# Census Formula Funding Flow %% census-comment-callout:start — time-limited notice, remove after 2026-11-02 %% > [!important] Comment on the proposed census rule by Monday, November 2, 2026 > The Census Bureau has proposed rewriting who gets counted in the census, and where. Anyone can comment before the period closes. > > **What the proposed rule would change** > - **Who counts toward House seats:** Only U.S. citizens and lawful permanent residents would count toward apportionment. As written, the rule would leave out international students, work-visa holders, refugees and asylees. > - **Where you count:** Your "usual residence" would be wherever you lawfully spent the most days between January 3 and April 1, as shown by your tax records. That test can put snowbirds, college students, people who moved and service members in the wrong place. > - **What the form can ask:** The census form could no longer ask about race and ethnicity, which it has asked in some form in every census since 1790. The American Community Survey would keep them. > > **How to comment** > 1. Use the official comment form on [Regulations.gov](https://www.regulations.gov/commenton/USBC-2026-0628-0001) (docket USBC-2026-0628). > 2. Write in your own words how the change would affect you or your community. Specific, original comments count for more than form letters. > 3. Submit by **November 2**. Comments become public record. The Bureau accepts anonymous comments if you enter "N/A" in the required name fields. > > **Need help?** [Hands Off Our Census](https://handsoffourcens.us/) is a nonpartisan coalition of census and data-policy groups. It has a [step-by-step comment guide](https://handsoffourcens.us/comment-guide.html), a [plain-language explainer](https://handsoffourcens.us/proposal.html) and a [Spanish-language page](https://handsoffourcens.us/comentarios.html). > > **Background:** [[2030 Census Residence Rule]] · [[How the Census Apportions Congress]] %% census-comment-callout:end %% The census is usually explained as a contest over political power, and it is. But the count also sets the terms on which federal money reaches states, counties, school districts and cities, every year for the ten years after it is taken. This page maps that second path: what the number becomes, which formulas consume it, and what each archetype does to a place that was counted badly. The scale is the [[Census Bureau]]'s own estimate: more than **$2.8 trillion** distributed in fiscal year 2021 through **at least 353** federal assistance programs that used Bureau data in whole or in part. That figure deserves a caveat the headline never carries — more than **$700 billion** of it was one-off COVID-19 assistance, so FY2021 is not a picture of the recurring stake. The structure underneath it is, and the structure is what this map is for. Companion maps: [[How the Census Is Actually Counted]] is the count itself; [[How the Census Apportions Congress]] is the seats path; [[How a Census Question Is Added or Removed]] and [[How a Federal Survey Is Approved]] are how the questions get on the form at all; [[Census Data Products Tree]] is what the count publishes. This page picks up where those stop, at the moment a published number becomes a cheque. ⭐ **The organizing finding is that the census enters the single largest formula in the system as a divisor rather than as a measure of need — and that inverts the penalty for being undercounted.** The [[Federal Medical Assistance Percentage]] decides what share of each state Medicaid dollar the federal government pays. It uses no census income data at all. It uses per capita personal income, a [[Bureau of Economic Analysis]] figure whose denominator is the Census Bureau's own midyear population estimate. A state's total income does not change when the census misses people; only the denominator falls. Same income over fewer people means **higher** measured per capita income, which the statute squares, which raises the state's share and lowers the federal one. **The undercounted state is penalised for looking richer than it is.