Jeff Reid NYC July 2026
Data / data

Work in America from the Centennial to 2026.

Tracing the work Americans do over the last 150 years through census microdata and fifty years of monthly labor statistics.

At the nation’s centennial in 1876, America was in the third year of a depression triggered by railroad overbuilding, one year from the first national labor uprising. At the bicentennial in 1976, the nation confronted stagflation, in between the oil shocks of 1973 and 1979, watching factory employment begin its long slide. As we celebrate America at 250, the anxiety is about the arrival of machines that write and reason, threatening white collar jobs. Looking at data about the work we do shows the tails on shifting labor markets is long.

Employment since the Civil War

The long view opens in 1850, a decade before the Civil War and a generation before the centennial. It shows the steep and steady decline of agricultural work, which peaked as a share of the workforce early in the 1800s — already down to 58% of workers when the chart begins — and in total employment around 1910, at some 12.5 million farm workers. Manufacturing peaked as a share of employment around 1960, at 27.5%, with total factory employment declining after 1980. White collar professional services work — health care, education, law, engineering — has grown steadily since the 1950s, reaching 28.8% of the total workforce by 2024.

What American workers do · by industry · 1850–2024
0%25%50%75%100%1870 · Emancipation1890 · census lost to fire1940 · “gainful worker” → “employed”18501875190019251950197520002024
Hover the chart for details · click a legend chip to pin a band

Filter the chart by sex, race, or region and the single national story splits into many: emancipation pulling four million people into the counted workforce in 1870, women’s work shifting from domestic service to offices, the South staying agricultural decades longer than the Northeast.

The chart is built from person-level records from every surviving decennial census since 1850, plus the American Community Survey for recent years, accessed through the University of Minnesota’s IPUMS project. Each census’s industry codes are harmonized to a single 1950-era classification so a blacksmith in 1860 and an auto worker in 1960 can sit in the same band. The census of 1890 burned in a 1921 fire, and before 1870 the count excluded enslaved workers.

The last fifty years, up close

Beginning in 1976, we have monthly labor microdata to work with, showing finer-grained trends through multiple recessions and the shock of Covid in 2020.

The 50 years since 1976 show a country moving starkly away from making things towards information and care work.

Winners & losers · change in share of employment · 1976 → today
-10-50+5Health Care+4.7Legal/Engineering/Misc Professional+3Information & Communications+2.5Business & Professional Services+2.5Social Services+2Construction+1.5Finance/Insurance/Real Estate+1.4Transportation+1.2Entertainment & Recreation+1.1Education+0.3Public Administration+0.2Utilities-0.1Retail Trade-0.4Mining-0.5Agriculture-1.4Personal Services-2.2Wholesale Trade-2.4Manufacturing-13.4
Percentage points of civilian employment · 1976 average vs. trailing 12 months · Making thingsMoving & sellingInformation & professionalCare & educationOther services

The biggest winner of the last half-century isn’t tech. It’s care: health care alone gained more share than any other industry, and with education and social services it forms the largest employment cluster in the country. Manufacturing’s loss dwarfs everything — nearly fourteen points, more than every other declining industry combined.

Five kinds of work · monthly · 1976 → today
0%25%50%75%100%19801990200020102020
Making thingsMoving & sellingInformation & professionalCare & educationOther services

Recessions hit different kinds of work differently. The Great Recession was a goods recession — construction and manufacturing shed millions while health care kept hiring straight through it. Covid inverted the pattern, cratering in-person services. And since late 2022, the industry AI was supposed to hollow out first has quietly kept growing:

Industry shares through recessions & shocks · 1976 → today
0%5%10%15%20%25%NAFTAdot-com peakChina joins WTOLehmanCovidChatGPT19801990200020102020
Pick up to 5 industries above · 12-month rolling averages · recessions shaded

Every industry, at a glance:

Every industry, 1976 → today · share of employment · own scale
Agriculture
peak 4.1%
4.1% → 2.7%
Mining
peak 1.1%
0.9% → 0.3%
Construction
peak 8.2%
5.9% → 7.4%
Manufacturing
peak 22.7%
22.7% → 9.2%
Transportation
peak 5.7%
4.4% → 5.6%
Utilities
peak 1.4%
1.4% → 1.3%
Wholesale Trade
peak 4.3%
4.1% → 1.7%
Retail Trade
peak 17.8%
16.5% → 16.1%
Information & Communications
peak 4.0%
1.4% → 4.0%
Finance/Insurance/Real Estate
peak 7.1%
5.4% → 6.8%
Business & Professional Services
peak 5.7%
2.8% → 5.3%
Legal/Engineering/Misc Professional
peak 6.5%
3.5% → 6.5%
Health Care
peak 11.6%
6.9% → 11.6%
Education
peak 9.7%
8.5% → 8.8%
Social Services
peak 3.0%
0.8% → 2.8%
Personal Services
peak 4.9%
4.9% → 2.8%
Entertainment & Recreation
peak 2.3%
1.1% → 2.1%
Public Administration
peak 5.1%
4.6% → 4.8%
12-month rolling averages · gray bars = recessions · each chart scaled to its own peak

Notes on the data

Employment by industry, 1850–2024, from IPUMS USA census microdata (1% samples; ACS from 2005), harmonized to the 1950 Census industry classification. Workers are gainful workers with a reported industry before 1940 and employed persons thereafter. Industry before 1910 is inferred from occupation. The 1890 census was destroyed in a 1921 fire. Before 1870, enumeration excluded enslaved workers; 1850 counted free males only. Armed forces are excluded throughout — the census did not consistently record industry for military personnel after 1940. Workers with industry not reported or not classifiable (~6% in 1900 and 1960) are excluded from shares.

Monthly data, 1976–present: IPUMS-CPS basic monthly samples (employed civilians, survey-weighted), industries harmonized on the 1990 census classification and shown as trailing 12-month averages — the CPS microdata is not seasonally adjusted. October 2025 is missing (the survey disrupted by the federal shutdown). Recession shading follows NBER dating.

Sources: Steven Ruggles et al., IPUMS USA: Version 16.0 (Minneapolis: IPUMS, 2025), https://doi.org/10.18128/D010.V16.0; Sarah Flood et al., IPUMS CPS (Minneapolis: IPUMS, 2025), https://doi.org/10.18128/D030.V12.0. Monthly context series: U.S. Bureau of Labor Statistics via FRED.

datahistorylabor