# When and how long Americans sleep (American Time Use Survey)

> Share of Americans asleep at each minute of the day, average sleep by age, and the 2003–2025 trend, computed from BLS ATUS time-diary microdata.

- Source: [American Time Use Survey 2003–2025, U.S. Bureau of Labor Statistics](https://www.bls.gov/tus/data/datafiles-0325.htm)
- Citation: U.S. Bureau of Labor Statistics. American Time Use Survey 2003–2025 Multi-Year Microdata Files (Activity summary file atussum-0325, Activity file atusact-0325). Released June 2026. https://www.bls.gov/tus/data/datafiles-0325.htm
- Sample: 258,954 respondents (one diary day each), 2003–2025; 39,586 respondents in the pooled 2021–2025 tables
- Coverage: United States, civilian noninstitutional population age 15 and over; diary days 2003–2025 (no diaries for Mar 18–May 9, 2020)
- License: Public domain: a work of the U.S. federal government (BLS), not subject to copyright in the United States (17 U.S.C. § 105); BLS asks that it be cited as the source, see https://www.bls.gov/opub/copyright-information.htm
- Retrieved: 2026-10-14
- JSON (all tables): https://karelbusta.dev/data/sleep/datasets/atus-sleep.json
- CSV, Share of Americans asleep, minute by minute (2021–2025) (1440 rows): https://karelbusta.dev/data/sleep/datasets/atus-sleep/by-minute.csv
- CSV, Average sleep by age (2021–2025) (16 rows): https://karelbusta.dev/data/sleep/datasets/atus-sleep/by-age.csv
- CSV, Average sleep per day, 2003–2025 (23 rows): https://karelbusta.dev/data/sleep/datasets/atus-sleep/trend.csv

## Tables

### Share of Americans asleep, minute by minute (2021–2025)

Weighted % of people 15+ whose primary activity at each clock minute was sleeping (ATUS code 0101: sleeping, napping and spells of sleeplessness), ATUS 2021–2025 pooled. One row per minute 00:00–23:59; one column per day type × age group. Minutes 04:00–23:59 are from the diary day, 00:00–03:59 from the following night. n: all=39586, all_15_24=2800, all_25_34=5285, all_35_44=6430, all_45_54=5379, all_55_64=6614, all_65p=13078, wd=19691, wd_15_24=1401, wd_25_34=2641, wd_35_44=3233, wd_45_54=2679, wd_55_64=3248, wd_65p=6489, we=19895, we_15_24=1399, we_25_34=2644, we_35_44=3197, we_45_54=2700, we_55_64=3366, we_65p=6589.

| Column | Type | Unit | Description |
| --- | --- | --- | --- |
| minute | number |  | Clock minute, 0 = 00:00 … 1439 = 23:59 |
| time | string |  | Clock time HH:MM (local time of the respondent) |
| all | number | % | % asleep (code 0101), all days, ages 15+; n = 39586 respondents |
| all_15_24 | number | % | % asleep (code 0101), all days, ages 15–24; n = 2800 respondents |
| all_25_34 | number | % | % asleep (code 0101), all days, ages 25–34; n = 5285 respondents |
| all_35_44 | number | % | % asleep (code 0101), all days, ages 35–44; n = 6430 respondents |
| all_45_54 | number | % | % asleep (code 0101), all days, ages 45–54; n = 5379 respondents |
| all_55_64 | number | % | % asleep (code 0101), all days, ages 55–64; n = 6614 respondents |
| all_65p | number | % | % asleep (code 0101), all days, ages 65+; n = 13078 respondents |
| wd | number | % | % asleep (code 0101), weekdays, ages 15+; n = 19691 respondents |
| wd_15_24 | number | % | % asleep (code 0101), weekdays, ages 15–24; n = 1401 respondents |
| wd_25_34 | number | % | % asleep (code 0101), weekdays, ages 25–34; n = 2641 respondents |
| wd_35_44 | number | % | % asleep (code 0101), weekdays, ages 35–44; n = 3233 respondents |
| wd_45_54 | number | % | % asleep (code 0101), weekdays, ages 45–54; n = 2679 respondents |
| wd_55_64 | number | % | % asleep (code 0101), weekdays, ages 55–64; n = 3248 respondents |
| wd_65p | number | % | % asleep (code 0101), weekdays, ages 65+; n = 6489 respondents |
| we | number | % | % asleep (code 0101), weekends & holidays, ages 15+; n = 19895 respondents |
| we_15_24 | number | % | % asleep (code 0101), weekends & holidays, ages 15–24; n = 1399 respondents |
| we_25_34 | number | % | % asleep (code 0101), weekends & holidays, ages 25–34; n = 2644 respondents |
| we_35_44 | number | % | % asleep (code 0101), weekends & holidays, ages 35–44; n = 3197 respondents |
| we_45_54 | number | % | % asleep (code 0101), weekends & holidays, ages 45–54; n = 2700 respondents |
| we_55_64 | number | % | % asleep (code 0101), weekends & holidays, ages 55–64; n = 3366 respondents |
| we_65p | number | % | % asleep (code 0101), weekends & holidays, ages 65+; n = 6589 respondents |

