AI Analytics

Technical writing

BLS American Time Use Survey: The Federal Dataset Behind How Americans Actually Spend Their Time

· AI Analytics
Federal DataBLSDemographicsLabor Economics

The Bureau of Labor Statistics American Time Use Survey is the only federal dataset that measures how Americans allocate every hour of the day across the full range of human activity — from sleep and paid work to childcare, volunteering, TV watching, and prayer. Launched in 2003, the ATUS asks roughly 10,000 Americans per year to reconstruct their prior 24 hours in precise detail. The result is a record that is simultaneously a demographic survey, a welfare economics instrument, and the most granular account the federal government produces of how American life is actually lived.

What ATUS Is and Where It Comes From

The American Time Use Survey is conducted by BLS in partnership with the Census Bureau. It launched in January 2003, making it one of the newer additions to the BLS survey portfolio. Before ATUS, there was no continuous federal time-use data source in the United States; researchers relied on ad hoc academic surveys, the most prominent of which was the 1965–1966 Americans' Use of Time study conducted by the Survey Research Center at the University of Michigan. ATUS filled that gap with a permanent, nationally representative, annually recurring time diary.

The survey population is the civilian non-institutional population age 15 and older — the same universe as the Current Population Survey, from which ATUS respondents are sampled. Households are first contacted as part of the CPS monthly household survey, then a random member of the household age 15 or older is selected for the ATUS interview two to five months later. Because ATUS respondents are drawn from CPS households, all standard demographic characteristics available in the CPS — labor force status, occupation, industry, education, household composition, earnings, and race — can be linked to the time-use diary.

Annual sample sizes have shifted over time. From 2003 through 2019, roughly 26,000 respondents were interviewed each year. Beginning in 2021, BLS reduced the sample to approximately 10,000 per year as part of broader CPS redesign and resource constraints. The smaller post-2021 sample affects the precision of subgroup estimates — particularly for demographic groups or detailed activity categories that represent a small share of the population — but the core methodology and activity coding remain unchanged.

The response rate is approximately 45% of contacted CPS households — substantially lower than the CPS itself. BLS publishes annual response rate documentation and applies weighting adjustments to account for nonresponse, but the relatively low response rate is a standing methodological concern. Research has found evidence of time-use nonresponse bias: busier individuals are harder to reach and may have different time-use patterns than those who complete the survey.

The Time Diary Methodology

The ATUS interview collects a 24-hour time diary covering the reference day — the day immediately preceding the interview. The reference period runs from 4:00 am to 4:00 am rather than midnight to midnight, because the 4 am boundary captures most people after sleep and before the morning routine, reducing the number of activities that span the diary boundary.

The respondent reconstructs their day in chronological order, answering a series of standardized questions for each activity episode: What were you doing? Where were you? Who was with you? The interviewer probes for concurrent activities — ATUS distinguishes primary activities (what the respondent was mainly doing) fromsecondary activities (what they were doing simultaneously). A person who listens to a podcast while commuting reports driving as their primary activity and listening as secondary. Most published ATUS statistics are based on primary activities, though the secondary activity data is available in the microdata and is used in caregiving research.

Each activity is coded using the ATUS Lexicon, a hierarchical six-digit classification system maintained by BLS and the Census Bureau. The Lexicon has approximately 400 distinct activity categories organized under 17 major categories, each major category assigned a two-digit tier-one code. A respondent who reports “cooking dinner” is assigned code 020201 (Food and Drink Preparation), which falls under tier-one code 02 (Household Activities). The Lexicon is revised periodically to reflect emerging activities — online shopping and social media use were added in later revisions — and BLS publishes a crosswalk between revisions to support longitudinal comparison.

In addition to the activity code, each episode is coded for location (home, respondent's workplace, someone else's home, restaurant, vehicle, outdoors, and so on) and for with whom (alone, with spouse or partner, with own children, with other family members, with coworkers, with friends). These companion codes make it possible to study not just what people do but the social and physical context of their activities, enabling research on social isolation, parenting behavior, and the geography of daily life.

The 17 Major Activity Categories

Every ATUS activity is ultimately classified into one of 17 major categories, each covering a domain of daily life. The published average hours data for the US population reveals the broad structure of American time allocation.

Personal care — the largest category by far, averaging roughly 9.5 hours per day across the population. The bulk is sleep: BLS estimates average sleep at approximately 8.84 hours including short daytime naps. The remainder covers grooming, health-related personal activities, and religious personal care.

