Relative Frequency Explained: Tables, Cumulative Frequency, and Probability

Relative frequency is the share of a whole that belongs to one category: the count for that category divided by the total count of all categories. It turns a tally — 12 red, 28 blue, 20 green — into shares — 0.20, 0.47, 0.33 — that can be compared across data sets of different sizes. The arithmetic is one division per category. The discipline is the table: every category listed once, the shares adding to 1, and the cumulative column ending exactly at 1.00. This guide covers the table, cumulative frequency, percents, and the relationship with probability.

One category, one division

The definition is the whole method. Count how many observations fall in the category, count the total, divide. A survey of 60 people in which 12 prefer red gives a relative frequency of 12 ÷ 60 = 0.20 for red. As a percent that is 20%. As a fraction it is 1/5. The three forms say the same thing, and the form to use is the one the reader can check: a share with its counts, such as “12 out of 60,” is verifiable by division in any form, while a share without its counts is an assertion, however it is written.

The share is bounded. It cannot be below 0 — a negative count is impossible — and it cannot exceed 1, because a part cannot exceed the whole. A relative frequency above 1, such as 1.1667 for a count of 70 against a total of 60, means the inputs are invalid: the part exceeds the whole, and the correct response is to check the count and the total, not to quote the figure as a share. The Relative Frequency Calculator warns in that case, and the warning should be treated as a data error, not a result.

Reordering the list does not change the shares. The relative frequency of each category depends only on its own count and the total, so sorting the table by count, by name, or by the order the data arrived in leaves every share untouched. What reordering does change is the cumulative column, which is order-dependent by design — the next section explains why that is a feature, not a flaw.

A list of categories, and why the shares add to one

A relative frequency table lists every category once, with its count, its relative frequency, and usually its percent. The counts add to the total; the relative frequencies add to 1; the percents add to 100. Those three sums are the table’s self-check, and a table that fails any of them has an error: a missing category, a double-counted observation, or a rounding slip. Build the table with the counts first, derive the shares from them, and check the sums before interpreting anything.

Rounding is the usual reason the displayed shares add to 0.99 or 1.01. The cure is to carry more decimals through the calculation and round only for display, and to note the rounding where the table is published. A table whose shares add to 0.99 with a footnote “shares may not add due to rounding” is honest; a table whose shares add to 0.99 with no explanation invites the reader to hunt for a missing category. Decimals and percents are the same share in two clothes: 0.20, 20%, and 1/5 are one number, and mixing forms inside one table without labelling them is the fastest way to confuse a reader.

Worked table. Colours chosen by 60 people: red 12 (0.20, 20%), blue 28 (0.4667, 46.67%), green 20 (0.3333, 33.33%). The counts add to 60; the shares add to 1.0000; the percents add to 100.00. Blue is the modal category — the largest share — and the table shows it at a glance, which is the table’s purpose: comparison without re-reading the raw tally.

Cumulative frequency: the running total

Cumulative relative frequency adds the shares down the table, in order. After red it is 0.20; after blue, 0.20 + 0.4667 = 0.6667; after green, 1.0000. It answers “at most” questions: the share in the first two categories combined is 0.6667, or two-thirds. For ordered data — ages, scores, income bands — the cumulative column is the natural way to read thresholds: the share at or below a value, the share in the bottom half, the median band. For unordered categories — colours, names, flavours — the cumulative column depends on the listing order and carries no meaning beyond the running sum, so it is often omitted.

The last cumulative value must be 1.00. It is the whole table, added up, and anything else means a category is missing or a share is wrong. That check is the reason the cumulative column exists in a table even when nobody asks an “at most” question: it is a second self-check, independent of the share sum, that catches the same errors from a different direction. If the cumulative column ends at 0.98, do not interpret the table — fix it.

Cumulative frequency without the “relative” is the same running total in counts: 12, then 40, then 60. The relative form divides by the total; the count form does not. Both are used, and the label must say which. A column headed “cumulative” with no qualifier is ambiguous, and the ambiguity matters: 0.67 and 40 are both correct cumulative figures for the second row, in different units.

