The numbers in circulation, and what their sources actually say

How to read this page

Three verdicts, and a single criterion for assigning them: did we open and read the primary source?

• Faithful — the source says exactly what it is made to say.
• Distorted — the figure exists, but it changed meaning along the way: a lost denominator, a dropped condition, a shifted comparison period, an omitted caveat from the author.
• Unverifiable — we found no primary source, or the one cited cannot structurally produce this figure.

One clarification that matters: "distorted" does not mean "wrong at heart." It means that this particular figure does not prove what it is made to prove. Our job is to verify, not to settle the debate.

And one French trap worth knowing before anything else: three incompatible arrangements share the same name — 32 hours paid as 35, 35 hours compressed into four days, and plain schedule flexibility. The surveys that measure them cannot be added together. Yet they systematically are, in the same paragraph.

Microsoft Japan

What to know first: the press release exists, the figure is in it — and Microsoft itself pulled it out of its "results" section nine days after publication.

"Microsoft Japan's productivity jumped 40%"

Distorted. In the Japanese text, the metric is called 労働生産性（売上/社員数） — "labour productivity (revenue / number of employees)" — up +39.9% compared with August 2018. It is a one-month accounting ratio, not a measure of work efficiency.

On 8 November 2019, Microsoft added this note to its own press release: "these figures are factual information, but they are not results obtained through this challenge alone; they stem from various factors." The figure was moved out of the section that presented the results. No secondary source mentions it.

Source: "ワークライフチョイス チャレンジ 2019 夏" effect measurement results published, Microsoft Japan, 31 October 2019, corrective note of 8 November 2019.

"The trial covered 2,300 employees"

Unverifiable. The press release does not mention any headcount. It does specify the arrangement, however — closing every Friday in August 2019, namely the 2nd, 9th, 16th, 23rd and 30th: five Fridays, not four as is often reported.

"Electricity use fell 23%, paper use 59%"

Distorted — not in the values (−23.1% and −58.7%), but in their baseline: these are compared with August 2016, while the productivity figure is compared with August 2018. Two different baselines, presented as a single block of results.

Three things rule out attributing anything to the freed-up day alone: meetings were capped at 30 minutes in the same move, there is neither a control group nor a published methodology, and August is the month of Obon — a Japanese seasonal slowdown.

Iceland

What to know first: the famous 86% measures a right of access obtained through collective agreements after the trials, not a trial outcome. And the arrangement tested was not a four-day week.

"86% of Icelandic workers have moved to a four-day week"

Distorted, three times over. The report's sentence reads: "roughly 86% of Iceland's entire working population has now either moved to working shorter hours or have gained the right to shorten their working hours." The rate measures the share of the workforce covered, as of June 2021, by collective agreements negotiated between 2019 and 2021.

In passing, and because the rule applies to us too: the wording "86% of the country's workforce are now working shorter hours or gaining the right to shorten their hours", which circulates as the report's own, does not appear in it. It comes from a press summary. We had quoted it here ourselves until a review caught it.

Three slippages stack on top of each other: the right to negotiate becomes an actual practice; a diffusion figure recorded after the trials becomes a trial result; and "shorter hours" becomes "four-day week."

"Iceland's pilot proves it works at national scale"

Distorted. Two trials, in the public sector only: Reykjavik City Council (2014-2019, about 2,500 participants) and the national government (2017-2021, about 440) — close to 2,940 people, not a country. The change went from 40 hours to 35-36 hours, often spread over five days. At most workplaces, the lasting reduction actually achieved was one hour or less.

The results themselves are good — productivity stable or rising, wellbeing markedly improved, pay maintained. But participating workplaces were volunteers, there was no control group, and none of this was measured in the private sector.

Source: Going Public: Iceland's Journey to a Shorter Working Week, Guðmundur D. Haraldsson (Alda) and Jack Kellam (Autonomy), June 2021.

The 2022 UK pilot

What to know first: the raw figures are almost always accurate. What gets lost along the way are the denominators, the difference between "share of employees reporting a fall" and "size of the fall" — and a sentence in which the authors themselves declare three of their own results not significant.

"Revenue rose 35%"

Distorted. The report states that revenue "stayed broadly the same over the trial period, rising by 1.4% on average, weighted by company size." The +35% is an annualised comparison with the same periods in previous years — it therefore mixes the effect of the scheme with the post-Covid recovery.

Neither figure covers all 61 companies: the report gives 23 organisations for the +1.4% ("across those 23 organisations who gave the data") and 24 for the +35% ("across the 24 companies that supplied sufficient data"). Two different sub-samples, and neither is the pilot as a whole.

"92% of companies kept the four-day week a year on"

Distorted — because two separate measures were merged.

The 92% figure is solid: 56 of the 61 companies were still going at the end of the trial, in December 2022, 18 of them permanently. It says nothing about a year on.

The one-year figure is 89%: at least 54 of the 61 organisations were still running the scheme and at least 31 had made it permanent — but only 28 of the 61 organisations took part in the follow-up study, a non-response bias the authors do not analyse.

