The census and apportionment: counting people to divide power

The census and apportionment are the linked processes by which the United States counts its population every ten years and divides seats in the House of Representatives among the states according to that count, as Article I, Section 2 of the Constitution requires.⁠¹ The original rule counted “three fifths of all other Persons,” meaning enslaved people, a rule replaced by the Fourteenth Amendment’s direction to count “the whole number of persons in each State”.⁠^1,2 Since then the count has been contested in Congress, which failed to reapportion after the 1920 census,⁠³ and in the Supreme Court, which required equal-population districts in 1964 and has since ruled on sampling, imputation, and a proposed citizenship question.⁠^5,8,9,11 The formula used to divide seats is explained in the glossary entry on apportionment and is not repeated here.

In plain words

The United States decides how many members of Congress each state gets by counting people. That count happens every ten years and is called the census.

From the start, the count was about power. The original Constitution counted enslaved people as three fifths of a person for this purpose. That gave slaveholding states more seats without giving enslaved people any rights. After the Civil War, the Fourteenth Amendment changed the rule to count every person.

The count has caused fights ever since. After the 1920 census, Congress could not agree on how to divide seats and simply did not do it for a decade. Inside states, some legislatures did not redraw their own districts for sixty years, so a small share of voters could elect a majority. In the 1960s the Supreme Court said districts must have roughly equal populations: one person, one vote.

The arguments continue today. Should the census use statistics to fill in people it missed? Should it ask about citizenship? Whom does it miss most often? A 2020 follow-up survey found that some groups were undercounted and others overcounted.

Any system that adds up people’s voices by place, including this framework, faces the same questions: who is counted, how are gaps handled, and who decides?

Defined terms

• Census (actual enumeration). The constitutionally required count of the population every ten years.⁠¹
• Apportionment. Dividing House seats among states by population. See apportionment.
• Redistricting. Drawing district lines within a state after apportionment. See gerrymandering.
• Malapportionment. Districts with very unequal populations, so that votes in some districts weigh more than in others.⁠⁶
• One person, one vote. The principle, drawn from Wesberry v. Sanders and Reynolds v. Sims, that districts should have substantially equal populations.⁠^5,6
• Apportionment base. The population counted for drawing districts: total population, citizens, or eligible voters.⁠⁷
• Imputation. Inferring characteristics or occupancy for a nonresponding address from nearby records.⁠⁹
• Net coverage error. The difference between the census count and an independent estimate of the true population.⁠¹⁰

Evidence

Law: the original bargain

Article I, Section 2 directed that “Representatives and direct Taxes shall be apportioned among the several States ... according to their respective Numbers, which shall be determined by adding to the whole Number of free Persons, including those bound to Service for a Term of Years, and excluding Indians not taxed, three fifths of all other Persons.” It required “the actual Enumeration” within three years of the first Congress and “within every subsequent Term of ten Years,” and capped representation at one member “for every thirty Thousand”.⁠¹

The clause tied two things together: taxes and seats. It also tied representation to people who could not vote. Counting enslaved people even partially increased the House delegations of states that denied them every political right.

Law: the Fourteenth Amendment

Section 2 of the Fourteenth Amendment, which the National Archives dates to 1868, provides that “Representatives shall be apportioned among the several States according to their respective numbers, counting the whole number of persons in each State, excluding Indians not taxed”.⁠² It also provided that if a state denied the vote to adult male citizens for reasons other than rebellion or crime, “the basis of representation therein shall be reduced” proportionally.⁠² The amendment counted every person, but by its terms it penalized only denial of the vote to men.

History: the decade without reapportionment

The House of Representatives’ own history office records that the House “failed to reapportion itself after the 1920 Census,” the only time in its history, as “a battle erupted between rural and urban factions” because the method in use gradually “caused smaller rural states to lose representation to larger urbanized states.” The Permanent Apportionment Act, signed June 18, 1929, fixed “the number of Representatives at 435” and “created a procedure for automatically reapportioning House seats after every decennial census”.⁠³ Making the process automatic took the decision out of the hands of the members whose seats were at stake.

Law: the reapportionment revolution

Inside states, legislatures that drew their own districts often declined to redraw them. In Baker v. Carr, decided March 26, 1962, the Court reviewed Tennessee’s use of a 1901 apportionment act after more than 60 years in which every reapportionment proposal had failed, and held: “We hold that this challenge to an apportionment presents no nonjusticiable ‘political question’”.⁠⁴

Two years later the Court set the standard:

• Congressional districts. In Wesberry v. Sanders, decided February 17, 1964, Georgia’s Fifth District had 823,680 people while the Ninth had 272,154. The Court held that Article I, Section 2 means “as nearly as is practicable one man’s vote in a congressional election is to be worth as much as another’s”.⁠⁵
• State legislatures. In Reynolds v. Sims, decided June 15, 1964, the Court found that about 25.1 percent of Alabama’s population lived in districts that could elect a majority of the state senate, with population ratios as high as about 41 to 1. It held that both houses of a state legislature must be apportioned by population, writing that “Legislators represent people, not trees or acres,” and rejected the analogy to the U.S. Senate.⁠⁶

Law: which population?