Work and work-related activities — averaged across all days (including weekends and non-working days for part-time and unemployed individuals), the population average is approximately 3.3 hours per day. On days when employed persons actually work, the average rises to 7.6 hours, consistent with a standard workday. Work-related activities include commuting, work-related socializing, and other activities done as part of the job.

Household activities — roughly 1.9 hours per day on average. Within this category, food preparation and cleanup account for approximately 33 minutes per day, making it the single largest household activity. Interior cleaning, lawn and garden care, and household management account for most of the remainder.

Caring for household members — includes primary childcare (focused caregiving attention to children in the household), adult care, and eldercare. This category is highly skewed: parents of young children spend far more time in it than the population average.

Eating and drinking — 67 minutes per day on average, covering all eating and drinking occasions regardless of location. Americans eat roughly half their meals at home; food purchased away from home is common on workdays.

Socializing, relaxing, and leisure — the broadest leisure category, averaging 5.0 hours per day. Television watching is the dominant subcategory (see below). Reading, attending or hosting social events, and arts and crafts are smaller components.

Sports, exercise, and recreation — population average of approximately 0.3 hours (18 minutes) per day. This is the mean across all adults including those who do no exercise on a given day; on days when exercise occurs it averages closer to 70 minutes.

Education — approximately 0.4 hours per day averaged across the full population. Students on school days average 3.5 hours. The category includes class time, homework, and extracurricular activities.

Traveling — roughly 1.2 hours per day, covering all travel except travel coded as part of a specific activity (work travel is part of the work category). Commuting is captured here.

The remaining major categories — caring for non-household members, consumer purchases, professional and personal care services, household services, government services and civic obligations, religious and spiritual activities, volunteering, and telephone calls — each account for smaller fractions of the average day, typically under 30 minutes individually.

The Gender Gap in Time Use

The most replicated finding in the ATUS literature is the persistent gender gap in unpaid household work and caregiving. Women average approximately 4.5 hours per day on household activities and caregiving combined (categories 2, 3, and 4); men average approximately 2.5 hours. The 2-hour daily gap amounts to roughly 730 hours per year — more than 18 full 40-hour work weeks of unpaid labor annually that women perform over and above men.

The gap has a counterpart in market work and leisure. Men average approximately 0.5 hours more per day in paid work. Men also average roughly 0.4 hours more per day in leisure, driven primarily by TV watching and sports. The “leisure gap” has narrowed since ATUS launched in 2003 but has not closed; as of the most recent data years, women continue to have slightly less total leisure time than men.

The COVID-19 shock of 2020 provided a natural experiment in how the gender gap responds to external disruption. ATUS data for 2020 showed that women experienced substantially larger increases in childcare and homeschooling time following school closures, while men experienced increases in leisure time during the same period. The gender gap in unpaid caregiving widened during the peak lockdown period and then partially but incompletely reverted.

The ATUS gender gap in unpaid household work is the primary empirical input to research on the “motherhood penalty” and “fatherhood premium” in wages. Economists have used ATUS to document that the differential time demands of parenthood on mothers versus fathers — not raw productivity differences — account for a substantial share of the earnings divergence between men and women following childbirth. Fathers, freed from additional household time burdens, may increase their labor market commitment after children arrive; mothers typically cannot.

Parental Time Use and the Intensive Parenting Trend

ATUS distinguishes two types of childcare. Primary childcare occurs when caring for a child is the respondent's main activity — reading to a child, supervising homework, taking a child to the doctor. Secondary childcare occurs when a child is present in the room while the respondent does something else — cooking dinner while a toddler plays nearby. Both types are captured in the ATUS data, and both matter for understanding the actual burden of parenting.

Parents of children under 6 average more than 2 hours per day of primary childcare. For mothers of children under 6, that figure is substantially higher. Secondary childcare adds additional time that is often invisible in analyses that focus only on primary activities.

One of the most striking findings from linking ATUS to earlier time-use surveys is the increase in parental time with children from 1965 to the 2000s and 2010s. Despite the rise of dual-earner families — which might be expected to reduce parental time as more hours were committed to market work — parents in the 2010s spend significantly more time in direct childcare than parents did in the 1960s. This is the “intensive parenting” trend: a shift in norms toward more engaged, child-centered parenting that has increased the time demands on parents even as labor force participation among mothers increased. ATUS provides the continuous federal time-series that anchors this finding for post-2003 years.

Working from Home

ATUS includes a question about whether any work was done at home on the reference day. This seemingly simple question has become one of the most policy-relevant items in the survey following the COVID-19 pandemic's restructuring of where Americans work.