Probability, percentage points, and four mistakes

Relative frequency is the empirical cousin of probability. Probability is the long-run share a model predicts; relative frequency is the share a sample actually showed. Toss a fair coin 100 times and the relative frequency of heads will be near 0.50 but rarely exactly 0.50; toss it 10,000 times and it will be nearer still. The law of large numbers is the claim that the observed share converges to the probability as the sample grows. A relative frequency is therefore evidence about a probability, not the probability itself, and a small sample’s share is weak evidence — 2 out of 4 is 0.50 with almost no information in it.

Percentage points versus percent change is the distinction that most often corrupts a share’s interpretation. A share rising from 20% to 25% rises by 5 percentage points, and by 25% in relative terms (5 ÷ 20). Both are correct; they answer different questions, and the one that is quoted without its label misleads. “Support rose 5%” for a 20% → 25% move understates it fivefold in the reader’s ear; “support rose 25%” overstates it if the reader hears points. State which, every time.

Four mistakes, collected. Counting an observation twice, or missing a category, so the shares do not add to 1. Dividing by the wrong total — the total of one subgroup instead of the whole table. Rounding intermediate shares and then adding the rounded figures, so the sum drifts. And reading a cumulative column on unordered categories as if the order meant something. A routine prevents all four: list every category once, divide by the grand total, carry full precision to the end, and check both sums — shares to 1, cumulative to 1.00 — before writing a word about what the table means.

Charts: bars for shares, pies for the whole

A relative frequency table becomes a chart by turning the shares into lengths or angles. A bar chart draws one bar per category, its length proportional to the share, and it is the right chart for comparing categories: the eye judges lengths well, the ordering can be by size, and the values can be labelled on the bars. A pie chart draws one slice per category, its angle proportional to the share, and it is the right chart for showing the whole: the eye judges the whole circle, and the slices’ shares of it. The two charts answer different questions — comparison against the whole — and the chart that answers the wrong question misleads however accurately it is drawn.

Bar charts tolerate many categories and small differences; pie charts tolerate neither. A pie with twelve slices is a puzzle, not a picture, and a pie whose slices differ by two percentage points shows no difference at all. A bar chart with twelve bars is readable, and a difference of two points is a visible difference in length. The rule is practical: fewer than six categories and a whole to show, a pie is acceptable; more categories, or a comparison to make, use bars. And never use a three-dimensional pie or bar: the perspective distorts the lengths and angles the chart exists to show, and the distortion is the chart’s only effect.

Label every chart with its counts as well as its shares. A bar labelled 20% without its count — 12 out of 60 — hides the sample size, and a share from a sample of 20 carries far less weight than the same share from a sample of 2,000. The table is the chart’s evidence, and the chart without the table is an assertion. Publish them together, with the total stated, and the reader can check the chart against the table in seconds — which is the point of both.

Two-way tables: joint and conditional shares

When two categorical variables are recorded together — sex and preference, age group and product, region and outcome — the table has two dimensions, and the relative frequencies split into three kinds. The joint share is the cell divided by the grand total: the share of everyone who is both in the row and in the column. The row-conditional share is the cell divided by its row total: the share within that row. The column-conditional share is the cell divided by its column total: the share within that column. The three answer different questions, and quoting one as if it were another is the commonest error in survey reporting.

Worked example. One hundred people were asked whether they prefer tea: 60 men, of whom 15 prefer tea, and 40 women, of whom 28 prefer tea. The joint shares: men who prefer tea, 15 out of 100, 0.15; women who prefer tea, 28 out of 100, 0.28. The conditional shares: among men, 15 out of 60 prefer tea, 25%; among women, 28 out of 40, 70%. The overall share who prefer tea is 43 out of 100, 43%. The conditional shares show the difference between the groups — 70% against 25% — which the joint shares, 0.28 against 0.15, understate, because they are diluted by the groups’ different sizes.

The question decides the conditioning. “What share of women prefer tea?” conditions on women: divide by the row total for women. “What share of tea drinkers are women?” conditions on tea drinkers: divide by the column total for tea, 28 out of 43, 65%. The two answers — 70% and 65% — are both correct, and they are answers to different questions, and a report that quotes one without saying which has not reported a finding but a fraction. State the denominator every time: “28 of the 40 women” is verifiable; “70%” alone is not.

Frequently asked questions