What it measures
Underemployment is the share of recent graduates working in a job that does not typically require a bachelor's degree. The "recent graduate" population is people aged 22 to 27 who hold a bachelor's degree or higher and are not currently enrolled in school. A field's underemployment rate is the weighted share of its employed graduates in non-degree jobs. All figures are national and population-weighted.
Data sources
- Outcomes: U.S. Census Bureau American Community Survey 1-year public-use microdata (PUMS), 2018 through 2024, retrieved through the Census Bureau's API. Variables used: age (
AGEP), educational attainment (SCHL), school enrollment (SCH), field of degree (FOD1P), occupation (SOCP), employment status (ESR), and the person weight (PWGTP). - Benchmark: the New York Fed, The Labor Market for Recent College Graduates, used to calibrate and validate the level of our estimates.
Defining a "college job"
We treat an occupation as a college-level job if at least 60% of its prime-age workers (ages 30 to 54, employed) hold a bachelor's degree or higher; otherwise it is a non-degree job, and a graduate working in it is counted as underemployed. This follows the New York Fed's plain-language framing of underemployment as working in a job where most workers do not hold a degree.
The 60% cutoff is not arbitrary. It is calibrated so that our overall 2023 rate reproduces the New York Fed's recent-graduate underemployment figure of roughly 41%. Once set, the cutoff is held fixed across all years.
Field of study
Field of study is the first reported field of bachelor's degree (FOD1P), mapped from the Census Bureau's 174 detailed field codes to the 73-category taxonomy used by our companion snapshot tool, so the two line up. The crosswalk assigns each detailed field to exactly one category.
Time frame and the 2020 gap
The series starts in 2018 because the Census Bureau changed its occupation-coding system that year (from the 2010 to the 2018 Standard Occupational Classification). Splicing the two systems together introduced a discontinuity that masqueraded as a trend, so we restrict the series to the consistent 2018-onward period. There is no 2020 data point because the Census Bureau did not release a standard 2020 ACS 1-year file, owing to pandemic-related data-collection problems.
Validation against the New York Fed
Beyond the overall calibration, the field-level estimates track the New York Fed's published by-major snapshot closely, with the same ordering across fields:
| Field | Our 2023 estimate | NY Fed snapshot |
|---|---|---|
| Criminal justice | 69.2% | 65.8% |
| Sociology | 57.1% | 52.0% |
| Communications | 49.2% | 53.0% |
| Psychology | 45.5% | 48.3% |
| Computer science | 23.0% | 19.1% |
| Nursing | 10.4% | 12.8% |
| All fields | 40.9% | ~41% |
Differences of a few points are expected because the New York Fed uses an occupation classification derived from O*NET education requirements, while ours uses the workforce-degree-share rule above. The levels and the ranking agree.
Testing for real trends
To separate signal from sampling noise, each field's 2018-2024 path is fit with a weighted least-squares trend, weighting each year by the inverse of its sampling variance. Standard errors treat underemployment as a proportion from the field's unweighted sample size, inflated by a 1.5 design-effect factor to account for the ACS's complex sample. A field's trend is reported as rising or falling only when it is significant at the 95% level. By this test, 14 of 73 fields show a statistically real trend, and nearly all of them point downward. Most single-year jumps in individual fields are within the margin of error and are not treated as signal.
Limitations
- Cross-sectional, not longitudinal. The ACS is a snapshot each year, so we measure how a field looks over time, not how individual graduates move. Claims about persistence at the person level (for example, that early underemployment tends to stick) come from other work, such as the New York Fed and Talent Disrupted, not from this series.
- Approximation. The workforce-degree-share rule approximates the New York Fed's O*NET-based definition; absolute levels can differ by a few points.
- Small samples. Fields with few graduates in a given year carry more year-to-year noise; those fields are flagged in the tool, and the significance test is the safeguard against over-reading them.
- First major only. Field of study uses the first reported field of degree; double majors and graduate fields are not separately modeled.
- Approximate standard errors. The 1.5 design-effect factor is a screen, not exact replicate-weight variance estimation.
Reproducibility
For each year, the build pulls two samples: recent graduates (for the field and occupation of each graduate) and prime-age workers (to classify occupations). It applies the fixed college-job classification, weights by the person weight, and aggregates to fields. The resulting figures are downloadable as a single file: underemployment-by-field.csv.