Descriptive Statistics Calculator
Paste a data set and get every summary statistic at once: where the centre lies (mean, median, mode) and how the data spread around it (range, standard deviation, quartiles). Perfect for grades, measurements, readings or any data you need to describe quickly.
Separate the numbers with spaces, line breaks or semicolons (a comma between values is tolerated too). The standard deviation is shown both as a population figure (÷ n) and as a sample figure (÷ n − 1).
Your data
Result
- Mean 22.500
| Count n | 10 |
| Sum | 225.000 |
| Mean (arithmetic) | 22.500 |
| Median (Q2) | 21.000 |
| Mode | 15.000 (×2) |
| Minimum | 12.000 |
| Maximum | 41.000 |
| Range (max − min) | 29.000 |
| 1st quartile Q1 | 15.750 |
| 3rd quartile Q3 | 26.500 |
| Interquartile range (Q3 − Q1) | 10.750 |
| Variance (population, ÷n) | 67.450 |
| Standard deviation (population, σ) | 8.213 |
| Variance (sample, ÷n−1) | 74.944 |
| Standard deviation (sample, s) | 8.657 |
| Coefficient of variation | 36.50 % |
mean = (Σxᵢ) / n · σ = √[ Σ(xᵢ − mean)² / n ]
Measures of location: mean, median and mode
A data set is summarised by two families of numbers. Location says where the centre sits. Spread says how far the values scatter. Get those two right and you already have a picture you can work with — a class, a lab run, a column exported from a spreadsheet.
The mean is the sum divided by the count. Everyone knows it; plenty of people over-trust it. One extreme value — a mistyped 410 instead of 41, a CEO salary dropped into a staff list — drags the average with it. The mean is the centre of mass, not the “typical” observation. Put simply, it answers: if we piled every value on a number line, where would the pile balance?
mean = (Σ xᵢ) / n
The median is the middle of the sorted list: half the observations sit below it, half above. Swap the 41 for 410 and the median barely twitches, because only rank matters, not how far the tail runs. That is why a “typical” salary, house price or commute time is almost always a median. The U.S. Census Bureau does the same for household income — more on that below. On an even count the median is the average of the two central values, which is what this page does.
The mode is the value, or values, that show up most often. On the default series 12 15 15 18 20 22 25 27 30 41 the mode is 15, twice. If every observation is distinct, there is no mode worth naming. Modes earn their keep on discrete or categorical data: most common shoe size, most common error code, the grade that appears on the most scripts. A set can be bimodal. This calculator lists every value that ties for the top frequency.
- Mean — centre of mass; uses every value; sensitive to outliers.
- Median — 50th percentile; uses rank; robust to extremes.
- Mode — most frequent; can be multimodal, or absent.
For a symmetric unimodal set the three sit on top of each other. For a right-skewed set — income, city sizes, waiting times — you get mode < median < mean. That ordering is itself a diagnostic. Paste the numbers, read the three, and you already know whether a tail is pulling.
How the values spread: range, standard deviation, IQR
Location without spread is a half-story. Two classes can share a mean of 72 and still be nothing like each other: one packed between 68 and 76, the other from 40 to 99.
The range is max − min. Fast, and it depends on two points alone. Change the 41 to 50 and the range jumps; the rest of the data have not moved. Fine as a first glance. Not what you quote in a report if anyone will act on it.
Variance is the average of squared deviations from the mean. Standard deviation is its square root, so it lives in the same units as the data — points, millimetres, dollars. A small σ means a tight cluster. A large one means a wide scatter. That said, the number is only as honest as the mean it is built around: if the mean is already being yanked by a tail, σ inflates with it. Pair the two, always.
σ = √[ Σ(xᵢ − mean)² / n ] · s = √[ Σ(xᵢ − mean)² / (n − 1) ]
Quartiles cut the sorted set into four slices of equal count. Q2 is the median. Q1 is the 25th percentile, Q3 the 75th. The interquartile range (IQR = Q3 − Q1) is the width of the middle half. Extremes do not enter the IQR, which is why box plots and “robust spread” talk in IQR rather than in σ. On the default ten values this page gives Q1 = 15.75, Q3 = 26.5, IQR = 10.75.
