Each month the ONS publishes labour-market statistics that become headlines: unemployment up, employment down, vacancies falling. Behind each headline is a particular source with its own strengths, and in the UK the sources have recently differed enough that the choice matters.
The four main sources
The Labour Force Survey (LFS). A household survey that provides the official unemployment rate, the employment rate and economic inactivity. It follows the internationally agreed definition: people are unemployed if they are without a job, have looked for work in the past four weeks and could start within two weeks. It captures self-employment and does not depend on tax or benefit records. Its weakness is that, as a survey, it depends on people responding, and response rates have fallen over recent years. The ONS has said that a higher response rate helps the precision of its estimates, but that caution is still needed when assessing change over time and when analysing more detailed estimates.
PAYE Real Time Information (RTI). HMRC data from employers’ payroll submissions, covering payrolled employees. Because it comes from administrative tax records, it is not affected by survey response. The ONS has said it regards RTI as the most reliable measure of the number of employees. It does not cover the self-employed, and early estimates are revised.
The claimant count. The number of people claiming unemployment-related benefits, which in the Universal Credit era includes some people in work on low earnings. It tells you about benefit use, not the unemployment rate, and policy changes can shift it without any change in the labour market.
The vacancy survey. An employer survey of job openings. It is useful as a read on demand for workers, though it measures vacancies at a point in time and can be volatile.
An example of the difference
For January to March 2026, the ONS reported that the number of payrolled employees fell by 94,000, or 0.3%, over the year. Looking at one source and ignoring the other gives only part of the picture. A survey-based unemployment rate and a payroll count can move in different directions for a period, particularly when survey quality is in question.
A survey being rebuilt
The ONS is developing a Transformed Labour Force Survey, an online-first design intended to replace the current LFS with a larger sample. In April 2026 the ONS said the earliest the headline labour-market statistics could move to it is 2027. Until then, users have to treat the existing survey figures with caution, and give weight to the payroll and vacancy data alongside them.
What the headline rate leaves out
Even a well-measured unemployment rate does not describe the whole labour market:
- Economic inactivity. People who are not working and not looking, for example because of long-term sickness, study or caring, are not counted as unemployed. A low unemployment rate can coexist with a large inactive population.
- Hours and pay. Employment can hold up while hours fall or real wages stagnate.
- Underemployment. Some people in work want more hours.
- Regional and demographic differences. National averages hide wide variation between regions and age groups.
How to read a release
- Use the ONS’s three-month averages rather than a single month, since single-month survey figures are noisy.
- Check the payroll count and the survey together, and note when they disagree.
- Look for revisions, which can be large for the latest months.
- Read the ONS’s quality notes, which now accompany each release.
These data matter to policy. The Bank of England watches the labour market for signs of wage pressure when setting rates, and the way it interprets weak data connects to what the inflation figures say.
This is general information, not financial advice. Figures reflect ONS statements at the time of writing and are revised regularly.
Sources: Office for National Statistics, UK labour market statistical bulletins (2026) and Labour Force Survey quality information; HMRC PAYE Real Time Information; ONS Transformed Labour Force Survey updates (April and July 2026).