The equal population rule left open which population. In Evenwel v. Abbott, decided April 4, 2016, challengers argued that Texas’s state senate districts, within 8.04 percent of equal total population, deviated by more than 40 percent when measured by eligible voters. The Court held that “a State may draw its legislative districts based on total population,” while stating that it “need not and do not resolve” whether a state may instead equalize voter-eligible population. Justice Thomas, concurring in the judgment, argued that the Constitution leaves the choice of base to the states; Justice Alito, also concurring in the judgment, would not have decided whether other bases are permitted and disputed the majority’s historical reading.⁠⁷

Law: how to count

• Sampling. In Department of Commerce v. U.S. House of Representatives, decided January 25, 1999, the Court held that “the Census Act prohibits the proposed uses of statistical sampling” to determine population for apportionment.⁠⁸
• Imputation. In Utah v. Evans, decided June 20, 2002, the Court upheld “hot-deck imputation,” in which the Census Bureau, as a last resort, inferred that an address with missing information had the characteristics of its nearest similar neighbor. Imputation raised North Carolina’s count by 0.4 percent and Utah’s by 0.2 percent, shifting one House seat from Utah to North Carolina. Justice Thomas, joined by Justice Kennedy, argued that “actual Enumeration” bars imputation; Justice O’Connor would have treated it as prohibited sampling.⁠⁹
• The citizenship question. In Department of Commerce v. New York, decided June 27, 2019, the Court held that the Constitution permits a citizenship question on the census and that the Secretary’s decision was reviewable, but found that the stated Voting Rights Act rationale “seems to have been contrived,” and affirmed the remand to the agency. The justices split differently on different parts of the opinion.⁠¹¹

Current data: who is missed

The Census Bureau’s 2020 Post-Enumeration Survey, released March 10, 2022, estimated a national net coverage error of minus 0.24 percent, not statistically different from zero. It estimated undercounts of 3.30 percent for the Black population, 4.99 percent for the Hispanic or Latino population, 5.64 percent for American Indians and Alaska Natives living on reservations, and 2.79 percent for children ages 0 to 4, and overcounts of 1.64 percent for the non-Hispanic White population and 2.62 percent for the Asian population.⁠¹⁰ A national count can be accurate in total while being inaccurate for particular groups.

Competing interpretations

Count everyone versus count the electorate. The Evenwel majority relied on constitutional history and practice to permit total population.⁠⁷ Supporters argue that representatives serve everyone in a district, including children and noncitizens. The Evenwel challengers argued that equal total population can leave voters with unequal weight. The Court did not decide whether a state may choose the second approach.

Enumeration versus estimation. Supporters of statistical methods argue that a headcount predictably misses some groups, as the 2020 coverage estimates show,⁠¹⁰ and that statistics can correct it. Critics, including the justices who would have barred imputation, argue that the Constitution requires an actual count and that statistical adjustment invites manipulation.⁠⁹ Current law bars sampling for apportionment but permits limited imputation.⁠^8,9

Who sets the rules. The 1920s deadlock and the decades of unredrawn state districts show what happens when the people who benefit from the current count decide whether to change it.⁠^3,4 The 1929 automatic process and the 1964 cases took some decisions away from incumbents. Justice Thomas’s Evenwel concurrence questions whether the Court had a sound basis for doing so.⁠⁷

Interpretation, this site’s reading. The census history shows that counting rules are never neutral, because every rule changes someone’s share. The defensible approach is a rule fixed in advance, published, and applied automatically, with known error reported openly.

Application to this framework

These are design implications, not claims about the framework’s performance.

1. Say who is counted. Every My Vote Forward aggregate should state its base: participants, verified residents, or another group. A priority from participants is not a priority of the population; see representation.
2. Fix rules before results. The 1929 Act made reapportionment automatic to remove incumbents’ discretion.⁠³ Aggregation rules in this framework should be published and frozen before a collection period opens; see published releases.
3. Report coverage error by group. The census reports undercounts by group.⁠¹⁰ A participation tool should report which groups are underrepresented among participants, not only the total; see measurement.
4. Be cautious with filling gaps. Imputation and sampling are contested even for the census.⁠^8,9 If results are ever weighted or modeled, the method should be disclosed and the unweighted figures published too.
5. Geography changes results. Reynolds and Wesberry show how boundaries change the weight of each voice.⁠^5,6 When results are reported by area, the areas should be standard public boundaries, published in advance.
6. Privacy and accuracy trade off. Published small-area data can reveal individuals. Protections such as small-group suppression, k-anonymity, and differential privacy reduce that risk at a cost in precision for small groups; see published releases.

Hypothetical example. A county reports priorities by its five supervisor districts. One district has twice the participants of another. If the county averages district results equally, each participant in the smaller district counts twice as much in the county total. Publishing both the participant-weighted and district-weighted figures shows readers the choice that was made.

Unresolved questions

• May a state draw districts by voter-eligible population instead of total population? Evenwel left this open.⁠⁷
• Can undercounts of particular groups be reduced without the statistical methods the Court has limited⁠^8,10?
• Should the House size, fixed at 435 since 1929, be revisited⁠³? That is a question for Congress.
• For non-binding participation tools, what base is fairest when people in one place are far more likely to take part than people in another?