How Do Outliers Affect Averages? Mean vs. Median

How do outliers affect averages?

Imagine asking how long a routine trip usually takes. Most trips may be similar, while one takes much longer because of a road closure. That unusually long trip might be perfectly real, but it can change the answer you get from an average. The important question is not simply whether the trip looks unusual. It is what kind of average you are using and what you want that average to describe.

In everyday speech, average often means the arithmetic mean: add the values and divide by how many there are. Another measure of the center is the median, the middle value after sorting the observations. An outlier is an observation unusually far from the rest. There is no universal distance that makes a value an outlier in every setting; context and the distribution matter. NIST’s discussion of measures of location distinguishes how measures of center respond to unusual values.

A worked example with synthetic trip times

Consider a synthetic, illustrative set of trip times, in minutes: 40, 42, 44, 46, and 48. Its mean is (40 + 42 + 44 + 46 + 48) ÷ 5 = 44 minutes. Its median is also 44 minutes, because 44 is the middle value when the times are ordered. These figures are invented solely to show the calculation; they are not measurements from a real route.

Now imagine a synthetic, illustrative alternative in which the last trip takes 128 minutes instead of 48. The ordered times are 40, 42, 44, 46, and 128 minutes. The mean becomes (40 + 42 + 44 + 46 + 128) ÷ 5 = 60 minutes, while the median remains 44 minutes. Replacing the illustrative 48-minute trip with the illustrative 128-minute trip raises the mean by 16 minutes. The middle position has not changed, so neither has the median.

The illustrative 128-minute trip should not automatically be deleted. Perhaps a road closure made it genuinely slow. Perhaps a recording error put the wrong time in the dataset. Those explanations call for different actions. The calculation tells us how the summaries respond; investigating the observation tells us what the data mean.

Why the mean moves and the median may not

Every observation contributes its numerical value to the mean. When one value rises substantially and the number of observations stays the same, the total rises, so the mean rises too. This sensitivity can be useful: if the question concerns the total time people spent traveling, a genuinely long trip matters. But it can make the mean a poor description of what a person should expect on a routine trip.

The median works differently. It is determined by position in the sorted data. Moving an already-high value farther upward does not necessarily change which observation sits in the middle. That makes the median less sensitive to the size of an extreme observation. It does not make the median immune to every change: adding observations, removing them, or changing their order can move it. OpenStax’s explanation of the mean and median describes this distinction.

Neither measure is automatically the correct one. A mean answers a question about the combined value spread across observations. A median identifies the midpoint of the ordered observations. They can both be useful, but they are not interchangeable descriptions of a dataset.

Check the value before choosing what to report

First, check whether the unusual observation belongs in the dataset. Look for a unit mismatch, a transcription mistake, a duplicate, or a measurement taken under different conditions. Correct a documented error when the original record supports a correction. Do not change a valid value merely because it makes the summary less tidy. A rare but genuine event is still part of the experience the data may need to represent.

Next, ask what the audience needs to know. If the purpose is to describe a typical observation in a skewed set, the median may be easier to interpret. If the purpose is to allocate resources based on the combined amount, the mean may be essential. For travel times, someone planning when to leave may care about the usual trip and the possibility of a delay. Reporting only one measure can hide part of that story.

It can help to show both the mean and median, alongside the spread of the observations and a plain-language note about the unusual value. A large gap between the two centers is a reason to inspect the distribution, not proof that a particular record is wrong. The choice of summary should follow the question and the quality of the data.

What changes when the dataset changes?

The influence of an outlier on the mean depends partly on how many observations share the calculation. The same unusual value generally has more influence when there are fewer other observations. Its position also matters for the median: an extreme value can leave the middle observation unchanged, while a change near the middle may alter it. That is why a statement such as “outliers do not affect the median” is too broad.

The meaning of “unusual” depends on the setting as well. A long journey may be routine on one route and exceptional on another. Pooling unlike routes can make an ordinary observation look extreme, or conceal a meaningful delay. Before interpreting a surprising value, confirm that the observations are comparable and that the measurement describes what the analysis claims it describes.

A useful sensitivity check is to calculate the summary with and without the questioned observation, while retaining the original data and clearly labeling the comparison. This shows how strongly a conclusion depends on that record. It is an investigation, not permission to discard inconvenient results. If the observation is valid and relevant, the final account should explain its effect.

A practical checklist

When an average seems surprising, identify which measure is being reported. Inspect the individual observations or a plot, check the unusual value against the original record, and consider whether the observations belong together. Then choose a summary that answers the actual question: the mean for a combined amount distributed across observations, the median for the midpoint, or both when each adds useful context.

Before sharing the result, make the treatment of unusual values explicit. Say whether a value was verified, corrected, excluded for a stated reason, or retained. If the mean and median tell noticeably different stories, report that difference rather than smoothing it away. The goal is not to find an average untouched by inconvenient observations; it is to give readers an accurate account of what happened and what a typical value can, and cannot, tell them.