Statistics, surveys and quality: statistics terms explained

The status labels, data sources and quality measures that tell you how much weight a published figure can bear.

Accredited official statistics

Official statistics that the Office for Statistics Regulation has independently reviewed and confirmed meet the highest standards of the Code of Practice for Statistics.

Accredited official statistics are official statistics that the Office for Statistics Regulation (OSR) has assessed and found to comply with the Code of Practice for Statistics, covering trustworthiness, quality and value. The label replaced the older term "National Statistics" in 2023; the meaning is the same.

What it tells you

The producer is independent in how it compiles and publishes the figures.
Methods and quality are documented and open to scrutiny.
Release dates are announced in advance.

Worked example

Suppose two hypothetical sources give different crime figures: one an accredited official statistic, the other a one-off departmental estimate. Accreditation does not guarantee the first is right in every detail, but it tells you its methods have passed independent review, which makes it the stronger starting point.

How it trips up claims

Accreditation can be suspended or removed if the OSR finds problems; OSR publishes lists of accredited, suspended and removed statistics.
Statistics without accreditation can still be official statistics, such as official statistics in development.

Also searched as: National Statistics, accredited statistics, what are accredited official statistics, kitemark

Official definitions: Office for Statistics Regulation: Accredited official statistics · OSR: Accredited official statistics policy · Office for Statistics Regulation: Code of Practice for Statistics

Official statistics in development

New or substantially changed official statistics that are still being tested and developed, and published early so users can give feedback. Formerly 'experimental statistics'.

Official statistics in development are official statistics that are new, or are being substantially changed, and are published while their methods are still being tested. The label replaced "experimental statistics" in 2023. Producers use it to involve users early and to be open about the stage of development.

What it tells you

The figures follow the Code of Practice for Statistics, but methods may change.
Larger revisions are more likely than for established series.
The producer should explain the limitations and invite feedback.

Worked example

A hypothetical new estimate shows 120,000 people in a category this year. A methods change next year, applied to the same period, gives 105,000. Both were published openly; the change reflects development, not a hidden correction.

How it trips up claims

Treat precise year-to-year changes in a developing series with care; check whether the producer recommends comparisons over time.
The label is a description of maturity, not a judgement that the figures are unreliable. Many series move on to become accredited official statistics.

Also searched as: experimental statistics, statistics in development meaning, official statistics in development definition

Official definitions: OSR: Official statistics in development · OSR: Guidance on producing official statistics in development

Pre-release access

Limited access to official statistics given to named officials and ministers shortly before publication, under rules set by law, so they can prepare responses.

Pre-release access is the practice of giving a small number of named people, such as ministers and their officials, sight of official statistics before they are published. For statistics produced by UK government departments it is governed by the Pre-release Access to Official Statistics Order 2008, which limits access to no more than 24 hours before release. Departments publish lists of who receives it. The devolved administrations have their own rules.

Why it matters for checking claims

Nobody with pre-release access may comment publicly on the statistics before the official publication time.
Statistics should be released at the pre-announced time (commonly 9:30am for ONS and many departments), so everyone gets the figures at once.
The ONS stopped giving pre-release access to its own statistics in 2017.

Worked example

A hypothetical bulletin is due at 9:30am on Wednesday. Officials named on the pre-release list may see it from 9:30am on Tuesday to prepare a briefing. A public statement quoting the figures before 9:30am Wednesday would be a breach the Office for Statistics Regulation could investigate.

Also searched as: pre-release, early access to statistics, pre-release access list, embargo statistics

Official definitions: Pre-release Access to Official Statistics Order 2008 (legislation.gov.uk) · GOV.UK: Pre-release access to official statistics, statement of compliance · Office for Statistics Regulation: Code of Practice for Statistics

Survey sample

The group of people or businesses selected to take part in a survey, chosen so that their answers can be used to estimate figures for the whole population.

A survey sample is the subset of a population that is selected and asked to take part. Results are weighted (grossed up) so that they represent the whole population. The Labour Force Survey and the Crime Survey for England and Wales are examples.

Worked example

If a hypothetical survey interviews 10,000 people from a population of 50 million, each respondent represents on average 5,000 people. If 400 report being unemployed, the estimate is about 2 million. Weighting adjusts this so that, for example, age and region match known population totals.

