What a Boxplot Hides About Distribution Shape

What a Boxplot Summarizes—and What It Cannot Show

A boxplot can make it easy to compare the center and spread of quantitative data. But its compactness comes at a cost: it displays only selected positions in the ordered data, not every observation. A boxplot may show where the middle half lies without showing whether those observations are evenly spread, piled up in one place, or separated by gaps.

To see the difference, consider three hypothetical sets of completion times, each measured in minutes. Their dotplots have visibly different patterns, even though all three sets produce the same five-number summary. The comparison is not a quirk of one unusual example. It illustrates a general limitation of a boxplot: many different data sets can share the same summary.

For the examples below, use the median-of-halves convention: for an even-sized data set, split the ordered values into a lower and an upper half; \(Q_1\) is the median of the lower half, and \(Q_3\) is the median of the upper half. This convention makes it possible to verify the summaries consistently. The key point is what happens after the summary is calculated: the boxplot retains those five values, not the full arrangement of observations.

One Summary, Different Dotplots

In each table below, a dot represents one hypothetical completion time. Dots stack vertically when observations have the same value. The horizontal position identifies the time in minutes; the stack height shows how many observations take that time.

Why the Five Values Do Not Determine the Shape

The examples share a summary because the summary depends on a few selected ordered positions. For eight observations under the convention used here, the minimum and maximum are the first and eighth values; \(Q_1\) is determined by the second and third values; the median by the fourth and fifth; and \(Q_3\) by the sixth and seventh. The observations can differ while those selected values, or the averages that define the quartiles and median, remain the same.

For instance, in the first set the middle pair is 12 and 16, while in the second it is 13 and 15. Both pairs average to 14. The second and third values in the first set are 12 and 12, while those in the second are 11 and 13. Both pairs average to 12. Similar changes in the sixth and seventh values still yield \(Q_3=16\). The summaries agree even though other values—and their arrangement—differ.

A boxplot can therefore conceal features that matter to an analysis:

• Clusters: groups of observations concentrated in different regions.
• Gaps: stretches of the measurement scale with few or no observations.
• Modality: the number and prominence of peaks in the distribution.
• Within-quartile patterns: whether observations in a quartile are spread across its interval or concentrated at only a few values.

The lengths of the box and whiskers do not supply those details. For example, the box from 12 to 16 minutes spans four minutes in all three examples. That equal length says that the quartile endpoints are the same; it does not say that the observations between those endpoints are distributed in the same way. A boxplot also cannot reveal that the first data set has a gap between its two main stacks.

Use the Graph That Answers the Shape Question

A boxplot is useful when the question concerns the median, quartiles, range, or comparisons of those summaries. If the question is whether observations cluster, whether there is a gap, or how values are distributed within a quartile, the dotplot contains information the boxplot has discarded. For small data sets, the dotplot can show every observation directly. The graphs answer different questions; one is not a more detailed rendering of the other.

This matters because conclusions based only on a boxplot can overstate what is known. Two groups might have matching boxplots but quite different internal patterns. If the purpose is to look for clusters or gaps, equal five-number summaries do not establish that the distributions have the same shape. Inspect a graph that preserves individual observations before making a claim about those features.

Common Mistake

Wrong reasoning: “The boxplots are identical, so the distributions must have the same shape.” This conclusion treats five selected values as if they described every observation. The three dotplots above show why that inference does not follow: the same five-number summary can accompany separated clusters, a broader pattern, or a central concentration.

Correction: Say only that the five displayed summary values are the same. To compare clusters, gaps, modality, or within-quartile patterns, examine a dotplot or another graph that shows more of the observations. A summary is useful precisely because it is compact, but that compactness means some information is absent.