Statistics › Section A · Sampling and data › Topic A.3

Frequency and relative frequency tables

Before any graph or average, raw data usually get organised into a table of counts. This topic builds frequency tables step by step and turns them into the proportions and running totals that answer real questions.

Counting first: frequency

A raw data set is just a pile of values, and the simplest way to start making sense of it is to count. The frequency of a value is the number of times it occurs. Roll a die 20 times and the frequency of 5 is how many rolls landed on 5; survey 20 shoppers and the frequency of "weekly" is how many said weekly. A useful warning carries over from sampling: two people drawing random samples from the same population will usually get different tables, and both can be correct. Tables describe the sample in hand; they do not have to match to be right.

Frequency tables for categorical data

A frequency table has one column for the values or categories and one for their frequencies. For categorical data the categories are usually listed in a sensible order - alphabetical for names, size order for ordered ratings like excellent, good, fair, poor - and the frequency column must total the sample size \(n\). Ordered ratings are still categorical: the order helps a reader scan the table, but the categories are labels, not numbers, so no arithmetic is done on them. The table's first job is answering "how many?" questions exactly, which is something no graph or average can do.

Relative frequency

A count tells you how many; a relative frequency tells you what share. The relative frequency of a value is its frequency divided by the total:

\text{relative frequency} = \frac{f}{n}

If 5 of the 20 shoppers answered "weekly", the relative frequency is \(5/20 = 0.25\), or 25%. Relative frequencies are proportions, so the column always totals 1 (100%) - apart from small rounding gaps, which is worth checking as a way of catching arithmetic slips. Report them as proportions or percentages, but pick one convention per table and say which.

Why percentages beat raw counts

Raw counts mislead whenever the groups being compared are different sizes. Suppose 300 of 1,200 customers at Store A returned within a month, while 120 of 400 customers at Store B did. Store A has more returning customers, but the fair comparison is the relative frequency: \(300/1200 = 0.25\) for Store A against \(120/400 = 0.30\) for Store B. Store B actually keeps its customers better. This is the same lesson as sampling: absolute numbers need a denominator before they mean anything, and a percentage carries its denominator with it.

Cumulative relative frequency

For data whose values have an order - ratings from poor to excellent, wait times in minutes - the cumulative relative frequency adds each relative frequency to all the ones before it. The entry for a value answers a whole family of questions in one number: what proportion of the sample is at or below that value. The final entry is always 1, or 100% within rounding, because the cumulative column has simply added up every share in the table. Cumulative columns only make sense when "at or below" is a meaningful question, which is why they fit ordered ratings and measured values but not unordered categories like blood type.

Grouped tables for measured data

Measured data such as heights or times take many distinct values, so a table with one row per value would be endless. Instead the values are grouped into equal-width intervals: 5-minute bins for waiting times, 2-centimetre bins for heights. One common rule: place the class boundaries so that no observed value can fall exactly on an edge, by giving them one more decimal place than the data (masses recorded to the nearest 0.1 kg could be grouped 49.95 to 51.95 rather than 50 to 52). Each value is counted in exactly one interval, a tally is usually kept while counting, and the frequencies again total \(n\). For example, waiting times recorded to the nearest minute could be grouped \(0.5\)–\(5.5\), \(5.5\)–\(10.5\) and \(10.5\)–\(15.5\): every possible value lands inside an interval, and none lands on an edge.

Worked example: hold times

A support line records how many minutes 20 calls were on hold: \(1, 1, 2, 2, 2, 2, 3\) minutes and so on. The finished counts: 1 minute, frequency 2; 2 minutes, 4; 3 minutes, 5; 4 minutes, 4; 5 minutes, 3; 6 minutes, 2. 1. Check the total: \(2+4+5+4+3+2 = 20\), matching \(n\). 2. Relative frequencies: \(2/20 = 0.10\), \(4/20 = 0.20\), \(5/20 = 0.25\), \(4/20 = 0.20\), \(3/20 = 0.15\), \(2/20 = 0.10\). 3. Check the column: \(0.10+0.20+0.25+0.20+0.15+0.10 = 1.00\). 4. Cumulative relative frequencies: 0.10, 0.30, 0.55, 0.75, 0.90, 1.00 - each entry is the previous one plus the row's relative frequency.

Worked example: reading the table

The hold-time table can now answer questions no single count could. a) What proportion of calls were answered in at most 2 minutes? Read the cumulative relative frequency at 2 minutes: 0.30, or 30%. b) What proportion took between 4 and 6 minutes? Subtract the cumulative entry below the range from the entry at its top: \(1.00 - 0.55 = 0.45\), or 45%. c) What proportion took more than 5 minutes? Everything minus the cumulative entry at 5 minutes: \(1 - 0.90 = 0.10\), or 10%. Check: the 4-, 5- and 6-minute rows hold \(4+3+2 = 9\) of 20 calls, and \(9/20 = 0.45\), which matches answer (b).

Common mistakes to avoid

First, a relative frequency column that misses 1 by more than rounding - re-add the frequencies before trusting anything downstream. Second, cumulative entries computed by adding in the wrong order; the column must grow monotonically to 1. Third, confusing frequency with relative frequency: 12 is a count, 0.30 is a share, and only the share travels between samples of different sizes. Fourth, comparing raw counts across groups of different totals. Fifth, grouping into intervals of unequal width, which makes counts hard to compare, since a wider interval collects more values simply by being wider. Sixth, expecting the rounded column to hit exactly 1.00 every time; a sum like 0.999 is rounding, not error.

Key takeaways

Frequency is a count; a frequency table pairs each value or category with its count, totalling \(n\). Relative frequency, \(f/n\), is the value's share of the whole, and the column totals 1. Percentages make groups of different sizes comparable - always compare shares, not raw counts. Cumulative relative frequency runs a total of shares and answers "at or below" questions; differences of cumulative entries answer "between" questions, and \(1 - \) cumulative answers "more than" questions. Measured data are grouped into equal-width intervals, often with boundaries no observed value can land on.

Flashcards

Flashcard

Frequency

The number of times a value occurs

A raw count for one value or category.

Test your understanding

Question 1 of 5

In a frequency table, a category's frequency is:

Question 2 of 5

40 students were surveyed and 12 named tennis as their favourite sport. The relative frequency of tennis is:

Question 3 of 5

Quiz scores for 50 students have relative frequencies 6 points: 0.10, 7 points: 0.24, 8 points: 0.36, 9 points: 0.20, 10 points: 0.10. What proportion scored at most 8?

Question 4 of 5

Quiz scores for 50 students have relative frequencies 6 points: 0.10, 7 points: 0.24, 8 points: 0.36, 9 points: 0.20, 10 points: 0.10. What proportion scored more than 7?

Question 5 of 5

Gym A saw 360 of its 1,500 members renew; Gym B saw 140 of its 500. Which gym has the higher renewal rate?

← Previous topicNext topic →