** Andrew Reamer of the George Washington Institute of Public Policy put a number on it in March 2018: across the 37 states then above the statutory floor, a median **$1,091 of federal money forfeited per person missed**, in a single fiscal year, across five HHS programs — ranging from $533 per person in Utah to $2,309 in Vermont. For Texas he modelled a one-percentage-point worse undercount as 251,456 more people missed, per capita income up $371, an FMAP of 58.05 falling to 57.28, and roughly **$292 million** gone in FY2015 alone. Those are his figures and his method, not the government's. The second finding follows from the first and is sharper. **The 50% statutory floor switches the penalty off for the states that need it least.** A state already at the minimum match cannot fall below it, so an undercount costs it nothing in FMAP — California and New York were in that group in Reamer's analysis. The penalty is therefore not distributed according to who was undercounted. It is distributed according to where a state happens to sit relative to a number Congress wrote in 1965. The third is about time. The Population Estimates Program carries the census base forward by adding births, deaths and migration to it. **The method adds change to a level; it never re-measures the level.** Nothing in that step could find a household the census missed and put it back, and there is no revision to a corrected level during the decade. The corrected series — the intercensal estimates — is produced once every ten years, *after* the next census: the revised 2010–2020 figures were released on 7 November 2024. By then the decade they describe had been funded and closed. The correction is a historical record, not a refund. > [!note] How to read this chart > `──▶` = the main path · `└─▶` = a branch · `◇` = a decision point · `⏱` = a statutory or recurring clock · `◉` = the point at which a census-derived number enters a calculation · `⊘` = no census number enters, foreclosed by the statute's own text · `◈` = a default that resolves against the place that was undercounted · `★` = changed or acted on in 2025–26 · `[cite]` = the controlling authority. Everything inside the fence is plain text; linked entities appear in the prose, Notes and See also. ``` CENSUS FORMULA FUNDING FLOW (as of 21 October 2026) │ From the decennial count to the grant cheque │ Legend: ──▶ main path └─▶ branch ◇ decision point ⏱ clock │ ◉ a census-derived number ENTERS the calculation here │ ⊘ no census number enters — foreclosed by the statute's own text │ ◈ a default that resolves AGAINST the place that was undercounted │ ★ changed or acted on in 2025-26 [cite] controlling authority │ ■ 0. THE COUNT — one measurement, once a decade │ The enumeration as of APRIL 1 of the year ending in zero. Everything │ below this line is that number plus arithmetic. [13 U.S.C. § 141] │ │ │ ├─ the apportionment counts ──▶ seats in the House. A different path, and │ │ the one everybody watches. │ │ │ └─▶ THE SAME COUNTS BECOME THE ESTIMATES BASE — and that is the money path, │ which nobody watches, and which runs for the whole decade after. │ ├─◆ A. THE BASE BECOMES A SERIES ⏱ annual, every year of the decade │ │ The Population Estimates Program carries the base forward by the │ │ COHORT-COMPONENT METHOD. Each July 1 estimate is: │ │ │ │ base + births − deaths ± migration │ │ │ ├─ THE METHOD ADDS CHANGE TO A LEVEL. IT NEVER RE-MEASURES THE LEVEL. │ │ There is no survey inside this step that could find a missed household │ │ and put it back. If the base is low, every estimate built on it is low. │ │ │ ├─ each VINTAGE replaces the entire prior series back to the base, so two │ │ vintages are not comparable to each other — only within one vintage │ │ │ └─★ the base itself was not finished until 2026. Post-2020 estimates ran on │ a BLENDED BASE — 2020 Census totals, 2020 Demographic Analysis, and the │ Vintage 2020 estimates — because census data were not available at the │ demographic detail the base needed. Vintage 2025, released 25 Jun 2026, │ was the FIRST to carry the 2020 Census's full age/sex/race detail into │ the 1 Apr 2020 base. Five years and three months into the decade. │ ├─◆ B. THE DERIVED LAYERS — two of them, and they are not alike │ │ │ ├─◆ B1. THE POVERTY LAYER ◉ census survey data │ │ │ Small Area Income and Poverty Estimates: a MODEL, not a count. │ │ │ Inputs: federal tax return records · SNAP participation · Bureau of │ │ │ Economic Analysis income · the American Community Survey · and the │ │ │ population estimates from Part A as the control total. │ │ │ │ │ ├─ outputs: states · counties · SCHOOL DISTRICTS, single-year │ │ └─ the school-district model adds confidential IRS personal income data │ │ and the School District Review Program's boundaries │ │ │ └─◆ B2. THE INCOME LAYER ◉ census POPULATION only — no census income │ │ PER CAPITA PERSONAL INCOME is a Bureau of Economic Analysis figure. │ │ Its numerator is BEA's own income measure. Its DENOMINATOR is the │ │ Census Bureau's midyear resident population estimate from Part A. │ │ │ └─▶ SO THE CENSUS ENTERS THE LARGEST FORMULA IN THE SYSTEM AS A DIVISOR, │ not as a measure of need. Hold that; Part E is what it does. │ ├─◆ C. THE THREE FORMULA ARCHETYPES │ │ Every formula grant that consumes census data is one of these, or a │ │ stack of them. The archetype decides what an undercount costs. │ │ │ ├─◆ C1. PER-CAPITA / POPULATION-SHARE ◉ head counts │ │ │ Money divided by relative population. The simplest design, and the │ │ │ only one where losing a person has an obvious, linear price. │ │ │ │ │ └─ example: 55% of a state's Surface Transportation Block Grant, after │ │ the Transportation Alternatives set-aside, must be obligated in │ │ areas "in proportion to their relative shares of the State's │ │ population"; the other 45% may go anywhere in the state. │ │ [23 U.S.C. § 133(d)(1)(A), (d)(1)(B)] │ │ │ ├─◆ C2. NEED-ADJUSTED / POVERTY-COUNT ◉ modelled poverty, not population │ │ │ Money divided by a count of a qualifying subgroup, usually weighted │ │ │ so that concentration is worth more than mere incidence. │ │ │ │ │ ├─ example: Title I-A of the Elementary and Secondary Education Act runs │ │ │ on FORMULA CHILDREN — overwhelmingly children aged 5-17 in poor │ │ │ families, from the Part B1 model, not from the decennial form │ │ │ │ │ ├─ FOUR formulas run at once off the same count, with different │ │ │ eligibility gates and different weighting: │ │ │ Basic ........... 10 children and a 2% rate ....... unweighted │ │ │ Concentration ... 6,500 children OR a 15% rate .... unweighted │ │ │ Targeted ........ 10 children and a 5% rate ....... weighted │ │ │ Finance Incentive 10 children and a 5% rate ....... two-stage │ │ │ │ │ └─◈ and then a HOLD HARMLESS floor of 85-95% of last year's grant, set │ │ by the district's own child rate. A district whose measured poverty │ │ falls is protected; a district whose real poverty was never │ │ measured in the first place has nothing to be held harmless at. │ │ │ └─◆ C3. INCOME-INDEXED MATCHING RATE ◉ census population as a divisor │ │ No pot is divided here at all. The census number sets the SHARE of │ │ each state dollar the federal government reimburses — so the money │ │ is open-ended and the census decides the exchange rate. │ │ │ ├─ the Federal Medical Assistance Percentage, in the statute's words: │ │ 100 per centum less the state percentage, where the state │ │ percentage bears the same ratio to 45 per centum as the SQUARE of │ │ state per capita income bears to the SQUARE of national per capita │ │ income [42 U.S.C. § 1396d(b), § 1905(b)] │ │ │ ├─ floor 50% · ceiling 83% · promulgated between 1 Oct and 30 Nov, on │ │ the average per capita income of the THREE most recent calendar │ │ years for which satisfactory data are available from the Department │ │ of Commerce, and then conclusive for four quarters │ │ [42 U.S.C. § 1301(a)(8)(B)] │ │ │ └─ ONE RATE, SEVEN PROGRAMS. The same percentages set federal matching │ for Medicaid, the TANF Contingency Fund, child support collections, │ child care mandatory and matching funds, Title IV-E foster care │ maintenance, adoption assistance and kinship guardianship │ assistance; the enhanced CHIP rate is derived from them. │ ├─◆ D. WHERE A CENSUS NUMBER DOES NOT DECIDE ⊘ │ │ Not every dollar on the "census-guided" list is guided by the census, │ │ and the biggest exception is one of the biggest line items. │ │ │ └─⊘ FEDERAL-AID HIGHWAY