### Average sleep by age (2021–2025)

Weighted average hours per day spent sleeping (ATUS code 0101, includes naps and sleeplessness), ATUS 2021–2025 pooled, by 5-year age band. Weekends include the 7 BLS holidays.

| Column | Type | Unit | Description |
| --- | --- | --- | --- |
| age_group | string |  | Age |
| age_min | number | years | Age from |
| age_max | number | years | null = open-ended (85+ / all ages); ATUS top-codes age at 85 |
| hours | number | h/day | Sleep, all days |
| hours_weekday | number | h/day | Monday–Friday diary days excluding BLS holidays |
| hours_weekend | number | h/day | Saturday, Sunday and BLS holidays |
| n | number |  | Respondents |
| n_weekday | number |  | Respondents (weekday diary) |
| n_weekend | number |  | Respondents (weekend/holiday diary) |

### Average sleep per day, 2003–2025

Weighted average hours per day spent sleeping, people 15+, by ATUS year. `hours` uses the BLS definition (code 0101) and reproduces BLS Table A-1 'Sleeping'.

| Column | Type | Unit | Description |
| --- | --- | --- | --- |
| year | number |  | Year |
| hours | number | h/day | Code 0101: sleeping + sleeplessness + sleeping n.e.c. (BLS published definition) |
| hours_010101 | number | h/day | Code 010101 only, excluding sleeplessness |
| minutes_010102 | number | min/day | Code 010102 (lying awake, insomnia) |
| n | number |  | Respondents |
| weight | string |  | Weight variable |
| note | string |  | Note |

## Caveats

- ATUS 'sleeping' is self-reported in a time diary, not measured. It includes naps, dozing, falling asleep, waking up and getting up, plus 'sleeplessness' (lying awake, tossing and turning). In practice it is closer to time in bed trying to sleep than to physiological sleep, which is why the averages (about 9 h/day) are higher than survey or actigraphy estimates of sleep.
- Only the primary activity is recorded. Someone who dozes off in front of the TV but reports 'watching TV' counts as awake.
- Each respondent reports one day. The minute-by-minute curve is the share of different people asleep at each minute, not a typical person's night.
- Diaries run from 4:00 a.m. on the diary day to 4:00 a.m. the next day. On the clock-time curve, 00:00–03:59 is the night after the diary day, so the weekend curve's early hours are Saturday and Sunday nights, not Friday night.
- People round reported times, so the curves have steps at the hour and half-hour (for example the drop at 7:00).
- 2020 is a partial year: data collection stopped from March 18 to May 9. Its value is not an annual average and is not comparable with other years. BLS did not publish an annual 2020 estimate.
- 2025 has no diaries for September 30–November 11 because of the federal shutdown, and 2013 has none for September 30–October 15. BLS weights the remaining days of the quarter to stand in for the missing ones.
- Sample sizes have fallen from 20,720 respondents in 2003 to 6,146 in 2025, so recent single years are noisier.

