Why Compare Shape?
When comparing groups, asking which has the higher center or greater spread does not tell the whole story. Two groups can have the same median and a similar range, yet one may be roughly symmetric while the other has a long tail. Or groups may have similar centers but differ in whether their values gather into one cluster or several. Those patterns matter because a single summary number can conceal how observations are distributed.
A shape comparison describes the overall arrangement of values in each group. It is most useful when the groups measure the same quantitative variable, in the same units, and are shown on a common horizontal scale. The goal is to report the visible patterns, not to infer why they occurred.
Recognizing Shape Features
A roughly symmetric distribution has a similar overall pattern on either side of its central region. Exact mirror symmetry is uncommon in real data, so “approximately symmetric” is usually the appropriate description. A skewed distribution instead has an extended tail in one direction. The tail is the part of the distribution that stretches away from the main concentration of observations; it is not necessarily where most values lie.
A cluster is a region where observations are relatively concentrated. A gap is a region with few or no observations separating concentrations or parts of the distribution. Tails can also differ in how far they extend and how gradually observations thin out. These features can coexist: a distribution might have two clusters and a long right tail, for example.
Look at the whole arrangement before naming a pattern. A single extreme-looking value does not, by itself, establish that a group is skewed. Likewise, small samples can look irregular simply because they contain few observations. Shape descriptions should reflect the overall evidence in the display and should be qualified when the pattern is uncertain.
Comparing Groups on a Common Display
A side-by-side display makes comparisons easier when both groups are plotted on the same scale. First check that the variable and units match. Then compare the broad features: Is each group approximately symmetric or skewed? Is one group more clearly clustered, or does it have a gap? Which tails extend farther? Finally, separate those observations from comparisons of center and spread.
Side-by-side boxplots, introduced in the earlier material on boxplots, provide a compact comparison. The box and median line can give clues about asymmetry, and the whiskers can show that one side extends farther. But a boxplot compresses the data into a few landmarks. It does not show every observation or reveal the detailed internal arrangement, so it cannot reliably display a gap or multiple clusters. When those features matter, use a display that preserves more information about the distribution. Do not claim that a boxplot shows a cluster that its summary marks cannot establish.
Worked Example: Same Median, Different Skew
Suppose two hypothetical groups of students receive quiz scores, in points. Their sorted scores are:
Group A: 2, 3, 4, 5, 6, 7, 8
Group B: 2, 3, 4, 5, 6, 7, 14
Each group has \(n=7\) observations, so the median is the fourth value in sorted order. For both groups, the median is 5 points. The ranges are:
Group A’s scores extend evenly from 2 to 8 around 5: for each value below 5 there is a corresponding value the same distance above 5. Group B has the same lower scores through 7, but its largest value is 14 rather than 8. The values therefore extend much farther to the right of the main run. Group B is right-skewed in this small example, while Group A is symmetric.
The shared median does not make the shapes the same, and the range difference is not itself the shape description. The distribution of Group B’s values, especially the extended high end, supplies the evidence for the right tail. With only seven scores per group, the conclusion should remain descriptive and cautious.
Why Center Alone Cannot Describe Shape
A short calculation illustrates why a center summary cannot replace a shape comparison. For a symmetric collection of \(2k\) values paired around a central value \(c\), write each pair as \(c-d_i\) and \(c+d_i\), where \(d_i\) is its distance from \(c\). Each pair sums to \(2c\), so the mean of all \(k\) pairs is:
This calculation explains how a balanced, symmetric arrangement can have its mean at \(c\). But knowing that the mean equals \(c\) does not reveal the distances \(d_i\), whether observations form clusters, or whether a tail extends farther on one side. Many different arrangements can share the same mean; likewise, groups can share a median and still have different shapes. A center is one feature, not a complete description.
Worked Example: A Gap Between Two Clusters
Consider hypothetical completion times, in minutes, for two groups finishing the same short course:
Group A: 10, 11, 12, 13, 14, 15, 16, 30, 31, 32
Group B: 10, 11, 12, 13, 14, 15, 16, 17, 18, 19
Both have ten observations, so each median is the average of the fifth and sixth ordered values. For both groups:
Their ranges differ:
In Group A, the first seven times run from 10 to 16 minutes, and the remaining three run from 30 to 32 minutes. There are no observations from 17 through 29 minutes, leaving a clear gap between the two clusters. Group B’s values instead form one consecutive run from 10 to 19. Thus, the groups share a median but differ in both spread and shape; the gap and two concentrations are the evidence for Group A’s distinctive shape. A median and range alone would not identify where the observations lie within those endpoints.
Compare Tails and Symmetry Carefully
A longer tail is a directional feature. If the observations taper farther toward high values, describe the group as right-skewed; if they extend farther toward low values, describe it as left-skewed. Avoid deciding the direction from the side where the display looks visually heavier: skew refers to the extended tail, not the side with more observations. When neither side clearly extends farther, “roughly symmetric” may be more appropriate than forcing a skew label.
Worked Example: Comparing a Left Tail With a Balanced Pattern
Two hypothetical groups record a measurement in millimetres:
Group A: 7, 8, 9, 10, 11, 12, 13
Group B: 0, 8, 9, 10, 11, 12, 13
There are seven values in each group, and the fourth ordered value is 10 in both. Their medians are therefore 10 mm. The ranges are:
Group A is balanced around 10: its values are three, two, and one millimetres below 10, paired with values three, two, and one millimetres above 10. Group B has the same upper values but one value extends much farther below the central group, to 0 mm. Its tail extends toward smaller measurements, so the pattern is left-skewed. This conclusion describes the arrangement shown; it does not establish why the low measurement occurred or whether it is an error.
Common Mistakes and Useful Language
A shape comparison is most informative when it states the feature, names the group, and gives the direction or location in context. For example: “Group B’s quiz scores are more right-skewed, with a longer tail toward high scores, while Group A’s scores are roughly symmetric.” This is more precise than “Group B looks different.” If the display is inconclusive, say so rather than overstating a pattern.
- Confusing tail direction with where values cluster. The group may have most observations at lower values and still be right-skewed if the tail extends toward higher values.
- Treating a shared median as proof of matching distributions. The median locates a center; it does not specify the arrangement of observations on either side.
- Calling a difference in range a shape description. Range describes the distance between the smallest and largest values. Describe skew, clusters, gaps, or tails separately.
- Claiming a boxplot reveals a gap or multiple clusters. A boxplot summarizes positions and does not show the full internal arrangement. Use an informative display before making such a claim.
- Explaining a pattern from the graph alone. A display can show a difference in shape, but it cannot by itself establish its cause or whether an unusual measurement is erroneous.
Check Your Understanding
Use the ideas in this tutorial to answer the following questions.
- In your own words, what makes a distribution right-skewed? Does the longer tail point toward larger or smaller values?
- Two groups have the same median, but one has two clusters separated by a gap. Explain why the shared median does not describe that shape difference.
- A side-by-side boxplot shows one group’s upper whisker is longer. What might this suggest, and what should you avoid concluding from the boxplot alone?
- Why is a common measurement scale useful when comparing the shapes of two groups?
- Write one sentence that compares groups by a visible shape feature while keeping that description distinct from a comparison of their centers.