APPORTIONMENT AMONG THE STATES HAS NO POPULATION │ TERM. Each state's share is its SHARE OF FY21 APPORTIONMENTS applied to │ the national total, then floored at 95% of its Highway Trust Fund │ contributions and at 1% above last year. [23 U.S.C. § 104(c)(1)] │ │ │ └─▶ the census decides which PARTS of a state get highway money, │ through C1 above. It does not decide which states do. A perfect │ count would not move one dollar across a state line here. │ ├─◆ E. THE TRANSMISSION OF AN ERROR — what a missed person actually costs │ │ │ ├─ under C1, the loss is intuitive: fewer heads, a smaller share. │ │ │ └─◈ UNDER C3 IT RUNS BACKWARDS, AND C3 IS THE BIG ONE. │ │ │ ├─ a state's total income does not change when the census misses │ │ people. Only the denominator falls. │ ├─ same income, fewer people ──▶ MEASURED PER CAPITA INCOME GOES UP │ ├─ higher per capita income ──▶ squared ──▶ a bigger state percentage │ └─ bigger state percentage ──▶ A LOWER FMAP ──▶ LESS FEDERAL MONEY │ │ │ └─▶ THE UNDERCOUNTED STATE IS PENALISED FOR LOOKING RICHER. │ │ ◈ AND THE FLOOR SWITCHES THE PENALTY OFF AT THE TOP. A state already │ at the 50% minimum cannot fall further, so an undercount costs it │ nothing in FMAP. The loss lands only on the states above the floor. │ └─◆ F. THE CORRECTION CLOCK ⏱ once per decade, afterwards │ ├─ the census is not adjusted for measured coverage error. The │ Post-Enumeration Survey measures the error; the counts stay as counted. │ ├─ the estimates are not revised to a corrected level during the decade. │ Part A has no mechanism for it. │ └─◈ INTERCENSAL ESTIMATES — the corrected series — are produced ONCE EVERY TEN YEARS, AFTER the next census. The revised 2010-2020 series was released 7 NOVEMBER 2024. │ └─▶ So the arithmetic is settled, and the money is gone, before the number it ran on is corrected. The correction is a historical record, not a refund. ⊘ No formula reaches back. ``` ## Notes - **The $2.8 trillion figure is the Census Bureau's, for fiscal year 2021, and more than $700 billion of it was COVID-19 assistance.** The Bureau's own analysis covered the decennial census, the [[American Community Survey]], the geographic programs that support both, and the annual [[Population Estimates Program]]. The twenty largest programs accounted for roughly 90% of the total. A separate 2023 count by the Project On Government Oversight put 338 programs and about $2.1 trillion against fiscal year 2020 — a different year, a different program universe, and a different author. Neither number is a steady-state figure, and this page does not assert one. - **"Per capita income" in the Medicaid formula is not a census statistic.** The statute directs the Secretary of Health and Human Services to use per capita income data from the Department of Commerce, which in practice means the [[Bureau of Economic Analysis]]. BEA states that it uses Census Bureau midyear resident population estimates as the denominator. That is the whole of the census's involvement in the largest formula on the page, and it is the reason the undercount penalty runs in the counter-intuitive direction. - **The three-year average is a statutory requirement, not a convention.** The rate must be promulgated between 1 October and 30 November each year, on the average per capita income for the three most recent calendar years for which satisfactory data are available, and is then conclusive for the following four quarters. The FY2027 rates were published on 28 November 2025. The specific calendar years behind any given year's rate are not stated in the notice itself and are not asserted here; Reamer's 2018 report records that the FY2015 rates used 2010–2012. - **Title I-A's poverty count comes from a model, not from the census form.** The [[Small Area Income and Poverty Estimates]] program combines federal tax records, SNAP participation, BEA income, the American Community Survey and the population estimates. Congressional Research Service analysis of the FY2020 cycle found roughly 96.7% of formula children were children aged 5–17 in poor families, and that the grants then being made rested on estimates published in December 2019 using calendar-year 2018 income — a lag of roughly two years in that cycle. No general lag rule is asserted. - **The highway exception is real and is easy to get wrong.