The boom in the denominator: what the UK’s productivity revival is actually made of

By Jun Du, Managing Director of The Productivity Institute

The UK’s productivity revival is, on the published evidence, mostly a fall in labour input. Over the two years to 2026 Q2, gross domestic product grew by 1.3 per cent a year against 2.0 per cent a year between 2016 and 2019, while hours worked grew by 0.2 per cent a year. The Resolution Foundation, whose Q3 outlook prompted this week’s coverage, reports that 77 per cent of the improvement in measured productivity growth comes from the slowdown in hours rather than any acceleration in output. Barbas, Valero and Van Reenen put it more guardedly: “Part of this improvement reflects weaker labour input growth, particularly in employee numbers.” Real gross value added rose 2.1 per cent across their six quarters while their preferred labour input measure fell 0.3 per cent.

This blog is not to dispute the finding that has driven the debate. Over the six quarters to 2026 Q1, the Labour Force Survey (LFS) recorded a rise of 377,000 employees while Pay As You Earn (PAYE) Real Time Information (RTI) recorded a fall of 133,000 (CEP, Table 2). The Office for National Statistics (ONS) recommends the RTI-based series, and on the present state of the LFS that is the right call. Nor did official statistics miss a productivity recovery. The ONS introduced its administrative measure in August 2024, recommended in June this year that users focus on it, and led on it this month.

The ONS’s account is more specific than the coverage suggests. Not that the LFS is wrong, but that enhancements introduced since January 2024 are themselves moving the series: “recent LFS measures of employment growth are likely to be temporarily higher while recent LFS measures of productivity growth are likely to be temporarily lower” (ONS, “Addressing uncertainty”). James Benford, the ONS’s Director General for Surveys and Economic Statistics and Deputy National Statistician, drew out the symmetry on LinkedIn. The same logic means productivity growth was flattered in the run-up to 2023. Before the enhancements the survey captured fewer employed people, so it overstated productivity growth then just as it understates it now — a reason for caution about any comparison straddling the break, including the two-period comparisons this debate rests on.

An improvement in measured productivity that arises because labour input stopped growing, in a period when output grew more slowly than in the late 2010s, is not an improvement in capability. The distinction matters, because many policy choices are made on the basis of it. Three qualifications bear on it, ahead of the ONS methods article on the component-based measure on 17 September.

The hours dependency is not temporary

The employment gap may well fade as the LFS is repaired, and the ONS expects it to. The dependency described here will not.

Every output-per-hour figure in this debate is a product of two terms, and only the worker count has been moved onto a better source. The ONS is explicit: “RTI does not collect actual hours worked. This means that the whole-economy hours worked for both RTI and LFS are calculated by multiplying LFS average hours worked with the number of workers from the RTI and LFS.” Self-employment is likewise taken from the LFS. The CEP paper describes the identical construction, and the Resolution Foundation concedes that its measure “only corrects for who is working, not how long they work for”.

Average hours are collected directly from LFS respondents, so this is not a circularity. But the hours term still rests on the sample whose employment estimates are in dispute, and correcting the headcount leaves the other half of the denominator untouched. The collection issue the ONS disclosed for 3 May to 10 June moved the administrative headline. Without it, growth in output per hour would have been 0.6 per cent rather than 0.7 per cent in the year to 2026 Q2.

What makes this structural is a sentence in the CEP paper’s data appendix: “HMRC recently announced they will not move forward with their plan requiring employers to report detailed employee working hours via RTI.” The route by which administrative data might have closed the hours gap is shut, which makes the ONS’s component-based work — labour input built from usual hours, overtime and leave — matter more, not less, than the current round of headline figures.

How much more is now on the record. On the Bank of England’s staff blog, Sophie Piton and Fabrizio Cadamagnani note that the ONS is moving to the component method precisely “to minimise the bias from the secular decline in Labour Force Survey response rates”. Initial estimates suggest the change will raise measured UK productivity growth by 0.4 percentage points a year over 2008 to 2019. A revision of that size does not adjust the recent numbers; it changes the post-financial-crisis slowdown, a significant and consequential fact in UK economic policy of the past fifteen years.

