Americans work fewer than 2 hours a day on average
Will the American Time Use Survey report that the civilian population aged 15 and over spent on average fewer than 2.0 hours a day on working and work-related activities?
Chance by Q4 2031
<1%
Cumulative probability
- Latest forecast
| Year | Q1 | Q2 | Q3 | Q4 |
|---|---|---|---|---|
| 2026 | 0% | |||
| 2027 | 0% | <1% | <1% | <1% |
| 2028 | <1% | <1% | <1% | <1% |
| 2029 | <1% | <1% | <1% | <1% |
| 2030 | <1% | <1% | <1% | <1% |
| 2031 | <1% | <1% | <1% | <1% |
Model reasoning
This is very unlikely. In 2025, Americans 15 and older averaged 3.32 hours a day of work, and the lowest annual figure on record since 2003 is that same 3.32. Falling below 2 hours would take a collapse in work bigger than the Great Depression within about four years. The labor market in 2026 looks normal. What would most change this forecast is a sudden, economy-wide wave of job losses, for example from very fast AI automation, starting in 2027–2028.
Current status
The latest Table 1, released June 25, 2026 for 2025, shows 3.32 hours/day for "Working and work-related activities," civilian population total. The 2024 figure was 3.43. Because only years ending after the 2026-09-16 opening date count, calendar year 2026 is the first eligible year. Its results are due in late June 2027. Recent ATUS releases came out on June 27, 2024, June 26, 2025 and June 25, 2026. That leaves five possible releases in the window, covering data years 2026 to 2030, each landing around Q2 of 2027 to 2031. No YES is possible before June 2027.
Base rate (computed)
I downloaded series TUU10101AA01011987 from BLS LABSTAT (https://download.bls.gov/pub/time.series/tu/tu.data.1.AllData):
- Annual values, 2003–2025 (2020 missing): 3.69, 3.66, 3.70, 3.75, 3.81, 3.73, 3.53, 3.50, 3.57, 3.53, 3.46, 3.59, 3.53, 3.61, 3.59, 3.57, 3.61, [2020 –], 3.50, 3.50, 3.56, 3.43, 3.32. The range is 3.32–3.81.
- Year-to-year changes: standard deviation 0.081 across 20 consecutive-year pairs. The largest drop was −0.20, in the Great Recession (2008→09).
- Quarterly series (TUU10101QA01011987): the lowest quarter on record is 3.12 (Q4 2020, COVID).
To get below 2.00 from 3.32, the figure has to fall 1.33 hours (−40%). That is about 7.3 standard deviations over five years under a random walk with the historical volatility. Roughly, the measure equals the share of the 15+ population that is employed times daily hours per worker. Below 2.0 would need about 40% of all work hours to vanish, which is beyond the Great Depression (unemployment near 25%). The reference class I used is "annual ATUS values, plus the deepest US labor shocks." Its base rate is zero observed cases, so all the probability sits in tail scenarios.
Current context
- Labor market (BLS API): in September 2026 the share of people 16+ with jobs was 59.2% (about 59.7% in 2025), unemployment was 4.2%, and average weekly hours were 34.4. That is a normal market, so the 2026 value will probably be about 3.2–3.4 and the June 2027 release can only resolve YES through a gross error.
- Survey continuity: ATUS is ongoing and BLS plans an AI module from January 2027 (Federal Register, July 10, 2026). The government is funded to December 11, 2026.
- Response rates have fallen to about 22%, but nonresponse bias of anything like 40% is implausible.
- AI forecasts: Metaculus's Labor Automation Hub median expects AI to cut overall employment by only 0.7% by 2030, and weekly hours from 38 to 34 by 2035.
Paths to YES and NO
YES needs one of:
- a collapse in US work worse than the Great Depression by 2029–2030, for example from very fast transformative AI with a demand collapse, or another catastrophe, while BLS keeps publishing a comparable Table 1;
- a publication error in that cell.
NO is the default. Discontinuing ATUS or changing its methods in ways that make the table non-comparable also leads to NO.
Strongest case for YES: the series has fallen 0.24 hours since 2023, and some forecasters expect AGI by 2027–28. Even so, removing 40% of hours within two to three years would require automating physical work at a speed robotics and institutions cannot plausibly reach. So this only justifies a small amount of tail probability in the 2029–2030 data years.
Hazard model
- Data year 2026: ~0.01%
- 2027: 0.02%
- 2028: 0.04%
- 2029: 0.08%
- 2030: 0.12%
Each hazard falls in Q2, with a sliver in Q3 to allow for a delayed release. The cumulative probability is about 0.3% by end-2031.
Resolution criteria
Question. By the last day of each quarter, has the American Time Use Survey reported that the civilian population aged 15 and over spent on average fewer than 2.0 hours a day on working and work-related activities?
Definitions.
- Statistic: the U.S. Bureau of Labor Statistics (BLS) American Time Use Survey (ATUS) annual-release Table 1, “Time spent in primary activities and percent of the civilian population engaging in each activity, averages per day by sex,” row “Working and work-related activities,” column “Average hours per day, civilian population — Total.” This is the annual average across all days of the week and the entire ATUS reference population aged 15 and over, including people not employed and people who did none of the activity on their diary day.
- Working and work-related activities: the ATUS composite category, including working, job-related activities, other income-generating activities, job search, and related travel and waiting time, as defined in the release's Technical Note. The narrower “Working” row is not the qualifying statistic.
- Fewer than 2.0 hours: a published Table 1 point estimate strictly below 2.00 hours, using its displayed precision. A value of 2.00 does not count; confidence intervals and independently calculated unrounded estimates do not determine resolution.
Resolves YES if BLS publishes a qualifying annual estimate below 2.00 hours by the quarter-end. An initial release or an official revision can qualify. The Observation period convention applies: the annual observation period must end after the opening date, so calendar year 2026 is the earliest eligible year. No attribution to AI is required. Once a qualifying publication establishes YES, later revisions do not reverse it.
Resolves NO if no qualifying estimate has been published by the quarter-end. Estimates limited to employed people, people who engaged in the activity, a demographic subgroup, weekdays, weekends, or a subannual period do not count. Neither third-party calculations from ATUS microdata nor figures from another survey count. A renamed or renumbered successor table counts only if it retains the same population, annual all-days averaging basis, and activity coverage.
Resolving source. BLS's official annual ATUS release Table 1, its archived editions, and BLS-issued revisions or equivalent successor tables. Category definitions are in the ATUS Technical Note. This is a named-source question: only BLS-published estimates qualify, rather than the broader Public confirmation sources in A2. Apply the Resolution window convention and the publication date under Resolution date (A3), not the reference year's end.