** Highways appear prominently among the largest "census-guided" programs, but state-level apportionment under the [[Bipartisan Infrastructure Law]] is a historical-share calculation: each state receives its share of FY2021 apportionments applied to the current national total, floored at 95% of its [[Highway Trust Fund]] contributions. Census population governs the sub-state suballocation of the Surface Transportation Block Grant, not the division among states. - **The 2020 count's coverage error was measured and left in place.** The Post-Enumeration Survey found a national net coverage error of −0.24%, not statistically distinguishable from zero, over statistically significant undercounts of 3.30% for Black residents, 4.99% for Hispanic residents and 5.64% for American Indian and Alaska Native residents living on reservations, against overcounts of 1.64% for non-Hispanic White residents and 2.62% for Asian residents. A national figure near zero is consistent with large offsetting errors by group and by place, which is exactly the distribution that formula funding is sensitive to and apportionment is not. - **The estimates base for this decade was not complete until mid-2026.** Post-2020 estimates ran on a "blended base" assembled from 2020 Census totals, the 2020 Demographic Analysis and the Vintage 2020 estimates, because census results were not available at the demographic detail the base required. Vintage 2025, released 25 June 2026, was the first to incorporate the 2020 Census's full age, sex, race and Hispanic-origin detail into the 1 April 2020 base, and put the national population at 341,784,857 as of 1 July 2025. - **A live rulemaking sits directly upstream of everything on this page.** The [[2030 Census Residence Rule]], published at 91 FR 57524 on 10 September 2026, would restrict the apportionment count to U.S. citizens and lawful permanent residents and redefine usual residence as the place where a person lawfully spent the most days in a 3 January–1 April enumeration period, evidenced by tax records. **The comment period was extended at 91 FR 63207 on 5 October 2026 and closes on 2 November 2026.** - **What the rule does not say is the question this map leaves open.** The proposed rule does not analyse federal funding formulas, and its only funding reference is a Regulatory Flexibility Act observation that any effect on small governmental entities "would be indirect." A footnote states that the Bureau maintains and publishes statistics separate from those used for apportionment and expects to continue doing so. Whether the formula and estimates path would then run on a full resident count while apportionment ran on a narrower one — and if so, which number the [[Bureau of Economic Analysis]] would divide by — is not addressed in the proposal. `[unconfirmed]` - **Open questions not resolved here.** Whether the Base Evaluation and Research Team recommendations that produced the Vintage 2025 base change also altered county-level and sub-county estimates materially `[unconfirmed]`; whether any formula program has ever been re-run against intercensal estimates after the fact `[unconfirmed]`; and what the current-decade equivalent of Reamer's per-person FMAP loss figures would be, which would require re-running his method on present data and is not attempted. ## See also - [[How the Census Is Actually Counted]] · [[How the Census Apportions Congress]] · [[How a Census Question Is Added or Removed]] · [[Census Data Products Tree]] · [[Census Data Access Crosswalk]] - [[Census]] · [[Census Bureau]] · [[American Community Survey]] · [[Population Estimates Program]] · [[Small Area Income and Poverty Estimates]] - [[Federal Medical Assistance Percentage]] · [[Medicaid]] · [[Elementary and Secondary Education Act]] · [[Block Grant]] - [[Bureau of Economic Analysis]] · [[Highway Trust Fund]] · [[Bipartisan Infrastructure Law]] · [[Department of Commerce]] - [[2030 Census Residence Rule]] · [[2030 Census]] · [[Local Update of Census Addresses]] · [[How a Federal Survey Is Approved]]