## Methodology

## What was measured

The American Time Use Survey (ATUS) is run by the Census Bureau for the Bureau of Labor Statistics (BLS). Each
respondent is one person age 15 or older, drawn from households finishing the Current Population Survey. They report
by phone what they did on the previous day, from 4:00 a.m. on that day to 4:00 a.m. on the interview day. Every
activity gets a start time, a stop time and a 6-digit code from the ATUS Coding Lexicon.

**Sleeping** means ATUS 2nd-tier code **0101**, the code BLS uses for its published "Sleeping" estimates (Table A-1
and news release Table 1, footnote: "Includes naps and spells of sleeplessness"). It covers:

- `010101` Sleeping. The 2025 lexicon examples include sleeping, napping, dozing, falling asleep, waking up and
  getting up.
- `010102` Sleeplessness. Examples include insomnia, lying awake, tossing and turning and counting sheep.
- `010199` Sleeping, n.e.c. This code is almost never used.

Including 010102 means the % asleep curve and the hours do count time spent lying awake in bed. We kept it so these
numbers use the same definition as every BLS publication, and so the area under the curve equals the average hours.
The effect is small: sleeplessness averages 1.7–5.6 minutes a day depending on the year (`minutes_010102` in the
trend table). The trend table also gives `hours_010101`, which leaves sleeplessness out.

## Processing (ingest/atus-sleep/build.ts)

Inputs are the BLS **2003–2025 multi-year** files, listed with their sha256 in `ingest/atus-sleep/manifest.json`.
The Activity summary file (`atussum_0325.dat`) has one row per respondent, with minutes per activity code, age,
diary day, holiday flag and weights. The Activity file (`atusact_0325.dat`) has one row per diary episode.

**Weights.** Each respondent's final weight is used. That is `TUFNWGTP` for every year except 2020. `TUFNWGTP` is
the multi-year weight, built with the 2006 method so that it is comparable across years. `TUFNWGTP` is missing for
2020, so 2020 uses `TU20FWGT`. That weight represents only Jan 1–Mar 17 and May 10–Dec 31, 2020 (User's Guide §7.2
and §9.2.2). No respondents were dropped, and every respondent had a positive weight.

**Estimators** (ATUS User's Guide §7.4):

- Average hours per day = Σ wᵢ·Tᵢ / Σ wᵢ / 60, where Tᵢ = t010101 + t010102 + t010199 minutes for respondent i.
- % asleep at minute m = 100 · Σ wᵢ·Iᵢ(m) / Σ wᵢ, where Iᵢ(m) = 1 if respondent i's diary episode covering minute m
  has code 010101, 010102 or 010199.

Both sums run over the respondents in the group. Episode spans come from the Activity file: an episode covers diary
minutes [`TUCUMDUR24` − `TUACTDUR24`, `TUCUMDUR24`), counted from 4:00 a.m. and with the last episode cut off at
4:00 a.m. The build checks that every diary covers exactly 1,440 minutes. It also checks that each respondent's
sleep minutes in the Activity file equal the Activity summary file. Finally it checks that the area under the
all-days curve (9.0219 h) matches the directly computed pooled average (9.02 h).

**Day type.** We follow the BLS definition. *Weekdays* are Monday–Friday diary days that are not holidays.
*Weekends and holidays* are Saturdays and Sundays (`TUDIARYDAY` = 7 or 1) plus the holidays flagged by
`TRHOLIDAY` = 1: New Year's Day, Easter, Memorial Day, the Fourth of July, Labor Day, Thanksgiving and Christmas.
ATUS samples weekend days more heavily, so about half of respondents have a weekend or holiday diary. The weights
restore each day's real share.