"Burnout fell 71%, stress fell 39%"

Distorted, and it is the single most mechanical error in the whole corpus. The report states "71% had reduced levels of burnout" and "39% of employees were less stressed." These are proportions of employees reporting an improvement, not the size of that improvement. Self-reported, before/after measures, with no control group.

"Absenteeism fell 65%, resignations fell 57%"

Distorted — not in the values, which are accurate (absenteeism from 2.0 to 0.7 days per employee per month; departures from 2.0 to 0.8 per 100 employees), but in the sentence that immediately follows them in the report:

"In part because of the small numbers in the sample, and in part because of labour market contingencies, we are unable to say that these three trends are statistically significant."

It is the clearest case in this whole exercise: the source disavows its own figure in the very next sentence, and that sentence is never repeated alongside it.

"The four-day week rebalances housework"

Distorted by omission. The report does find that time spent by men on childcare rose twice as much as for women (27% versus 13%) — then adds: "but the share of housework between these two genders stayed almost exactly the same (68% reporting no change for men and women)." Coverage keeps the first half of the sentence and cuts the second, even though the authors flag it themselves with a "but."

"X% reduction in the mental-load gap between men and women"

Unverifiable. The report publishes two figures not broken down by sex: 54% of employees found it easier to balance work and housework, 60% found it easier to balance paid work and caring responsibilities. The gender dimension is added afterwards.

Sources: The Results Are In: The UK's Four-Day Week Pilot, Autonomy / 4 Day Week Global / Boston College / University of Cambridge, February 2023; Making It Stick: The UK Four-Day Week Pilot One Year On, Autonomy.

Spain

What to know first: the scheme is documented, its results are not.

"The Spanish pilot delivered conclusive results"

Unverifiable. What is established concerns the scheme itself: a grant programme from the Ministry of Industry, Trade and Tourism, worth about €50 million, targeting around 250 industrial companies moving to 32 hours. By May 2023, only 41 applications had been submitted.

No evaluation report with a stated methodology and public data could be located, and the dedicated ministry page returns an error when accessed. The Spanish results circulating in the French-language press can therefore only be traced to isolated company testimonials.

"The Valencia pilot demonstrated the benefits of the four-day week"

Distorted — because this was not a company-level trial. In 2023, Valencia grouped local public holidays onto four consecutive Mondays, producing a four-day week for 360,000 workers.

The reported effects are mixed, and coverage keeps only half of them: lower stress, better air quality, children's wellbeing — but also higher tobacco and alcohol consumption and lower retail sales, offset by hospitality and tourism. None of this is quantified in any accessible publication.

France — the surveys, and actual adoption

What to know first: no opinion poll measures actual adoption. And the three arrangements that share the name "four-day week" are not measured by the same questions.

"77% of French people support the four-day week"

Distorted, twice over. The question asked was: « Vous personnellement, seriez-vous favorable ou pas favorable à l'instauration de la semaine de 4 jours, 32 h payées 35 h, dans les entreprises publiques et privées en France ? » [Personally, would you be in favour or not in favour of introducing the four-day week, 32 hours paid as 35, in public and private companies in France?]

The result is 70% for French people as a whole; the 77% figure covers only working respondents, 57% of the sample, or about 575 people — hence a margin of error of around ±3.3 points, per the report's own table.

And the condition "32 hours paid as 35," which changes everything, disappears from the headlines: support for a fully paid reduction in working time becomes support for "the four-day week" in general.

Source: Les Français et la semaine de 4 jours, Ifop for Politis, no. 120718, survey of 12-13 March 2024, 1,008 people.

"6 in 10 French employees (64%) are ready to adopt the four-day week"

Distorted. The question asked about a wish for « plus grande flexibilité dans l'organisation de leurs horaires, avec la possibilité de les condenser sur une semaine de 4 jours » [greater flexibility in organising their hours, with the option of condensing them into a four-day week]. This is an appetite for flexibility, relabelled as support for the four-day week.

The survey was conducted in-house by the ADP Research Institute, with no third-party institute, and no detailed methodology or margin of error published.

"Only 5% of French companies have adopted the four-day week"

Unverifiable — and we do not repeat it. This figure appears in the same ADP release, with no source or reference at all, and it is incompatible with the document's own methodology: the survey questions a panel of 1,951 employees, never companies. Such a design cannot structurally produce a company adoption rate.

The only official orders of magnitude available come from Dares: 3.4% of non-farm private-sector companies with 10 or more employees mainly apply a weekly cycle of fewer than five days (2022), and about 10,000 employees were affected as of January 2023 — on the order of 0.05% of private-sector employees.

"More than 1 in 3 employees (36%) are ready to take a pay cut to work 4 days"

The figure is faithful; its retelling is not. The trade-off was a 5% pay cut, and that figure — without which the claim makes no sense — disappears from almost every retelling.

Source: The New Human Age, ManpowerGroup, survey of November 2022, 1,001 employees in France.

