China's national sleep averages do not agree

Ask how long people in China sleep and the published answers sit 59 minutes apart. One series says 6.85 hours a night, another 7.84, one year apart, each describing the whole country.

Seven rows follow, six surveys: two device series, four questionnaires.

Six averages for one country

Two rows are one survey counted twice: one 2024 fieldwork, published in the agency’s bulletin and again in a journal with the PSQI score. The two papers even disagree on their field windows.

The four questionnaire series sit between 7.24 and 7.84 hours. The two device series sit under 7. The instrument explains more of the spread than the year does.

What each instrument counts

The wearable series comes from the sleep society with a device maker’s health platform, and describes platform users who already wear the hardware: the 2026 edition drew 10 percent of its data, by stratified random sampling, from more than 250,000 users across 31 provinces aged 18 and over, and reports night sleep only. Any average from it describes device-wearing adults who opted in.

The 2024 survey is the closest thing here to a national probability sample: one provincial-level division per region, ten county-level units in each, three districts or townships, two communities or villages, 120 households per community and one resident aged 15 or over, drawn by multistage stratified cluster randomisation with weights built from the 2020 census. Its confidence interval is roughly ±0.08 hours around 7.24, which covers the sampling error of the design, not the gap to 6.85.

The chronic-disease surveillance asks one question — how long do you sleep in total on a typical day — recorded to the minute, and weights to the 2010 census rather than the 2020 one. A typical-day total is not nightly sleep, and the paper does not say whether naps sit inside it. The Kadoorie baseline does: it asks how many hours you typically sleep per day, including naps.

The sleep report series runs about six thousand respondents, mostly online panels, and publishes a sleep index alongside duration. The Kadoorie interview is a one-off baseline from 2004–2008.

Duration in bed is not duration asleep. In the 2024 survey, 22:08 to 6:18 is 8 hours 10 minutes; minus 27 minutes of sleep latency, 7.71 hours, against a published mean of 7.24, so about 28 minutes go to waking. The 2026 white paper implies the same: 00:10 to 7:27 is 7 hours 17 minutes against its stated 6.97, a 19-minute gap. Both are my arithmetic on their published means.

One series, four samples, one lookalike

The report series published with the Chinese Academy of Social Sciences is the most quoted Chinese sleep series; its instrument changed under it.

The 2022 wave is a household survey with probability-proportional sampling; the 2021 and 2023 waves are online panels. The 2025 row is a different book: the index report’s 2026 edition, its fourteenth year, released with a university’s AI group. Reading five rows as one trend line assumes the instrument stayed the same. It did not.

The insomnia question shows the seam. The share reporting no insomnia in the past month is 35.6 percent for 2022, 89.6 percent for 2023 and 34.1 percent for 2024, in the same series. A jump from 35.6 to 89.6 and back to 34.1 is more consistent with a change in respondents or in how the question was read, and no document says which.

The claim that Chinese people sleep an hour and a half less than a decade ago rests on the same series, whose coverage says the average is nearly 1.5 hours shorter than ten years earlier, naming no baseline year or figure (CCTV, 2022-03-18). The index’s method changed inside that decade: the 2013 edition computed duration as wake-up time minus bedtime, the 2014 edition asked respondents to report it directly, and that report says the two years are not comparable (2014 index report PDF, fieldwork Oct–Dec 2013, 8,286 samples).

The same word on five different rulers

Prevalence is where definitions come apart worst. The most repeated number in Chinese sleep coverage is 38.2 percent.

One percentage travelled a long way: announced in 2016 as the incidence of sleep disorders, quoted in 2019 as the adult insomnia rate with a headcount, repeated in 2023 as sleep-wake disorders with a global comparison. Each version holds only if the earlier wording is dropped.

The government’s own indicator shows why one number is hard to pin down. The 2019 action plan defines 失眠现患率 as the share who screen positive on a sleep-related scale and puts its 2016 baseline at 15 percent, with the 2022 and 2030 target a slowing rise, not a level. It recommends 7–8 hours of daily sleep for adults, against a 6.5-hour baseline its footnote dates to 2017 and attributes to no survey.

The 2025 guideline names the condition it counts: insomnia disorder, marked by difficulty falling asleep, staying asleep or waking early, together with impaired daytime functioning (National Medical Journal of China 105(34):2960, 2025-09-09). A survey question about insomnia in the past month, a PSQI cutoff, and that clinical definition each produce a different rate from the same population.

Adolescents: thresholds, not averages

For schoolchildren there is no ambiguity about the target. The 2019 action plan lists, as an advocacy indicator, at least 10, 9 and 8 hours a day for primary, junior and senior secondary pupils. The education ministry repeated the same three figures as requirements in 2021 (教基厅函〔2021〕11号, signed 2021-03-30).

