An interactive history of U.S. labor force participation

If you want to know how the labor market has changed over time, you usually look at the unemployment rate or maybe the employment-to-population ratio. But while those summary statistics are important, they don’t tell us about what people outside the labor force are doing. Are they in school? Acting as a primary caregiver? Disabled? Retired from the workforce?

The chances a worker is in any of those roles at a specific age during their life has changed quite a bit over the years. Inspired by Matt Bruenig of Demos, we looked at the trends in labor force status by age since 1975, using data from the Current Population Survey.

The interactive graph below shows the share of U.S. workers at different ages who are:

• Employed part-time or full-time
• Officially unemployed
• Disabled
• In-home caregivers
• Students in school
• Retired

Methodology

The data assembled span three versions of the Current Population Survey, with new surveys being instituted in 1989 and 1994. All three surveys feature a labor force participation item that is generated based on responses to a series of yes/no questions on the survey. This variable is called ESR, LFSR, and PEMLR, respectively, on the three versions of the survey. A second variable—called major activity, or MAJACT, on the first two surveys and PENLFACT on the post-1994 survey—was used to distinguish between certain categories of non-labor force respondents. Finally, a question on total hours worked was used to distinguish full-time workers from part-time workers.

The results are fairly consistent across surveys for certain age groups but there are important discrepancies. Most notably, the pre-1989 survey did not allow respondents to specifically identify themselves as retired. Instead, the “other” category included retirees. The wording and question order of the 1989-1993 survey appears to bias respondents in favor of choosing “carer” over “retired,” so another break in the retired series is evident in 1994. Minor changes in the survey may also have contributed to the uptick in respondents identifying as “disabled” in the most recent version of the survey.

This project’s github includes the Python code that was used to analyze the raw monthly CPS data, including our survey-weighting procedure and all coding decisions made.