The first nation in the world to take a regular population census, the United States has been counting its population every 10 years since 1790—as required by the U.S. Constitution (Article I, Section 2).1 The first U.S. census was conducted by 16 U.S. marshals and their 650 assistants. It took them 18 months to visit households and compile the final tally of 3.9 million people, including nearly 700,000 slaves.

The population of the United States today is nearly 85 times larger than it was during that first census. The land area of the nation has changed. Technology and social structures have changed. And while the primary purpose of the decennial census remains the same—determining the number of seats each state occupies in the U.S. House of Representatives—uses of the data have grown.

The 1790 census included information on “Free White males of 16 years and upward” (to assess the country’s military potential).2 Today an accurate count of residents helps to shape important infrastructure investments, such as hospitals, schools, roadways, bridges, and railways. Detailed population information is also critical for emergency response during disasters.

Enumeration Methods Have Changed with Technology

Over the decades, census-taking switched from a task of the U.S. marshals to one of specially-trained enumerators (who took over in 1880), and from paper-and-pencil tabulation, to punch-cards, to electronic data collection.

Early censuses were taken by going door to door. In those early years—when literacy was low—enumerators asked questions and recorded information about each occupant in a household. Later, they began offering a mail-back form for those who did not want to respond in person. It was not until 1960 that mail self-response became the primary census data collection mode, but some communities, like those in remote Alaska, are still enumerated in person today.

Continuing the tradition of changing methods for changing times, the 2010 Census provided field workers with handheld electronic devices to capture address data, but still relied on paper data collection for nonresponse follow-up (when a trained Census worker visits an address for in-person data collection if the form for that address was not already mailed in).

Making yet another technological leap, the 2020 Census is designed to be conducted primarily via internet self-response. While paper questionnaires will still be mailed and in-person enumeration will be conducted for those households who do not respond, the Census Bureau expects that most households will submit their 2020 form online.

Questions Evolve in Response to Societal Change

The census questionnaires have changed every decade. In most cases the changes involved requesting more detailed information, but sometimes the modifications simply reflected prevailing social and political currents. For example, the number of racial categories used in the census has fluctuated considerably over the years. Groups identified by geography (such as Asians, Pacific Islanders, and Aleutian Islanders) have been listed as races, together with groups defined by skin color (blacks and whites). The racial categorization of some nationality groups has also changed over time. Asian Indians were included in the white race in the 1970 Census but were counted in the Asian and Pacific Islander category starting with the 1980 Census. The 1970 Census was the first to ask U.S. residents whether they were of Hispanic origin. And beginning with the 2000 Census, Americans were given the choice of marking all “race” categories with which they identified, resulting in the first decennial counts of multiracial persons.

As living arrangements became more complex, the question that asks how each household member is related to the householder added more response categories, including one for “unmarried partner” to reflect the increase in cohabitation.

In every decennial census from 1940 to 2000, two questionnaires were used to collect information: a “short form” with only basic questions such as age, sex, race, and Hispanic origin; and a “long form” that included about 50 additional questions on socioeconomic and housing characteristics. Only a subset of households received the long-form questionnaire—about one in every six in 2000. However, the 2020 Census—like that of 2010—will be a short form-only census. This is because the decennial long form has been replaced by the American Community Survey (ACS). The ACS is a nationwide, continuous survey designed to provide reliable and timely demographic, housing, social, and economic data every year. The ACS replaced the long form beginning in 2010 by collecting long form-type information throughout the decade rather than only once every 10 years.

Hard-to-Count Populations

Decades of research have shown that the decennial census is very accurate, but (like population censuses in other countries) it is subject to both undercount and overcount errors that differ by age, sex, and race. The 2010 Census was no exception. Despite the best efforts and careful planning of Census Bureau staff, the direct, physical enumeration of the U.S. population is imperfect.

Part of the challenge in counting the population accurately is that some people are harder to count than others. People who lack a permanent address are less likely to complete a census form than people who have a permanent address. Similarly, language barriers, distrust of government, and frequent moves tend to make certain groups harder to count. On the other side of the spectrum, some people may be counted more than once. For example, those who own more than one home may submit a census form for each address, and children away at college may be counted at both their college and parental home.

In 2010, the Census Bureau estimated that their total overcount was fairly small (about 36,000 people, or 0.1 percent of the population), but that over/under-counts varied by age, race, and other characteristics.3 Both the 2000 and 2010 census tended to undercount renters and overcount homeowners. Young children tend to be undercounted, while older adults tend to be overcounted.4

The Census Bureau works to reduce over-/under-counts with each census. One new tool for the 2020 Census is the Response Outreach Area Mapper (ROAM), which can be used by community groups to identify local areas of potential undercount and target outreach to those neighborhoods.

Population and Cost Continue to Increase

As the total population count soared from 76 million in 1900 to more than 281 million by 2000, the cost of conducting the decennial census rose from about 16 cents per person to more than $16 per person.5 For the 2010 Census, the rate increased to more than $38 per person. The mailout/mail-back questionnaires, first used extensively in the 1960 Census, drastically cut back the need for enumerators to go door to door. In 2010, questionnaires were mailed to nearly all households, yet the Census Bureau still employed more than 600,000 temporary workers to help carry out that Census.6

The 2010 Census cost about $12.3 billion, but the 2020 Census is projected to be the most expensive ever, at approximately $15.6 billion, an increase of 27 percent.7The population, however, is projected to increase by just under 8 percent between 2010 and 2020 (see table). While cost continues to rise faster than population, the increase in expense from 2010 to 2020 is expected to be lower than that of censuses for the past several decades because of anticipated savings from online response and streamlined field operations.

United States Population and the Cost of the Census

Census Population % Increase from Previous Census Cost % Increase from Previous Census
1980 226,542,199 11.4% $3,000,000,000
1990 248,718,301 9.8% $4,700,000,000 57%
2000 281,421,906 13.1% $9,400,000,000 100%
2010 308,745,538 9.7% $12,300,000,000 31%
2020* 332,555,000 7.7% $15,600,000,000 27%
*Projected population by July 1, 2020 and estimated cost for Census 2020.
*Projected population by July 1, 2020 and estimated cost for Census 2020.
Source: U.S. Census Bureau.