Cumulative Frequency

⚡ Key Takeaways

Cumulative frequency is the running total of frequencies up to and including a given value or class in a dataset. In words: it tells you how many observations fall at or below a certain point. It applies anywhere data is grouped or ordered — statistics, quality control, exam grading, demographics. Formalised through the development of descriptive statistics in the 19th–20th centuries, it underpins medians, quartiles, and percentiles.

💡 The Core Insight

A frequency distribution describes data locally (how much sits in each bin). Taking a running sum converts it into a global, positional description: at every value you instantly know what fraction of the data lies below. This single transformation turns counting into ranking, and ranking is exactly what medians, quartiles and percentiles require.

DefinitionCumulative Frequency

Suppose a dataset of $n$ observations is sorted into ordered classes (or distinct values) with upper boundaries $x_1 < x_2 < \dots < x_k$ and corresponding frequencies $f_1, f_2, \dots, f_k$ . The cumulative frequency $F_j$ up to the $j$ -th class is the total number of observations falling at or below $x_j$ :

1

\quad F_j = \sum_{i=1}^{j} f_i

Derivation hint: it is a definition — read it as "add up all frequencies from the first class through the $j$ -th". Here $f_i$ is the frequency of the $i$ -th class, and the symbols are:

$n$ = total number of observations,
$k$ = number of classes,
$x_j$ = upper boundary of the $j$ -th class,
$F_j$ = cumulative frequency to class $j$ .

The final cumulative frequency equals the total:

2

\quad F_k = \sum_{i=1}^{k} f_i = n

Derivation hint: every observation belongs to exactly one class, so summing all frequencies counts each observation once. A useful recursive form lets you build the table one step at a time:

3

\quad F_j = F_{j-1} + f_j, \qquad F_0 = 0

Derivation hint: split the sum in [1] into the first $j-1$ terms (which equal $F_{j-1}$ ) plus the last term $f_j$ .

DefinitionRelative and Percentage Cumulative Frequency

Dividing by the total gives the relative cumulative frequency, an estimate of the proportion of data at or below $x_j$ :

4

\quad G_j = \frac{F_j}{n}, \qquad 0 \le G_j \le 1

Derivation hint: divide [1] by $n$ ; since $F_k = n$ , the largest value is $G_k = 1$ . Multiplying by 100 yields the percentage cumulative frequency $P_j = 100\,G_j$ , the basis of percentiles.

TheoremMedian from Cumulative Frequency

For grouped data, the median lies in the class where the cumulative frequency first reaches or exceeds $n/2$ . It is estimated by linear interpolation within that class:

5

\quad \text{Median} = L + \left(\frac{\tfrac{n}{2} - F_{<}}{f_m}\right) h

Derivation hint: assume observations are uniformly spread across the median class of width $h$ ; the fraction of that class needed to reach the $n/2$ -th observation is $(\tfrac{n}{2}-F_{<})/f_m$ . where $L$ = lower boundary of the median class, $F_{<}$ = cumulative frequency before the median class, $f_m$ = frequency of the median class, $h$ = class width.

ProofInterpolation Formula for the Median

We derive equation [5] step by step. 1. Locate the median position. With $n$ ordered observations, the median sits at position $n/2$ along the cumulative count. We find the class whose cumulative frequency first reaches $n/2$ ; call it the median class with lower boundary $L$ , frequency $f_m$ , width $h$ . 2. Count what lies below. Let $F_{<}$ be the cumulative frequency of all classes before the median class. Then the median observation is the $(\tfrac{n}{2} - F_{<})$ -th observation within the median class. 3. Assume uniform spread. We cannot see individual values inside a grouped class, so we assume they are evenly distributed across the interval $[L, L+h]$ . Then $f_m$ observations occupy width $h$ . 4. Find the fractional position. The $(\tfrac{n}{2}-F_{<})$ -th of the $f_m$ observations sits at a fraction

\frac{\tfrac{n}{2}-F_{<}}{f_m}

of the way across the interval. 5. Convert fraction to value. Multiply by width $h$ and add the lower boundary $L$ , giving equation [5]. $\blacksquare$

∎

🧠 Intuition

Think of the cumulative frequency curve as a "filling tank". As you scan rightward through the data, the tank fills: each class pours in its frequency, and the curve rises by that much. It never falls — cumulative frequency is monotonically non-decreasing, because you can only add, never subtract, observations. The steepness of the curve at any point reflects how densely packed the data is there: a steep rise means many observations in a small range; a flat stretch means a gap. The whole curve always climbs from $0$ up to $n$ .

mathematician · Francis Galton | 1822–1911

Galton, a Victorian polymath and cousin of Charles Darwin, pioneered the graphical and quantitative study of distributions. He coined the term ogive for the cumulative curve (borrowing from the S-shaped architectural arch), introduced percentiles and the median as practical summaries, and used cumulative methods to study human variation. His work, alongside that of Karl Pearson, turned the cumulative frequency curve from a curiosity into a standard statistical tool.