The coefficient of variation is σ relative to |mean|, as a percent. It lets you compare spread on different scales: exam scores out of 20 versus out of 100, or two sensors in different units. When the mean is zero the CV is undefined — the page prints a dash rather than invent a ratio.
- Range — two extremes; cheap and brittle.
- σ / s — typical distance from the mean, same units as the data.
- IQR — middle 50 %, ignores the tails.
- CV — relative spread; compare sets that do not share a scale.
Population or sample? n versus n − 1
If the numbers in the box are the whole population you care about — every student in this class, every batch this week, all 50 states — divide the sum of squares by n. That is σ, the population standard deviation.
If they are a sample from something larger, and you want to estimate that larger spread, divide by n − 1. That is Bessel's correction, and the result is s. Using n would systematically undershoot the population variance; the n − 1 version is unbiased for the variance. (The SD itself stays slightly biased. Textbooks mention it, then move on. For the work this page does, s is the sample number you want.)
AP Statistics, intro stats, the TI-84's 1-Var Stats: they all print both. σx is the ÷ n column. Sx is the ÷ n − 1 column. The exam wants Sx whenever the stem says “a random sample of…”. In practice the numerical gap is tiny once n is in the hundreds, and it is the whole story when n = 8. This page shows both so you can pick, rather than guess which convention was used.
- Census of a class, a day's production, a closed list — use ÷ n (σ).
- Survey, experiment, random sample meant to stand for a wider group — use ÷ n − 1 (s).
- Unsure — report both, as this page does, and say which one you will use downstream.
Mind you, the mean is always ÷ n. Bessel's correction is only for variance and standard deviation. Nobody divides the mean by n − 1. A common homework mix-up is to “correct” everything; leave the average alone.
Quartile conventions: R-7, Excel INC vs EXC
There is no single “the” quartile. Hyndman and Fan listed nine interpolation recipes. This calculator uses linear interpolation, method R-7: the same default as Excel's QUARTILE.INC and PERCENTILE.INC, as R's quantile(..., type = 7), and as recent NumPy. The position of percentile p is (n − 1)p, then the page interpolates between the two neighbouring order statistics.
h = (n − 1) · p · Q = x(⌊h⌋) + (h − ⌊h⌋) · (x(⌊h⌋+1) − x(⌊h⌋))
Excel also ships QUARTILE.EXC / PERCENTILE.EXC, the exclusive method. It places p at (n + 1)p and needs n ≥ 4 before Q1 and Q3 even exist. Tukey's hinges, used by some box-plot routines, split the sample at the median and take the medians of the two halves. On a large set they sit on top of each other. On ten observations they can differ by a point or two — enough to fail a “match the key” check, not enough to change a decision.
Several conventions exist for quartiles. This page uses linear interpolation (R-7, Excel QUARTILE.INC, NumPy). Exclusive methods and Tukey's hinges may give slightly different Q1/Q3 on small sets. Match the convention of whoever will read the number before you hunt a bug.
On the default ten values, R-7 gives Q1 = 15.75 and Q3 = 26.5. A lab partner on EXC will not agree. Google Sheets' QUARTILE follows INC. If a screenshot from another tool disagrees with this page by a few tenths, check INC versus EXC first.
Why U.S. household income is reported as a median
Every September the Census Bureau publishes household income. The headline is a median, not a mean: $80,610 in 2023 (CPS ASEC, 2023 dollars). The mean sits well above $100,000. Both numbers are true. They answer different questions.
Income is right-skewed. A long tail of high earners pulls the mean up; the median stays with the household in the middle of the sorted list. Quote the mean and you describe a country richer than the one most families actually live in. Quote the median and you describe the typical household. Journalists who write “average income” and then paste the mean are mixing the two — on purpose or by accident; either way the reader is misled.
The same split shows up in house prices (the NAR median existing-home price), in salaries inside one firm, in hospital length of stay, in anything with a hard floor near zero and no real ceiling. Whenever a handful of large values can exist, ask for the median first. Use the mean when you will multiply by n — total payroll, total tax base, expected value in a model. All the same, do not pick the one that tells the nicer story.
