Box Plot Calculator — Five Numbers

A box plot compresses a whole dataset into five numbers and a map of its troublemakers. Paste your data below and this calculator returns the five-number summary — minimum, first quartile, median, third quartile, maximum — plus the interquartile range and every point the 1.5 × IQR rule flags as an outlier worth investigating.

Five Numbers, Honestly Computed by Hand’s Method

John Tukey designed the box plot in the 1970s as exploratory data analysis — a sketch you could compute by hand that shows a distribution’s centre, spread, skew and suspects at a glance. The computation this calculator performs is Tukey’s own: sort the data; the median splits it; the first and third quartiles are the medians of the lower and upper halves; the interquartile range (Q3 − Q1) spans the middle half of your data; and the fences at Q1 − 1.5 × IQR and Q3 + 1.5 × IQR mark the boundary beyond which points are flagged as outliers — not deleted, not condemned, flagged for the investigation the note demands. The box in the eventual drawing runs Q1 to Q3 with a line at the median, and the whiskers reach to the most extreme points still inside the fences. Everything else about the plot — its orientation, its grouping, its comparative rows of boxes — is presentation; these five numbers and the fences are the substance, and they are what this page computes, exactly, from whatever data you hand it.

Worked Example: Ten Test Scores

The default data — 12, 15, 14, 10, 18, 22, 9, 16, 14, 20 — sorts to 9, 10, 12, 14, 14, 15, 16, 18, 20, 22. The five-number summary reads off directly: minimum 9, maximum 22, median 14.5 (the average of the middle pair 14 and 15), Q1 12 (median of the lower five), Q3 18 (median of the upper five). The IQR is 18 − 12 = 6 — half the class scored within a 6-point band — and the mean, printed for comparison, is exactly 15.0, slightly above the median because the 22 pulls it that way (a one-line lesson in skew, delivered by two statistics disagreeing politely). The fences sit at 12 − 9 = 3 and 18 + 9 = 27, so nothing flags: this class has no outliers, just a spread. Change one score to 38 and watch the machinery earn its keep — the median barely shrugs while the max and the outlier flag tell the story. That shrug is the entire argument for robust statistics in one gesture: the centre holds its nerve while the fences do the shouting, each statistic doing the job it was built for.

The Quartile Methods Disagreement (Read This Before Comparing Software)

Quartiles are the rare elementary statistic with no universal definition, and honest tooling admits it. Tukey’s median-of-halves (used here) is the classical hand method; Excel’s QUARTILE.INC interpolates percentiles linearly; QUARTILE.EXC uses a different plotting position; R alone offers nine quantile types. On large samples the methods converge to trivia; on a class of ten they can differ by a point — enough to make two correct programs appear to contradict each other. The defence is the one this page practices: name the method (Tukey, in the note, in the FAQ, everywhere), reproduce it exactly, and let the reader audit. For the box plot’s purpose — a robust sketch — the differences matter less than the fences’ stability, and any reader comparing tools should compare like methods with like, a discipline the internet’s calculator ecosystem would benefit from adopting wholesale.

Comparing Groups: The Plot’s Real Career

One box is a summary; five boxes side by side are an analysis. The grouped box plot — a row of boxes per class, treatment, month or machine — is the working scientist’s fastest honest comparison: medians line up at a glance, spreads confess which group is consistent and which is chaos, skew shows in each box’s asymmetry, and outliers identify themselves for follow-up by group rather than polluting a pooled verdict. The five numbers per group come from exactly the computation this page performs, repeated per group with the discipline that the same method and the same fences govern every box (mixed methods make mixed pictures, and the comparison becomes theatre). Run your groups through the calculator one at a time, tabulate the five numbers, and the chart you draw from them will be Tukey’s own — statistics as sketching, with the arithmetic showing and the conclusions earned.

