Labour Force Survey quality update: January 2026

• Response levels on the Labour Force Survey (LFS) have shown clear improvement because of the interventions to improve quality, with Wave 1 response levels now very close to their pre-coronavirus (COVID-19) pandemic level, although response levels for Waves 2 to 5, though increased, are still slightly below their pre-coronavirus level.

• The composition of respondents according to different characteristics has also shown some improvement since late 2023, however, representativeness is lower than before the pandemic and has not improved for all characteristics; we continue to progress our work on administrative data, with a view to improving our understanding of possible bias in our surveys.

• Analysis of respondent attrition between waves indicates that respondents who leave the survey early tend to work more hours, while those who complete more waves generally work fewer hours at the start and experience a greater reduction in hours over time; weighting adjustments help to reduce, but do not fully eliminate, this effect.

• The expected impact of population revisions on LFS headline estimates since mid-2022 is driven by strong population growth in the year to mid-2023, but weaker growth in the year from mid-2024 to mid-2025 than currently used in LFS weighting.

• Coherence between the LFS and other labour market data sources, though improved, remains a challenge, and we continue to monitor this closely; detail on the coherence of the latest estimates can be found in the January 2026 Labour market overview bulletin.

• As part of our ongoing work to strengthen understanding of coherence across labour market sources, we are also reviewing the impact of the revisions applied to initial estimates from our administrative data sources, Real Time Information (RTI) and Claimant Count; more details can be found in this article.

• Caution is still advised when assessing change over time periods (particularly those affected by operational changes) and when analysing more detailed estimates; as the quality of the LFS has improved, it is likely these quality improvements have affected reported statistics.

• The Transformed Labour Force Survey (TLFS) remains the long-term solution for collecting labour market data, with a short longitudinal “Core” labour market-focused survey that went live in July 2025; this is complemented by a separate cross-sectional “Plus” survey to provide wider socioeconomic, household and local data.

For many years, household surveys both in the UK and in comparable countries have been facing the challenge of falling response rates. Coupled with challenges in collecting and processing survey data since the start of the coronavirus (COVID-19) pandemic, quality concerns became acute for Labour Force Survey (LFS) data collected in 2023. This led to the suspension of publications based on LFS data, and the withdrawal of accredited statistics status for publications based upon LFS and Annual Population Survey (APS) data from 2024 onwards.

In response to these concerns, several changes were made to the operation and processing of LFS data since late 2023. These have been detailed in previous articles.

Since October 2023, the Office for National Statistics (ONS) has made several changes to address quality concerns with the LFS. The five-wave structure of the LFS means that some of these changes can take at least 15 months to fully feed through into survey estimates and at least 18 months to feed through into measures of quarterly change.

Estimates from January to March 2025 include the full effect of the improvements in LFS data collection and sampling methods introduced from January 2024. However, since then, we have increased the number of interviewers for the LFS, which has continued to increase the number of responses to the survey. Consequently, estimates may be subject to the effect of these further improvements, which may have an ongoing impact on the survey. An increased volatility will remain in the LFS estimates for mid-2023 and throughout 2024, so we would advise additional caution when interpreting survey change measures. We recommend using LFS estimates as part of our suite of labour market indicators, alongside workforce jobs, Claimant Count and Pay As You Earn (PAYE) Real Time Information (RTI) estimates.

Throughout this article, we refer to a mixture LFS data for Great Britain and for the UK. The operational responsibility for data collection is split between the Office for National Statistics (ONS, for Great Britain) and the Northern Ireland Statistics and Research Agency (NISRA, for Northern Ireland). The ONS has responsibility for publishing statistics at a UK level. Data regarding the collection of the survey often refers to Great Britain, while references to published statistical measures often refer to the UK as a whole.

This article provides an update to our most recent Labour Force Survey quality update: September 2025 article published in September. We cover data up to the period of July to September 2025, to align with the latest data covered in our Labour Force Survey performance and quality monitoring report: July to September 2025 methodology, published in November 2025. It also provides an update on how the quality of the LFS has evolved recently, considering the impact on response levels, the characteristics of respondents, and the published statistics. It will focus on the LFS, but given the linked design of the two surveys, impacts on the LFS also affect the size and quality of the APS.

