Mean, Median and Mode Calculator

What each one actually measures

They are all called averages, and they answer three genuinely different questions.

The mean is the balance point. If the values were weights on a ruler, the mean is where it balances. Every observation contributes in proportion to its distance, which makes the mean the most informative of the three when the data is well behaved — and the most fragile when it is not. Move one value far enough and the mean follows it anywhere.

The median is the middle. Half the observations sit on each side. It does not care how far away anything is, only how many are on which side, which is why it barely moves when an extreme value appears. Its breakdown point is 50% — the theoretical maximum — and nothing else here comes close.

The mode is the most common. It is the only one of the three that must be a value that actually occurred, and the only one that works on data with no numeric meaning at all. It is also the least stable on continuous measurements, where exact repeats are largely accidental.

A quick illustration with the tool’s presets. On 4, 5, 6, 6, 6, 7, 8 all three are 6. Change the last value to 80 and the median stays at 6, the mode stays at 6, and the mean jumps to about 16 — a “typical” value larger than all but one observation. Nothing is broken; the mean is doing exactly what it is defined to do.

Reading skew straight from the gap

This is the part most calculators leave out, and it is the most useful thing on the page.

Mean above median means a right tail. Something up at the high end is pulling the balance point away from the middle. Incomes, house prices, city populations, response times, insurance claims and word frequencies are all like this — most values modest, a few very large.

Mean below median means a left tail. Rarer, and it shows up in bounded measurements near their ceiling: exam marks on an easy test, time-to-completion for a task nearly everyone finishes, component lifetimes with early failures.

Mean equal to median means roughly symmetric. Not necessarily normal — symmetry is a much weaker claim than normality, and a bimodal distribution can be perfectly symmetric — but at least not lopsided.

Pearson’s second skewness coefficient turns that into a comparable number by scaling the gap: 3(mean − median) divided by the standard deviation. Dividing by s makes it unit-free, so a value of 0.8 means the same thing on salaries in rupees and on marks out of 100. Beyond roughly ±1 is strong skew.

The reason to prefer this over a moment-based skewness is that you can compute it in your head from two numbers you already have, and it is far less sensitive to a single extreme value than a measure built on cubed deviations. It is an indicator, not a test.

The mode problem, and what most tools get wrong

The mode is the most awkward of the three, and the awkwardness is usually hidden rather than reported.

There can be no mode at all. If every value occurs exactly once, no value is most frequent. That is not the same as the mode being zero, and it is not the same as every value being a mode. This page says “none”.

There can be several, and that is usually the headline. Two clear peaks almost always means two populations have been mixed — two machines, two shifts, two customer segments, adults and children. A single “average” then describes nobody, and the right response is to split the data rather than summarise it.

This is where the tooling actively misleads. Excel’s MODE returns one value — the first it encounters — and gives no indication that others tied. A genuinely bimodal dataset looks unimodal, and the single most important feature of the data disappears into a cell. Newer Excel has MODE.MULT, which returns them all, but MODE is what people reach for and what older sheets contain. This page reports every tied value and flags multimodality explicitly.

On continuous data the mode is nearly meaningless. Measure anything precisely enough and no two values repeat, so any mode you do find is an artefact of rounding. For continuous measurements the useful analogue is the peak of a histogram or density curve, which depends on the bin width you chose — a real decision, not a property of the data.

Which one to report

A short decision procedure that covers most real cases.

Symmetric, no extreme values: the mean. It uses all the information and it is what every downstream method — standard deviations, t-tests, confidence intervals — expects as input.

Skewed or containing extremes: the median. This is why national statistics agencies publish median household income and median house prices rather than means. A handful of very large values would otherwise produce an “average” that is above what most people experience, which is technically correct and practically misleading.

Categories, or the most likely single outcome: the mode. The most common blood type, the most-chosen answer, the most frequently ordered size. A mean shoe size is a real number and a useless one for stocking a shop.

When they disagree substantially: report more than one. Giving both the mean and the median costs a few characters and tells the reader immediately that the distribution is lopsided. A summary that quotes only the flattering one is the most common way statistics are used to mislead without anything false being said.

When the mean is actively misleading

Three concrete cases, all common.

Salaries. Load the salaries preset: eight ordinary figures and one large one. The mean lands well above what most of those people earn, so “the average salary here” describes a level nobody in the middle is at. This is exactly why pay transparency reporting uses medians and quartiles.

Averaging averages. The mean of several group means is not the overall mean unless the groups are the same size. Three classes averaging 60, 70 and 80 do not give a school average of 70 if the classes hold 10, 10 and 100 students. That needs aweighted average, and getting it wrong is one of the commonest errors in reporting.

Rates and ratios. The arithmetic mean of growth rates or speeds is wrong. Driving out at 30 km/h and back at 60 km/h does not average 45; it averages 40, because you spend twice as long at the slower speed. Rates need theharmonic mean, and compounded growth needs thegeometric mean.

The pattern in all three: the arithmetic mean assumes every observation should carry equal weight and that adding the values is meaningful. When either assumption fails, the mean is not approximately right — it is answering a different question.

Ordinal and categorical data

Which centres are even defined depends on what kind of data you have, and this is worth being strict about.

Categorical data — blood types, countries, product categories — has no order, so only the mode exists. There is no middle of an unordered set and no meaningful sum.

Ordinal data — rankings, satisfaction scales, small/medium/large — has order but no guaranteed equal spacing. The median and the mode are both well defined. The mean is where it gets contentious: computing 3.7 from a five-point Likert scale assumes the gap between “agree” and “strongly agree” is the same size as the gap between “neutral” and “agree”, which nobody has established.

In practice mean Likert scores are reported constantly and are often useful for comparing groups. The honest position is that it is a pragmatic convention rather than a sound one, and that a median plus the full distribution of responses says more and assumes less.

Interval and ratio data — temperatures, lengths, counts, money — supports all three. Note that the tool will compute a mean for anything you paste, including data where it should not be computed. That judgement is yours, and no calculator can make it for you.

Sources and methodology

Definitions, and the reporting conventions cited above.

• NIST/SEMATECH e-Handbook — measures of location, including the definitions used here
• US Census Bureau income statistics — an example of median rather than mean being the published figure, for the reason this page gives
• Microsoft documentation for MODE — which returns a single value, the behaviour this page deliberately does not copy
• Weighted Average Calculator — for when groups are different sizes and a plain mean is wrong
• How we calculate — methodology

Method. Modes are found by frequency count and every tied value is returned, sorted — the suite asserts that a bimodal set reports both values rather than the first, and that an all-unique set reports no mode rather than an arbitrary one. The skew direction is judged against a threshold relative to the standard deviation rather than an absolute gap, so a difference of 2 is not called skew on salaries and ignored on exam marks; symmetric data is required to report as symmetric rather than being given a spurious direction. Pearson’s second coefficient is checked against a hand calculation. That engine is verified on every change against 44 assertions. The count and the per-case breakdown are published on the formula verification page.

Related calculators

Where this goes next:

More in Math, or browse all calculators.