It is rewarding to read a cogent article (Sullivan, 2020) that appropriately stresses the accomplishments of generations of civil servants who administered the decennial counts. As Teresa Sullivan emphasizes, the census is the foundation of our system of representative and democratic government. It is well known that the population count determines the number of people who represent a state in Congress and in the Electoral College. The composition of Congress from 2023 to 2033 will be ascertained by the April 2020 count. The number of electors from each state who will select presidents when they meet in late 2024 and 2028 will be similarly determined. Examining demographic changes since 2010, it is likely that Florida and Texas will gain two seats in this year’s enumeration while Arizona, Colorado, Montana, North Carolina, and Oregon will gain one. Nine states—all in the Rust Belt or Northeast except for Alabama—will lose one representative.
Sullivan (2020) shrewdly adds “(and Republic)” to the title of her article. James Madison, one of the founding fathers, argued that the gathering of population data through the decennial census would allow Congress to “adapt the public measures to the particular circumstances of the community” (Reamer, 2018). Early in the 19th century questions were added to the census about manufacturing and, later, about literacy, school attendance, and place of birth. After the Civil War, the census enumeration schedules were laden with numerous questions about mortality, fertility, employment, and national origins. Beginning in the Depression Decade (1929–1941), Congress followed the recommendations of Madison and appropriated funds to address crucial problems that local governments faced, including housing, unemployment, health care, and poverty. From the 1930s until 1980, Congress again and again expanded these programs and developed new ones addressing child health, food scarcity, and precollege education. During the Nixon presidency, many older cities approached bankruptcy as their industries closed and their populations plummeted. Congress responded with revenue sharing, in which a small fraction of federal tax revenue was distributed to 31,000 local governments on the basis of their population size and their economic status as indexed by their tax revenue. The population count was the key to what local governments received.
Andrew Reamer at George Washington University established a Counting for Dollars program to ascertain and describe how federal funds are currently distributed to state and local governments on the basis of data derived from the decennial census. In fiscal 2018, approximately $1.5 trillion dollars of federal spending was allocated to states on the basis of data linked to or derived from the census count for local areas. For example, federal funds to support the Head Start Program are distributed to states on the basis of Census Bureau estimates of the number of children under age 5 in a state living in impoverished households. The Women, Infants, and Children feeding program allocates federal dollars to states on the basis of Census Bureau estimates of a state’s proportion of the national total of children under age 5 in households with incomes below 185% of the poverty line. The Supplemental Nutrition Assistance Program (food stamps) distributes federal dollars on the basis of information about poverty rates and unemployment in local areas (Reamer, 2019).
Congress—not the census—determines the amount of federal funds to be distributed but census-derived data determine the amount that a state obtains. From 1800 through 2000, decennial censuses reflected Madison’s wise suggestion about gathering extensive data. By 1930, the enumeration form asked adults 32 questions (Ruggles et al., 2019). Since 1940, most of the social and economic questions were asked of a sample. Collecting data became more difficult and response rates declined as more and more questions were added. In 2010, householders were asked just a few basic questions about age, sex, household relationship, race, and Hispanic origin. The 2020 enumeration will ask householders only nine questions and other occupants of that household just seven.
If the census asks no social and economic questions, where does Congress obtain data to allocate $1.5 trillion in federal funds annually? In 2003 the Census Bureau initiated the American Community Survey, which now interviews occupants of three million households every year. That survey asks householders 28 questions about housing and occupants 48 questions about their social and economic characteristics, including almost a dozen about earnings and income. The sampling frame for the American Community Survey is linked to the decennial census counts. If the census misses a substantial proportion of residents in remote rural areas or in high-poverty neighborhoods in cities, those residents will likely be underrepresented in the American Community Survey and the states’ characteristics, such as the poverty level, the number of poor children, and median household income will be inaccurate. There are a variety of federal programs that send funds to states on the basis of their unemployment rates. The Census Bureau, in their Current Population Survey, gathers data from about 60,000 households each month focusing upon employment and earnings. Again, the sampling for this labor force survey is linked to decennial census counts. A major undercount of areas of high unemployment in the census may lead to erroneous reports of unemployment in the state, likely costing some states their fair share of federal funds. Some states also have programs that are even more directly linked to local population counts. In Michigan, for example, 10 percent of the revenue collected by the state’s sales tax is returned to local governments on the basis of their population count in the census.
If undercount rates were the same for all groups and for all states, there would be no misallocation of the $1.5 trillion in federal funds. But they are not identical for all groups. The most sophisticated analysis of coverage in Census 2010 reported that children under age 5 were undercounted by 5% while persons 20 to 24 and those over age 60 were overcounted. This comes about because college students and ‘snowbirds’ are frequently counted twice. The White population was overcounted by 0.8% but the Hispanic population was undercounted by 1.5%, African Americans by 2.1%, and Native Americans on reservations by 4.9%. Residents of owner-occupied units were overcounted by 0.6% but those in rental units were undercounted by 1.1%. (O’Hare, 2019, chap. 4; U. S. Bureau of the Census, 2012) Differential undercount in Census 2020 may result in many states not obtaining federal funds that they rightfully deserve.
