Where Are Federal Workers Employed? It Depends on the Dataset
Over the past several weeks, I have been investigating how changes in the US federal government will affect states and localities using local-level data from the Office of Personnel Management (OPM). To further this work, many have pointed me to a December 2024 report from the Congressional Research Service (CRS) which uses the Census Bureau’s American Community Survey (ACS) data as another source of local federal worker counts (the Metropolitan Council of Governments has a nice dashboard that uses all ACS data).
I have found, however, that the two data sources have stark differences that can affect analysis trying to determine the implications of federal workforce changes on areas around the country.
Overall, US federal government employes 2.3 million civilian workers, 2.1 million uniformed personnel, and 700,000 “government enterprise” employees, based on OPM data and tabulated by the Congressional Budget Office. Of those 2.1 million uniformed personnel, about 1.3 million are on active duty with about 166,000 stationed abroad (including US territories) and the remaining 767,000 are reservists, most of whom are in the Army. In total, these 5.1 million federal employees were paid (in salary and benefits) about $271 billion in 2022.
The OPM data that I’ve been using is based on administrative records of the federal workforce. These counts don’t include location data for certain federal employees, such as uniformed military personnel, “government enterprise” workers like US Postal Service employees, and certain security positions like people working at the Central Intelligence Agency or National Security Agency. The data that do include location information are recorded at the employee’s place of work (“duty station”), not place of residence. So, if you work at OPM in Washington, DC but commute from your home in Northern Virginia, the location record in the OPM data would be Washington, DC.
The CRS reports the number of federal civilian employees in each congressional district using the ACS, a monthly survey of about 300,000 households. Here, people self-report their occupation and their location—that same OPM employee would show up in Northern Virginia in the ACS rather than in DC. Further, the ACS doesn’t have the same security or uniformed military personnel restrictions as the OPM data, and as a result, more federal workers show up in the ACS. However, because the ACS is a household survey, it is also subject to reporting error; for example, if contractors or freelancers mistakenly report they work for the federal government rather than a nonprofit or state or local government. And because it is a household survey, the CRS report also includes margins of error for each Congressional district, which generates a low and high estimate of federal workers in each area.
These differences result in large discrepancies between the two datasets. The state-level OPM estimates (I do not include Puerto Rico and other US territories in this analysis) show a total federal civilian workforce of 1,946,491 people (in March 2024). By comparison, the district-level ACS data show a total federal civilian workforce of 4,385,408 people, with a range of 3.6 million to 5.2 million based on the margins of error. Comparatively, the ACS data include location information for 2.3 times as many federal workers as the OPM data.
If we aggregate the ACS data to the state level, we see that location information exists for more federal workers in every state using the ACS data as opposed to the OPM data. The smallest state-level ratio difference between the two datasets is Maine, where the OPM data show there are about 12,200 federal workers and the ACS data show 17,400 workers, a ratio of 1-to-4. At the other end of the spectrum, the number of federal workers in Connecticut is nearly seven times larger in the ACS data than in the OPM data. Given the proximity of Connecticut and New York City, this difference may be driven by commuting patterns and the ways the two datasets record location information. Even so, there are about 3.5 times as many workers in the ACS data than in the OPM data (269,110 and 75,795) in both New York state (187,592 and 53,394) and New Jersey (81,518 and 22,401).
For most states, the ratio of ACS to OPM is somewhere between 2 and 3. For eight states, the ratio is below 2, and for seven states, the ratio is between 3 and 4. Only Connecticut has a ratio above 4 (view the interactive map).
Unsurprisingly, Washington, DC bucks the state-level trend given the differences in sample definitions—especially work location versus residence location. DC is the only area where more people are counted as federal workers in the OPM data than in the ACS data. According to the OPM data, 162,000 federal civilian workers work in Washington, DC, but less than half as many live in DC (71,000) according to the ACS data. The margin of error is fairly narrow, suggesting a total range of about 10,000 workers.
Given the different sampling definitions, the OPM and ACS data are like apples and oranges. Both datasets can paint a picture of geographic dispersion of federal workers, but they are not like-for-like substitutes.
Each data source has its own advantages and disadvantages, but care should be used when combining or comparing them to get a sense of where the federal workforce lives and how changes at the federal level might affect local areas. If exploring the recent impacts of federal workforce reductions—which have primarily affected non-uniformed military personnel—the OPM data may be a better source. If looking at aggregate impacts of changes of the federal workforce on housing prices or commuting patterns, the ACS may be the better bet. In either case, changes to what and how federal government agencies publish and collect their data have and are changing dramatically in the face of the President’s recent executive orders, which will have important impacts on how researchers explore the impacts of the federal workforce on the economy.
Podcast: Data for a Continent: Inside the European Correspondent’s Visual Journalism
In this week’s episode of the podcast, I chat with Mandy Spaltman and Sebastian Graeff, co-leads of the data team at The European Correspondent, a volunteer-driven journalism initiative covering news across Europe. They discuss how their team of passionate designers and analysts craft daily data visualizations, their creative process, the tools they use, and the challenge of maintaining a cohesive style while allowing artistic freedom. Hear how they balance storytelling and design, and what the future might hold for interactive and multilingual content.
Disruptive Data Leadership Summit
Excited to announce that I’ll be speaking at the Disruptive Data Leadership Summit. This summit is all about building people-first data teams and leading with impact—and I can’t wait to share insights alongside 20+ other amazing speakers.
🎟 Join me for free! >>
Things I’m Reading
Books, Blog Posts, and Papers
The First Signs: Unlocking the Mysteries of the World's Oldest Symbols by Genevieve von Petzinger
There’s a little repetition here in the middle of the book, but the beginning and end—where the author focuses on her actual work—are really interesting.
Testing Perceptual Accuracy in a U.S. General Population Survey Using Stacked Bar Charts by Rice et al.
A dataset for plain language adaptation of biomedical abstracts by Attal et al.
Dots versus density: the impact of crime mapping techniques on perception of safety, police performance and neighbourhood quality by Wuschke et al.
The Best Alternative to Stacked Bar Charts I've Ever Seen! (with Sebastine Amede & Darragh Murray) by Kevin Flerlage
TV, Movies, and Music
The Residence, Netflix
YES PLEASE! Honestly, I didn’t love the first episode, but stuck with it and it was terrific.
Hacks, Max
Good ol’ fun.
Young Sheldon
I have about one season left of Suits, so I need another show I can “watch” when I workout. This might be it. But maybe not.
