eduKate Singapore

VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Next useful routeCanonical hubHELP

Mathematics in Bukit Timah | Histograms, Frequency Density and Unequal Class Widths

—

by

eduKate Asia
Frequency-density histogram of 100 observations: intervals 0 to 5, 5 to 15, 15 to 20 and 20 to 40 minutes have heights 4, 3, 2 and 2, and frequencies 20, 30, 10 and 40.

A bar covering twenty minutes looks wider than one covering five minutes. Does its height have to be taller to represent more observations? In a frequency-density histogram, the answer is no. The area of each bar, not its height alone, represents its class frequency. This distinction becomes especially important when the class widths are unequal.

This Mathematics in Bukit Timah guide uses a wholly fictional set of 100 recorded activity durations to show how to construct and interpret a histogram. No participation figures, student behaviour, local travel times or class results are actual measurements from a Bukit Timah school or tuition centre. The invented context makes statistics tangible; the mathematics depends on the table and graph, not the neighbourhood name.

The quick example contains four continuous duration classes. From 0 to under 5 minutes, frequency is 20. From 5 to under 15 minutes, frequency is 30. From 15 to under 20 minutes, frequency is 10. From 20 to under 40 minutes, frequency is 40. Their widths are 5, 10, 5 and 20 minutes, so their frequency densities are 4, 3, 2 and 2 observations per minute.

Article contents

Original worked frequency-density histogram with invented 100 observations in unequal intervals of width 5, 10, 5 and 20, demonstrating that rectangular bar area represents frequency.

Quick answer: frequency density is frequency divided by class width

A histogram represents numerical observations grouped into intervals. If its vertical axis shows frequency density, the bar’s height is frequency ÷ class width.

Multiplying that height by the class width gives back the frequency. Therefore bar area = frequency in the ideal graph scale.

This is why a broad, low bar can represent more observations than a narrow, high bar. Look for the horizontal width and vertical density together, not merely the tallest rectangle.

Why Bukit Timah is a setting rather than a data source

A family near Sixth Avenue or Beauty World might discuss schoolwork durations, reading sessions or hypothetical waiting times. Such quantities are familiar examples of continuous data measured in minutes.

But this article does not claim that 100 local pupils completed an actual questionnaire. The frequency table is an intentionally constructed classroom dataset, and all discussions of its shape refer only to those invented values.

Responsible statistical teaching requires naming the source of a dataset. A precise chart can be numerically flawless and still mislead when a hypothetical sample is incorrectly presented as community evidence.

Start with the four frequency classes

Imagine a teaching dataset consisting of 100 fictional session durations in minutes. The intervals and counts are:

  • 0 ≤ t < 5: frequency 20.
  • 5 ≤ t < 15: frequency 30.
  • 15 ≤ t < 20: frequency 10.
  • 20 ≤ t < 40: frequency 40.

Each observation should belong to one interval. The left endpoint is included and the right endpoint is excluded in this convention, preventing an observation of exactly 15 minutes from being placed in both neighbouring classes.

The frequencies add to 20+30+10+40=100. That total is an essential audit check before plotting any bar.

Class width is not class frequency

The first interval has width 5 minutes and contains 20 observations. The second has width 10 minutes and contains 30. The fourth has width 20 and contains 40.

A pupil who copies the frequencies 20, 30, 10, 40 as histogram heights is building a different representation from a frequency-density histogram. Equal height units would make the wide classes disproportionately large in visual area.

Write the width in its own table column. Then calculate density as a separate step instead of guessing from which bars should look important.

Worked example 1: calculate the first two densities

For 0≤t<5, class width is 5−0=5 minutes. Frequency density is 20/5=4 observations per minute.

For 5≤t<15, width is 15−5=10 minutes. Density is 30/10=3 observations per minute.

Thus the first bar is taller despite representing fewer observations. Its area is 5×4=20; the second bar area is 10×3=30. The apparent contradiction disappears once the reader interprets area rather than height.

Complete the density column

For 15≤t<20, width is five and frequency is ten, so density is 2. For 20≤t<40, width is twenty and frequency is forty, so density is also 2.

The density list is therefore 4, 3, 2, 2. The two final bars have equal heights even though one contains ten observations and the other forty.