As Sullivan (2020) stresses, the professional integrity of the professional staff of the Census Bureau provides an excellent defense for the accuracy of the census. She mentions an important challenge facing Census 2020, namely, a possible reluctance by many to be enumerated, perhaps because of a distrust of the government or a fear that what is reported will not remain confidential. During World War II, a War Powers Act gave some federal agents access to presumably confidential information about the Japanese population from Census 1940 (Anderson, 2015, pp. 192–199; Okamura,1981; Minkel, 2007). To prevent such disclosures, Congress in 1954 enacted Title 13 of the U. S. Code. This makes it a serious federal offense for anyone to release information reported to the Census Bureau by individuals or by businesses for 72 years. So far, there have been no documented violations of the confidentiality promises of Title 13. The Census Bureau is devoting efforts to prevent any hacking in Census 2020 and is, apparently, preparing for a denial of service attack. As Sullivan points out, were there a successful hack of Census 2020 or an effective ‘denial of service’ attack, the response rate might plummet. Census Bureau director Steven Dillingham has given assurances that his agency computers will work and be secure (Schneider, 2019).
There are two other challenges for Census 2020. The Census Bureau hopes that most individuals will respond by using the Internet or their cell phone. Printed forms will be available upon request. The American Community Survey asks whether householders have a computer and broadband access. There are substantial racial, income, and geographic differences. For example, in 2017, 90% of households headed by an Asian had broadband service, 83% for White households, but only 76% for African Americans and 71% for Native Americans. Among households with incomes in excess of $150,000, 92% have broadband access, but only 70% of households with incomes below $20,000. In Seattle, 92% of households subscribed to broadband service, in Detroit, only 67%. (Data obtained from the University of Minnesota’s Integrated Public Use Sample website [Ruggles et al., 2019].) The Census Bureau will send hard copies of the enumeration forms to those residing in areas where the American Community survey found broadband use was rare and where cell phone service is unavailable.
What percent of households will respond using the Internet, telephone, or hard copy? The Census Bureau scheduled three pretests for Census 2020 but, funds were cut and the only pretest took place in Providence, Rhode Island. The response rate was 55%. Presumably, with the help of a national advertising campaign and the promotional activities of thousands of local “complete count” committees, the response rate in April 2020 will be higher. The Census Bureau is currently planning for a response rate of 60.5%. There are approximately 137 million housing units in the country. The Census Bureau’s estimated response rate suggests that 54,000,000 housing units will have to be visited by an enumerator, some of them several times. That will demand a large labor force. In 2010, the Census Bureau’s employment peak employment was 700,000 during the late April/May enumeration span. The unemployment rate in April 2020 may be as low as 3.5%. Will it be possible to recruit the qualified enumerators required to visit more than 50 million residences?
This brings up the question of how to handle the many housing units that will be visited several times by an enumerator who concludes that the residence is probably occupied but he or she cannot contact anyone for information. In the past, the Census Bureau permitted enumerators to take information from neighbors. And in some cases, information from a similar occupied residence in the neighborhood was used as if it were information for the residence where no one could be contacted. This was known as “hot deck” imputation. The constitutionality of this procedure was challenged after Census 2000. Utah lost a congressional seat to North Carolina by fewer than 900 people. The Utah attorney general observed that if imputed individuals were removed from the counts in both states, Utah would retain the congressional seat it lost to North Carolina. Litigation about the imputation of people to the census count used for congressional apportionment went to the Supreme Court. In Utah v. Evans (2002), the Supreme Court ruled that the use of imputation to produce the apportionment count was constitutional.
To reduce costs in the 2020 enumeration, the Census Bureau will minimize the number of times an enumerator visits a residence that appears occupied but whose residents did reply by Internet, phone, or paper and cannot be contacted. They intend to use federal records to impute residents in such situations. The names and addresses of most all people living in the country are found in some federal statistical system. Social Security is the most inclusive such system since even young children have to have Social Security numbers to be claimed as dependents. Medicare, Medicaid, the Veterans Administration, and Railroad Retirement Board also have statistical information about millions of residents. Census Bureau surveys now contact more than 5 million households each year. Current planning calls for using information from their federal records to complete the census count. This has not been done previously, so we do not know if it will be a highly effective method for minimizing undercount or whether it will increase the number of people double counted. The Census Bureau’s analysis of Census 2010 revealed that 16 million were omitted from the count, but this was offset by 10 million erroneous enumerations—many of them people counted twice—and 6 million imputations. Thus, the net undercount was about 30,000, or an undercount rate was just .01% (O’Hare, 2020). If Census 2020 shows that a state loses a congressional seat to another state by a small number and that resulted from more imputations in one state than the other, a federal lawsuit is likely.
