National Statistical

There has been debate recently about which source of information on jobs and working patterns should be used to best estimate changes in productivity. In this blog Richard Heys looks at the different sources and explains the approach the ONS has settled on.

As ONS Deputy Chief Economist, with responsibility for its productivity statistics, I am always on the lookout for fresh data sources that we could use to understand new perspectives on our productivity statistics – either to complement or challenge our existing published data.

In a world of household surveys, business surveys and administrative data covering the labour market, the opportunities to produce a myriad of alternative combinations are vast.

However, the reality of being an analyst means asking oneself a few pertinent questions. Chief among them is this: if I want to compare different versions of the same analysis, how can I tell if they are truly comparable?

Here specifically for productivity: are we working towards the same concept? A measure of output per hour will inevitably look different to a measure of output per worker – unless the share of people working part-time, and the precise distribution of hours worked remains constant. But once we get beyond this and agree all our measures target, say, output per worker, we then need to ensure we have the same concept of “workers”.

And this is where the recent flurry of media activity comparing measures derived from the Labour Force Survey (LFS) and HM Revenue and Customs’ Pay As You Earn (PAYE) records hits the metaphorical methodological rocks.

Put simply, the LFS, being a household survey, is designed to capture a representative sample of those living in the UK households aged 16 or over. That includes the employed, the self-employed, the unemployed and the economically inactive (effectively those neither working nor looking for a job). PAYE records, on the other hand, capture only the Pay-As-You-Earn records of all those employed in the UK. The differences are immediately obvious: for example, the LFS captures the self-employed, while PAYE does not. It will soon begin to capture high-earning self-employed individuals through ‘Making Tax Digital’, but it will take time for this mechanism to be rolled out to all self-employed individuals.

So, one can use PAYE as a data feed for the number of employees. Given the recent, well-publicised, challenges around the LFS and its low response rate, there is merit in testing exactly that. This is why, for the past few quarters, the ONS has published exactly this comparison in its productivity bulletins. But how have we solved the problem of how to measure the self-employed?

Well, effectively there are two ways of capturing this concept. The first is self-assessment tax returns and the second is the LFS. So, what are the pros and cons of the two? Tax returns are collected with a lag, published with an even longer lag and represent an annual aggregate, thus making the production of a quarterly series difficult. Nevertheless, particularly in 2023 when the LFS sample was at its lowest, it’s an important comparator. The LFS, by contrast, gives us quarterly data on a timely basis, giving us the only way to analyse the most recent years.

If one wants to understand recent history there are few, if any, functioning alternatives to the LFS, unless one excludes the self-employed completely, which would provide only a partial picture. So, assuming one wants to include this group, what does the comparison – which can be seen in Figure 4 of today’s productivity article – show?

Effectively, a small number of stories appear. Firstly, up to Quarter 3 2020 the two indexed series move closely together. From the start of our time series until mid-2022, the PAYE‑derived index generally showed slightly stronger growth relative to its 2023 level, reflecting that PAYE data report fewer workers in each period. During the period when the LFS sample was weakest, the two series converged, but since the LFS recovery began, the PAYE‑derived index has again shown somewhat stronger growth relative to its 2023 baseline.

So, what do they tell us? Are there materially different productivity growth trends on show? I would argue that, no, there are not. Throughout the time series the two series have displayed very similar medium-term trends in growth. Given one is an administrative source and the other a survey which at times struggled with its sample size, the commonalities in the stories these two datasets are showing us offers encouraging signs in terms of the reliability of our understanding of recent UK productivity performance. Does the recent spike in growth in the PAYE-derived estimate in the latest quarter suggest a significantly rosier trend in the near future using this source?

Again, possibly not: it may just reflect the longer-term trend re‑establishing itself, with the PAYE‑derived index showing slightly stronger growth in output per worker since 2023 than the LFS‑derived series. It is important to note that because both series are indexed (2023=100), this difference reflects relative changes rather than absolute productivity levels. We will continue to monitor this relationship in coming quarters and will flag to users any substantive shifts in the growth patterns these data reveal.

Keeping these data under observation is key to determining how best to report the productivity statistics. In our article today we have, as previously, transparently released both LFS and PAYE-derived estimates to give users as much detail as we can. Making Tax Digital may change how we model the PAYE-derived data in the future as more self-employed workers come under PAYE, so we will keep these measures under review to ensure we continue to provide the best data we can on this key macroeconomic topic.

This post was edited on 13 November 2025 to correct a couple of misprints and clarify our description of the relationship over time between the LFS and PAYE series.