An accurate reader should be able to explain this immediately: one equal-height rectangle is four times as wide, so its area and frequency are four times as large.

Worked example 2: recover frequencies from a histogram

Imagine the bar spanning 20 to 40 minutes has height 2 on a frequency-density scale. Its width is 20 minutes.

The frequency is width×density=20×2=40 observations. The preceding class spans 15 to 20, with width 5 and the same density 2, so its frequency is only 5×2=10.

The bars are equally tall but not equally important to the sample total. A student who reads them as both frequency two has used the density-axis values as counts.

Why histogram bars normally touch

For continuous or grouped quantitative data, adjacent class intervals can share an endpoint, such as 0–5 and 5–15. Their bars typically touch because the horizontal scale represents a continuous range.

A category bar chart is different: its groups may be unordered labels such as reading, music or technology. Gaps can emphasise those separate categories.

The physical act of drawing rectangles is similar, but the horizontal axis has a different meaning. Students should identify the variable type before choosing a chart.

Continuous boundaries must be handled consistently

Suppose an invented time is exactly 5 minutes. Under the half-open convention, it belongs to 5≤t<15, not 0≤t<5.

Without a clear boundary convention, an exact endpoint might be counted twice or omitted. The same issue occurs with exam marks, measured lengths and age classes.

Sometimes school tables use whole-number inclusive class labels such as 1–5 and 6–10. Their appropriate continuous boundaries must be inferred from measurement resolution or explicit assumptions, not silently treated as identical to half-open duration intervals.

Worked example 3: where does 20 minutes belong?

Our third class is 15≤t<20 and the fourth is 20≤t<40. An observation recorded as exactly 20 minutes belongs to the fourth class.

A record of 19.999 minutes, if measurement precision allows it, belongs to the third. No observation at 20 belongs to both.

This illustrates why the word “under” is worth printing explicitly. A student should not have to guess whether boundary measurements are included.

A class with zero frequency still has a defined interval

A continuous class from 40 to under 50 minutes might have frequency zero in some invented dataset. Its density would be 0 divided by 10, giving zero.

The interval remains part of the horizontal scale if included in the chart, but its bar has zero height and area. It adds zero to the sample count.

Do not confuse a zero-frequency class with an actual observation of duration zero. These represent different facts about the dataset.

A histogram’s height is a rate of frequency across the interval

Frequency density has units of observations per unit of the horizontal variable. If the horizontal axis is minutes, the vertical axis is observations per minute.

This does not mean a student arrives or completes a task at that particular moment once per minute. It is a graphical density used to assign total observation count to an interval.

Students should understand the unit as part of the area relationship: minutes × observations/minute = observations. Unit cancellation explains why bar area corresponds to a count.

Worked example 4: convert density units carefully

Suppose a duration class covers 0.5 hour and contains 30 observations. Frequency density per hour is 30/0.5=60 observations per hour.

The same interval covers 30 minutes, so density per minute is 30/30=1 observation per minute. Both heights describe the same class count on axes using different units.

Multiplying 60 observations/hour by 0.5 hour gives 30 observations; multiplying 1 observation/minute by 30 minutes gives 30. The units verify consistency.

Frequency versus relative frequency

Frequency is a count of observations. Relative frequency divides the count by the sample total, producing a fraction or percentage of the whole.

For our sample of 100, the four relative frequencies are 20%, 30%, 10% and 40%. Those percentages sum to 100%.

The corresponding relative frequency densities would be each proportion divided by class width. Relative-frequency histogram area represents proportion rather than raw count. The same construction principle applies, but the vertical unit changes.

Worked example 5: a relative-frequency density

For the 20–40-minute class, proportion is 40/100=0.4 and class width is 20 minutes.

Relative-frequency density equals 0.4/20=0.02 per minute. The bar area 20×0.02=0.4, equivalent to 40% of the sample.

If the vertical axis uses percentage points per minute, the numerical height would be 40/20=2 percentage points per minute. Label units rather than mixing 0.02 and 2 without explanation.

Cumulative frequency gives an ordering perspective

Cumulative frequency adds classes from left to right. For our fictional table, cumulative counts are 20 at 5 minutes, 50 at 15, 60 at 20 and 100 at 40.