Sampling error

A different random sample would give a slightly different answer. This sampling variability is why survey estimates come with confidence intervals. Larger samples give narrower intervals, but small subgroups, such as one region or age band, have fewer respondents and wider uncertainty.

How it trips up claims

A small monthly change in a survey estimate can be within sampling variability; check whether it is statistically significant.
If certain groups do not respond, the sample may not represent them well even after weighting (see response rate).

Also searched as: sample survey, sample size, sampling error, survey weighting

Official definitions: ONS: Labour Force Survey user guidance · ONS: Uncertainty and how we measure it · ONS: Crime in England and Wales, latest bulletin

Response rate

The share of people or businesses selected for a survey who actually take part. Lower response rates can make estimates less precise and risk bias.

The response rate is the proportion of the selected survey sample that completes the survey. Producers publish it as a quality indicator.

Worked example

A hypothetical survey selects 20,000 addresses and achieves 8,000 interviews: a response rate of 40%. If response falls to 20%, only 4,000 interviews remain. The confidence intervals widen, and if the people who stop responding differ from those who stay, the estimate can be biased even after weighting.

Why it matters

Response rates to many household surveys have fallen over time in the UK and other countries.
The ONS publishes Labour Force Survey performance and quality reports showing response rates and achieved sample sizes, and has reweighted and developed the survey to address lower response.

How it trips up claims

Sharp movements in a survey with a low response rate deserve a check against other sources, such as administrative data like payroll records.
Producers sometimes flag particular estimates as less reliable during periods of low response; read the bulletin's quality notes.

Also searched as: survey response rate, non-response bias, LFS response rate, achieved sample

Official definitions: ONS: Labour Force Survey performance and quality monitoring reports · ONS: Labour Force Survey user guidance

Administrative data

Information collected by organisations while running a service, such as tax records, benefit claims or hospital activity, and then used to produce statistics.

Administrative data are records created in the course of running a service, not collected for statistics. Examples include HMRC PAYE payroll records, DWP benefit claims, NHS hospital records, school census returns and police recorded crime. Producers turn these into official statistics.

Strengths and limits

They often cover everyone using the service, so they give detail for small areas and groups that surveys cannot.
They only record what the system records. Changes in rules, recording practice or IT systems can change the numbers without any change in the real world.

Worked example

Suppose hypothetical police recorded crime rises 10% in a year after a force improves how it logs reports. The rise may reflect better recording rather than more crime. That is why the ONS reports recorded crime alongside the Crime Survey for England and Wales, a sample survey that is not affected by recording practice.

How it trips up claims

Check the methodology notes for changes in rules or recording before reading a trend.
The OSR publishes a toolkit on quality assurance of administrative data, which producers use to explain known limits.

Also searched as: admin data, administrative records, register data, recorded data

Official definitions: OSR: Administrative data quality assurance toolkit · ONS: Crime in England and Wales, latest bulletin

Provisional figures

Early estimates published before all data are in, so users get timely figures. They are expected to be updated when later data arrive.

Provisional figures are first estimates released quickly, before all source data are available. They are labelled provisional so users know they may change. Later versions may be called revised, final or confirmed.

Worked example

A hypothetical provisional count of deaths for a month is 40,000, because some deaths are registered weeks after they occur. When late registrations arrive, the count becomes 41,200. Comparing the provisional figure for this year with a final figure for last year would understate this year's total.

Where you will see it

ONS long-term international migration estimates are published as provisional and updated later.
Monthly public sector finances and early GDP estimates are provisional and routinely revised.
Many health and crime releases flag the latest period as provisional.

How it trips up claims

Compare provisional with provisional, or final with final, where possible.
A headline built on the newest provisional month may change by the next release; check the producer's revisions information.

Also searched as: provisional data, preliminary estimates, first estimate, flash estimate

Official definitions: ONS: Long-term international migration, provisional, latest bulletin · ONS: Revisions

Revisions

Changes to previously published figures, made when more complete data, improved methods or updated seasonal factors become available. A normal part of official statistics.

