The Digital Whale

Work

Remote and hybrid work

The randomised evidence on home working points in three directions at once, and the headline share of remote workers depends on which survey is cited.

Between a fifth and a third of American workers do some work from home, depending entirely on which survey is asked, and the randomised evidence on whether home working raises or lowers output does not point one way. A 2015 call-centre experiment found a large performance gain. The same research team's 2024 hybrid trial found no performance effect at all, though quitting fell by a third. A randomised trial of data-entry workers in India found the opposite sign again. All three are well-conducted experiments, and they disagree because the work differs. The most repeated line on this subject — that remote work is 13% more productive — is the call-centre number applied to jobs it was never measured on. For a practical commercial comparison with this kind of workplace measurement, see online timesheets.

The national numbers are three different questions

There is no single figure for how much work happens at home, because the main instruments ask incompatible questions. The US Bureau of Labor Statistics runs two. Its monthly household survey asks whether a person teleworked or worked at home for pay in the reference week.

2026

22.2% of employed people in the US teleworked or worked at home for pay in July 2026, 34.1 million of 153.4 million at work.

Direction: No detectable effect. Strength of evidence: Strong.

US Bureau of Labor Statistics, Current Population Survey Table A-41, 2026Current Population Survey, 153.4 million people at work, July 2026

Caveat Counts any paid work at home in the reference week, mixing occasional evening email with fully remote jobs, and the series only begins in October 2022.

Its time-use survey asks people to account for a diary day instead, producing a higher number for a different reason, and showing the education gradient clearly.

2026

35% of employed Americans did some or all of their work at home on days they worked in 2025, rising to 51% among those with a bachelor's degree or higher against 19% for those with high school only.

Direction: Increase. Strength of evidence: Strong.

US Bureau of Labor Statistics, American Time Use Survey 2025 Results, 2026American Time Use Survey, 2025 results, released 25 June 2026

Caveat A diary measure of any work at home on the diary day, conceptually different from the telework question, so the two BLS series must not be plotted together.

The most-quoted series of all is not a government statistic. WFH Research's Survey of Working Arrangements and Attitudes measures the share of paid full days worked at home, and it is an opt-in online panel.

2026

About 26% of paid full days in the US were worked from home in July 2026, roughly flat since 2023 after falling from the 2020 peak.

Direction: No detectable effect. Strength of evidence: Mixed.

Barrero, Bloom & Davis, WFH Research updates release, 2026Opt-in online panel of US residents aged 20-64 earning $10k or more, reweighted to CPS margins

Caveat Not a probability sample, and it consistently reports higher home working than government surveys do.

Britain's position is worse documented than it looks. The Office for National Statistics reported 28% of working adults hybrid, 16% working from home only and 56% travelling to work only, from 6,670 Opinions and Lifestyle Survey respondents covering September 2022 to January 2023 (ONS, 2023). That remains its flagship article on the split, and it is now over three years old.

A call centre in 2010 is not an office today

The experiment that started the argument randomised volunteers at a Chinese travel agency into home and office working. The task was routine phone work with individually countable output, which is why the effect was measurable at all.

2015

Home working raised performance 13%, made up of 9% from more minutes worked per shift and 4% from more calls per minute, and roughly halved attrition, while cutting promotion conditional on performance.

Direction: Increase. Strength of evidence: Strong.

Bloom, Liang, Roberts & Ying, Quarterly Journal of Economics, 2015Randomised trial, Ctrip call-centre employees, volunteers only

Caveat Routinised, individually measurable work with self-selected volunteers, and the promotion penalty is almost always dropped when the 13% is quoted.

The hybrid trial found retention, not performance

Nine years later the same lead author ran a larger randomised trial on graduate engineers and marketing staff at the same firm, testing two work-from-home days rather than four.

2024

Two home-working days produced no detectable effect on performance grades, lines of code or promotions, while cutting quit rates by a third, from 7.20% to 4.80%.

Direction: No detectable effect. Strength of evidence: Strong.

Bloom, Han & Liang, Nature, 2024Randomised trial, 1,612 Trip.com employees, two years of follow-up

Caveat One Chinese online travel firm with strong management systems, testing a two-day hybrid design, so it says nothing about fully remote arrangements.

Retention is the durable finding across both trials; performance is not.

The clearest counterweight comes from India

The third experiment reverses the sign, in a setting that isolates the mechanism: poor home infrastructure, no supervision and piece-rate output counted to the keystroke.

2023

Workers assigned to work from home were 18% less productive than office-assigned workers, with two-thirds of the gap appearing immediately.

Direction: Decrease. Strength of evidence: Strong.

Atkin, Schoar & Shinde, NBER Working Paper 31515, 2023Randomised experiment on data-entry workers, India

Caveat Low-skill piece-rate data entry with poor home infrastructure, so it generalises to professional work no better than Bloom's trials do.

Read together, the three trials say that the effect of home working on output is a property of the task, the worker and the home, not of remote work as a category. A gain, a null and a loss are all real results, and anyone quoting one as the answer has chosen it. What the tools themselves change is covered in what workplace technology measures, and the difficulty of measuring output at all in why productivity resists measurement.

Return-to-office mandates have not been causally estimated

Claims that mandates cause senior staff to quit or destroy shareholder value trace to two unrefereed 2024 working papers on S&P 500 firms by Ding, Ma and co-authors, posted on SSRN as Return-to-Office Mandates and Return to Office Mandates and Brain Drain. Their effect sizes are not quoted here: the primary documents could not be verified, they are not peer reviewed, and mandate timing is not random. Firms often mandate after a bad year, the exact confound that would produce the reported pattern with no causal effect.

The short version

  • 22.2% of employed Americans teleworked in July 2026 on the household survey, against 35% who did some work at home on a diary day in 2025 — different questions, not a trend.
  • The 2015 Ctrip trial's 13% performance gain came from volunteers doing routine, individually countable phone work, and the same paper found promotion chances fell.
  • The 2024 randomised hybrid trial found no performance effect, and cut quitting by a third.
  • A randomised trial of Indian data-entry workers found home-assigned staff 18% less productive, which is the same design pointing the other way.
  • The evidence has a hard limit: no peer-reviewed study has causally estimated what a return-to-office mandate does, and Britain's official hybrid split dates from early 2023.