These totals describe the number of observations below each stated upper boundary. In particular, 50 of the 100 observed durations are under 15 minutes in this constructed table.

A histogram shows the distribution class by class; cumulative frequency shows how totals build across ordered intervals. They are related but not identical graphs.

Worked example 6: exactly how many observations are below twenty?

The first three classes cover 0≤t<20. Their frequencies add to 20+30+10=60.

Therefore 60 observations are below 20 minutes, and 40 are 20 minutes or above but below 40. Both conclusions are exact for the grouped frequency table.

No individual durations inside a class are needed to count the complete classes. That changes when the question asks for a threshold in the *middle* of a class.

A threshold inside a grouped class requires an assumption

Suppose the question asks how many durations are below 10 minutes. The first class contributes exactly 20, but the 5–15 class spans both below and above ten.

We know thirty observations occupy 5≤t<15, but we do not know their precise locations within that range. The exact total below ten is therefore not determined by the grouped counts alone.

Assuming uniform distribution within the class would allocate half of those thirty to 5–10, giving an estimate of 20+15=35. It remains an estimate, not an observed exact count.

Worked example 7: interpolate within a class

To estimate the number below 10 minutes, use the cumulative count of 20 at the start of the 5–15 class.

Ten is halfway across that ten-minute interval. Under the explicit uniform-within-class assumption, estimate half of its thirty observations below ten: 15. Hence estimated cumulative count is 35.

If the original thirty records were mostly clustered at 14 minutes, the real count would be much smaller. The model assumption is essential to the estimate.

The tallest bar is not always the modal class

For equal class widths, the tallest histogram bar corresponds to the class with the greatest frequency. For unequal widths, the greatest density identifies the tallest bar, while the greatest frequency belongs to the largest bar area.

In our dataset, the first class has highest density 4. However, the widest final class has greatest frequency 40. Statements about “modal class” require the school’s defined convention and a clear distinction between peak density and most populous class.

A careful solution can report both without ambiguity: the densest interval is 0–5 minutes, whereas the largest frequency count belongs to 20–40 minutes.

Worked example 8: the bar-height trap

Compare the 0–5 class and 20–40 class. Their densities are 4 and 2 respectively, so the first bar is twice as tall.

But their counts are 20 and 40, so the final class contains twice as many observations. A reader focusing only on heights would reverse the frequency ranking.

The correction is to calculate or visually compare rectangular areas. This is one of the most important habits in reading unequal-width histograms.

A bar chart with counts as heights may be useful elsewhere

An ordinary frequency bar chart can assign one category or group to each separated bar and set height proportional to frequency. That chart answers legitimate count-comparison questions.

However, if the horizontal bar widths are proportional to unequal continuous class intervals and bars touch, treating heights as counts misrepresents how histogram area relates to frequency.

The lesson is not that one chart form is inherently dishonest. It is that the visual encoding must match the quantity readers are expected to compare.

A histogram is not a scatter plot

A scatter plot records paired numerical observations, such as minutes studied and a quiz score for the same person. A histogram groups values of one quantitative variable into intervals.

Our fictional duration histogram does not show whether longer sessions led to higher marks or were preferred by particular students. That would require paired data or additional information.

This is a distinct skill from Mathematics in Bukit Timah: scatter plots and correlation, which studies associations between two variables.

Histograms summarise data but sacrifice detail

The table says twenty observations lie in 0–5 minutes. It does not reveal whether they all lasted one minute, were evenly spread or clustered near four minutes.

Different original datasets can produce exactly the same grouped frequencies and histogram. Therefore the histogram cannot reconstruct every individual record.

Students should avoid describing an exact mean, median or maximum observation when only group intervals are supplied, unless the question provides sufficient extra information.

Worked example 9: two datasets sharing one class count

Imagine ten observations all in 5≤t<15. One possible set could have all ten values equal to 6 minutes; another might have all ten equal to 14 minutes.

Both give a class frequency of ten in 5–15, yet their exact means are 6 and 14. The grouped histogram cannot distinguish them.

This counterexample shows why a grouped-data mean usually needs midpoint estimation rather than a claim of exactness.

