Can a center lie where no observation occurs?

Yes, means and conventional even-sample medians need not be observed values.

Can there be more than one mode?

Yes, tied highest frequencies or multiple density peaks can occur.

Which measure is always best?

None; the variable and purpose determine which summary is informative.

AI Glossary / Statistics and probability / Intermediate

Measures of central tendency (center or central location)

Definition

Measures of central tendency, also called measures of center or central location, are summaries intended to represent a typical or central value of a distribution. The most familiar are the arithmetic mean, median, and mode. They define typicality differently: the mean averages numerical magnitudes, the median describes the middle of an ordered distribution, and the mode identifies the most frequent value or category.

What it means

A single number can make a large dataset easier to discuss, but there is no universal definition of its center. An organization asking about a typical customer experience may need a different summary from one calculating total resource demand per customer.
The choice depends on the variable, distribution, and question. Two datasets can have the same center and very different variability or shape. A useful summary therefore identifies the measure used and places it alongside enough information to avoid hiding important differences.

How it works

For the arithmetic mean of a finite numerical dataset, add the values and divide by their count. Each observation contributes its magnitude. When observations have justified weights, use their weighted sum divided by the total weight and explain what those weights represent.
For the conventional numerical sample median, order the values. With an odd count, take the middle observation; with an even count, average the two middle observations. More generally, a median has at least half the probability or observations at or below it and at least half at or above it. Ties mean there need not be exactly half strictly on each side.
For an empirical mode, count how often each value occurs and identify the greatest frequency. There can be several modes. For continuous distributions, a population mode describes a density peak; a sample's most repeated recorded value may instead reflect rounding or binning.
Check the measurement scale. Unordered category names can have a mode, but their arbitrary numerical codes do not have a meaningful arithmetic mean. Ordered categories can support a median category without assuming that adjacent categories are equally spaced.
Inspect a plot and a measure of spread. Distinct clusters may make a single central value an inadequate description, even when it is calculated correctly.

Three practical examples

Customer support: a team reports median waiting time to describe the middle experience and mean waiting time when examining aggregate waiting burden per contact.
Product planning: a retailer uses the modal requested size for a popularity question while retaining the full size distribution for inventory decisions.
Personal budgeting: someone compares mean monthly spending with the median to understand the effect of occasional large purchases, without treating either statistic as the complete budget.

Think of it this way

Think of three ways to choose a representative place in a crowd. The mean acts like a balance point for numerical positions. The median identifies a middle position in the ordered crowd. The mode identifies the most crowded position. These can be different places, and none tells you how widely the whole crowd is spread.

Worked example

Seven recorded waiting times, in minutes, are 2, 3, 3, 4, 5, 6, and 26. Their sum is 49, so the mean is 49/7 = 7 minutes. The fourth observation is four, making the median four minutes. Three occurs twice while every other value occurs once, so the mode is three minutes.
Now replace the final value, 26, with 96. The sum increases by seventy to 119, and the mean becomes seventeen minutes. The median remains four and the mode remains three. Their stability in this example shows resistance to this particular change, not proof that they can never change when data are altered.
Consider a second dataset: 0, 0, 10, and 10. Its mean and conventional median are both five, but nobody has a value of five. Its two empirical modes are zero and ten. Reporting only five would hide the two separated groups. Reporting the distribution makes that structure visible.

Illustrative case study

Fictional case: Brook Service reports a single average resolution time to its management team. The word average is left undefined, causing different departments to compare different figures. A review also finds a small set of prolonged cases that matters operationally.
The team specifies mean and median explicitly, adds a high percentile, and separates major case types. It preserves long cases after verifying them rather than deleting them to improve the summary. This fictional example illustrates clearer reporting; it does not claim a particular statistic automatically produces better service.

Common misconceptions and mistakes

Do not assume the mean is always the most representative number or that the median is always preferable whenever data are skewed. The decision question still matters. For example, the mean connects directly with a total divided by a count.
Do not assume symmetry alone forces mean, median, and mode to coincide. A symmetric distribution can have two separated modes, and some population distributions lack a finite mean. Statements about all three coinciding need additional conditions.
Do not mistake absence of repeated continuous sample values for proof that the underlying population lacks a density mode. Do not average group medians to obtain an overall median, or average group means equally when the intended observation weights differ.

Business use

Specify the population, observation unit, time period, weighting, and missing-data treatment before choosing a central summary. A customer-weighted mean can differ from a transaction-weighted mean even when both calculations are correct.
For business communication, give the chosen measure its proper name and units. Pair it with spread, sample size, and relevant segments. When presenting model predictions, also identify whether the training objective targets a conditional mean, median, or another quantity; a forecast described only as typical can conceal a consequential difference.

Check your understanding

for 1, 2, 2, 3, and 12, what are mean, median, and mode?

Answer: four, two, and two.

can a set of unordered product categories have a mode?

Answer: yes, the most frequent category.

do equal means imply similar distributions?

Answer: no; spread, shape, and subgroup structure can differ substantially.

Related terms

Median
Mode
Quantile
Dispersion
Skewness
Arithmetic mean
Weighted mean
Multimodal distribution
Measurement scale

Learn next

Next, compare measures of spread and read histograms alongside central summaries. Then study weighted aggregation and how prediction losses determine which aspect of a conditional distribution a model estimates.

Conditional distribution

Sources

Australian Bureau of Statistics, Measures of central tendency: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/measures-central-tendency NIST/SEMATECH, Measures of Location: https://www.itl.nist.gov/div898/handbook/eda/section3/eda351.htm

Previous: Measurement model Next: Mechanism design