Five deliveries take 2, 3, 3, 4 and 28 days. Their mean is 8 days, but four of the five arrive within 4 days. If a customer asks how long delivery usually takes, “8 days” gives the wrong impression. The median and mode are both 3 days, and each answers a different version of the question: What’s the middle delivery time? What delivery time occurred most often?

That’s why choosing an average starts with the question, not the calculation. The mean reflects every value and is useful when the total matters. The median describes the midpoint without letting an unusually large or small value dictate the result. The mode identifies the most frequent value or category. None is the right answer for every dataset.

Illustrative image: mean median mode

Mean, median and mode at a glance

Measure How to find it Best suited to Main limitation
Mean Add the numerical values and divide by how many there are Summarizing totals per observation, or numerical data without influential extremes Unusually high or low values can pull it away from what most observations look like
Median Sort the values and find the middle; with an even count, average the two middle numbers Describing a typical value when data are skewed or contain outliers Says little about how far values lie from the middle
Mode Find the value or category that occurs most often Identifying the most common size, choice, response or repeated number May be uninformative when values rarely repeat or several values tie

These are three measures of central tendency: ways to summarize where data sit. They aren’t interchangeable. Before selecting one, decide whether “typical” means a share of a total, a midpoint or the most common observation.

Choose the mean when every value should count toward the result

The arithmetic mean is the familiar add-and-divide average. For the five delivery times, the calculation is (2 + 3 + 3 + 4 + 28) ÷ 5 = 8 days. The 28-day delivery has as much right to be in that calculation as any other: it happened. The issue is what the result is meant to describe.

Suppose a delivery manager needs to report the total time spent across deliveries as a per-delivery figure. The mean does exactly that. Multiplying the mean of 8 days by five deliveries recovers the total of 40 days. The median cannot be used that way. Likewise, if a teacher needs a class’s total points divided by the number of students, the mean answers the question directly.

The mean is also a clear choice when numerical values are reasonably balanced on either side of the center and no small number of extreme observations dominates. For 2, 3, 3, 4 and 3 days, the mean is 3 days; it gives a fair picture of the group. In a roughly symmetric, single-peaked distribution, the mean and median will often be close. NIST’s discussion of measures of location also shows why that agreement can break down when data have long tails or extreme values.

An outlier does not make the mean incorrect. It can make the mean unsuitable for a particular claim. If the aim is to describe what a customer is likely to experience, “average delivery: 8 days” hides the fact that most of these deliveries were much faster. If the aim is to account for all 40 delivery-days, leaving out the 28-day result would hide something else.

Choose the median when the middle matters more than the extremes

To find the median, put the values in order. The delivery times already are: 2, 3, 3, 4, 28. The third value, 3 days, is the median. If the longest delivery took 50 days instead of 28, the median would still be 3 days. The mean would rise from 8 to 12.4 days.

The median is therefore often the better headline for a “typical” home price, salary or wait time when a few values are much higher or lower than the rest. It doesn’t ignore those observations; it answers a question about position rather than magnitude. OpenStax’s explanation of the median and outliers makes this distinction clear: extreme values can greatly change a mean without changing which observation sits in the middle.

With an even number of observations, there is no single middle entry. For delivery times of 2, 3, 4 and 28 days, average the two middle values: (3 + 4) ÷ 2 = 3.5 days. That result need not be a delivery time anyone experienced.

The median is resistant to the size of an extreme value, not immune to every change in the data. Add or remove observations and the middle position may move. Nor does a median describe the total or reveal whether values are tightly packed. Two groups can share a median even if one has far more unpredictable delivery times.

Choose the mode when frequency is the point

The mode of 2, 3, 3, 4 and 28 days is 3 days because it occurs twice; every other value occurs once. Here the mode agrees with the median, but that agreement is not guaranteed.

Frequency is especially useful when the data are categories. A store choosing which shoe size to stock most heavily needs to know which size sells most often, not the arithmetic mean of sizes. A survey team may want the most selected answer. For categories such as preferred payment method, a mean has no meaningful interpretation, while a mode does.

The mode needs more care with measurements. If recorded delivery times are 2.9, 3.0, 3.1 and 3.2 days, no exact value repeats. Yet the times plainly cluster near 3 days. Rounding them or grouping them into intervals might reveal a useful peak, but the answer will depend on the chosen rounding or interval widths. State those choices rather than presenting a grouped mode as an exact observed value.

A dataset can also have two or more modes. If two shoe sizes tie for the highest sales, report both rather than inventing a single winner. And “most common” does not necessarily mean “common to most”: a mode chosen by 30% of survey respondents still leaves 70% choosing something else. The mode can summarize numerical values or categories, but its frequency should be reported when that context matters.

Let skew and outliers inform the choice—not dictate it

Skew describes a distribution with a longer tail on one side. In right-skewed data, a few high values can pull the mean above the median. With a longer left tail, the mean is often below the median. The delivery example has a long right tail: one 28-day result pulls the mean to 8 days while the median stays at 3.

That pattern is a clue, not a rule for choosing one number. A long tail can contain meaningful observations. For budgeting across all deliveries, the mean captures their full effect. For telling customers what the middle delivery experienced, the median is more useful. OpenStax’s examples of skewed distributions show how the mean tends to move toward the tail; what to report still depends on the question.

Don’t delete an outlier just because it makes the mean awkward. First check whether it is a recording mistake, an observation from a different group or a genuine but unusual result. A mistakenly entered 280-day delivery should be corrected. A real 28-day delay belongs in the data, even if the median is the better measure of a typical delivery.

It’s also possible for one central number to describe neither of two distinct groups well. Imagine a service with many 2-day deliveries and many 20-day deliveries, but few in between. A mean near 11 days might describe almost nobody’s experience. Separate summaries for the two delivery groups—or a display of the full distribution—would be more helpful than arguing over a single average.

Check what the number leaves out

A quick selection process works better than a blanket rule such as “use the median for outliers”:

  1. Identify the data. For numerical measurements, consider the mean and median. For categories with no natural numerical meaning, look at frequencies and the mode.
  2. Name the question. If you need a total divided across observations, use the mean. If you need the middle observation, use the median. If you need the most frequent value, use the mode.
  3. Inspect the distribution. Check for long tails, isolated extremes, ties and separate clusters before settling on a headline number.
  4. Show context when one number is thin. A median alongside a range, or a mean alongside a median, can reveal more than either alone.

Be careful with ordered categories, too. Survey responses such as “poor,” “fair” and “good” have an order, so a middle response may be meaningful. But assigning them numbers and calculating a mean assumes the gaps between responses have numerical meaning. For categories with no order, such as payment method, even a middle category is undefined. The distinction between ordered and unordered categories matters before any average is calculated.

Finally, an average is only as useful as the data and display around it. Five deliveries are enough to demonstrate the calculations, not enough to establish a dependable service-wide pattern. When results are shown in a chart, check the axes, scale and missing context as well as the headline average.

The useful question isn’t “Which average is best?” It’s “What do I need this number to tell someone?” Use the mean for the contribution of every numerical value, the median for the midpoint and the mode for the most frequent result—and show more of the data when none tells the story alone.