Estimating the mean from class midpoints

A conventional estimate of the mean from grouped data assigns every observation in each class the class midpoint. For intervals 0–5, 5–15, 15–20 and 20–40, the midpoints are 2.5, 10, 17.5 and 30 minutes.

Multiply midpoint by frequency for each class, add the products and divide by total frequency.

This is an approximation because actual values within each class are unknown. The method should be labelled as an estimated grouped mean, not the exact mean of the original hundred observations.

Worked example 10: the grouped mean estimate

Use the four frequency–midpoint products: 20×2.5=50, 30×10=300, 10×17.5=175 and 40×30=1200.

Their sum is 1725 minute-observations. Divide by 100 observations to obtain an estimated mean of 17.25 minutes.

The arithmetic is exact for the midpoint approximation. The real sample mean could differ because the exact durations inside each class were not supplied.

Class width affects the precision of an estimated mean

A very broad class such as 20–40 loses more detail about individual durations than a narrow class such as 15–20.

Replacing all forty observations in 20–40 with midpoint 30 makes an assumption about their average inside that interval. If they were clustered near 20, the estimated contribution would be high; if clustered near 40, it could be low.

Wider grouping can simplify visual reporting but reduce the precision with which certain summaries can be recovered. This trade-off should be explained rather than hidden.

A median class can be identified from cumulative counts

For one hundred observations, the median position lies between the fiftieth and fifty-first ordered values. Cumulative frequency reaches 50 at the upper boundary of the second class and 60 at the upper boundary of the third.

The fiftieth observation belongs to 5–15, while the fifty-first belongs to 15–20 under the class conventions. The exact median is the average of those two unobserved original values, so the grouped table does not determine it uniquely.

A common interpolation approximation may put the median at the boundary 15, but that is an estimated value under a distributional convention, not a recovered exact individual median.

Worked example 11: why the median need not be exactly fifteen

Suppose the fiftieth invented observation is 14.8 minutes and the fifty-first is 15.2 minutes. Their median is 15.0.

Another compatible dataset might have the fiftieth at 5.1 and fifty-first at 19.9. Its median is 12.5. Both datasets can share the class totals.

Thus the grouped counts alone do not prove the exact median equals 15. Interpolation offers an estimate when such modelling is explicitly requested.

When can frequency density be zero?

A class interval with no observations has frequency zero and density zero, provided class width is positive.

There is no positive-height bar in that interval. A chart can still preserve the horizontal scale so readers see the empty range.

Removing the interval and sliding the neighbouring bars together would change the representation of the numerical variable, potentially hiding an important gap in the data.

Frequencies cannot be negative

Counts of observations are nonnegative whole numbers. Class widths for valid intervals are positive. Frequency densities are therefore nonnegative in ordinary histograms.

If a calculation produces a negative frequency from positive width and density data, a subtraction or reading error has occurred. If frequency is noninteger when the dataset is supposed to contain exact counts, check whether the displayed density was rounded.

An approximate density may yield a product slightly different from an integer; distinguish rounding effects from impossible counts.

Worked example 12: read a rounded bar height

Suppose a fictional histogram bar covers width 8 minutes and its displayed frequency density is approximately 2.6 observations per minute, rounded to one decimal place.

Multiplying 8×2.6 gives 20.8, not an exact whole-number count. This shows the displayed density may have been rounded and does not by itself justify reporting precisely 20.8 observations.

The original exact count or a higher-precision height would be needed to determine the integer frequency reliably.

A frequency-density axis needs a clear label

A graph with vertical numbers 1, 2, 3, 4 but no label invites readers to interpret them as counts or rates. A proper histogram using unequal classes should identify the density scale.

In our invented chart, label the vertical axis “Frequency density (observations per minute)” and horizontal axis “Duration (minutes)”.

Those two labels make the multiplication logic recoverable. The numerical values alone are insufficient to ensure a reader understands what each bar means.

Do not invent a population claim from a sample histogram

Even a correctly computed histogram based on one fictional set of 100 observations says nothing about the wider Bukit Timah population.

In a real dataset, representativeness depends on selection, measurement conditions and exclusions. A histogram describes the distribution of values within its defined dataset, not automatically everyone who might share the activity.