Revisions are changes to figures already published. They happen because early estimates rely on incomplete data, because seasonal adjustment is re-estimated, or because methods are improved. Producers set out revisions policies and often publish analysis of how large past revisions have been.

Worked example

Hypothetical first-estimate quarterly GDP growth is 0.1%. Three months later, with more survey and tax data, it is revised to 0.3%. A year later, after annual benchmarking, it becomes 0.2%. Each version was the best estimate at the time.

Types of revision

Scheduled: planned updates as more data arrive, such as later GDP estimates.
Methodological: changes to how a series is compiled, usually announced in advance.
Corrections: fixing errors, which producers must explain under the Code of Practice.

How it trips up claims

A claim made using an older vintage of data may no longer match the current published series. Note the date you accessed a figure.
Small changes near zero, such as whether a quarter shows slight growth or contraction, are especially sensitive to revision.

See how to understand revisions.

Also searched as: revised figures, data revisions, why do statistics get revised, revisions policy

Official definitions: ONS: Revisions · ONS: Quarterly national accounts, latest bulletin · Office for Statistics Regulation: Code of Practice for Statistics

Confidence interval

A range around an estimate that shows how much it could vary because it comes from a sample. A 95% interval is the usual standard.

A confidence interval is a range around an estimate that reflects sampling uncertainty. A 95% confidence interval means that if the survey were repeated many times with fresh random samples, about 95% of the intervals calculated would contain the actual population value.

Worked example

A hypothetical survey estimates unemployment at 1,500,000 with a 95% confidence interval of plus or minus 80,000, so the range is 1,420,000 to 1,580,000. If the next estimate is 1,530,000, the change of 30,000 is well inside that range and may simply reflect sampling.

How it trips up claims

Two estimates whose intervals overlap may not be meaningfully different. A formal test of statistical significance on the change is the better check.
Smaller groups have wider intervals, so regional or subgroup rankings are often less certain than they appear.
Confidence intervals cover sampling error only. They do not capture non-response bias or measurement problems.

ONS bulletins and the Crime Survey for England and Wales note when changes are statistically significant.

Also searched as: margin of error, 95% confidence interval, CI, sampling variability

Official definitions: ONS: Uncertainty and how we measure it · ONS: Labour Force Survey user guidance

Statistical significance

A test of whether a difference between estimates is larger than would be expected from sampling variation alone. It does not mean the difference is large or important.

A difference is statistically significant when a statistical test indicates it is unlikely to be the result of random sampling variation alone. Official bulletins commonly use a 5% significance level, linked to 95% confidence intervals.

Worked example

A hypothetical survey shows 20% of people experienced an issue last year and 21% this year. With a large sample, a one-point rise might be significant; with a small sample, the same rise might not be. Significance depends on the size of the change and the sample, not only on the change.

What it does and does not mean

Significant does not mean large, important or caused by a particular policy.
Not significant does not prove there was no change; the data may be too limited to detect it.
Testing many comparisons will produce some significant results by chance.

How it trips up claims

A claim that something "rose" based on a non-significant change in a survey goes beyond what the producer reports. The Crime Survey for England and Wales, for example, states whether each change is statistically significant.
Significance tests apply to sample data, not to complete counts such as registered deaths.

Also searched as: statistically significant, significance test, p-value, what does statistically significant mean

Official definitions: ONS: Uncertainty and how we measure it · ONS: Crime in England and Wales, latest bulletin

Headline figure

The main summary number that a statistical release leads with, such as annual CPI inflation or quarterly GDP growth. Other measures in the same release can tell a different story.

A headline figure is the main measure a statistical producer leads with in a release. Examples: the annual CPI rate in the inflation bulletin, quarter-on-quarter real GDP growth, or the ILO unemployment rate for people aged 16 and over.

Why producers choose a headline

A single lead figure helps users find the most used measure quickly and keeps reporting consistent from month to month. The full release usually contains many more measures.

Worked example

In a hypothetical month, headline CPI inflation falls from 3.5% to 3.0%, while core inflation stays at 3.2% and services inflation rises. "Inflation fell" describes the headline; the other measures add context.