- Typical household, typical worker, typical sale — median.
- Total dollars in the system — mean × n.
- Skew diagnostic — report both; mean ≫ median flags a right tail.
A normal distribution is symmetric, so its mean and median coincide. Income is a standard example of a quantity for which the bell is the wrong picture. The median is not a timid substitute for the mean. It is the right centre for that shape.
The 1.5 × IQR rule for outliers
Tukey's rule of thumb — the one drawn on a box plot — flags a point as a mild outlier if it sits more than 1.5 × IQR below Q1 or above Q3, and as an extreme outlier past 3 × IQR.
lower fence = Q1 − 1.5 × IQR · upper fence = Q3 + 1.5 × IQR
On the default series, IQR = 10.75, so the fences sit at 15.75 − 16.125 = −0.375 and 26.5 + 16.125 = 42.625. The 41 is inside. It is the maximum, not an outlier. Replace 41 with 50 and it crosses the fence. The arithmetic is mechanical. The judgement is not.
The rule is a flag, not a death sentence. A legitimate 50 in a class of 10 is still a 50. Do not delete it because a formula frowned. Do look at it: transcription error, different population, heavy tail. In a quality-control chart you investigate. In a grade book you keep it and maybe switch from mean to median for the “typical” score.
The 1.5 is a convention Tukey liked because, on a true normal sample, it flags roughly 0.7 % of points as outliers in total (about 0.35 % on each side) — rare enough to be interesting, common enough to show up in homework. Some fields use 3 × IQR and ignore the mild band. State which fence you used. “We dropped outliers” with no rule attached is not analysis.
Weighted mean: when not every row counts equally
The ordinary mean treats every observation as 1. Plenty of real averages do not. A GPA weights each course by credit hours. A survey weights each respondent by how many people they stand for. A price index weights each item by its expenditure share. The formula is the same idea as the mean, with weights wᵢ in both the numerator and the denominator.
x̄w = (Σ wᵢ xᵢ) / (Σ wᵢ)
If every weight is 1 you recover the ordinary mean. If you average already-averaged groups — class means of different sizes, monthly rates of different lengths — and you forget the weights, you get Simpson's trap: a number that is not the mean of the pooled data. Three sections of 20 students at 70 % plus one section of 80 at 90 % is not a class mean of 80 %.
This page does not take weights. Paste the raw values, repeated as they occur, and the unweighted mean is correct. For a GPA or a weighted index, multiply each value by its weight yourself. The median of a weighted sample is a different object (you walk the cumulative weights until 50 %). Do not mix it with this page's unweighted median.
Still, a lot of “the average American…” claims are weighted, whether the author says so or not: household surveys, labour-force statistics, index funds. When you reuse a published average, ask what the weights were. Unweighted rows of a convenience sample are not the same object.
Reading a box plot from the five-number summary
A box plot is the five-number summary drawn as a picture: minimum, Q1, median, Q3, maximum — with the 1.5 × IQR fences deciding which “maximum” is a whisker and which is a plotted outlier. From this page you already have every piece. Sketch the box and you can see skew that a single mean hides.
- The box runs from Q1 to Q3; its width is the IQR, the middle 50 %.
- A line (or a gap) inside the box is the median. Off-centre median ⇒ skew.
- Whiskers reach the most extreme points still inside the fences.
- Points beyond the fences are plotted individually — the 1.5 × IQR rule, drawn.
A long right whisker with the median hugging Q1 is the income shape again. A symmetric box with equal whiskers is the picture a bell curve would draw. Neither picture shows sample size, nor multimodality. Two well-separated clumps can produce a respectable-looking box. Pair it with a histogram or a dot plot when n is small enough to stare at. When n is huge, the box is the summary you can actually hold.
Software disagrees about whiskers. Some always draw min to max (a “quartile plot” without outlier dots). Some use 1.5 × IQR. A few use 3 × IQR. If you publish the figure, say which. The numbers on this page stay the same either way; only the drawing convention changes.