Must class intervals have equal width?

No, but charts and comparisons must account for unequal widths.

Does the class midpoint equal every observation in the class?

No. It is a representative value used in some approximations.

Can a value belong to two adjacent classes?

Not in a properly defined single grouping; endpoint rules should assign it uniquely.

AI Glossary / Statistics and probability / Beginner

Class interval

Definition

A class interval is a defined range of numerical values used to group observations in a frequency table or histogram. Each interval has boundaries and an explicit rule for including values at those boundaries. Properly constructed intervals do not overlap, so an observation belongs to only one class in the grouping.

What it means

Grouping turns a long list of measurements into a manageable summary. Instead of displaying every waiting time, a report might show counts for times below five minutes, five to below ten, and ten to below twenty. The intervals describe the grouping, while frequencies describe how many observations fall inside. A class interval is not a confidence interval: it organizes observed values rather than expressing uncertainty about an estimated parameter.

How it works

Choose boundaries appropriate to the variable, measurement precision and reporting purpose. Ensure the intervals cover the values to be summarized without unintended gaps or overlap. State whether each endpoint is included, for example using [5,10) for values at least five but less than ten. Count observations or compute relevant summaries within each class. Equal-width intervals simplify comparisons, but unequal widths may be useful when the data span different scales. In a histogram with unequal widths, use frequency density or probability density so bar area, rather than height alone, represents the count or proportion.

Three practical examples

Service reporting: summarize waiting times in clearly labeled ranges while retaining boundary rules for values exactly at five or ten minutes.
Customer analysis: present a distribution of purchase amounts without exposing every individual transaction.
AI evaluation: group document lengths or prediction errors to inspect performance across ranges, checking that the grouping does not conceal important tail behavior.

Think of it this way

Think of sorting measured objects into labeled containers. The container boundaries determine where each object goes. Wider containers can hold more objects simply because they cover a wider range, so comparing their contents requires attention to width as well as count.

Worked example

Consider ten waiting times in minutes: 2, 4, 5, 7, 9, 10, 12, 15, 18 and 19. Use intervals [0,5), [5,10) and [10,20). Their counts are two, three and five, with widths five, five and ten. Values five and ten belong to the second and third classes respectively. For a frequency-density histogram, bar heights are count divided by width: 0.4, 0.6 and 0.5 observations per minute. The third class has the largest count but not the highest density. For a probability-density histogram, divide again by ten, giving heights 0.04, 0.06 and 0.05 per minute; multiplying by widths gives areas 0.2, 0.3 and 0.5, which sum to one. If only the grouped counts remain, a midpoint estimate of the mean is [2 × 2.5 + 3 × 7.5 + 5 × 15]/10 = 10.25 minutes. The exact mean from the original values is 101/10 = 10.1 minutes. The difference illustrates information lost through grouping.

Illustrative case study

Fictional case: Birch Service Operations reports response times using broad intervals. Managers initially interpret the tallest bar as the most concentrated range, but analysts discover that the final interval is much wider than the others. They revise the chart to use density and include the percentage exceeding the operational target. The underlying ungrouped data are retained for quantiles and modeling. This fictional case illustrates clearer distribution reporting and makes no claim that the service itself became faster.

Common misconceptions and mistakes

Class limits written as “0 to 5” and “5 to 10” are ambiguous unless the treatment of five is stated. A histogram's appearance can change substantially with boundary placement and width; one display should not be treated as proof of the distribution's shape. Group midpoints are approximations for individual values, not recovered measurements. Open-ended classes, such as “100 or more,” have no finite midpoint without additional assumptions. An interval with zero observations is different from an interval omitted because data were not collected. For rounded discrete data, class labels and underlying measurement boundaries may require different interpretations.

Business use

Businesses should choose intervals that support the decision while avoiding selective boundaries that exaggerate a preferred story. Preserve consistent definitions for comparisons across periods, or clearly explain changes. Report counts or percentages alongside sparse ranges when useful. In machine-learning pipelines, binning is a modeling transformation that can lose information and should be fitted and validated appropriately. Do not assume that a convenient dashboard grouping is also the best input representation for a predictive model. Retain original values whenever permitted so later analyses are not limited to grouped approximations.

Check your understanding

Q: Why is the third histogram bar lower than the second when using density in the worked example?

Answer: A: Its five observations occupy a ten-minute interval, giving 0.5 per minute, while the second contains three observations in five minutes, giving 0.6 per minute.

Related terms

Frequency distribution
Histogram
bin width
class boundary
class midpoint
frequency density
grouped data and discretization

Learn next

redraw the example with equal-width intervals and then shift their boundaries. Compare the visible shape and identify which conclusions remain stable. Calculate exact summaries before grouping, then compare them with midpoint approximations to see what information is lost.

Sources

Sources: Australian Bureau of Statistics, class interval definition: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/statistical-terms-and-concepts-glossary ; ABS, frequency grouping and nonoverlapping intervals: https://www.abs.gov.au/websitedbs/d3310114.nsf/home/Basic%2BSurvey%2BDesign%2B-%2BAnalysis ; ABS, calculating summaries from grouped data: https://www.abs.gov.au/AUSSTATS/abs%40.nsf/bb8db737e2af84b8ca2571780015701e/17bb49c5c8191b10ca2571fe007d69dc!OpenDocument .

Previous: Class incremental learning Next: Classical decomposition method