How it trips up claims

A claim may use a different measure from the headline (for example RPI or a single component) without saying so.
Headline labour market figures cover three-month periods, not single months (see rolling average).
Headline debt figures exclude public sector banks; other debt measures exist.

See how to read an ONS bulletin.

Also searched as: headline rate, headline inflation, main figure, lead measure

Official definitions: ONS: Consumer price inflation, latest bulletin · ONS: Labour market overview, latest bulletin

Sample size

The number of people, households or businesses that took part in a survey. Larger samples give more precise estimates, but size alone does not remove bias.

Sample size is the number of units, such as people, households or businesses, whose responses an estimate is based on. Official surveys report the achieved sample, which can be far smaller than the number of people approached because of non-response.

Why it matters when checking a claim

Precision falls fast in sub-groups. A survey of tens of thousands of households can still have only a few hundred responses for one region, age group or ethnic group. Estimates for small groups have wide confidence intervals, and producers may suppress them (see data suppression).
Big is not the same as representative. A large self-selected online poll can be less reliable than a smaller random sample. See selection bias.
Look for the quality notes. ONS bulletins and methodology pages give achieved sample sizes and response rates, and the Labour Force Survey has regular performance and quality reports.

Also searched as: sample size meaning, how big a sample, achieved sample, small sample

Official definitions: ONS: Labour Force Survey user guidance · Office for Statistics Regulation: Code of Practice for Statistics

Labour Force Survey (LFS)

The ONS household survey behind the UK's headline employment, unemployment and inactivity figures.

The Labour Force Survey (LFS) is the ONS's large continuous household survey of the UK labour market. It is the source for the headline employment rate, ILO unemployment and economic inactivity figures, as well as NEET estimates and data on hours worked.

Why it matters when checking a claim

Response problems. LFS response fell sharply from 2020. Because of quality concerns, LFS-based statistics lost their accredited status and were labelled official statistics in development for a period; the ONS changed this designation to official statistics in August 2026. Check the current label in each release.
Reweighting. The ONS periodically reweights the survey to new population estimates, which can revise the level of employment by large amounts.
Rolling quarters. Monthly releases show three-month averages, so consecutive releases overlap by two months.
A transformed survey. The ONS has been developing a Transformed Labour Force Survey to replace it; any switch may create a break in the series.

See the labour market statistics guide.

Also searched as: LFS, Labour Force Survey uk, Transformed Labour Force Survey, TLFS, LFS response rate

Official definitions: ONS: Labour market overview, UK, latest bulletin · ONS: Labour Force Survey user guidance

Correlation and causation

Two things moving together (correlation) does not show that one causes the other (causation). Other factors, chance or reverse causation can explain the pattern.

Correlation means two measures tend to move together. Causation means a change in one actually produces a change in the other. Official statistics usually show correlations; establishing cause needs a design that rules out other explanations, such as a trial, a natural experiment or a careful comparison group.

Why it matters when checking a claim

"Since the policy began" claims. A figure that changed after a policy started may have changed for other reasons, such as the economy, population or a change in how data are recorded.
Confounding factors. Areas with more of one thing (say, police officers) may also have more crime because both reflect population density.
Reverse causation. The direction can run the opposite way to the claim.
Look for evaluations. Government evaluations following HM Treasury's Magenta Book guidance state what they can and cannot attribute to a policy.

See how to check a statistic.

Also searched as: correlation is not causation, correlation vs causation, causal link, confounding

Official definitions: HM Treasury: The Magenta Book (guidance on evaluation) · Office for Statistics Regulation: Code of Practice for Statistics

Selection bias

A distortion that arises when the people or cases included in data differ systematically from the population a claim is about.

Selection bias arises when the people, places or cases that end up in a dataset are not representative of the group a conclusion is drawn about. It can come from who chooses to respond to a survey, which cases get recorded, or which examples are picked to make a point.

Why it matters when checking a claim

Self-selected polls. Online or phone-in polls capture people motivated to take part. Official surveys use random samples and weighting to reduce this problem.
Non-response bias. When fewer people answer a survey, those who do may differ from those who do not. The ONS has published work on bias in the Labour Force Survey for this reason.
Recorded cases. Police recorded crime covers only crimes reported to and recorded by the police, so it reflects reporting behaviour as well as crime.
Picked examples. A handful of local cases does not describe the national picture; look for the full official series.

