# Census Undercount The **census undercount** is the gap between a census count and the true population, measured after the fact by the Census Bureau's post-enumeration survey (PES) — and the vault tracks it because the undercount is not random. It concentrates in the same populations each decade, systematically undercounting the communities that lean Democratic and overcounting the ones that lean Republican, which means the census error is a structural advantage for the right that requires no one to alter a single published number. The undercount shifts House seats, Electoral College votes, and hundreds of billions in federal funding in a predictable direction every ten years — and because there is no correction downstream, the error compounds for a full decade until the next census. ## The differential undercount Every census misses some people and double-counts others. The politically consequential fact is that the errors are directional. The 2020 Post-Enumeration Survey — released in May 2022 — found that the census: **Undercounted**: Black residents by 3.30 percent, Hispanic/Latino residents by 4.99 percent, American Indian and Alaska Native residents on reservations by 5.64 percent, and people who reported "Some Other Race" by 4.34 percent. **Overcounted**: non-Hispanic White residents by 1.64 percent and Asian residents by 2.62 percent. These errors are not new. The PES for the 2010 census found similar patterns — a 2.06 percent net undercount for Black residents and a 1.54 percent net undercount for Hispanic residents — though the 2020 errors were substantially larger. The consistency is the point: the census has undercounted Black and Latino populations in every modern enumeration, and overcounted White populations in recent ones. The differential undercount is not a one-time error but a structural feature of how the census operates. ## Why it matters: [[Apportionment]] and funding The census determines two things that shape political power for a decade: the [[Apportionment]] of House seats among the states, and the distribution of over $2 trillion per year in federal funding through programs that use census-derived population data (Medicaid, Title I education grants, transportation funding, and hundreds of other formula-based programs). Because undercounted populations — Black, Latino, and Indigenous residents, young children, renters, people in irregular housing — concentrate geographically in urban areas and certain states, the undercount transfers political representation and federal resources from those communities to the predominantly White, suburban, and rural communities that are overcounted. In 2020, this was not abstract: analysis by the Urban Institute estimated that the differential undercount likely cost states with large minority populations at least one House seat and shifted billions in annual federal funding. The constitutional framework makes this uncorrectable. The Supreme Court held in [[Utah v. Evans]] (2002) that the Census Bureau may use statistical imputation for within-state redistricting purposes, but Congress has since 1999 prohibited the use of statistical sampling or adjustment for the apportionment count itself. The PES measures the error; it does not change the seats. [[How the Census Apportions Congress]] — the vault's explainer on the equal-proportions method — documents the mathematical machinery that converts these flawed counts into seat allocations. ## The citizenship question as an undercount amplifier The most direct attempt to weaponize the undercount was the Trump administration's effort to add a citizenship question to the 2020 census. The theory was explicit in the files that emerged during litigation: a citizenship question would depress response rates among immigrant and Latino households — households already prone to undercount — which would shift [[Apportionment]] toward older, Whiter, more Republican-leaning communities. The Supreme Court blocked the question in [[Department of Commerce v. New York]] (2019), holding that Commerce Secretary Wilbur Ross's stated rationale (Voting Rights Act enforcement) was "pretextual." But the fear effect was real regardless: the Census Bureau's own research showed that the mere public controversy over the question suppressed self-response rates among Latino and immigrant communities in the 2020 census, even though the question was never actually asked. The undercount in those communities in 2020 was worse than in 2010 — and the citizenship-question controversy is one plausible explanation. [[Trump v. New York]] (2020) addressed a related but distinct question: whether the president could direct the Census Bureau to exclude undocumented immigrants from the apportionment base entirely. The Supreme Court dismissed the challenge as premature, and the Biden administration rescinded the order before it could take effect. But the attempt revealed the strategic logic: if you cannot count fewer people, try to exclude the people you did count. ## The connection to [[Gerrymandering]] and [[Shelby County v. Holder]] The census undercount interacts with two other structural mechanisms the vault documents. [[Gerrymandering]] uses the census data as the baseline for drawing districts — if the data itself is skewed by the differential undercount, the districts drawn from it inherit that skew. And [[Shelby County v. Holder]] (2013) eliminated the preclearance regime that had provided the most effective check on minority vote dilution — which means that the communities most affected by the undercount also have the least protection against being gerrymandered into irrelevance. The [[One Person One Vote]] principle requires districts of roughly equal population, but the equal-population requirement is only as accurate as the census data that defines "population." A differential undercount that systematically undercounts minority communities and overcounts White communities means that the populations being equalized are wrong — and wrong in a direction that consistently advantages the same political coalition. ## The structural advantage The vault tracks the census undercount as a structural advantage rather than a conspiracy because no one needs to do anything wrong for it to operate. The census misses people who are hard to count — people who distrust government, people who move frequently, people in nontraditional housing, people who do not speak English — and those populations correlate with race, ethnicity, and partisanship. Outreach funding, field-office locations, the length of the response window, digital accessibility, and question design all influence the undercount rate, and decisions about those inputs are made by the political appointees who run the Commerce Department. The [[Fourteenth Amendment]]'s requirement of an "actual Enumeration" creates the obligation, but the operational decisions that determine the accuracy of that enumeration are matters of budget and policy that change with each administration. The result is a system in which the census error functions as a ratchet: the communities that are undercounted receive less representation and less funding, which makes them harder to count in the next census, which produces a larger undercount, which reduces their representation further. Breaking the cycle would require either statistical correction (prohibited for apportionment) or a sustained investment in enumeration infrastructure that targets the hardest-to-count populations — an investment that is politically contingent on the very representation the undercount distorts. ## See also [[How the Census Apportions Congress]] · [[Apportionment]] · [[Utah v. Evans]] · [[Department of Commerce v. New York]] · [[Trump v. New York]] · [[One Person One Vote]] · [[Gerrymandering]] · [[Shelby County v. Holder]] · [[Fourteenth Amendment]] · [[Reconstruction Amendments]] · [[Donald Trump]] · [[Stephen Miller]]