Coherence between the LFS and other data sources has improved, but remains a challenge, and we continue to monitor this closely. Our latest views on interpreting LFS data and the labour market narrative will remain in the monthly Labour Market Overview bulletin.

The changes referred to in Section 2: Background to Labour Force Survey quality continue to have a clear positive effect on the number of achieved responses to the LFS. The overall level of response has increased steadily since the low point in July to September 2023.

Considering the number of responses for the UK, including imputation (which is the same as the number of cases available to analyse in a dataset), there were 75,757 responses to the LFS in July to September 2025. This is an increase of 31,519 from the low point in July to September 2023. However, this does remain 8,305 responses below the figure seen in October to December 2019.

Breaking this down by wave of response, as reported in our LFS performance and quality monitoring report, we have seen Wave 1 response levels almost recover to pre-coronavirus (COVID-19) pandemic levels. There were 8,345 household interviews conducted in July to September 2025, compared with 8,963 in October to December 2019. Waves 2 to 5 have also increased but remain further from their pre-coronavirus position, with 25,678 responses (including imputed cases) in the latest period, compared with 26,823 at the end of 2019.

We have also seen improvements to response rates to the LFS, although they are more moderate by this measure. Further information on survey response levels and response rates can be found in our LFS performance and quality monitoring report.

This higher LFS response level helps to increase the precision in our estimates. This is shown by the confidence intervals for our headline labour supply estimates, which are presented in Figure 1. Across all three headline rates, these have narrowed since mid- to late-2023, which suggests improved precision compared with the periods where response rates were at their lowest. However, confidence intervals remain wider than before the pandemic, reflecting the trends in response.

Detailed estimates will continue to see greater volatility compared with more aggregated measures. However, this has always been a feature of LFS-based analysis to some extent, when all else is equal. The lower achieved response level for the latest periods, compared with response levels before the pandemic, will make this volatility more acute. It is also likely that variation in response rates has affected non-response bias in the survey results.

For example, we can consider estimates of redundancies or estimates of young people not in employment, education or training. Both concepts focus on much smaller population subgroups compared with the headline rates, and therefore continue to see relatively large quarterly changes, despite improved sample sizes.

Changes to the composition of respondents, according to a range of characteristics, is important when considering the quality of Labour Force Survey (LFS) statistics. This is highlighted in our published LFS performance and quality monitoring report in November 2025.

Ideally, those responding to the LFS would be broadly representative of the overall population for which we are estimating. However, we know that the likelihood of responding to surveys varies regarding several different characteristics. For this reason, we apply a range of statistical methods to our data, particularly with respect to weighting and non-response bias adjustments. These aim to ensure that our estimates are as representative as possible.

LFS weighting uses several characteristics, including:

• age
• sex
• location of respondent
• housing tenure (included since the coronavirus (COVID-19) pandemic)

Our non-response bias adjustments operate at the household level and focus on Indices of Multiple Deprivation and Output Area Classification.

We can analyse a range of characteristics to understand how the composition of respondents has shifted over time. We will focus on a selection of these in this article.

These analyses show that for a range of these characteristics, we have seen relatively large compositional shifts in recent years within our unweighted datasets. Initially, these moved the composition away from benchmarks such as Census 2021, followed by improvement for some characteristics as recovery actions were taken. The most significant of these recovery actions fed into point-in-time estimates for January to March 2025. However, we continue to see modest changes in the composition of respondents for some characteristics in the latest quarter (July to September 2025). Missing data, accounting for fewer than 1% of responses, have been excluded from the analysis presented in this section.

In each of the following subsections, we compare the unweighted distribution of respondents between different categories over time. We specifically compare July to September 2019, July to September 2023, January to March 2025 and July to September 2025.

