02 Section A, Key question 3 - Statistics and content analysis (part 2)

Official and non-official statistics

The source of a statistic and the method used to produce it are separate questions.

Statistics are numerical summaries, such as a number of claimants or the percentage of respondents reporting discrimination. They are quantitative data because they express quantities. Qualitative data concern meanings, descriptions and experiences: an interview account of feeling excluded, for example. Neither type is automatically more accurate. A numerical count may measure the wrong thing, while a detailed account may accurately reveal a particular person's experience without showing how widespread it is.

Official statistics are produced within the official statistical system, including by the Office for National Statistics (ONS) and government departments. Non-official statistics are produced outside that system, for example in a charity's own research, an academic project or a business report. Check the particular publication's status and producer: a private organisation can collect data on behalf of an official body. NatCen, for example, conducts some official surveys as well as its own British Social Attitudes research. NatCen describes these roles.

The production method matters too. Administrative data are records created while an organisation carries out its work, such as processing benefit claims. A survey estimate is calculated from answers obtained from a sample, normally with adjustments intended to estimate a wider population. Official statistics can use either method. The ONS Claimant Count uses benefits administration records; Labour Force Survey unemployment estimates use survey answers and an internationally agreed definition. These measure different populations and cannot simply be treated as interchangeable unemployment totals. ONS explains the distinction.

For a sociologist reusing a published statistical table, it is secondary data: somebody else collected the underlying information for an earlier purpose. This can save time and provide national or historical coverage that a small new project cannot obtain. However, the original categories may not match the new research question. A charity's client count might illuminate demand for its services while omitting people who need help but never approach that charity. Non-official evidence is not automatically unrepresentative, just as official evidence is not automatically a census of everyone.

Measurement and statistical quality

A useful inequality statistic needs a clear measure and a population appropriate to the question.

Operationalisation means translating an idea into something observable. Poverty, for example, can be measured through income, lack of necessities or reported difficulties meeting costs. These indicators overlap but are not identical. In the Department for Work and Pensions' Households Below Average Income publication for financial years ending 1995–2025, relative low income means having net equivalised disposable household income below 60% of that year's median. Equivalisation adjusts income for differences in household size and composition; the median is the middle value in an ordered distribution. This is a specified income indicator of living standards, rather than a direct count of every experience of hardship. Comparisons must also keep the choice of before or after housing costs consistent. DWP sets out the measure.

Standardisation means using consistent definitions and procedures. It supports reliability, meaning consistency of measurement, and makes comparison easier. It does not mean that a reliable series must show unchanged results: real circumstances can change. Validity concerns whether the measure captures the phenomenon being investigated. Benefit records can reliably count registered claims while being an incomplete measure of unemployment. Some unemployed people do not claim, and some people included in the Claimant Count do not meet the survey definition of unemployment. Changes to eligibility can alter counts even without an equivalent change in people's employment circumstances.

Categories are socially constructed in the sense that people choose and apply definitions; they are not labels supplied automatically by society. This does not make the resulting evidence fictional. Transparent definitions, consistent collection and published quality checks allow researchers to judge what it can support. The England and Wales Census 2021 ethnic-group question added a Roma tick-box compared with 2011. The appearance of a new category does not mean the population previously did not exist. ONS describes the overall ethnic-group variable as broadly comparable, while explaining the category changes. A comparison of a newly separated group needs more care than a comparison of unchanged broad categories. See the ONS variable guidance.

Coverage asks who is included. Administrative records may cover the registered cases very well but exclude unregistered people. Surveys also face sampling uncertainty, because only part of the population is observed, and non-response bias, where respondents differ systematically from people who do not reply. Imagine an original, invented survey of insecure employment in which workers with the least predictable hours are hardest to contact. Their omission could make insecurity appear less common. A larger sample of the easier-to-contact workers would not necessarily solve this problem. Weighting adjusts respondents' contributions to an estimate, but cannot guarantee that every unmeasured difference has been corrected.

NatCen's British Social Attitudes 2024 survey illustrates why comparison requires reading the technical report. It invited people through sampled addresses, offered online or telephone completion and widened its scope to include Northern Ireland and people aged 16–17. Its frame covered private households, excluding institutional residents and addresses absent from the frame. A comparison with earlier surveys of adults aged 18 and over in Great Britain therefore needs an equivalent population; NatCen provides a specific comparison weight for that purpose. Its report also explains non-response adjustments. The original technical report makes these decisions visible.

The practical judgement is to ask whether dates, geography, population, definitions and collection procedures match. National statistics are often valuable for finding patterns across class, gender, ethnicity or age, but a broad category can conceal variation within it. Numbers alone also give limited access to how inequality feels or how people understand it.

Counts, percentages and change

A count becomes a useful comparison only when its denominator is clear.

The denominator is the total against which a count is compared. The following dataset is entirely invented for teaching. Two surveys in an imaginary town ask different samples of employees aged 18 and over whether their employer changed a shift's start time with less than 24 hours' notice during the previous four weeks. Pay means gross annual employment pay; the £25,000 boundary is an invented grouping choice, not an official poverty threshold. Assume the wording and grouping rules are the same in both years.

In 2024 each group has 40 respondents reporting a short-notice change. This does not mean the experience is equally common in each group: 40 out of 200 is 20%, whereas 40 out of 400 is 10%. The lower-pay group has twice the reported proportion. The denominator here is respondents within each pay group, not all employees in the town or all 600 respondents combined.

For the lower-pay group, the share rises from 20% to 30%. The percentage-point increase is 30 − 20 = 10 percentage points. The percentage increase relative to the original share is (30 − 20) ÷ 20 × 100 = 50%. Saying “a 10% increase” would describe a different calculation: a 10% increase on 20% would give 22%.

The higher-pay share rises by 5 percentage points, from 10% to 15%, but its relative increase is also 50%. The lower-pay group therefore has the larger percentage-point increase, while both groups have the same proportional increase. In 2025 the gap between them is 30 − 15 = 15 percentage points. These are descriptions of this invented responding sample, rather than estimates proven representative of a real town.