Box plots compress a distribution into five statistics

What it is

The classic box plot starts from a five-number summary: minimum, first quartile, median, third quartile, and maximum. The box spans the interquartile range, its line marks the median, and whiskers describe the outer range. A Tukey-style plot often stops each whisker at the most extreme observation inside a fence and plots farther observations separately, so its rendered marks are not always determined by only five values. The title highlights the compression: location, central spread, and tail cues remain, while density, modality, gaps, and individual records cannot be reconstructed.

Why it happens

Medians and quartiles partition observations by rank. They are generally more resistant to extreme values than means and standard deviations and give groups a compact shared grammar. Yet identical quartiles can arise from unimodal, bimodal, heaped, or gapped data. Whiskers and outside points describe tails only under the chosen convention. Unequal halves of a box may suggest asymmetry but do not establish population skewness. Compression serves summary comparison by mapping many data-generating processes to nearly the same marks.

Studying it

Generate datasets with matching five-number summaries but different shapes, then compare box plots, raw points, histograms, ECDFs, and violins on modality, tail, location, and spread tasks. Manipulate sample size, skew, mixture proportion, discretization, missingness, and whisker convention. Measure accuracy, time, confidence, and claims that infer shape from summaries. In realistic studies, record statistical experience, display size, and access to a data table. A study testing median lookup alone cannot support conclusions about shape comprehension.

Where it stops holding

A box plot is neither an assumption-free reconstruction of a population nor a normality test. Raw points or density views are more direct for one group when shape matters; compression helps when many groups share limited space and the task concerns location and spread. Discrete, bounded, or zero-inflated data may yield collapsed boxes or overlapping points that reflect real structure. Violins, notches, and variable-width boxes add information but also add bandwidth, interval, or width conventions that require explanation.

Applying it

• Define the box, whiskers, and separately drawn points in the caption; do not call Tukey-fence whiskers the minimum and maximum.
• When shape affects a decision, overlay jittered raw observations or offer per-group histograms or ECDFs. Disclose any sampling of large data.
• Keep units and scales consistent, and provide medians, quartiles, sample sizes, and an accessible data summary without requiring hover.
• Test with cases that share summaries but differ in modality. Claims of “unimodal,” “normal,” or “no gaps” signal that shape evidence is missing.

Related

• Same group: U2.09.2 Whisker and outlier rules are not unique; state them with the chart · U2.09.3 The box plot's strength is comparing many distributions side by side · U2.09.4 Sample size does not show in the box; annotate it separately · U2.09.5 Small-sample quartiles are unstable; box width is uninterpretable
• Nearby: U2.08.2 Box plots hide multimodal structure · U2.08.3 Showing the raw point distribution is more reliable
• Search terms: box plot · five-number summary · Tukey hinges · distribution shape