For each characteristic, we also include the equivalent distribution from Census 2021 for England and Wales. This acts as a benchmark against which to consider the distributions from the LFS. Legitimate differences can arise for a variety of reasons, because of:

• differences in the questions asked
• differences in population coverage
• genuine change in the population as time progresses

Some characteristics are relatively static, while others can vary more over time. This makes a fixed benchmark less useful.

Age

There is good alignment between the unweighted age distributions in the July to September 2019 LFS and Census 2021 for England and Wales. By July to September 2023, the distribution of the LFS has changed, with those aged 50 years and over accounting for 60.2% of LFS respondents compared with 46.9% of census respondents. This fell to 53.2% in January to March 2025 and fell slightly further to 52.3% in the latest period July to September 2025.

Housing tenure

Housing tenure shows a similar pattern. There are differences in the distribution of the tenure type between Census 2021 and the LFS in July to September 2019, but these are relatively moderate compared with more recent periods.

A housing tenure adjustment was added to LFS weighting during the pandemic in response to large changes in estimates. Analysis of weighted data has shown the tenure adjustment is still required to maintain quality and robustness of the estimates.

Disability

The pattern for disability differs to age and housing tenure. In July to September 2019, 19.8% of LFS respondents reported a disability, compared with 15.3% of people in the census. By July to September 2023, the gap between the census and LFS had widened, with 23.0% of LFS respondents reporting a disability; this then peaked in January to March 2025 at 23.8%. In the latest quarter, there has been a small fall, with 23.6% of LFS respondents reporting a disability.

Differences in disability distribution remain after weighting. As LFS estimates are consistently above the census, users should be cautious about comparing estimates from the two sources.

Country of birth

In July to September 2019, 83.4% of LFS respondents were born in the UK. This has been gradually decreasing over time towards the Census 2021 estimate of 78.5%. In July to September 2025, 81.7% of LFS respondents were born in the UK.

Trends by country of birth are subject to more natural variation than other characteristics discussed in this section. Shifts seen during these time periods may reflect real changes among the respondents. However, these shifts could have also occurred because the composition of the responding population has changed, following survey recovery measures.

Summary of composition analysis

Unweighted distributions can be out of alignment with benchmarks such as the census, and this is one reason why we apply weights to our data. Weighted distributions for those characteristics controlled for within weighting more closely align with the census.

However, a more representative unweighted sample means that weighting makes it less likely that the data include bias that cannot be controlled for. We remain confident that weighting works as intended for each of these controlled-for characteristics.

Though we have seen some recent improvement for some characteristics, differences in unweighted distributions persist compared with the pre-coronavirus (COVID-19) pandemic period, for example, for housing tenure. This reinforces the ongoing need for the tenure adjustment introduced during the pandemic, and for the ongoing monitoring of these distributions.

For disability status, the unweighted distribution of response has remained inconsistent with benchmarks implied by Census 2021 for England and Wales. We have therefore considered whether there could be further adjustments to LFS weighting to target additional characteristics such as disability status. The main limitation here and elsewhere is a lack of appropriate external benchmark to which weights can be calibrated. This is particularly relevant for those characteristics that are known to vary more over time, where a point-in-time benchmark would not be appropriate.

Recent improvements in the representativeness of the LFS sample have meant that we are now focusing on developing the Transformed Labour Force Survey (TLFS). This is also because of the practical limitations of attempting further improvements to the LFS. For this reason, we are not recommending any further adjustments to the weighting approach for the LFS.

Relevant learning from this investigation will inform the ongoing assurance and development of the equivalent methods being applied to the TLFS. This approach has been assured and endorsed by stakeholders, users and academic experts.

Hours worked is a variable on the Labour Force Survey (LFS), which has been recognised by Office for National Statistics (ONS) and Organisation for Economic Co-operation and Development (OECD) research over a number of years as being affected by biases that are mitigated by the ONS through the use of wider data sources, for example, to produce productivity statistics. The following analysis provides an improved understanding of some of the drivers of this.