"67% of working French people would like to work 4 days a week"

Unverifiable, and instructive. The wording used specifies « même en conservant le même volume horaire hebdomadaire, en échange de journées plus longues » [even while keeping the same weekly hours, in exchange for longer days]: this figure measures appetite for a compressed week — the exact opposite of what the Ifop poll measures. Yet the two are regularly placed side by side as if they confirmed each other.

We could not locate the original 2023 Indeed / OpinionWay release: neither the exact question, nor the quotas, nor the margin of error are accessible.

"Nearly 10% of French companies have trialled or are considering the four-day week"

Unverifiable. This claim is attributed to "Dares 2023" and matches no Dares figure we could find. The real orders of magnitude are given above, and the gap is considerable: 10,000 employees is 0.05% of private-sector employees.

The climate argument

What to know first: the figure is faithfully copied from the report. What changes is its status — a conditional projection for a horizon that has now passed becomes a present-tense fact.

"The four-day week would cut the UK's carbon footprint by 127 million tonnes a year"

Faithful to the source, distorted in its status. The report states: "shifting to a four-day working week without loss of pay could shrink the UK's carbon footprint by 127 million tonnes per year by 2025."

It is a conditional scenario, with a 2025 horizon — a horizon that has now passed, and the transition never happened. It is repeated in the present tense as an established effect.

The calculation itself is worth knowing: a linear regression on the 2009-2018 consumption footprint, the assumption of a move from 42.5 to 32 hours over four years, and an elasticity drawn from a 2012 paper — applied to the entire national footprint, retirees and children included, even though the hours reduction only applies to full-time workers. No rebound effect is modelled.

"That is the equivalent of taking every car in the UK off the road"

Faithful to the source, a shaky comparison — and the report says so itself in a footnote: the emission factor used is that of the average American vehicle, whose mileage and emissions are noticeably higher than the UK fleet's. On top of that there is a mismatch in what is being compared: a consumption footprint (46% imported emissions) set against tailpipe emissions.

Source: Stop the Clock: the environmental benefits of a shorter working week, Platform London for the 4 Day Week Campaign, May 2021.

"A Massachusetts study finds almost 30% less carbon footprint"

Distorted, and the gap is on the part that matters. The 2012 paper distinguishes two effects for a 10% reduction in hours:

• with GDP held constant, i.e. pay maintained: −8.6% carbon footprint, and a non-significant effect on territorial emissions;
• with GDP shrinking, i.e. lost income: −14.6%.

The 30% order of magnitude is only reached by using the second case, or an extrapolation that a 2019 report itself introduces with "extrapolating for the sake of argument." In other words: the figure assumes a fall in income — exactly the opposite of the "without loss of pay" version defended by those who cite it.

"The four-day week would boost the birth rate"

Unverifiable. No mention of fertility, birth rates, or natality in any of the primary reports we consulted. No causal study identified to date.

What survives verification

This page tears a lot down. Stopping there would be dishonest: checking figures is only useful if you also say what holds up. Here are the best-established results to date — with their limits, because they all have some.

The strongest study dates from 2025, published in Nature Human Behaviour by Wen Fan and Juliet Schor: 141 organisations, 2,896 employees, six countries, and — this is what sets it apart from everything else — 12 control organisations. The observed effective reduction is about 5 hours a week. Burnout falls from 2.83 to 2.38 on a 1-to-5 scale, job satisfaction rises from 7.07 to 7.59 out of 10. The authors flag the limits themselves: volunteer companies and therefore not representative, no randomisation, self-reported data, a sample dominated by professional services.

Persistence over time is the most robust fact in the file. In the UK, 56 of 61 companies carried on at the end of the trial — a complete denominator, a simple measure, no interpretation required. In Portugal, 37 of 41 companies kept the arrangement. In Iceland, productivity held steady or improved, at constant cost. Companies that were losing by it would not have continued.

Two objective rather than self-reported measures are worth knowing, because they escape the bias of self-assessment: the 2024 German pilot, run with the University of Münster, measured against a control group 38 extra minutes of sleep per week and nearly 1,850 more steps per week.

What nothing establishes to date: a revenue gain attributable to the scheme (the UK trial measures +1.4%, which is to say stability), an effect on carbon emissions at constant income, and any transfer beyond professional services — the published trials are overwhelmingly concentrated there.

The detail of this work, with its sources, is in our reference page on the four-day week.

And the legal question?

It is handled elsewhere, with the same demand for sources: what the law actually says, labour-code article by labour-code article, for France and six other countries.

What we could not verify

In the interest of honesty, here is the list of what remains open — and why:

• the Dares PDF by Pauline Grimaud on the four-day week in company agreements, blocked by an anti-bot mechanism: it would pin down the counting method behind the 10,000 employees and the exact definition of the 3.4%;
• the AEF dispatch where a figure of 8.6% appears whose subject escapes us (subscriber-only access) — we therefore do not use it;
• the original 2023 Indeed / OpinionWay release, never located;
• the L'ObSoCo-Actineo barometer, encountered only once, never traced back to its source;
• the Spanish ministry page, inaccessible, and any evaluation that may have been published since.

This list will get shorter as sources reopen. We would rather publish it than paper over the gaps with smooth phrasing.