Did you know? The tallest bar in a histogram does not always represent the largest number of people. When the intervals on the horizontal axis have different widths, a short, broad rectangle can contain more observations than a tall, narrow one. This is a useful lesson in statistical literacy: a picture can be perfectly drawn and still be read incorrectly.
For unequal-width histogram bars, frequency is represented by bar area, not height alone. Divide each class frequency by its interval width to find frequency density, use that density as the bar height, and multiply width by height to recover the count. A histogram groups numerical measurements into intervals; frequency is the number of observations within each interval, and the interval width is the difference between its boundaries.
Many school exercises begin with equal-width intervals, where comparing bar heights happens to work. The mistake appears when a learner carries that shortcut into a graph with 5-, 10- and 20-minute classes. This guide builds the skill from a diagnostic question through a complete example, an independent check, and decisions a learner can explain to someone else.
Start with the question that exposes the misconception
Draw two rectangles on squared paper. Rectangle A is 2 squares wide and 5 squares high; rectangle B is 5 squares wide and 3 squares high. Ask, ‘Which is taller?’ A is taller. Then ask, ‘Which has more square units of area?’ A has 2 × 5 = 10 square units; B has 5 × 3 = 15. Height and area give different answers. That is the whole histogram trap in a form a pupil can see without formulas.
Check a second misconception: if an interval runs from 15 minutes up to but not including 35 minutes, is its width 35, 20 or 21? The width is 35 − 15 = 20 minutes. The width is a difference between boundaries, not a count of printed integer labels. State the boundary convention so that an observation at exactly 15 minutes goes in one class, never two.
Worked example: waiting times at a fictional student exhibition
Imagine that volunteers timed the waiting period, in minutes, for 61 visitors to an entirely fictional school exhibition. The invented grouped data below are for practising histogram interpretation; they are not actual visitor records or a report about any Singapore school. Every waiting time is assumed non-negative and less than 50 minutes, measured on a consistent basis. The intervals use the convention ‘at least the left boundary, below the right boundary’.
- 0 ≤ t < 5 minutes: 10 visitors; width = 5 minutes; frequency density = 10 ÷ 5 = 2 visitors per minute.
- 5 ≤ t < 15 minutes: 18 visitors; width = 10 minutes; density = 18 ÷ 10 = 1.8 visitors per minute.
- 15 ≤ t < 35 minutes: 24 visitors; width = 20 minutes; density = 24 ÷ 20 = 1.2 visitors per minute.
- 35 ≤ t < 50 minutes: 9 visitors; width = 15 minutes; density = 9 ÷ 15 = 0.6 visitors per minute.
Now imagine four adjacent bars whose horizontal spans match those actual interval widths. Their heights are 2, 1.8, 1.2 and 0.6 on a vertical axis labelled frequency density (visitors per minute). The first bar is the tallest, but its area is 5 × 2 = 10. The third bar is lower, yet its area is 20 × 1.2 = 24. So the 15–35-minute class contains the largest count, while the 0–5-minute class has the highest density per minute. Those answers are not contradictory: they answer different questions.
Check the complete graph: 5 × 2 + 10 × 1.8 + 20 × 1.2 + 15 × 0.6 = 10 + 18 + 24 + 9 = 61 visitors. If the reconstructed count were 60 or 62, we would return to the boundaries, divisions, vertical scale or transcription before interpreting a pattern.
Why density is not a physical speed or a probability
The phrase ‘2 visitors per minute’ here is a graphical count density: 10 visitor observations are grouped into a 5-minute-wide waiting-time interval. It does not mean that exactly two people arrived during each minute or that a service desk served two people per minute. The measurement being grouped is each visitor’s waiting time. Labels must keep this distinction visible. Calling the vertical axis ‘number of visitors’ would be inaccurate for this unequal-width version because its bar heights are densities.
For a histogram of scores in marks, density would have units students per mark. For distances in kilometres, it might be observations per kilometre. The graphical rectangle’s area then recovers the number of observations. These ‘per unit’ labels are a way of keeping the mathematics honest, not a claim that physical people occupy the printed graph.
What happens if a learner uses frequency as height?
Suppose a student builds four bars with heights 10, 18, 24 and 9 but retains widths 5, 10, 20 and 15. The rectangles would have areas 50, 180, 480 and 135 square graph units. The shapes now visually exaggerate the broad 20-minute class simply because it stretches farther across the x-axis. Such a drawing is not a correctly scaled frequency-density histogram. If equal-height interpretation is wanted, regroup into equal-width intervals from the original observations, where possible, or display a labelled frequency table alongside the unequal-width histogram.
A helpful correction is to write two columns before drawing: frequency and frequency ÷ class width. Label which one is actually plotted on the vertical axis. A child who can explain why 24 ÷ 20 = 1.2 has learned more than the formula; they are protecting the reader from a misleading visual comparison.
The inverse skill: read the bar to recover the data
Many questions supply a graph rather than a table. A 12-minute-wide bar with height 1.5 observations per minute represents 12 × 1.5 = 18 observations. A bar from 40 up to 45 minutes with height 3.2 represents 5 × 3.2 = 16 observations. To verify, divide the count back by the width: 16 ÷ 5 = 3.2. This two-way operation is the independent check.