Before 2020, ATUS data showed that approximately 23% of employed Americans did some work from home on any given workday. This figure had been rising slowly throughout the 2010s as telecommuting became more common in professional and knowledge-work occupations, but it remained a minority experience concentrated in higher-education, higher-income workers.

In 2020, the share of employed Americans who worked from home on workdays jumped to 42% — an increase of nearly 20 percentage points in a single year, driven by office closures and public health restrictions. By 2021 and 2022, the figure settled at roughly 28–30%, well above the pre-pandemic baseline. That persistent structural shift — roughly 5–7 percentage points above the pre-pandemic trend — represents millions of workers whose daily time geography has permanently changed. ATUS is the primary federal data source for tracking this shift at the population level rather than through employer surveys or proprietary mobility data.

Leisure Time and Television

Americans average approximately 2.8 hours per day watching television, making it the single largest leisure activity — larger than socializing, reading, exercise, and all other leisure activities combined. TV watching has been the dominant American leisure activity throughout the ATUS period and shows only modest decline despite the growth of streaming and mobile screen time (which ATUS codes as “computer use for leisure” rather than TV watching).

Television watching is strongly negatively correlated with educational attainment and income. Adults without a high school diploma average substantially more TV time than college graduates. The pattern is not that college-educated Americans have less leisure — in fact, college graduates tend to have slightly more total leisure time than high school graduates once work hours are accounted for — but they allocate it differently. More-educated Americans direct leisure time toward reading, socializing with friends, exercise, and cultural activities; less-educated Americans allocate proportionally more to television.

This “leisure inequality” debate — whether it is the quantity or quality of leisure that varies by socioeconomic status — has been energized by ATUS data. Some economists argue that the leisure time of lower-income Americans is less satisfying or less health-promoting than the leisure time of higher-income Americans, even when total hours are comparable. The ATUS Well-Being Module (discussed below) provides direct evidence on this question through reported mood during activities.

Sleep and Health

The average ATUS respondent reports approximately 8.84 hours of personal care per day, the majority of which is sleep. But the ATUS sleep figure is not equivalent to nightly sleep duration. It includes daytime naps, time spent in bed but not asleep, and other personal care activities that respondents bundle into the morning and evening periods. Sleep researchers who use ATUS typically estimate actual overnight sleep at approximately 7.6 hours after adjusting for these components — consistent with other sleep survey data but lower than the 8+ hours the raw ATUS figure implies.

Sleep duration in ATUS varies systematically by demographic characteristics. Black Americans report shorter sleep on average than white Americans, a finding consistent with health disparity literature connecting sleep deprivation to elevated risk of cardiovascular disease, diabetes, and mental health conditions. Workers in shift occupations — nurses, truck drivers, manufacturing workers on night shifts — report fragmented or shortened sleep patterns. Parents of young children report less sleep than non-parents, with the effect largest for mothers in the first years after birth.

The connection between sleep, income, and health has been an active area of ATUS-based research. Lower-income Americans are more likely to work multiple jobs, longer hours, or irregular shifts — all of which compress sleep time. ATUS provides the time-diary evidence that connects labor market conditions to the most fundamental health behavior.

ATUS Special Modules

Several years of ATUS include supplemental modules that collect data beyond the standard time diary. These modules transform ATUS from a time-accounting exercise into a tool for welfare economics and policy evaluation.

The Well-Being Module, fielded in 2010, 2012, 2013, and 2021, asks respondents to rate a random selection of their activities on four experiential dimensions: how happy they felt during the activity, how meaningful it was, how stressed they were, and how much physical pain they were in. These ratings provide a direct measure of subjective well-being anchored to specific activities — enabling comparisons of, for example, how pleasant commuting feels relative to childcare, or whether employed people experience more positive affect during work or leisure. The Well-Being Module data is the primary US federal evidence base for the economics of happiness as applied to daily activities.

The Eating and Health Module, fielded in 2006–2008 and 2014–2016, collected detailed data on eating behavior: what foods were eaten, whether meals were eaten while doing other things (screen time while eating), and physical activity measures. The module links ATUS time-diary data to health outcomes data, supporting research on the relationship between time constraints, diet quality, and obesity.

The Leave Module, fielded in 2011 and 2017–2018, collected information on paid leave availability and use: whether workers had access to paid sick leave, vacation, and family leave, and whether they had taken leave in the reference period. The Leave Module data has been used to evaluate the welfare effects of paid-leave mandates and the demographic distribution of leave access.