Figure

An ogive (cumulative frequency curve) for exam marks, showing how the median and quartiles are read off

ExampleBuilding a Cumulative Frequency Table

Given: Test scores of 40 students grouped as follows.

Class	Frequency $f$
0–20	4
20–40	11
40–60	13
60–80	8
80–100	4

Steps: Apply the recursive rule [3], $F_j = F_{j-1} + f_j$ .

Up to 20: $F = 4$
Up to 40: $F = 4 + 11 = 15$
Up to 60: $F = 15 + 13 = 28$
Up to 80: $F = 28 + 8 = 36$
Up to 100: $F = 36 + 4 = 40$

Check with [2]: $F_5 = 40 = n$ . ✓

Mark ≤	Cumulative frequency
20	4
40	15
60	28
80	36
100	40

\boxed{\text{Cumulative frequencies: } 4,\ 15,\ 28,\ 36,\ 40 \text{ students}}

ExampleEstimating the Median by Interpolation

Given: Use the table above. $n = 40$ , so $n/2 = 20$ .

Steps: 1. The cumulative frequency first reaches or exceeds 20 in the class 40–60 (cf jumps from 15 to 28). This is the median class. 2. Identify quantities: $L = 40$ , $F_{<} = 15$ , $f_m = 13$ , $h = 20$ . 3. Apply equation [5]:

\text{Median} = 40 + \left(\frac{20 - 15}{13}\right)\times 20 = 40 + \frac{5}{13}\times 20

4. Compute: $\frac{5}{13}\times 20 = \frac{100}{13} \approx 7.69$ . 5. Median $\approx 40 + 7.69 = 47.69$ .

\boxed{\text{Median} \approx 47.7 \text{ marks}}

⚠️ Common mistake

Plot at the upper class boundary, not the midpoint. A "less than" cumulative frequency table counts everything up to and including the upper boundary of each class. So when drawing an ogive, each cumulative frequency must be plotted against the class's upper boundary (e.g. 20, 40, 60...), not its midpoint. Plotting at midpoints shifts the entire curve and corrupts every median and quartile read off it. Also remember to start the curve at the lower boundary of the first class with cumulative frequency 0.

Counterexample

A student claims "cumulative frequency can go up or down depending on the data." This is false. Because each step adds a non-negative frequency (equation [3]), the cumulative frequency can never decrease — it is monotonically non-decreasing. Consider frequencies $5, 0, 3$ : the cumulative values are $5, 5, 8$ . Even when a class has frequency $0$ , the cumulative frequency stays flat (5 → 5); it does not drop. Any "falling" cumulative frequency table contains an arithmetic error.

connection

Cumulative frequency sits at the crossroads of several ideas. It is the discrete analogue of the Cumulative Distribution Function in probability, where the running sum becomes a running integral. The reading-off of Quartiles and Percentiles from an ogive directly generalises the Median construction. It complements the ordinary Frequency Distribution (the two are linked by Finite Differences: differencing the cumulative table recovers the frequencies). The S-shaped ogive previews the Normal Distribution's cumulative curve, and the spread read from quartiles feeds into the Interquartile Range and Box Plot.

🔬 Where this appears

⏳ History

- 1654 — Pascal and Fermat lay groundwork for probability; the idea of accumulating outcomes appears implicitly.

- 1662 — John Graunt's Natural and Political Observations on the London Bills of Mortality builds early life tables, effectively cumulative survival counts.

- 1693 — Edmond Halley constructs a more rigorous life table from Breslau data, summing deaths cumulatively by age.

- 1830s–40s — Adolphe Quetelet promotes the statistical study of the "average man", popularising frequency distributions.

- 1875 — Francis Galton introduces the ogive and the use of the median and quartiles read graphically.

- 1885 — Galton formalises percentiles and uses cumulative curves to study human stature and inheritance.

- 1895 — Karl Pearson systematises descriptive statistics, including cumulative frequency methods, in his lectures and papers.

- 1905 — Max Lorenz introduces the Lorenz curve, a cumulative-frequency tool for inequality.

- 20th century — Cumulative frequency tables and ogives become standard in school and university statistics curricula worldwide.

open-problem

The interpolation formula [5] rests on the uniform-distribution-within-class assumption, which is rarely exactly true. If data clusters near one end of a class, interpolated medians and percentiles carry systematic error — yet there is no universal "best" interpolation rule for grouped data; competing conventions (linear, quantile types) yield slightly different answers, and statistical software offers up to nine definitions. A related practical subtlety is bin-width choice: too wide and you lose resolution, too narrow and the curve becomes jagged and unstable. How to choose optimal class boundaries from raw data, balancing smoothness against fidelity, remains a genuine question in nonparametric statistics (kernel density estimation is the modern continuous successor).

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also