Box plots hide multimodal structure

What it is

A box plot summarizes a distribution with a median, quartiles, whiskers, and conventionally marked points, but it does not encode how probability mass is arranged inside the intervals. Unimodal, bimodal, flat-topped, or gapped samples can therefore produce identical or similar boxes and whiskers. This is a summary–shape mismatch. “Hides multimodality” does not claim that every box conceals several modes; it means that a box summary alone can neither establish nor rule them out.

Why it happens

Reducing observations to a few quantiles is a many-to-one mapping: different local densities and arrangements share the same quantile locations. Whiskers describe range or a rule based on interquartile distance, and outlier marks depend on that convention; neither recovers modal location. Histograms, density plots, and raw points restore shape evidence through bins, smoothing, or record positions, each with its own parameter or overlap costs. A box plot's strength is a compact repeated grammar for comparing location and spread, not reconstructing one group's full distribution.

Studying it

Studies can construct samples with similar quantiles but different modality and compare boxes alone, boxes plus raw points, histograms, and density plots. Separate median or IQR comparison, mode detection, heaping or truncation detection, and group ranking. Measure errors, time, confidence, and false claims that similar summaries imply equal distributions. Control sample size, scale, bin or bandwidth, jitter, and whisker convention because these affect the competing displays. Apparent modes in finite samples are uncertain; visual separation does not replace mixture modelling or domain validation.

Where it stops holding

When the task is robust location, spread, or scanning many groups, boxes may be more compact than detailed shape displays; every task need not carry every raw point. At small samples, both quantiles and apparent modes fluctuate, so show observations or sample size rather than treating either shape as the population. At large samples, points overplot and a binned or density view may be preferable. Missingness, censoring, weights, and duplicates affect both summaries and shape and require explicit treatment. Although filed under “Histograms and distributions,” this leaf examines the box summary as a contrast to shape displays; whiskers, sample size, and cross-group box-plot practice receive fuller treatment in the dedicated box-plot group.

Applying it

• Separate summary-comparison from shape questions. Add raw points, a histogram, or a documented density view when modes, heaps, truncation, or mixed populations affect the decision.
• State sample size, whisker rule, weighting, and missing-data treatment. Do not automatically label points beyond whiskers as errors or exceptional cases.
• On small screens or with many groups, let a compact box overview open a group-level detail. Give keyboard and screen-reader users quantiles, range, sample size, and a shape summary.
• Validate with data having similar quantiles but different shapes. If readers equate similar boxes with equal distributions, revise the display or narrow the claim.

Related

• Same group: U2.08.1 Histogram bin width changes the conclusion · U2.08.3 Showing the raw point distribution is more reliable
• Nearby: U2.09.1 Box plots compress a distribution into five statistics · U2.09.3 The box plot's strength is comparing many distributions side by side
• Search terms: box plot multimodality · distribution shape · summary-shape mismatch · quantile summary