Accessing ATUS Data

BLS publishes ATUS microdata through the ATUS program page at bls.gov/tus. The data are distributed as fixed-width or comma-delimited text files organized into several linked data files. The main files are:

All ATUS analysis requires weighted estimation. The person-level weight (TUFINWGT) in the Respondent file adjusts for the CPS survey design, ATUS subsampling within CPS households, and nonresponse. Unweighted ATUS statistics are not nationally representative and should not be published.

IPUMS-ATUS, hosted at ipums.org/time-use through the University of Minnesota, provides harmonized cross-year ATUS extracts with consistent variable coding across the full 2003–present period. IPUMS has reconciled changes in activity codes, demographic variable definitions, and survey design across ATUS waves, making it substantially easier to construct consistent longitudinal analyses than working from the raw BLS files. IPUMS-ATUS also provides activity codes crosswalked to the Harmonized European Time Use Study (HETUS) codes, enabling international comparisons.

Python: Weighted Gender Gap in Time Use from ATUS Microdata

The following script downloads the ATUS Activity and Respondent files for the most recent data year from BLS, merges them on the case identifier, computes weighted average minutes per day by sex for each of the 17 major activity categories, calculates the gender gap in unpaid household work and caregiving versus paid work, and displays the top 5 activities by weighted mean for men and women separately.

import requests
import pandas as pd
import io, zipfile

# Download ATUS microdata for the most recent available year from BLS
# We need two files: the Activity file and the Respondent file
# BLS ATUS data: https://www.bls.gov/tus/datafiles.htm
YEAR = "2022"

BASE_URL = "https://www.bls.gov/tus/datafiles/"

def download_atus_zip(filename):
    url = BASE_URL + filename
    resp = requests.get(url, timeout=180)
    resp.raise_for_status()
    with zipfile.ZipFile(io.BytesIO(resp.content)) as zf:
        # Find the .dat file inside the zip
        dat_files = [n for n in zf.namelist() if n.endswith(".dat")]
        if not dat_files:
            raise FileNotFoundError("No .dat file found in " + filename)
        with zf.open(dat_files[0]) as f:
            return pd.read_csv(f, sep=",", dtype=str, low_memory=False)

# --- Download activity file ---
# Contains one row per activity episode per respondent
act = download_atus_zip("atusact_" + YEAR + ".zip")

# --- Download respondent file ---
# Contains one row per respondent with demographic and weight fields
resp_df = download_atus_zip("atusresp_" + YEAR + ".zip")

# Standardise column names to lowercase
act.columns = [c.strip().lower() for c in act.columns]
resp_df.columns = [c.strip().lower() for c in resp_df.columns]

# Key columns:
#   TUCASEID  -- unique case identifier (links activity to respondent)
#   TUACTIVITY_N -- activity sequence number
#   TUTIER1CODE -- 2-digit major activity category (01-17)
#   TUACTDUR24 -- duration of activity in minutes
#   TUFINWGT   -- respondent person weight (in respondent file)
#   TESEX      -- sex (1=male, 2=female, in respondent file)

# Convert duration to numeric
act["tuactdur24"] = pd.to_numeric(act["tuactdur24"], errors="coerce")
act["tutier1code"] = pd.to_numeric(act["tutier1code"], errors="coerce")

# Sum minutes by respondent and major category (primary activities only)
# tutier1code 1-17 are the 17 major ATUS activity categories
act_primary = act[act["tutier1code"].between(1, 17)].copy()

act_sum = (
    act_primary
    .groupby(["tucaseid", "tutier1code"])["tuactdur24"]
    .sum()
    .reset_index()
    .rename(columns={"tuactdur24": "minutes"})
)

# Merge respondent weights and sex onto activity sums
resp_slim = resp_df[["tucaseid", "tufinwgt", "tesex"]].copy()
resp_slim["tufinwgt"] = pd.to_numeric(resp_slim["tufinwgt"], errors="coerce")
resp_slim["tesex"] = pd.to_numeric(resp_slim["tesex"], errors="coerce")

merged = act_sum.merge(resp_slim, on="tucaseid", how="inner")

# ATUS activity category labels (major tier 1 codes 1-17)
CATEGORY_LABELS = {
    1:  "Personal care (sleep, grooming)",
    2:  "Household activities",
    3:  "Caring for household members",
    4:  "Caring for nonhousehold members",
    5:  "Work and work-related activities",
    6:  "Education",
    7:  "Consumer purchases",
    8:  "Professional and personal care services",
    9:  "Household services",
    10: "Government services and civic obligations",
    11: "Eating and drinking",
    12: "Socializing, relaxing, and leisure",
    13: "Sports, exercise, and recreation",
    14: "Religious and spiritual activities",
    15: "Volunteering",
    16: "Telephone calls",
    17: "Traveling",
}