As LFS response rates have fallen, the unweighted total weekly hours worked have also declined. However, the weighted series shows a relatively stable trend, aside from the coronavirus (COVID-19) pandemic period, with a small increase in total hours worked between 2017 and 2025. The stability of the weighted hours worked series indicates that corrective data collection actions and application of the weighting framework are, in the measurement of hours worked, mitigating against declining response rates.

Impact of wave attrition of employed respondents on hours worked

Analysis of attrition was conducted to understand the impact of declining response rates between waves on hours worked. By examining patterns of survey completion and the relationship to reported hours worked, the analysis aimed to assess whether attrition introduces systematic bias.

Between January to March 2017 and July to September 2025, 37% of employed people completed only one wave of the survey. Of people who were employed both at their first and final response to the survey, only 15% completed all five waves. Imputed responses were not included in the count of completed waves.

Employed people who leave the survey before completing all five waves tended to work more hours at the point of their first response than those who go on to complete more waves of the survey. In general, the number of survey waves completed is negatively correlated with average hours worked.

All respondent groups saw a reduction in hours worked between the first and final wave. In addition to working fewer hours at their first wave, people who complete more waves have a larger reduction in weekly average hours between their first and final wave, in comparison with people who complete fewer waves.

The prevalence rates of full-time workers, males, young people, and people without a disability are higher in the groups where respondents leave the survey before completing all five waves. For those who complete more waves of the survey, there are higher rates of part-time work and people with a disability.

Of those who complete only one wave of the survey, 13.6% were disabled. In comparison, of those who complete all five waves, 17.1% were disabled. Similarly, of those who only complete one wave of the survey, 23.5% worked part-time while 25.5% of those who complete all five waves worked part-time.

For those who complete more than one wave, the prevalence of disability and part-time work increases between their first and final wave. The biggest increases in disability and part-time work were for those who complete all five waves.

Looking at the impact of weighting on hours worked estimates, analysis shows weighting increases the average hours worked at both first and final waves. The fall in hours worked between first and final waves, however, is less than in the unweighted estimates.

In summary, the analysis indicates that attrition is not random, with respondents who leave the survey early tending to work more hours initially. There is also a higher prevalence of full-time workers, males, younger people and people without a disability among those who do not complete all waves. Those who complete more waves generally work fewer hours at the start and experience a greater reduction in hours over time, alongside an increasing prevalence of part-time work and disability. Weighting adjustments help to reduce, but do not fully eliminate, this effect.

During 2026 and into early 2027, we intend to reweight the Labour Force Survey (LFS) along with the other associated datasets based on LFS interviews, such as the Annual Population Survey (APS), the LFS and APS longitudinal datasets, and LFS and APS household datasets. This reweighting will bring estimates in line with updated information on the size and composition of the UK population.

Currently, our LFS-based datasets and accompanying estimates are on a range of different population bases, varying by dataset and period. LFS three-month person-weighted datasets, used for the headline labour market measures, are the most up to date, having been reweighted in 2024, for periods from 2019 onwards, to the population information available then. However, these datasets were not reweighted for periods from 2011 to 2019, with headline series scaled to population changes rather than constructed from reweighted datasets. Outside of the scaled headline series, other aggregates continue to have a discontinuity prior to that reweighting.

Other datasets, such as APS and household datasets, were not updated as part of that reweighting.

Our intention is to bring all the family of LFS datasets up to date, with weights based on a single set of population time series. This will put all weighted datasets on a consistent basis and remove any weighting-based discontinuities from the time series.

While this process will have a varying impact on different types of dataset, we will focus this indicative impact analysis on the LFS headline series published in dataset tables A02 SA (employment, unemployment and economic inactivity for people aged 16 years and over and aged from 16 to 64 years, seasonally adjusted) and A05 SA (employment, unemployment and economic inactivity by age group, seasonally adjusted).