Also searched as: sampling bias, non-response bias, self-selection bias, selection bias meaning

Official definitions: ONS: Exploring the use of external data to assess for observed bias in Labour Force Survey estimates · HM Treasury: The Magenta Book (guidance on evaluation)

Survey weighting

Adjusting survey responses so the sample matches the known make-up of the population, such as by age, sex and region.

Weighting adjusts the influence of each survey response so that the sample as a whole matches the population it represents. If young men are under-represented among respondents, each young man's answers are given more weight. Official surveys usually weight to population estimates by age, sex and area.

Why it matters when checking a claim

Weights depend on population estimates. When population estimates are revised, surveys may be reweighted and published levels can change, sometimes by hundreds of thousands, as has happened with the Labour Force Survey.
Weighting cannot fix everything. It corrects for characteristics that are known for the population, not for differences in attitudes or circumstances between people who respond and those who do not.
Price indices use weights too. In CPI, each item's weight reflects its share of household spending, updated each year. See weighted average.

Also searched as: weighting, weighted survey data, survey weights, reweighting

Official definitions: ONS: Labour Force Survey user guidance · ONS: Higher-level aggregation and weights in consumer prices

Outlier

A value that is unusually far from the rest of the data. It may be a genuine extreme or an error, and it can distort averages and trends.

An outlier is a value that sits far away from most of the other values in a dataset. It may be real, such as one very large company or one exceptional month, or it may be a recording error.

Why it matters when checking a claim

Averages. A single outlier can move the mean substantially while leaving the median almost unchanged.
Start and end points. Comparing with an unusual year, such as 2020, can make a change look much larger or smaller than the longer trend. See base effect.
Local figures. In small areas, one event (such as a single large incident) can produce a rate that looks extreme.
Producers check for them. The ONS uses outlier detection when compiling statistics such as consumer prices, and bulletins often explain when one-off events have affected a figure.

Also searched as: outliers, outlier meaning, extreme value, anomaly in data

Official definitions: ONS: Outlier detection for grocery scanner data in consumer price statistics · Office for Statistics Regulation: Code of Practice for Statistics

Data suppression

Withholding or rounding small numbers in published tables so that individuals cannot be identified, or because estimates are too unreliable to publish.

Data suppression is the practice of hiding or rounding some values in published statistics. Producers do it for two main reasons: to protect confidentiality, so that no individual can be identified from a small count, and to avoid publishing survey estimates that rest on too few responses to be reliable. Suppressed cells are usually shown with a symbol such as c, x, * or [c].

Why it matters when checking a claim

Totals may not add up. Rounding and suppression mean that rows can differ slightly from the published total.
Missing is not zero. A suppressed value should not be treated as nil.
Small-area data. Figures for a single ward, school or LSOA are the most likely to be suppressed or rounded, for example to the nearest 5.

Each table's notes explain the symbols and the rule used.

Also searched as: statistical disclosure control, suppressed data, small numbers rounding, what does c mean in statistics tables

Official definitions: ONS: Comparison of post-tabular statistical disclosure control methods · Office for Statistics Regulation: Code of Practice for Statistics

Census

A count of every person and household in a country on a set day, usually held every ten years. The UK census is run separately in each nation.

A census attempts to count every person and household in a country at a single point in time, with questions on topics such as age, ethnicity, religion, health, housing and work. In the UK it is run separately: by the ONS in England and Wales and by NISRA in Northern Ireland (both held in March 2021), and by National Records of Scotland (held in March 2022).

Why it matters when checking a claim

Census day snapshot. Results describe the population on one day. The 2021 census took place during pandemic restrictions, which affected some topics, such as travel to work.
Not UK-wide on one date. Combining Scotland's 2022 results with the 2021 results for the rest of the UK mixes dates.
Population estimates update it. Between censuses, mid-year population estimates roll the figures forward.
Question wording changes between censuses, so check comparability notes.

See where to find UK census data.

Also searched as: census 2021, UK census, Scotland census 2022, census data

Official definitions: ONS: Census · Scotland's Census