What do Labour Force Survey proxy responses tell us about rising long-term sickness in the UK?
By Josh Martin and Ben Baumberg Geiger
The apparent rise in long-term sickness has become one of the most closely watched developments in the UK since the COVID-19 pandemic. Policymakers, researchers and commentators are keen to understand why more people report being out of the labour market because of ill health, and whether this reflects a genuine deterioration in health or problems in the data used to measure it. Our new ESCoE discussion paper focuses on this question, looking at the UK Labour Force Survey (LFS).
The LFS is the key data source used to track economic inactivity and long-term sickness and generate UK labour market statistics. Yet it has faced considerable challenges in recent years, including sharply falling response rates. These developments have raised questions about whether the observed increase in long-term sickness might be partly a statistical artefact.
Our paper focuses on two aspects of survey methodology that have received little attention: proxy responses and survey mode. While concerns about response rates have been widely discussed, less attention has been paid to who actually answers LFS questions and how they are answered. More broadly, the paper highlights the importance of understanding how economic statistics are produced from surveys, rather than treating them as straightforward observations of reality.
The LFS samples households rather than individuals. As a result, one household member is often allowed to answer questions on behalf of another person in the household, known as ‘proxy responses’. While proxy responses have always been part of the survey design, their prevalence is higher than we might expect and has increased over time.
We find that proxy responses now account for around 40% of working-age observations in the LFS, compared with roughly 35% before the pandemic (see chart below). Among young people aged 16–24, the figures are much higher still – more than 75% of responses are proxies. In practical terms, this means that most of what we know about young people from the LFS is actually what their parents or guardians think they know.
Chart 1 – Proportion of proxy responses (%) in the UK LFS, 2002 Q1 to 2025 Q4, by age group
As well as an increase in proxy responses, the LFS has also seen changes in ‘survey mode’ – that is, how the interview is conducted. Most LFS interviews are face-to-face in the first survey wave, with households followed for four further waves, mostly over the phone. This changed during the COVID-19 pandemic, where face-to-face interviews were suspended until October 2023. Face-to-face interviews tend to have higher response rates and may lead to more reliable responses (as we review in more detail in the paper).
To investigate whether changing response patterns can explain the observed increase in long-term sickness over recent years, we compare trends using several alternative versions of the LFS data. We examine:
The results are reasonably consistent. Regardless of which cut of the data we use, indicators of long-term sickness increase after the pandemic (see chart below). Looking only at personal responses produces a trend that closely resembles the headline numbers. Restricting attention to wave 1 interviews reduces the size of the increase somewhat but does not remove it. Proxy responses often show a slightly smaller increase, but the direction of travel remains the same.
In short, the observed rise in long-term sickness is not simply a product of changing survey methods. While methodological choices affect estimates of the precise magnitude of the increase, they do not overturn the central conclusion that self-reported ill health has risen substantially.
Chart 2 – Proportion of people aged 16-64 economically inactive with main reason of long-term sickness (%), four-quarter rolling averages, 2003 Q1 to 2025 Q4
We also investigated whether proxy respondents report about health conditions differently from individuals reporting about themselves. Using the longitudinal structure of the LFS, we found that proxy respondents tend to under-report some long-term health conditions. Interestingly, the effect is strongest for broader measures of long-term and work-limiting health conditions, but there is much weaker evidence for under-reporting by proxies of economic inactivity due to long-term sickness. This might suggest that proxy respondents are reasonably good at identifying severe health conditions associated with labour market withdrawal, but less successful at identifying less visible health limitations.
Understanding important economic and social developments, such as the rise in long-term sickness, requires high quality statistics, often based on surveys. Survey results depend on survey design, respondent behaviour, weighting procedures and methodological choices. Understanding these processes is critical for enabling correct interpretation of statistics and assessment of their reliability.
The increasing importance of proxy responses is a useful illustration. For some groups, particularly young people, proxy interviews dominate the data. Researchers using LFS microdata should therefore consider examining personal and proxy responses separately, especially when studying subjective concepts such as health and well-being. The paper argues that trends among young people may be particularly sensitive to these issues. This is because proxy responses are so prevalent and because many proxy-reported young people are temporarily living away from home, for example while attending university.
The findings are also relevant for the ongoing development of the Transformed Labour Force Survey (TLFS). As the ONS moves towards an online-first approach to labour market data collection, it will be important to consider the role and quality of proxy responses as well as mode effects. Changes in the prevalence of proxy reporting could affect data quality and complicate comparisons between future statistics and historical series.
Our overall conclusion is that the methodological issues considered – proxy responses and survey mode – do not appear to explain the increase in reported long-term sickness and health-related inactivity since the pandemic. Proxy responses and survey mode influence measured levels and may affect estimates at the margin, but they do not explain away the post-pandemic rise in reported ill health. This finding does not validate the LFS in its entirety or fundamentally explain the increase in reported long-term sickness.
At the same time, the paper emphasises the value of scrutinising how statistics are produced. The quality of economic analysis depends not only on sophisticated modelling or theory, but also on careful attention to measurement. Understanding the strengths and limitations of our statistical systems is therefore essential if economic statistics are to continue informing public debate and policymaking effectively.
ESCoE blogs are intended to promote discussion and debate around economic measurement and statistics. The views expressed are those of the authors and should not be taken to represent the views of ESCoE, the Office for National Statistics, the Bank of England, the Monetary Policy Committee, or affiliated institutions.