The UK’s flagship labour force survey may be getting more reliable

The U.K’s. Office for National Statistics’ monthly labour report came out earlier this week, and midway down the page was a positive development: the ONS’s Labour Force Survey (LFS) data was seeing much higher-quality responses.

Since 2022, the ONS has faced a lot of questions about the reliability and accuracy of its labour market statistics after responses to its Labour Force Survey (LFS) fell from 47.6% in June-August 2013 to as low as 14.6% a decade later. Response rates fell especially quickly during the COVID pandemic, which meant that the ONS was seeing large gaps in its labour force coverage. I was even asked about these concerns when testifying in front of a House of Lords committee back in October of 2022.

Policymakers rely on this data to make highly consequential policy decisions, including whether to raise interest rates. If the data is off, then policy responses have no hope of fixing the problem.

Response rates and quality control are perennial problems with survey data: I saw that firsthand when my house was visited a few years back by a contracted surveyor for a company the ONS contracts with to carry out its face-to-face surveys.

I, of course, had to participate - and convinced two of my three housemates to do so as well. The survey asked participants about their perceptions of and experience with crime in their local areas: after I told my surveyor that I usually felt safe in my neighborhood of Finsbury Park, she made an expression of surprise and said, “Really?” in a disbelieving tone (I’m sure I don’t have to tell you why this isn’t the ideal response to a survey subject!)

Statisticians are well aware of these sorts of idiosyncrasies and use techniques to adjust for them: sampling a sufficient number of people, reweighting data based on demographics, using many different interviewers, and triangulating their data with other data sources that may prove more reliable. In the ONS’s case, they also surface more reliable tax administrative data from PAYE employees, although it doesn’t cover the entirety of the population. If this data broadly reflects their LFS results, they can be more confident in both data sources. That’s why private sector data sources like Deel are valuable, even if they’re not randomly selected and don’t cover 100% of the population.

But if responses fall too quickly, it’s much more difficult for the ONS to properly reweight its data. That was the fear in 2022-2023. And these types of falls are often not random: the composition of respondents likely changed between 2019 and 2023. For instance, imagine the following scenario: the unemployed are more likely to respond to face-to-face visits than the employed are, but both are equally likely to respond to phone calls. Prior to the pandemic, the survey’s weights would have downweighted the responses of the unemployed to correct for this tendency. In a post-pandemic world, where the employed were suddenly more likely to respond to surveys, but the survey weights were the same as before would mean that the population of unemployed would look smaller than it actually was.

In response to the post-pandemic drop, the ONS made some immediate changes: they reintroduced face-to-face interviewing in October of 2023, they nearly doubled the number of households they were contacting, they changed their re-weighting methodology in December of 2024 to account for compositional changes in interviewees, and they recruited more field staff to increase their interview capacity.

The results are visible: the number of responses (77,927) has risen significantly, nearly back to the numbers seen in October-December 2019 (84,062), immediately before the pandemic.

However, this increase has mostly been driven by the increased number of households contacted, rather than improvements in response rates: pre-pandemic Wave 1 response rates hovered around 50-55%, whereas 2025 saw rates closer to 30-35%. This sustained drop could still be cause for concern, as the composition of respondents has changed: respondents appear to skew older and wealthier than pre-pandemic.

The changes in statistical weights should go some way towards rectifying this, but it’s far from perfect. We don’t know the exact population demographics that make up a given region or industry, and that means we can’t know if the weights are correct: the random sampling of the LFS should fix this, but if the response is non-random and/or if there aren’t enough responses for a given demographic, random sampling might not be enough. Fortunately, the ONS is working on more fundamental reforms to the LFS, which you can read about here. The ONS plans to implement these beginning in 2027.