The national education monitoring programme reports against them as shares, not means.

The ministry’s own commissioned monitoring in December 2020 across 10 provinces reported means instead: 9.5 hours for primary pupils and 8.4 for junior secondary, described by the ministry as showing insufficient sleep “to varying degrees” (briefing, 2021-04-02).

A third version comes from the society’s 2019 youth white paper, on 6–17-year-olds: 62.9 percent slept less than 8 hours, and 81.2 percent of 13–17 year olds (Xinhua, 2019-03-21).

So a fifth of grade 4 pupils clearing 10 hours, a ministry mean of 9.5 hours and 62.9 percent sleeping under 8 hours can all be true at once: two thresholds, three instruments.

Group numbers cannot be read against national ones

• The 2026 white paper reports university students sleeping 18 minutes longer than the whole sample, at 7.2 hours, with 21 percent falling asleep after 02:00 and 32 percent waking after 09:00 (reporting, 2026-03-23).
• The 2025 white paper, as summarised by a report index, puts the shortest night at 36–45 years (6 hours 44 minutes), with 26 percent under six hours and 39 percent waking twice or more (summary page, 2025-03-19).
• The 2025 sleep report puts middle-income respondents at 6.85 hours against 7.84 for its whole sample — one hour inside one instrument — and mothers at 7.68 against 7.95 for non-mothers (t=4.76, p<0.001) (publisher’s press material, 2025-03).
• In the 2024 national survey, women aged 18–44 slept longer than men (7.66 against 7.49 hours) while women aged 45–64 slept less (6.82 against 6.97), and poor sleep quality rose from 11.3 percent at 15–24 to 41.0 percent at 65 and over (journal version, 2026-04-09).

The figure 6.85 hours appears twice as two different things: a device average, and one survey’s middle-income subgroup.

Method

Everything here comes from documents read on 2026-10-12, or from arithmetic on their figures. I have no first-hand sleep data.

• National surveys: the China CDC Weekly report of the 2024 survey and its journal version, the chronic-disease surveillance trend paper, and the Kadoorie baseline paper.
• Report series: publisher abstracts for the 2023 and 2024 editions, press accounts of the 2021, 2022, 2025 and 2026 editions, and the index report’s own 2014 edition. These are books; abstracts and reporting were all I could read.
• Wearable white papers: official media reporting for both editions, plus a report index’s summary for the 2025 subgroup figures. Neither white paper’s file was readable to me.
• Rules and targets: the 2019 action plan, from the health commission’s indicator-table PDF and from a copy of its full text; the 2021 education ministry notice and its briefing; the education monitoring reports.
• Definitions: the 2025 insomnia guideline’s abstract, the review that compares 38.2 percent with a global rate, and the two papers that define prevalence by scale score.
• Two arithmetic checks: time in bed minus sleep latency minus reported duration, for the 2024 national survey and the 2026 white paper.

What I could not check

• The 38.2 percent figure’s original instrument. I found it announced at a 2016 press event and repeated since, with no page giving that survey’s sample, question or age range.
• The 15 percent baseline. The indicator table prints it against 2016 and states no sample, instrument or question wording.
• The 6.5-hour baseline. The table’s footnote dates unlabelled baselines to 2017 and names no survey.
• “More than 510 million” people with sleep disorders. Press accounts attribute it, alongside the 38.2 percent rate, to a 2023 big-data report I could not find published (2024-03-05). It is a headcount with no stated denominator.
• The 2025 white paper at source. Its 6.85-hour mean reaches me through official media reporting; the 36–45 subgroup and the 26 and 39 percent shares reach me through a report index summarising the white paper. The society’s page for that edition returned no content when I read it.
• The 2026 white paper at source. Its numbers here come from the health ministry’s newspaper and from technology-press reporting of the launch. I found no publisher PDF or abstract.
• The wearable samples’ composition. Neither edition’s reporting gives the sex split or age distribution of its sample, so a device average cannot be reweighted against a national one.
• The 2025 index figures (7.16 h, 63.15). Both are from press reporting of the index report; I found no publisher abstract for that edition and could not confirm who ran the fieldwork.
• Whether the 2024 national survey is the same fieldwork as the society’s screening programme. The papers and the screening announcement share a lead author, and the papers list the society among their funders. No document I read states the link outright.
• A unit error. Press coverage of the 2026 white paper reports an annual economic loss from sleep problems of “over 240 million yuan” (health ministry newspaper, 2026-03-23). At national scale that reads like a misplaced unit. I left the figure out rather than guess.