How the Census Could Be Improved Before 2030

The census is the most consequential statistic in American life. It decides how many seats each state holds in Congress, how district lines get drawn, and how trillions of federal dollars flow to states and towns. It is also imperfect, and the imperfections are well documented, mostly by the Census Bureau itself. With planning for the 2030 count already underway, this is the window in which the next census gets better or does not. Here is what the record says goes wrong, and the main ideas on the table for fixing it.

What the census gets wrong

The 2020 census counted the national total almost exactly right. The problem is who was miscounted underneath that total. The Bureau's own Post-Enumeration Survey found that some groups were undercounted while others were counted twice, and the errors ran in opposite directions by race and by age.

The youngest Americans fare worst of all over time: the under-5 undercount of 5.46 percent, about 1 million children, was the highest recorded since 1950, and it got worse between 2010 and 2020. Because funding formulas run on these counts, the communities that get missed are also the ones that lose money for the following decade.

Fix one: use the records the government already has

The most promising idea, and the centerpiece of the Bureau's own 2030 research agenda, is administrative records: tax filings, Social Security records, Medicare and Medicaid enrollment, and other data the government already holds. According to the Government Accountability Office's review of 2030 preparations, the Bureau plans to expand the use of these records to count people who never respond, to add people a household response missed, and to remove people counted twice. Done well, that directly targets the known failure points, young children left off forms and renters who fall through the cracks, without a single extra knock on a door. It is also cheaper: the 2020 count cost roughly $14 billion, much of it spent sending workers to follow up with nonresponders.

Fix two: make the technology boring and reliable

2020 was the first census most people could answer online, and self-response worked, about two-thirds of households responded on their own. The Bureau is now consolidating its patchwork of collection systems into a single platform called DICE, budgeted at about $1.08 billion through 2033. The catch, per the GAO's audit of the program, is that the modernization schedule is not reliable, and the watchdog has told the Bureau to fix its cost and schedule estimates. A census is a one-shot operation. The technology has one job, to work on the first try in April 2030, and the years to guarantee that are these ones.

Fix three: count children like it matters

The young-child undercount has its own causes and needs its own fixes: form design that prompts for babies and shared-custody kids, outreach through schools, clinics, and child care programs, and using birth records to check the count as it happens. Advocacy groups like First Focus are pressing for exactly this kind of targeted investment in 2030. The payoff is concrete. Census-guided programs from Medicaid and CHIP to WIC and Head Start serve young children, and a state that misses its kids underfunds them for 10 years.

Fix four: keep the count stable and out of politics

Accuracy also depends on things that are not technical. Research improvements get adopted, delayed, or dropped as administrations change, as happened this year when the Bureau reverted to 1997 race and ethnicity standards for the 2027 American Community Survey, walking away from a combined question that its own testing showed measured Hispanic and Middle Eastern populations more accurately. We covered that reversal in detail here. Whatever one thinks of any single decision, whipsawing the methodology erodes the thing a census runs on, which is public trust and respondent participation. Stable funding through the decade, not just a surge in year nine, is the other unglamorous ingredient every review, including the GAO's, keeps returning to.

The stakes

None of these fixes are exotic. Use data the government already holds, ship reliable technology, design the form for the people it misses, and keep the operation steady. The 2030 census will decide apportionment, redistricting, and funding until 2040, and its accuracy is being determined now, in research programs and budget lines almost nobody watches.

The census is also the foundation of everything on this site. For the basics, see what the census is for and why it runs every 10 years, or explore what the current data shows for any place with the compare tool.

Sources

The 2020 accuracy figures are from the U.S. Census Bureau's Post-Enumeration Survey estimates of undercount and overcount and its young-children undercount analysis. The 2030 improvement plans and critiques are from the Government Accountability Office, including "2030 Census: Preparations Are Underway with Changes to How the Count Takes Place", the 2020 coverage-errors report, and the IT modernization schedule audit. The young-child funding stakes draw on First Focus on Children.