This indicative impact analysis is informed by information from mid-year estimates and national population projections published since the reweighting in 2024. It uses aggregate information by sex and age band to scale the component series, based on the pre-existing employment, unemployment and economic inactivity proportions from the LFS for those different subpopulations. The final subnational population projections used in the reweighting may reflect later information than that used to inform the national population projections that are the basis of this analysis.

As this is the method that is used in constructing the modelled series between 2011 and 2019, and there have been only minor changes to population estimates for those periods since that work was done, an analysis of those time periods would produce very similar answers to those currently published. This analysis therefore focuses on the more recent periods. In particular, it looks at the change in population estimate paths over the last few years compared with the populations and projections that were used in the 2024 reweighting.

Recent population estimates have been particularly affected by revisions to patterns of international net migration. Although the projections that the current LFS weighting is based on allowed for high net migration in 2023 and 2024, they were lower than our current estimates for those periods. However, that higher level of migration was also projected into 2025, whereas latest estimates are not quite as strong. Consequently, population estimates from mid-2022, through 2023 and 2024, are higher than the previous populations used to weight the survey, but by mid-2025, the population forecasts come back more in line.

This pattern is largely reflected in our indicative estimates for the employment level over that time. The higher population numbers, with stronger growth from mid-2022 to mid-2024 are expected to result in initially higher employment levels and growth, with levels coming closer in line by the end of the period. Because the upward revisions in population numbers are stronger in age groups from 25 to 49 years, that have higher employment rates, the revisions are expected to be proportionately higher, but not enough to have a significant impact on employment rates.

With the revisions being strongest in age bands with higher employment rates and lower economic inactivity rates, the impact on economic inactivity levels is expected to be less significant than the employment revisions. Estimates of unemployment are not expected to be significantly changed by reweighting.

At their largest impact, around mid-2024, the new population estimates are expected to revise employment, unemployment and economic inactivity levels, for people aged 16 to 64 years, by around 120,000, 5,000 and 20,000, respectively. However, from mid-2024 to mid-2025, we see lower growth in the populations that we will use for reweighting, than we do in those currently used to weight the LFS. This results in population estimates for mid-2025 that are near our existing ones, with the size of the corresponding revisions therefore expected to reduce greatly.

Administrative data such as Pay As You Earn (PAYE), Real Time Information (RTI) and Claimant Count data play an important role in the labour market story. The latest data point is an early estimate that is more uncertain and subject to more revisions than the rest of the time series. Here we explain some work we have carried out on the impact of the PAYE RTI flash estimates and the nature of the recent Claimant Count revisions we have seen. This is part of our ongoing work to strengthen the understanding of coherence across labour market sources.

Revisions to PAYE RTI estimates

HM Revenue and Customs (HMRC) and the Office for National Statistics (ONS) publish estimates of payrolled employees monthly. Eligible employers must report payroll information, including all employees’ pay, tax and deductions to HMRC each time an employee is paid. This provides an estimate of the number of individuals on payrolled employment in any month, which is one of the suite of indicators published on the UK labour market.

An initial flash estimate is published, which is available with a lag of around 15 days. Because there is a trade-off between timeliness and accuracy, this flash estimate is subject to revision given that it is based on a larger proportion of imputed returns than estimates of previous months. For instance, there are likely to be people who have been in employment but for whom HMRC has not yet received corresponding RTI submissions. This could be because of missing submissions or submissions returned with missing data. It could also be because the payment frequency cycle of that employment does not align with the extraction dates. In generating the estimates, these missing data are imputed. A revised estimate is published the following month, by which time more data have been received.

Figure 14 shows monthly snapshots from 2025 of how many people were estimated to be in payrolled employment at each point in time. More recent estimates are of a higher accuracy, reflecting how these incorporate new returns to replace initial imputed values. It shows how there can be some uncertainty in flash estimates, which is why we are explicit in our communications that early estimates are subject to revision. If we compare the flash and revised estimate, there has been a tendency for upward revisions to the level of those on payrolled employment of around 14,500 over the period July 2022 to October 2025. This represents approximately 0.05% of those who are in payrolled employment.

