In Spartanburg, South Carolina, a letter carrier could deliver the post and ask a household to complete a census interview on the same round. In Huntsville, Alabama, postal employees doing census interviews had to finish their postal shifts first. The distinction was the point of the 2026 US census test. The Census Bureau wanted to know whether people who already know local streets could help it reach households that had yet to answer, and what it would take to supervise that work.
The bureau completed the test on 31 August 2026 in selected parts of the two metropolitan areas. September reports from federal inspectors describe what happened on some of those doorsteps, including interviews that staff had to repeat after their equipment failed to synchronise. Their observations give the public an early look at how the trial worked. They leave open the bigger question of whether involving postal workers would save money or improve the next national count.
How a test across six places became two
The original announcement in July 2024 named Colorado Springs, parts of western North Carolina and western Texas, and tribal lands in Arizona alongside Huntsville and Spartanburg. Officials wanted to try ways of reaching people historically missed by the census. For example, the proposed Arizona test included the Fort Apache and San Carlos reservations. The North Carolina site included the Qualla Boundary, home to the Eastern Band of Cherokee Indians.
On 2 February 2026 the bureau narrowed the plan to two sites and put postal workers at the centre of it. Both remaining areas offered urban and rural addresses, and some homes that are difficult to identify. Neither offered a substitute for fieldwork on the two Arizona reservations. A judgement about how well a letter carrier can locate an address on a South Carolina route says little about how to design a count on tribal lands.
What households were asked to do
On 1 May the bureau invited households at roughly 81,000 housing units in the Huntsville area and 73,600 in the Spartanburg area to respond. Those numbers describe addresses invited. They tell us neither how many people lived at them nor how many interviews were completed. The initial questionnaire was available online in English. Participants could use a computer, phone or tablet. Telephone interviews and paper replies were unavailable as ways to respond on their own.
The bureau used questions from the American Community Survey, including questions about age, education and citizenship. It estimated that the online form would take about 40 minutes. The test therefore tells us what those households were invited to answer in 2026. Decisions about the content of the 2030 questionnaire are separate. From 1 June, workers began visiting households that had yet to respond. They asked a shorter set of questions in person. Households could still respond online while visits continued.
The July reminder to residents said an address could receive up to four reminders by post after the invitation. The letter contained a unique census identifier. Visits could take place in the evening or at weekends. The option to keep responding online mattered for a household where everyone was out when a worker called. A notice left at the door could give the household a way to complete the form before another visit.
Two jobs on a single route
In Huntsville the bureau hired postal workers directly for census duties outside their usual shifts. An address could receive visits from both a postal employee working for the census and another census taker. In Spartanburg participating carriers remained postal employees, took the census confidentiality oath and collected information during their rounds. There the household’s location determined which kind of worker would call. The bureau described the arrangement in its March pilot announcement.
For residents hearing personal questions from a familiar carrier, the distinction between delivery work and census work mattered. The bureau’s instructions on confidentiality applied to postal participants in both places. They had to pass background checks, complete training on handling answers and using government devices, and take a lifelong oath to protect respondents. The bureau said individual answers could be used only for statistical purposes.
This design gives the agencies two different staffing problems to examine. In Alabama, can someone available for a second job interview residents at times when they are home? In South Carolina, can a carrier fit interviews around mail delivery, and can a census supervisor see enough of the work to help? Simply comparing a total for each city would mix staffing arrangements with differences between neighbourhoods and the timing of visits.
What inspectors saw in the field
A Commerce Department inspector general report issued on 23 September describes observations of 15 bureau employees attempting 205 interviews in June. From observations and discussions, inspectors identified 55 instances of technological problems. Seventeen involved synchronising information collected from households. Some interviewers had to restart the application and, in some cases, repeat interviews. At other times supervisors assigned cases manually when the software failed to give work to an available employee.
The inspectors’ separate report on interviewing procedures recorded 12 instances in which staff failed to leave a visit notice containing the case identifier. Inspectors also saw employees use personal phones to find assigned addresses. In two instances, locations displayed by the official mapping application failed to match the correct address or an actual building. These details help explain how a seemingly simple count of repeat visits could include return trips caused by faulty equipment or missing information.
In an alert issued on 24 September, inspectors described a further difficulty in Spartanburg. Census supervisors interviewed in June had limited contact with the 20 carriers conducting interviews on their rounds. Supervisors reported that they could not see case notes or how long carriers spent on census work. Inspectors gained access to observe carriers directly in August. Their September alert concerned the June visit and left the assessment of those later observations to a fuller evaluation.
What the evidence can settle before 2030
The Government Accountability Office identified another gap. Following a Commerce Department review, the bureau reduced or dropped ten of the 19 activities originally planned. Changes affected experiments with administrative records and an online response option for people in university accommodation. In its report issued on 4 June, the watchdog recommended that the bureau test the omitted features before finalising the census design. The department agreed with that recommendation.
Anyone looking for a new population figure for either city will need another source. The bureau says it will produce no statistical data products from the exercise. An evaluation could still report how many attempts it took to finish an interview or what teams spent travelling between addresses. Those are measures of how the count works.
The next major field test on the bureau’s timetable is a 2028 dress rehearsal. The eventual national count determines how seats in the House of Representatives are divided among states and supplies data for drawing districts and allocating public funds. A useful cost comparison from this summer would need the hours spent delivering, interviewing, supervising and correcting cases, as well as information about which households each approach reached. The number of doorbells rung answers only a small part of that question.