This is a useful link to the earlier Bukit Timah tree-sampling and bias guide. Describing collected observations and generalising to a population are different statistical tasks.

Chart shapes can mislead when axes are cropped

If the horizontal axis omits some intervals or the vertical scale is unclear, the drawing may exaggerate differences. This issue overlaps with general graph-reading discipline but is especially damaging when density and frequency are confused.

The solution is to read the full set of axes, check class boundaries, calculate bar areas and compare them with tabulated counts.

A student can often detect a flawed histogram by asking whether the sum of bar areas corresponds to the stated sample total.

Worked example 13: confirm the total from the chart

Our four histogram bars have areas 5×4=20, 10×3=30, 5×2=10 and 20×2=40.

Sum the areas: 20+30+10+40=100 observations. This recovers the entire sample count from the graphical dimensions.

If a pupil instead sums bar heights 4+3+2+2=11 and reports eleven observations, the error is interpreting a density sum as a count.

A probability can be estimated from histogram frequencies

If one observation is selected uniformly at random from the hundred fictional records, the probability that its duration lies from 20 to under 40 minutes is 40/100=0.4.

The probability of a duration under 15 minutes is (20+30)/100=0.5. These use whole-class counts, so no within-class interpolation is required.

For a threshold such as ten minutes, however, exact probability is not determined from the grouped table alone because the 5–15 class straddles the threshold.

Worked example 14: probability below twenty

There are exactly 20+30+10=60 observations below twenty minutes under the stated half-open classes.

A uniformly selected fictional record therefore has P(duration<20)=60/100=0.6. Its complement is P(duration≥20)=40/100=0.4.

The probabilities add to one because the two events divide the defined hundred observations. This connects histograms to sample spaces without claiming that every minute value is equally likely.

Choosing intervals can influence the appearance

Suppose we regroup the original 100 observations into different duration classes. The shape of the histogram may change because class boundaries and widths change.

Different groupings can expose or conceal clusters, provided the underlying raw data support those groupings. The grouped table alone does not allow arbitrary recalculation into new overlapping intervals without the original records.

A responsible analyst chooses meaningful intervals and reports them rather than shifting boundaries solely to make a preferred story look stronger.

Class width is not always measured in time

Histograms can group masses, heights, ages, temperatures or scores, provided the horizontal variable and class intervals are appropriately defined.

For mass classes measured in kilograms, frequency density has units of observations per kilogram. The same bar-area principle recovers counts.

The formula does not depend on the Bukit Timah duration story. The transferable concept is a count distributed across an interval of a quantitative axis.

Frequency polygons and cumulative curves answer other questions

A frequency polygon can connect appropriate class frequency points in a grouped distribution, depending on the convention. A cumulative frequency curve shows the accumulated count below increasing thresholds.

Neither should be confused with the frequency-density histogram itself. Each representation has different axes or reading rules.

Students should practise saying what a chart encodes before calculating from it. This prevents a method learned in one diagram from being applied incorrectly in another.

Primary preparation: tables, counts and equal intervals

Primary learners can begin with small frequency tables and bar charts, using equal-width categories or simple groups suited to the syllabus.

Have students count observations, check totals and compare which category has more. They should first understand what a count represents before being introduced to frequency density.

An unequal-class histogram can be an enrichment idea for a capable learner, but it is not a replacement for foundational multiplication, division and graph reading.

Lower Secondary progression: read the continuous axis

Learners can move from tables into grouped numerical data, noticing that measurements occupy ranges rather than category names.

Review interval notation, class width and density multiplication as prerequisites. The teacher should explicitly distinguish a histogram from an ordinary bar chart.

The question “Does bar height or bar area represent frequency?” should be answered before any advanced interpolation or grouped-mean calculation.

Upper Secondary extension: uneven widths and estimates

For students whose syllabus includes frequency density, unequal class widths make a valuable diagnostic. The student should calculate and reverse the density formula confidently.

Grouped means, interpolation and distribution analysis should be introduced with their assumptions. They should not be presented as exact recovery of individual observations that were never supplied.

The exact content in a particular SEC Mathematics or Additional Mathematics course should be checked against the relevant syllabus.