There is a correlation of 0.976 between the flash and revised estimate of the 12-month change in payrolled employment for the period July 2023 to October 2025. The mean revision is 0.08 percentage points, which indicates there is a tendency for a slight upward revision to the flash estimate. Revisions are time-varying as there is dispersion around this mean revision.

We have estimated the correlation between the flash estimate of the 12-month change and the revisions to flash estimates, which provides a check on whether there is a cyclicality to the revisions. From this, there is no strong evidence of cyclicality in the revision performance over this period, but it is one that we will continue to monitor.

We will continue to be explicit in our communications that there is some uncertainty in the flash estimates. We will also explore how we might be able to improve how we communicate uncertainty, including providing real-time updates on the revision performance of the flash estimate of the change in payrolled employment. We will consider whether it is best to focus on the revised estimate in our communications, which is a more reliable indicator of later estimates.

Another consideration is whether it is more appropriate to lead on the 12-month change in payrolled employment, where there is lower volatility than in the monthly change. We will continue to keep stakeholders informed of any work on this topic of data uncertainty.

Revisions to ONS Claimant Count estimates

Historically, the Claimant Count has been based on a single standalone benefit within the benefit system, most recently Jobseeker’s Allowance (JSA). Since the introduction of Universal Credit, the Claimant Count has included elements of both JSA and Universal Credit, with the current Claimant Count being predominantly made up of claimants of Universal Credit.

Universal Credit is a combined benefit, covering several different types of claimants in different work and personal circumstances with the nature of their claim determined by factors such as work status, earnings and health. The nature of a claim can change over time. The full nature of a claim, based on all these factors, for a particular point in time, may not be available when we take an initial extract from the system used to produce Claimant Count statistics.

For example, work capability assessments may still be ongoing meaning that a judgement on a claimant’s ability to have worked may not have been reached. Consequently, this leaves the Claimant Count, for a given period under Universal Credit, with a potential for notable revisions between the initial extract being taken and a revised extract from the Universal Credit system a month later. The initial extract from the benefit system does, however, reflect the state of that system at that point in time.

Recent revisions to the Claimant Count between the initial and revised extracts have tended to be downwards and of a similar magnitude. The ONS is in discussion with the Department for Work and Pensions (DWP) regarding the nature of these revisions, whether there is anything that can be done to bring greater clarity to the initial extract to address the issue and whether an adjustment can be applied to improve the accuracy of the initial estimate. Figure 15 shows the revision between the initial Claimant Count figure published each month and the revised estimate published one month later.

The analysis in this article has focused on impacts at the UK level only. While impacts will be similar across the different areas of the UK, they will not necessarily be uniform. We plan that future iterations of this article will cover quality and impacts for levels below the UK.

Our published plans for economic statistics and survey improvement and enhancement reinforce the importance of the Transformed Labour Force Survey (TLFS) for delivering high-quality labour market statistics. We aim to transition to our published headline labour market statistics in November 2026, although this may extend into 2027 if our (or users’) assessment of quality requires more data to be collected and assessed. Updates will be published quarterly in our Labour market transformation article.

We continue to progress our work on administrative data, including linking HM Revenue and Customs (HMRC)’s Pay As You Earn (PAYE) Real Time Information (RTI) to the TLFS and the Labour Force Survey (LFS) with a view to improving our understanding of possible bias in our surveys. We are aiming to publish initial findings from this work early this year. Alongside this, we have continued to progress our work on a linked employer-employee dataset (LEED). 

We are also launching a new project, in collaboration with the Economic Statistics Centre of Excellence to explore how we can use a range of methods to combine at an aggregate level different labour market indicators improving our understanding of the labour market. 

Following the letter from the ONS Permanent Secretary to Penny Young, Deputy Chair of the UK Statistics Authority, the ONS has received feedback on the importance of Annual Population Survey (APS) data to users and are working through options for future delivery. We can reassure users that no decision will be made to pause the APS without further engagement and then a formal public consultation. We will keep users informed as decisions are made in the coming months.