Notches, Violins and the Plot’s Descendants

Tukey’s sketch has bred a family, and meeting two heirs clarifies the original. The notched box plot cuts a waist into the box around the median (roughly median ± 1.58 × IQR ÷ √n) as a built-in significance visual — boxes whose notches do not overlap differ in median at about the 95% level, turning the sketch into a test you can read across a room. The violin plot wraps the box in a mirrored density curve, restoring the shape information the five numbers discard (bimodality above all — two humps that no quartile can confess). Both descendants still carry Tukey’s five numbers at their core, computed exactly as this page computes them; the family’s growth is a tribute to the sketch’s honesty about what it omits, and the calculator’s five numbers remain the entry ticket to every fancier drawing.

Related Tools

More from the Statistics Calculators hub: the Descriptive Statistics Calculator for the full moment-by-moment account of the same data, and the IQR Calculator if your dataset already lives there.

Frequently Asked Questions

What does the box in a box plot actually show, then?

The middle 50% of the data: the box’s bottom edge is Q1 (a quarter of the data lies below), its top edge is Q3 (three quarters below), and the line inside is the median. Points plotted beyond the whiskers are the fence-breakers — the outliers this calculator lists by value. The visual grammar is why the plot survives: five numbers and a suspect list, readable across a room, comparable side by side for as many groups as the page holds — the grouped comparison (classes, treatments, months) is the box plot’s killer application, and its numbers are all computed exactly as this page computes them.

Are outliers errors I should simply delete?

Almost never reflexively. The fence rule flags points for investigation: some prove to be data-entry errors (the 380 that was 38.0), some are genuine extremes (the day the website’s traffic tripled), and deleting genuine extremes fabricates a tidier world than the one your decisions must live in. The professional sequence is flag, investigate, document, decide — with sensitivity checks (does the conclusion change without them?) rather than silent removal. Tukey, who invented the flag, treated outliers as the most interesting observations in the dataset; the calculator’s note channels him deliberately.

Why 1.5 × IQR for the fences, exactly?

Tukey’s choice, calibrated to roughly the ±2.7 sigma region of a normal distribution — outside which a Gaussian point is genuinely unusual (about 0.7% of normal data flags) while heavy-tailed real data flags at the informative rate the exploratory method wants. A stricter 3 × IQR fence exists in the same tradition and is sometimes drawn as the “far out” boundary. The 1.5 figure is a convention with a rationale, not a law of nature, and the calculator prints the fences’ values so you can apply any multiple your field prefers to the same quartiles.

Median or mean — which centre should I report, then?

The box plot’s whole philosophy votes median: it ignores the outliers the fences flag, so a single wild point cannot drag the reported centre the way it drags the mean (the worked example’s 15.0 versus 14.5 is the gentle version; incomes and house prices are the violent one). Report both when they disagree — the gap between them is itself a statistic, the signature of skew — and reach for the mean when the total matters (budgets divide by means; typical values are medians). The calculator prints both precisely so the argument can happen in the open.

How many data points do I need for a box plot, minimum?

The calculator insists on four (below that, quartiles are theatre), and the sketch becomes meaningful around ten to twenty; the method’s robustness grows with n while its charm — the hand-checkable five numbers — survives any size. Very large datasets keep the plot honest (the fences fill with the expected fraction of a big sample’s tails) and very small ones deserve the caution that two of their five numbers are single observations. As ever, the summary is a beginning: plot the points themselves beside the box whenever the audience’s decision rides on the shape.

What if my data has many repeated values in it?

Ties are handled naturally by the median machinery (the worked example’s twin 14s sit at the centre without fuss), but a dataset with heavy repetition — survey Likert scales are the classic — can produce a zero IQR when more than half the responses share one value, collapsing the box to a line and flagging everything else as an outlier. That result is the method truthfully reporting that this data wants a bar chart of frequencies instead: ordinal categories are counted, not quartiled. Knowing which summary a dataset deserves is the statistical literacy the calculator’s FAQ, at its best, quietly teaches.