**Age.** Age is `TEAGE`, which BLS top-codes: since 2005, `TEAGE` = 80 means 80–84 and 85 means 85 and older.

**Pooling.** The `by-minute` and `by-age` tables pool the five latest years, **2021–2025**. All of them come after
the 2020 suspension, and pooling gives stable curves. The pooled sample is 9,087 + 8,136 + 8,548 + 7,669 + 6,146 =
39,586 respondents. Raw weights are pooled without rescaling. Each year's weights add up to that year's
person-days, so every year counts in proportion to its population.

## Tables

- **by-minute.** One row per clock minute (`minute` 0–1439, `time` 00:00–23:59), with 21 columns of % asleep:
  day type (`all`, `wd` = weekdays, `we` = weekends & holidays) × age group (15+, `15_24`, `25_34`, `35_44`,
  `45_54`, `55_64`, `65p`). Values are rounded to 0.1 percentage point. Clock minute c corresponds to diary minute
  (c − 240) mod 1440, so 04:00–23:59 is the diary day and 00:00–03:59 is the following night. Each column's
  respondent count is in its description. Counts range from 1,399 (weekend, 15–24) to 39,586 (all).
- **by-age.** Average hours of sleeping per day (2 decimals) for all days, weekdays and weekends & holidays, by
  5-year age band (15–19 … 80–84, 85+) plus "All 15+", with the respondent count for each.
- **trend.** Average hours of sleeping per day for each year 2003–2025, people 15+ (`hours`, code 0101). Also
  `hours_010101`, `minutes_010102`, the respondent count, the weight variable used, and a note for the years with
  data gaps (2013, 2020, 2025).

## Verification against BLS published estimates

`trend.hours` was compared with the "Sleeping" row of BLS Table A-1 for each published year
(https://www.bls.gov/tus/tables/a1_2003.pdf … a1-2025.pdf, all persons, "Average hours per day, civilian
population"). Shown as published / this dataset (unrounded):

| Year | BLS A-1 | This dataset | | Year | BLS A-1 | This dataset |
|---|---|---|---|---|---|---|
| 2003 | 8.57 | 8.567 | | 2015 | 8.83 | 8.834 |
| 2004 | 8.55 | 8.546 | | 2016 | 8.79 | 8.787 |
| 2005 | 8.62 | 8.625 | | 2017 | 8.80 | 8.803 |
| 2006 | 8.63 | 8.627 | | 2018 | 8.82 | 8.816 |
| 2007 | 8.57 | 8.574 | | 2019 | 8.84 | 8.841 |
| 2008 | 8.60 | 8.602 | | 2021 | 8.95 | 8.951 |
| 2009 | 8.67 | 8.672 | | 2022 | 9.02 | 9.016 |
| 2010 | 8.67 | 8.665 | | 2023 | 9.07 | 9.071 |
| 2011 | 8.71 | 8.714 | | 2024 | 9.04 | 9.036 |
| 2012 | 8.73 | 8.731 | | 2025 | 9.03 | 9.033 |
| 2013 | 8.74 | 8.738 | | | | |
| 2014 | 8.80 | 8.797 | | | | |

All 22 published years match to the published two decimals. 2005 sits exactly on a rounding boundary (8.6250).
BLS's single-year 2003–2005 tables used that year's weighting method rather than the 2006 method behind
`TUFNWGTP`, so tiny differences are possible in those years. Code 010101 alone would *not* match: it is 0.03–0.09 h
lower. BLS published no annual 2020 Table A-1.

The weekday/weekend split was also checked for 2025. BLS news release Table 2 reports 8.76 h on weekdays and
9.68 h on weekends and holidays. The same code run on the 2025 respondents alone gives 8.756 and 9.680. That
split is not a table in this dataset; the check only confirms the day-type definition.

## Built from this dataset

- Ranking: [How much do Americans sleep at each age?](https://karelbusta.dev/data/sleep/sleep-by-age)
- Fact: [Americans spend more time asleep than they did 20 years ago, not less](https://karelbusta.dev/data/sleep/americans-sleep-more-than-in-2003)