Sullivan’s (2020) article is a paean honoring diligent civil servants who administered 24 decennial enumerations. It is appropriate to add that census staff were largely responsible for speeding the development of what we know today as information technology. Faced with the challenge of rapidly tabulating answers to 30 questions for 50 million residents to be enumerated in 1890, census officials organized a competition for the design of machines that would tabulate data accurately and quickly. The winner, Herman Hollerith, created a machine that allowed a clerk to look at a census enumeration form and enter the information as a series of punches on a hard stock card. He also created an electrically operated machine that would rapidly tabulate those punch cards, providing counts of people and their characteristics. For almost three-quarters of a century, data processing replied upon the machines that Hollerith—one of the founders of IBM—developed for Census 1890.
In preparation for Census 1950, the bureau sought to improve technology. In the mid-1940s, John Mauchly and Presper Eckert at the University of Pennsylvania designed a programmable computer for use by the military. Census Bureau officials were convinced that the machine could be useful for the census and so the bureau began cooperating with Mauchly and Eckert and supported efforts to make their computer useful for tabulating data from the 38 questions to be posed to 152 million people in Census 1950. The result was the Universal Automatic Computer (UNIVAC), a huge machine that used thousands of tubes and data input on tape, but it was the ancestor of the computers we use today (Ruggles & Magnuson, in press).
In the 1950s, Census Bureau clerks still entered data from the questionnaire onto cards and that information was then transferred to tapes that were read by the UNIVAC computer. To save personnel costs and improve accuracy, the Census Bureau designed an enumeration form for 1960 in which most questions were answered by having respondents fill in a small circle. The bureau developed a Film Optical Sensing Device (FOSDIC), which scanned the enumeration form and transferred the marks onto the tapes that the computer could process. The needs of the Census Bureau and the ingenuity of their staff over the last century and a half were key motivating forces in the development of modern information technology.
The census continues to be a solid foundation for our federal statistical system. There are likely to be glitches and confusion about aspects of Census 2020, just as there have been in previous enumerations. As Sullivan (2020) notes, the professional staff at the Census Bureau has a long and distinguished history of dedication to the accuracy of the data they release.
Anderson, M. J. (2015). The American census: A social history (2nd Ed.). New Haven, CT: Yale University Press.
Minkel, J. R. (2007, March). Confirmed: The U. S. Census Bureau gave up names of Japanese Americans in World War II. Scientific American. Retrieved from https://www.scientificamerican.com/article/confirmed-the-us-census-b/
O’Hare, W. P. (2019). Differential undercounts in the U. S. Census: Who is missed? Cham, Switzerland: Springer Open.
O’Hare, W. P. (2020). Understanding who was missed in the 2010 census. Washington, DC: Population Reference Bureau. Retrieved from https://www.prb.org/understanding-who-was-missed-in-the-2010-census/
Okamura, R.( 1981). The myth of census confidentiality. Amerasia Journal, 8, 111–120. https://doi.org/10.17953/amer.8.2.jk115283m54v4313
Reamer, A. (2018). Census derived datasets used to distribute funds. Counting for Dollars 2020, Report #4. Washington, DC: George Washington University. Retrieved from https://gwipp.gwu.edu/sites/g/files/zaxdzs2181/f/downloads/Counting%20for%20Dollars%20%234%20Census-derived%20Datasets%20rev%2005-19.pdf
Reamer, A. (2019). Comprehensive accounting of census guided federal spending (FY17). Counting for Dollars 2020, Report #7. Washington, DC: George Washington University. Retrieved from https://gwipp.gwu.edu/counting-dollars-2020-role-decennial-census-geographic-distribution-federal-funds
Ruggles, S., Flood, S., Goeken, R., Grover, J., Meyer, E., Pacas, J., & Sobek, M. (2019). IPUMS USA: Version 9.0 [data set]. Minneapolis, MN: IPUMS. Retrieved from https://doi.org/10.18128/D010.V9.0
Ruggles, S., & Magnuson, D. L. (in press). Census technology, politics, and institutional change, 1790–2020. Journal of Social History.
Schneider, M. (2019, December 9). Census Bureau head says IT concerns are being remedied. Federal Times. Retrieved from https://www.federaltimes.com/it-networks/2019/12/09/census-bureau-head-says-it-concerns-are-being-remedied/
Sullivan, T. A. (2020). Coming to our census: How social statistics underpin our democracy (and republic). Harvard Data Science Review, 2(1).
Utah v. Evans. (2002). 635 U. S. 452. Retrieved from https://caselaw.findlaw.com/us-supreme-court/536/452.html
U. S. Bureau of the Census. (2012, May 22). Census Bureau releases estimates of undercount and overcount in the 2010 census [Press release]. Washington, DC: Bureau of the Census. Retrieved from https://www.census.gov/newsroom/releases/archives/2010_census/cb12-95.html
This article is © 2020 by Reynolds Farley. The article is licensed under a Creative Commons Attribution (CC BY 4.0) International license (https://creativecommons.org/licenses/by/4.0/legalcode), except where otherwise indicated with respect to particular material included in the article. The article should be attributed to the authors identified above.