If a chart displays relative frequency density, the rectangle area represents a proportion rather than a raw count. For a dataset with 61 people, a relative area of 24/61 corresponds to 24 observations; without the sample size, it would only tell us the proportion. Read the y-axis title and any legend before multiplying anything.
Can we calculate an exact number inside half of a bar?
Not from grouped counts alone. In our 15 ≤ t < 35 class, 24 visitors wait somewhere within a 20-minute span. If someone asks how many waited between 25 and 35 minutes, the original table does not say. Under a stated uniform-within-the-class estimate, half the width would suggest about 12. But all 24 could in principle be concentrated in one half, so 12 is an estimate, not a measured fact. Distinguishing an assumption from evidence is the real learning transfer.
The same caution applies to calculating an exact median or average from grouped data. A class midpoint can support a reasonable estimate, but it does not recreate missing individual measurements. Do not turn a convenient model into invented observations.
Guided practice that becomes independent understanding
Stage 1 — compare rectangles before numbers
Give one 4-square-wide bar of height 3 and one 2-square-wide bar of height 5. Have the learner say ‘the second is taller; the first has greater area, 12 versus 10’. Only then introduce the histogram vocabulary. If they immediately point to the tallest bar, ask what the horizontal span means. The target is an articulated distinction, not a guessed answer.
Stage 2 — reconstruct a hidden frequency
Provide a 6–14-minute interval with plotted density 2.5 observations per minute. The width is 8 minutes, so frequency is 8 × 2.5 = 20 observations. Ask for a unit check, a sentence naming the interval, and an inverse division 20 ÷ 8 = 2.5. Next remove the calculator and repeat with a width of 4.
Stage 3 — independent transfer problem
A fictional reading-time study groups pupils into intervals 0–4 minutes (8 pupils), 4–10 minutes (15 pupils), and 10–22 minutes (18 pupils). Ask for class widths, densities, tallest bar, largest group and total sample. A fully checked answer is widths 4, 6, 12 minutes; densities 2, 2.5, 1.5 pupils per minute; tallest density bar 4–10; most pupils 10–22; and total 41 pupils. A correct answer must explain why ‘tallest’ and ‘largest group’ differ.
How parents and teachers can diagnose without over-teaching
- Checks only heights: ask what the bar width represents and have the learner colour its rectangle in squared units.
- Gets a wrong count from a given bar: check whether they multiplied by the true interval width or the printed endpoint.
- Counts a boundary twice: use inequalities such as 5 ≤ t < 15 and 15 ≤ t < 35, then place a value of exactly 15.
- Claims a precise count within a partial interval: ask whether the source supplied individual observations or only grouped totals.
- Omits a vertical-axis unit: have them say aloud ‘observations per minute’ and then explain what multiplication gives.
A useful progress record has three independent checks over separate practice sessions: correctly identifying unequal widths, reconstructing all frequencies without being prompted, and writing a one-sentence limitation for partial intervals. A learner who can perform all three on a new chart has stronger evidence of transfer than someone who merely remembers the words ‘frequency density’.
What the source material supports—and what this article does not claim
The OpenStax statistics treatment of histograms provides the foundations of frequencies, numerical intervals and grouped displays. A focused statistics explainer on unequal class widths derives the relation frequency density = frequency ÷ width and the area rule. The 61-person data and every assessment example above were invented and checked arithmetically for this lesson. This is a transferable statistical reading skill, not a claim about which question currently appears in a particular Singapore examination.
Frequently asked questions
Is the highest bar the modal class?
When the vertical axis is frequency density, the tallest bar has the highest density over its interval. In unequal-width graphs, the class containing the most observations may be wider and shorter. Check what the question means by ‘most common’ and explain your interpretation instead of treating height and total frequency as identical.
Should the bars always touch?
A histogram groups numerical ranges along a continuous scale, so adjacent nonempty intervals usually meet at shared boundaries. A missing interval can be represented with a zero-height span, rather than shifting other classes together. A categorical bar chart is a different kind of display, and its categories need not form a continuous scale.
Why not just use the same widths for every interval?
Equal widths simplify reading and are often preferable when raw data allow it. But available data may already be grouped into unequal intervals, and a legitimate analysis should preserve their boundaries rather than inventing a new distribution. Density makes a fair graphical comparison possible.
Does the area formula reveal every original value?
No. It recovers the count in each interval, not the individual waiting times or scores. Grouping permanently hides detail unless the original observations remain available.
The core aim is to understand what the picture can prove
For further practice with graphical claims, compare reading a line chart with a missing legend and reading a stacked area chart without confusing a coloured band with the total. The wider eduKate Learning System and the immutable Clementi Mathematics teaching reference place diagnosis, explanation, progressive practice and independent verification at the centre of learning. The child’s final test is simple: they can look at an unfamiliar unequal-width histogram, reconstruct the counts, and say what the graph does not tell them.