mean where useful but documents its denominator and scope. The improvement comes from matching summaries to decisions, not from searching for a single number that can describe every aspect of service quality.

Common misconceptions and mistakes

Do not confuse a statistic with a population parameter merely because both are expressed in the same units. A sample mean and a population mean can have different values. Nor is every number in a report an inferential estimate: an identifier or a budget target is not automatically a statistic in this sense.

An unbiased estimator need not be close to the truth in a particular sample, and a stable statistic can still be systematically misleading for a target population. Avoid presenting a calculated value without its denominator, units, reference period, or definition. A small p-value associated with a test statistic is not a substitute for reporting the size and uncertainty of an effect.

Business use

Statistics turn operational records into quantities that support comparison and action. Businesses benefit when definitions are consistent, auditable, and aligned with the decisions being made. Maintain a metric dictionary specifying the calculation, population, exclusions, timing, and owner. When comparing teams or periods, verify that the same statistic is being calculated from comparable data rather than assuming matching labels guarantee matching meaning.

Check your understanding

A sample contains 80 inspected items, of which 12 are defective. Identify the sample proportion and the corresponding population target.

Answer: The statistic is 12/80 = 0.15. A possible target parameter is the defect proportion in the defined population from which the items were sampled, subject to the sampling design and measurement definition.

Related terms

Learn next

Learn the difference between samples and populations, then compare estimators and their sampling distributions. Practice defining a business metric precisely before calculating it.

Sources

NIST/SEMATECH, Statistical Engineering Handbook glossary: https://www.itl.nist.gov/div898/handbook/glossary.htm

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

The official glossaries provide terminology for statistics, estimates, and descriptive summaries. All ticket data, calculations, and the fictional case are original teaching examples.

Previous: Stationary sequence Next: Statistical assembly