The source difference and the concept difference

Much of the apparent disagreement is not disagreement. The CEP and Resolution Foundation estimates are not independent. Both substitute administrative employment for the survey count in the same numerator, over slightly different windows. The informative comparison is within the ONS’s own four estimates, which vary along two dimensions at once.

Read across the rows: the source difference is about one percentage point on either concept, and on the ONS’s own reading part of that is temporary. Down the columns, the concept difference is 0.6 to 0.7 percentage points on either source. That gap is arithmetically the growth of average hours per worker, which was therefore rising at around that rate — the same output gain spread over more hours per person looks smaller. Much of what separates the optimistic reading from the pessimistic one is therefore not a data dispute but a choice of denominator. The headline contrast of “1.6 per cent against 0.2 per cent” sets a per-worker figure built on administrative data against an LFS-based one, and so runs the source and concept differences together.

What the decompositions cannot resolve

If labour input has stopped growing while output holds up, one candidate explanation is that less productive firms are closing and their workers being reabsorbed by better ones. This is the Resolution Foundation’s own argument from February, when Greg Thwaites noted that job losses from exiting firms in 2024 were the highest since 2011 and that “creative destruction has two parts, and so far we’ve mainly got the latter”.

The August analysis rebuts compositional explanations at sector and occupation level. Hospitality’s share of employee jobs is unchanged at 6.6 per cent, and jobs in below-median-paid occupations grew 3.4 per cent against 1.4 per cent in higher-paid ones. Both hold on their own terms. Neither reaches the between-firm mechanism, which operates within sectors and occupations: a weak hospitality firm closes, its staff are hired by a stronger one, the sector’s employment share is unchanged, and measured output per job rises. In a decomposition run on aggregates that appears as a within-industry contribution, indistinguishable from genuine improvement. This is an identification problem rather than an error, and a better labour input measure will not fix it. The occupational evidence also stops short: drawn from the Annual Survey of Hours and Earnings for the year to April 2025, its window closes on 6 April 2025, the day the employer National Insurance and National Living Wage increases took effect.

Evidence that can speak to the mechanism is already published and largely absent from the discussion. The ONS’s business dynamism statistics show the firm exit rate at 14 per cent in 2024 against entry of 12.6 per cent — exit ahead of entry for a second year, reversing 2012 to 2019, when entrants created more jobs than exiters destroyed. Job reallocation fell to 19.8 per cent, lower than in 2001 in every industry. Destruction is running ahead of creation in an economy churning less overall — a configuration in which measured productivity can rise without any firm becoming more productive. Payroll data point the same way: employment in accommodation and food services fell 2.8 per cent in the year to July 2026, against a whole-economy fall of 0.3 per cent.

The strongest evidence on the other side is the CEP paper’s real-wage check, in which several deflated earnings measures, including median hourly earnings excluding overtime, rise alongside the administrative productivity series. That support is genuine but not decisive, since a shake-out concentrated in low-paid employment raises median as well as mean earnings.

Why this is not only a measurement question

The two explanations imply different economies. If the improvement reflects firms getting better, it is a capability story that should show up in the external position as widening comparative advantage. If it reflects firms exiting, the same aggregate improvement is consistent with a narrowing production base and a more exposed one.

The external position is where that difference shows. The UK has offset a goods deficit with a services surplus since the 1980s, so the structure is not new; the divergence in growth is. Services export volumes stood 13 per cent above 2019 by 2023 while goods exports remained below their 2018 peak, and the services surplus has risen from 4.5 to 6.4 per cent of GDP since 2019. Whether the firms driving the measured productivity improvement are the same firms driving that export growth is unknown. Answering it needs firm-level data — the same evidence the identification problem above requires.

What would be the steps forward

Three things. The component-based hours work due on 17 September, the most likely route to an hours estimate not inherited from the LFS. The reconciliation findings due in autumn 2026, which should establish how much of the employment gap is survey repair. And a firm-level treatment of entry, exit and reallocation: the business dynamism statistics give the aggregate shape, and the Business Structure Database and Annual Business Survey can separate within-firm improvement from between-firm reallocation, which the sectoral evidence cannot.

The researchers involved have been careful about their own claims in ways the coverage has not. But we should be precise about what has been established. That measured productivity has improved is now well evidenced. That UK firms have become more productive is not the same proposition, and on the published data it remains, for the moment, undetermined.