Five mistakes and their different repairs

One learner may use frequency as vertical bar height when class widths differ. Another may subtract class endpoints incorrectly. A third may confuse the density axis with an actual count.

A fourth may double-count an observation on a class boundary. A fifth may infer an exact median from grouped data that support only an estimate.

These are different interpretation failures. A good small-group tutorial locates the first wrong representation before asking the student to perform more calculations.

A twenty-minute home diagnostic

Minutes 1–5: Give the four class widths 5,10,5,20 and frequencies 20,30,10,40. Ask for the total of 100 observations.

Minutes 6–10: Divide frequency by width to obtain densities 4,3,2,2 and plot touching bars to scale.

Minutes 11–15: Point to the two height-2 bars and ask why they have frequencies 10 and 40. The learner should refer to widths 5 and 20.

Minutes 16–20: Ask for the number below 20 minutes, which is exactly 60, and below 10 minutes, which is not exactly determined without the within-class data.

A seven-day practice plan

Day one reviews class intervals and boundary conventions. Day two calculates widths. Day three calculates density. Day four checks frequency as bar area.

Day five reconstructs cumulative counts. Day six estimates a grouped mean and names the approximation. Day seven changes the class widths and gives an unfamiliar histogram with one misleadingly tall narrow bar.

Use the same dataset in several representations before changing numbers. Understanding the connection between table and bar area is more important than drawing dozens of decorative charts.

Five challenge questions with checked answers

A. A class width is 5 and frequency is 25. Density? Answer: 25/5=5.

B. A bar spans width 12 and density 3. Frequency? Answer: 36.

C. Our original 20–40 class has density 2. Its frequency? Answer: 20×2=40.

D. Our original four classes contain how many observations below 15 minutes? Answer: 20+30=50.

E. Is the exact number below 10 minutes determined by the original grouped table? Answer: No; the 5–15 class includes values both below and above ten.

Frequently asked questions

Why can a shorter histogram bar represent more observations?

Because bar area, width multiplied by height, represents class frequency in a frequency-density histogram. A wide low bar can have more area than a narrow tall bar.

Is a histogram identical to a category bar chart?

No. A histogram groups a quantitative variable into intervals, while a bar chart often represents separate categories.

Why do histogram bars touch?

Adjacent intervals in a continuous scale have shared boundaries. Touching bars communicate the continuity of that variable.

Can the exact average be found from a grouped histogram?

Usually not without the original individual records. A midpoint-based mean is an estimate under a stated approximation.

What is frequency density measured in?

Counts per unit of the horizontal variable, such as observations per minute when duration is measured in minutes.

Is the modal class always the widest bar?

No. Width, frequency and frequency density represent different quantities. State whether the comparison concerns greatest count or greatest density.

Are these 100 observations real Bukit Timah pupil records?

No. Every duration and frequency is fictional classroom data.

Can a zero-frequency interval be omitted without changing the graph?

Removing its space may distort the horizontal quantitative scale. A zero-height interval may still need to be shown.

How do parents recognise real understanding?

A learner can convert between frequency and density, explain why unequal widths matter and identify which conclusions from grouped data are estimates rather than exact facts.

For Singapore progression, consult the SEAB 2027 G3 Mathematics syllabus directory and MOE Primary Mathematics syllabus. The illustrative histogram is educational artwork and the data are not official statistics.

Continue with Mathematics in Bukit Timah: mean, median, mode and outliers to distinguish exact and estimated descriptive statistics, and Mathematics in Bukit Timah: misleading graphs for broader chart-reading safeguards. The immutable Clementi Secondary 1 Mathematics tutorial reference provides the series’ teaching emphasis on explicit units, staged working and independent reasoning.

The habit to keep is to read the horizontal interval, calculate its width, check the vertical-axis unit, interpret the rectangle’s area and identify which inferences require assumptions about data inside the class. A histogram becomes genuinely useful when the reader can explain those steps.

The other three Mathematics in Bukit Timah guides from this batch: Circle sectors, arcs and ring areas · Linear inequalities and feasible regions · Counterexamples and mathematical proof. Each guide develops a different mathematical skill using defined examples, verified calculations and independent checks.

Discover more from eduKate Singapore

Subscribe now to keep reading and get access to the full archive.

Continue reading