A new report from the U.S. Government Accountability Office (GAO) warns that the Census Bureau, having narrowed the scope of its 2026 Census Test, may finalize important parts of the 2030 count’s design before it has evidence that those methods actually work. Released June 4, 2026, the report describes that the Bureau has reduced planned test sites from six to two and 19 planned operational activities to nine, leaving fewer opportunities to evaluate the new approaches the agency hopes to use at the end of the decade. These preparations matter far beyond the test itself, because the once-a-decade count determines how congressional seats are apportioned, how district lines are drawn, and how hundreds of billions of dollars in annual federal funding are distributed to states and communities.
The four sites the Bureau dropped from the 2026 test were Colorado Springs, Colorado; Tribal Lands within Arizona; Western North Carolina; and Western Texas. According to the report, these four sites were cut because they were not part of the Bureau’s original U.S. Postal Service pilot, leaving Huntsville, Alabama, and Spartanburg, South Carolina, as the two remaining sites. The four canceled locations included a tribal-lands site and several rural areas, which tend to be among the harder-to-count places, so their removal is likely part of why GAO flagged a loss of useful test data.
The 2026 Field Test
The 2026 Test is the first of two major field tests on the road to the 2030 count; the second is a dress rehearsal in 2028, and its purpose is to determine whether the new methods, processes, and technologies the Bureau identified in early research perform well enough to carry into the actual census. In practice, the test pairs online self-response with in-person follow-up, and the scaled-down version is built around a pilot that uses U.S. Postal Service staff to conduct that in-person counting at the two remaining sites.
Data collection for the revised two-site test runs across the spring and summer of 2026: residents at the Huntsville and Spartanburg sites are asked to respond online from May 1 through August 31, 2026, with census workers following up in person to count households that do not respond from June 1 through August 31, 2026. The Bureau has said it intends to complete the test by September 30, 2026, the close of the federal fiscal year.
The Bureau told GAO that it refocused the smaller test on two priorities: piloting the use of U.S. Postal Service workers for in-person counting and refining field infrastructure, staffing, and training at the two remaining sites in Alabama, and South Carolina. Scaling back, however, also pushed the test’s response and field-counting timelines later and cut overall hiring goals by more than 90 percent. GAO’s central concern is that, with fewer methods put to the test, the Bureau could lock in its 2030 design without knowing whether the changes improve or weaken the count, an outcome the watchdog links to potential higher costs, lower data quality, and diminished public confidence.
What the Test No Longer Covers
Of the 19 operational activities the Bureau originally planned to evaluate in the 2026 Test, 10 were fully or partially set aside. Several of the dropped items go to methods the Bureau hopes will modernize and economize the 2030 count, among them offering an online self-response option to residents of college dormitories, gauging how accurately new administrative-data modeling can count certain nonresponding households without a doorstep visit, and using the same field staff to enumerate both group quarters and ordinary housing units. Leaving these procedures untested matters because the Bureau may still rely on them in 2030 without first knowing how well they work; GAO points to a concrete precedent, noting that the agency ran into trouble submitting group-quarters data electronically during the 2020 Census, a feature for which it had reduced testing during the 2018 Test. The full set of the 10 reduced or removed activities is listed below.
Removed Entirely
In-Office Enumeration Model Input Data
In-Office Enumeration Model Accuracy
In-Office Enumeration Model Results
New Methods for Processing Addresses Without an Identifier
Group Quarters (GQ) Internet Self-Response Option
Combined Field Operations
Multi-Operational Field Staff for Group Quarters and Housing Unit Cases
Reduced (scaled-back)
Near Real-Time Processing
Ability to Integrate Multiple Operation Processing Segments in Near Real Time
Functionality of Near Real-Time Data Processing
Two further findings speak to how accurate the eventual census count is likely to be. The questionnaire used in the test, adapted from the American Community Survey, takes about 40 minutes to complete compared with roughly 10 minutes for the 2020 form, and GAO points to prior research indicating that longer questionnaires tend to lower response rates. At the same time, total Bureau staffing fell about 16 percent between January and October 2025, and the agency has identified mission-critical skills shortages, including in data science and cybersecurity, even as a workforce assessment meant to size up its 2030 needs sat paused and incomplete as of February 2026. A workforce that is both smaller and short on key skills, GAO cautions, could leave the Bureau without the people or expertise to run an effective enumeration.
For local governments in particular, the practical weight of these findings lies in the cost of an inaccurate count. The federal funding that supports local services, the population figures officials use for planning, and the representation a community receives all flow from census data, so quality problems at the national level land squarely on states and localities. Local jurisdictions also work directly with the census, for example, by reviewing address and boundary records and encouraging residents to respond. The GAO report does not examine that local role, but it arguably becomes more important as reduced federal testing and a smaller Bureau workforce increase the chance of errors that the national count might otherwise catch.
GAO Recommendations
GAO made two recommendations: that the Bureau research and test the activities and design features it removed from the 2026 Test before finalizing the 2030 design, and that it determine its workforce needs in time to close skills gaps, both of which the Department of Commerce accepted. Reporting on the findings, the technology outlet FedScoop noted that a companion GAO report issued the same day concluded that the Bureau’s enterprise IT modernization schedule was unreliable and that Commerce concurred with those recommendations as well. The Bureau has said that discussions about additional small-scale testing ahead of 2030 are ongoing.
Find us on:
