Overlap and Separation Between Two Distributions

Overlap Adds Another View of Group Differences

In Comparing Distributions, you learned to compare shape, center, spread, and unusual features. Overlap adds a useful question: across what values do the groups share observations, and where are their values more distinct? Two groups can have different centers while still containing many observations in a shared range. Conversely, a modest difference in centers can look more pronounced when each group has little spread.

Overlap is not a single summary statistic with one universally accepted calculation. It is a way to describe how two distributions line up on a common scale. The display matters: dotplots and histograms show where observations or intervals are concentrated, while side-by-side boxplots show the ranges and middle halves. As in Comparing Two Histograms on the Same Scale and Side-by-Side Boxplots Compared, compare displays only when their scales are aligned.

A shared range is a starting point, not a complete description. The minimum-to-maximum ranges may overlap even if most observations cluster apart. Boxplot boxes may overlap only a little, even when whiskers extend through a much wider shared range. A histogram or dotplot can reveal whether observations are concentrated in the shared region or only a few values reach it.

A practical comparison uses three layers: first check the full ranges, then compare the central regions, and finally inspect the shape and concentration of the observations in any shared region. State which layer supports your conclusion. “The ranges overlap” is a specific claim; “the groups are almost the same” is much broader and may not be supported.

What Overlap Can Suggest

When two groups overlap substantially, a value from one group may also be quite plausible in the other. Knowing an individual observation’s value may therefore not clearly identify which group it came from. When distributions are more separated, values may be more useful for distinguishing the groups—but overlap and separation alone do not guarantee that every observation can be classified correctly.

The comparison is about distributions, not a rule for every individual. Even when one group generally has larger values, some observations in that group may be below observations in the other group. Say that a group’s values tend to be higher or that its distribution is shifted higher when the display supports that description. Avoid saying that every member of one group has a higher value.

Overlap also does not determine whether an observed group difference is statistically significant. That question requires an appropriate inference procedure and information about how the data were collected. Nor does a difference between groups by itself show that group membership caused the difference. In this tutorial, use overlap as a descriptive feature alongside shape, center, spread, and unusual values.

Comparing Shared Ranges

For a quick first check, identify each group’s minimum and maximum, then find the numerical interval that both ranges cover. If one group’s range is from 14 to 27 centimeters and the other’s is from 18 to 32 centimeters, both ranges include values from 18 to 27 centimeters. The ranges overlap on that interval.

The width of a shared interval can help describe its size in the variable’s units. In this example, the interval’s width is \(27-18=9\) centimeters. But that width is not a percentage of the observations that overlap, and it does not show how densely either group’s observations fall in the interval. For that, inspect the individual observations or a suitable graph.

Be careful with endpoints and discrete data. If measurements are recorded as whole numbers, an interval from 18 to 27 includes those recorded values, but the arithmetic width is 9 centimeters. That is different from counting the 10 possible whole-number values 18, 19, and so on through 27. Say whether you are describing interval width or counting recorded values.

Overlap Between the Middle Halves

A side-by-side boxplot offers another view. Each box extends from \(Q_1\) to \(Q_3\), so it represents the middle half of that group’s observations. If the boxes overlap, the middle halves share a numerical interval. If the boxes barely touch or do not overlap, the central portions appear more separated, even if the whiskers share values.

The box overlap is about the interval shown by the boxes. It is not a count of how many observations from each group are in that interval: a boxplot does not show the individual data values. And a box that is twice as wide as another does not mean it contains more observations. Each box represents the middle half of its own group.

As in Comparing Spread Using IQR and Standard Deviation, compare matching measures when discussing spread. The IQR describes the width of a box; the overlap between boxes describes how their middle-half intervals line up. These are related but different observations. You can report both without treating one as a substitute for the other.

Using Individual Values to Describe a Shared Region

When raw data or a dotplot is available, you can count how many observations from each group fall in a shared numerical interval. This makes the description more specific than range overlap alone. State the interval, count observations in it for each group, and, if useful, report each count as a fraction or percentage of that group.

This calculation answers a limited question: what proportion of each group’s observed values lies in the specified interval? It does not by itself measure every aspect of overlap. Two distributions might put the same proportion in an interval but arrange those observations differently within it. Describe any visible clustering or gaps as well.

Common Mistakes and AP Exam Tips

• Treating range overlap as complete similarity. Overlapping minimum-to-maximum intervals do not show where most values lie. Look at the boxes or the shape and concentration in a dotplot or histogram.
• Confusing the width of an interval with the number of observations in it. Subtracting endpoints gives a width in the variable’s units. Counting observations requires raw data or a display that shows them.
• Calling box overlap a percentage of observations. Boxplots show quartiles, not individual values within the box. Describe the shared interval between the boxes, not an unsupported count or percentage of observations.
• Assuming overlapping groups have identical centers. The medians or means may still differ. Compare the centers directly, then describe the overlap as a separate feature.
• Claiming no individuals can be confused when groups look separated. Even an apparent separation in a sample does not guarantee separation for every person or in a larger population. Keep the conclusion tied to the observations shown.
• Turning a descriptive pattern into an inference or causal claim. Overlap does not provide a p-value, establish statistical significance, or show that one group characteristic caused higher or lower values.

A full-credit comparison names both groups and the variable, identifies the specific region that overlaps or appears separated, and supports the statement with values from the display. For example: “The middle-half intervals overlap from 50 to 54 minutes, although the Meadow median is 8 minutes higher than the Creek median.” That is more precise than saying only that the groups are “different” or “about the same.”