Mean, Median, Mode and Standard Deviation: Choosing the Right Summary
A page of numbers is not information. Thirty test scores, two hundred reaction times, a season of rainfall — each is raw until it is summarised, and the summary is a choice with consequences: the mean says one thing about a class, the median another, and the standard deviation decides whether either of them can be trusted. Students are taught to compute all four and rarely taught to choose among them, which is the only part the real world asks for.
This article covers what each summary measures, the situations where each misleads, and the population-versus-sample distinction in standard deviation that syllabuses love to test and textbooks love to mumble.
Key takeaways
- The mean shares the total equally; the median locates the middle; the mode reports the most frequent — three different questions, three different 'averages'.
- Outliers drag the mean and leave the median unmoved: skewed data (incomes, scores with one disaster) belongs to the median.
- Standard deviation is the typical distance from the mean — small means consistent, large means scattered.
- Population SD divides by n; sample SD divides by n−1, correcting the bias of estimating from a sample.
Three averages, three questions
The mean answers: if this total were shared equally, what would each get? Add the values, divide by the count. The median answers: what does the middle case look like? Sort the values and take the centre one (or the mean of the central pair). The mode answers: what happens most often? Count frequencies and take the winner, if there is one. Each is honestly called an average in everyday speech, which is exactly why statistics insists on the names: 'the average score was…' is an unfinished sentence until the speaker says which average, and on contested data — wages, house prices, exam results — the choice of average is often the argument itself.
The mean's weakness is its fairness: it shares everything, including the freak values. One score of 8 among a class of 80s drags the mean down noticeably while describing nobody — not the class, not even the unfortunate student, whose other scores were fine. The median walks past the freak undisturbed, which is why official statistics quote median incomes and median house prices: half above, half below, immune to billionaires and disasters alike. The working rule: when data is roughly symmetric, mean and median agree and either serves; when they disagree noticeably, the data is skewed, the median is the honest centre, and the gap itself is a finding worth reporting.
A centre without a spread is half a description. Two classes can share a mean of 70 while one scored uniformly 65–75 and the other swung between 30 and 100; teaching them 'the same' would be a category error, and so would reporting them the same. Range (max − min) gives the crudest spread in one subtraction. Standard deviation gives the refined one: roughly, the typical distance of values from their mean, computed by squaring each deviation (killing the signs), averaging, and square-rooting back. Squaring punishes distant values extra, so SD notices outliers the range merely includes.
The population/sample divide lives inside that averaging step. If your data is the entire group you care about — this class's scores, full stop — divide the squared deviations by n: the population standard deviation. If your data is a sample standing in for a larger group, the sample's own mean is itself an estimate that sits suspiciously close to the sampled values, biasing the spread downward; dividing by n−1 instead inflates the result just enough to correct the bias. School syllabuses test the distinction by name (σ versus s) and by denominator; data software often defaults to the sample version, which surprises students comparing hand calculations of class data.
The mode earns its keep where numbers are not really quantities: shoe sizes stocked by a shop, the most common blood type in a clinic, the modal exam grade in a report to governors. It is the only average available for categories, and the honest reports admit its limits — data with no repeated value has no mode, data with two peaks is bimodal and the pair should be reported, because forcing one 'most common' onto a split dataset hides the split, which is usually the story.
In practice, a competent summary is a sentence with four numbers in it: count, a centre chosen for the data's shape, a spread, and a note on skew or outliers when present. 'Twenty-eight scores: median 74, mean 71, SD 9 — one very low score pulls the mean below the median.' That sentence could run a parents' evening. Producing it takes a minute with a calculator and a lifetime's worth of judgement about which figures the situation demands — judgement that is, fortunately, mostly rules that can be learned, and this article has just listed them.
A worked example
A class's scores: 72, 85, 90, 85, 64, 91, 85, 78. Count 8. Mean: 650 ÷ 8 = 81.25. Sorted: 64, 72, 78, 85, 85, 85, 90, 91 — median = (85 + 85) ÷ 2 = 85. Mode = 85 (three times). Range = 27. The median and mode agree at 85 while the mean trails at 81.25: the low pair (64, 72) drags it, exactly the skew signature. Population SD ≈ 8.7. Reported honestly: 'a typical score is 85; the mean is lower because of two weak results' — a sentence a teacher can act on.
Mistakes that cost marks
Reporting the mean for skewed data because it is the famous one. If mean and median differ by a lot, the mean is describing the outliers' influence, not the typical case. Compute both; report the one that answers the question, and mention the other.
Using the sample SD for a whole population (or the reverse). The denominator is not a preference: n for the complete group, n−1 for a sample. Check your calculator's or software's default before comparing figures.
Claiming a mode for data that has none. All-distinct values have no most-frequent value; inventing one from the middle value confuses mode with median and forfeits the mark and the meaning simultaneously.
Questions, answered
Which average do exam boards report for a cohort? +
Usually the mean for marks (it uses every mark's information) alongside distributions and grade boundaries that reveal the shape. For public-facing statistics about people — pay, prices — medians dominate, for the outlier reasons above. The chooser's question is always the same: typical case (median) or equal share (mean)?
What does a standard deviation of zero mean? +
Every value identical — no spread at all. It is the consistency limit, and in real data it usually signals that somebody measured the same thing repeatedly, or that the data needs a second look.
Is a big standard deviation bad? +
It is descriptive, not a judgement: a big SD means scattered values. For a factory's bolt lengths, scatter is failure; for a class's project scores on an open brief, scatter may be the design working. Context grades the number; the number merely reports.
Put it into practice
Summarise a real set of your own scores and watch the mean and median disagree — or refuse to — in the flesh.
Open the Statistics Calculator →