# Compute weighted average minutes per day by sex and major category
def weighted_mean(group):
    w = group["tufinwgt"]
    m = group["minutes"]
    if w.sum() == 0:
        return float("nan")
    return (m * w).sum() / w.sum()

results = (
    merged
    .groupby(["tesex", "tutier1code"])
    .apply(weighted_mean, include_groups=False)
    .reset_index()
    .rename(columns={0: "avg_minutes_per_day"})
)

results["sex"] = results["tesex"].map({1: "Men", 2: "Women"})
results["category"] = results["tutier1code"].map(CATEGORY_LABELS)
results["avg_hours"] = (results["avg_minutes_per_day"] / 60).round(2)

# --- Gender gap: household activities (cat 2) + caring (cats 3+4) vs. paid work (cat 5) ---
unpaid_cats = [2, 3, 4]
paid_cats = [5]

def gender_gap(sex_label, cats):
    sub = results[(results["sex"] == sex_label) & (results["tutier1code"].isin(cats))]
    return sub["avg_minutes_per_day"].sum()

print("=== ATUS " + YEAR + " Gender Gap: Unpaid Work vs. Paid Work ===")
print()
for sex in ["Men", "Women"]:
    unpaid = gender_gap(sex, unpaid_cats)
    paid   = gender_gap(sex, paid_cats)
    print(sex + ":")
    print("  Household activities + caregiving: " + str(round(unpaid / 60, 2)) + " hrs/day")
    print("  Paid work:                         " + str(round(paid / 60, 2)) + " hrs/day")
print()

# --- Top 5 activities by sex (weighted mean hours/day) ---
print("=== Top 5 Activities by Weighted Average Hours/Day ===")
for sex in ["Men", "Women"]:
    top5 = (
        results[results["sex"] == sex]
        .nlargest(5, "avg_minutes_per_day")
        [["category", "avg_hours"]]
        .reset_index(drop=True)
    )
    print()
    print(sex + ":")
    print(top5.to_string(index=False))

A few notes on the implementation. ATUS microdata files are distributed as ZIP archives containing comma-delimited text files. The activity file contains one row per episode, so summing duration within a respondent-category grouping converts episode-level data to respondent-level daily totals before weighting. The person weight TUFINWGT is stored in the Respondent file and must be merged onto the activity-level data. The weighted mean uses TUFINWGT as frequency weights, which is the correct approach for producing nationally representative population averages. Because ATUS files and variable names can change slightly between annual releases, analysts should consult the BLS data dictionary for the specific year they are using.

Connecting ATUS to Other Federal Datasets

Because ATUS respondents are sampled from CPS households, the dataset links naturally to the broader federal data ecosystem. CPS labor force status, occupation (SOC), and industry (NAICS) variables are available for each ATUS respondent, enabling comparisons of time use by occupation or industry that no other federal dataset supports. A software developer and a truck driver both in the labor force appear identically in QCEW wages data; in ATUS, their daily time patterns differ dramatically.

For demographic context — household income, poverty status, housing costs, and commuting patterns — the American Community Survey provides county-level and PUMA-level data that overlaps substantially with the demographic variables in ATUS. ACS cannot replicate ATUS time-diary data, but it complements it at the geographic level by providing large-sample estimates for demographic subgroups that ATUS cannot support.

For occupational wage data to pair with ATUS time allocations by occupation, the BLS OEWS program provides median hourly and annual wages by SOC code. Combining OEWS wages with ATUS time-use patterns enables implicit valuation of unpaid household production: the market-replacement-cost approach to valuing homemaking and caregiving uses OEWS wages for comparable occupations (cooks, childcare workers, housekeepers) to assign a dollar value to time spent in unpaid household activities.

For county-level employment and wage data by industry — the establishment-side complement to ATUS's worker-side time perspective — see BLS QCEW: The County-Level Employment and Wages Dataset Behind Every Local Economic Analysis.

For occupational wage benchmarks by SOC code that can be paired with ATUS time allocations by occupation, see BLS OEWS: The Occupational Employment and Wage Statistics Behind Every Salary Benchmark.

For demographic context on household composition, educational attainment, and income distribution that complements ATUS subgroup analysis, see Census ACS: The American Community Survey and the Federal Demographic Dataset Behind Every Policy Decision.