The Supreme Court and the Census (2 of 3)

As noted by Nathaniel Persily, “In the Constitution itself, the census is ‘about’ representation, money, and race, so we should not be surprised to learn that courtroom controversies over the census have persisted with respect to these three themes.” For a brief review of the history of census litigation and controversy see Swanson and Walashek. This post covers two recent Supreme Court cases that illustrate the political significance of the decennial census.

Context

A significant issue in court cases deals with how the census should count. The issue of how to count derives in part from the Census Bureau’s post-census estimation of the accuracy of the decennial enumeration. The first such analysis occurred after the 1940 census. It estimated the black undercount at 8.4%. The 1940 census counted 12.9 million blacks but the actual number was more like 14.1 million. The nonblack (white) undercount was estimated at 6.3 million (5%). The total undercounts for all races was 7.5 million.

With the mailout-mail back census of 1960 (the 1950 census was the last one done completely by door-to-door enumerators) the Census Bureau worked diligently to try to reduce the undercount. Remember, also, the 1962 Supreme Court Baker v. Carr decision initiated the reapportionment revolution. It was also in the 1960s that federal programs proliferated often using census data to distribute federal funds. About $40 billion dollars in federal funds were distributed to state and local governments using census data after the 1980 census. By 1989, the amount of funds so distributed climbed to nearly $60 billion. Throughout the 1990s an average of $56 per person annually was distributed using at least in part census data. (Swanson and Walashek)

In 1950, the white undercount was estimated to be 3.8% and black, 7.5%. The undercounts continued to decline for the 1960 census, 2.7% for whites and 6.6% for blacks; for 1970, 2.2% for whites and 6.5% for blacks; for 1980, 0.8% for whites and 4.5% for blacks. However, the 1990 census reversed this trend. The white undercount rose to 1.3% and the black undercount rose to 5.7%. For the first time, the Census Bureau provided an estimate for the Hispanic undercount, 5.0%, and for American Indians and Alaska natives, 4.5%. (see GAO/GGD-98-103) Throughout these decades the undercount problem centered primarily in urban areas.

The undercount issue rose in prominence as the federal government distributed an increasing amount of federal to states and localities using census data. For example, the city of Hartford, Connecticut has the sixth-highest rate of undercounting among cities in the 1990 census. The undercount was about 6,500 persons. This undercount resulted in the city losing a seat in the state legislature and receiving about $5 million less each year from the federal government. Receipt of state funds was also affected by the undercount. At this time, Hartford was ranked as the eighth poorest city in the country.

The increased undercount of the 1990 census generated what would become a significant Supreme Court decision. In 1991 Congress enacted legislation directing the Commerce Secretary to use the National Academy of Sciences to study methods that would reduce the undercount. Two of the three panels created by the academy concluded “that the differential undercount cannot be reduced to acceptable levels at acceptable costs without the use of . . . a statistical sampling procedure that adjusts census results to account for undercount in the initial enumeration.” (Swanson and Walsh).

Subsequently, the Census Bureau decided to use statistical sampling to supplement the traditional enumeration methods. After this plan was put forward, however, the Republican-controlled Congress enacted legislation prohibiting the use of sampling in determining the apportionment of House seats. Republicans feared the loss of House seats and opposed sampling because most of the persons undercounted tended to vote Democratic. Democrats, however, supported sampling because it probably would increase the number of Democratic representatives in the House.

President Clinton vetoed the bill. Congress then enacted legislation, which Clinton signed, that directed the Census Bureau to consider proposed methodologies for conducing the 2000 Census. The Census Bureau then announced plans to use two forms of statistical sampling in the 2000 Census to address the undercount problem. Two objectives generated this decision. One was reduced costs. The costs of the census had ballooned dramatically since 1950 even discounting inflation and population growth. The efforts at non-response follow-up generated much of this increased cost. The other objective was to increase the accuracy of the census count by reducing the differential undercounts.

Can the Census Bureau use sampling in enumerating the population?

In 1998, two law suits challenged the legality and constitutionality of the proposal. Both district courts, the District Court for the Eastern District of Virginia and the District Court for the District of Columbia, ruled the Bureau’s plan violated the Census Act and permanently enjoined the plan to use sampling for purposes of apportionment. The Supreme Court combined the cases on appeal in the Dept. of Commerce v. United States House of Representatives. The Court decided the case in January 1999.

Justice O’Connor wrote the majority opinion joined by the four other conservative justices, Rehnquist, Scalia, Kennedy, and Thomas. The Court ruled that the Census Act prohibited the use of sampling for purposes of reapportionment. In making this decision the Court had to reconcile two seemingly disparate parts of the Census Act. Section 141(a) of the act said the Secretary of Commerce shall take a decennial census of the population “in such form and content as he may determine, including the use of sampling procedures and special surveys.” Section 195 states, “Except for the determination of population for purposes of apportionment of Representatives in Congress among the several States, the Secretary shall, if he considers it feasible, authorize the use of the statistical method known as ‘sampling’ in carrying out provisions of this title.”

Perhaps in layman’s terms, the Court had to decide something like this: A family has three cars, a Buick, Ford, and Chevrolet. The senior son is responsible for getting the cars inspected. The father says to the son, “except for the Chevrolet when you take the cars to get them inspected you shall have a second set of keys made for each car.” Is the son prohibited from making a second set of keys when he takes the Chevrolet in for state inspection? The Court ruled yes. If the father wanted the Chevrolet to get a second set of keys he would have said so. The Court’s reasoning was based on its interpretation of how the census had been historically undertaken.

After the Department of Commerce case, the Census Bureau planned to release two sets of data from the 2000 census. One would be the enumerated head count that must be used to apportion House seats among the states. The second would be the statistically adjusted data using sampling in part. This adjusted data set could be used for redistricting within states and for other purposes. However, due in part to the advent of a Republican administration that was opposed to sampling, it took a court order to have the adjusted data released.

The adjusted data showed that the 2000 census missed about 3.3 million persons. But the Census Bureau had a disclaimer, saying that “the estimates dramatically overstate the level of undercoverage” of the census. Many saw the disclaimer as a political move, in part because the Census Bureau never explained its rationale for the disclaimer. Others opined that 6 million people were missed in the initial count but more than 3 million were counted twice. The duplication was thought to be the dramatic increase in vacation and second homes. An August 2001 study of the census by the Census Monitoring Board estimated that a 3.3 million undercount would mean a loss of $4.1 billion in federal funds for 31 states; (see LaFluer)

What constitutes sampling?

The 2000 census provoked another Supreme Court case, Utah et al. v. Evans. The State of Utah challenged the Census Bureau’s use of “hot-deck” imputation in the enumeration of the country’s population. Imputation is a methodology that infers the population count and characteristics of missing units or households when no determination can be made after the standard non-response follow-up. In these cases, the Census Bureau infers that the address or unit about which it is uncertain has the same population characteristics as those of its geographically closest neighbor of the same type (apartment unit or single-family dwelling, for example).

Utah sued the Department of Commerce Secretary (Evans) because the imputation increased North Carolina’s population by 0.4% while increasing Utah’s population by only 0.2%. Consequently, Utah lost a House seat while North Carolina gained a House seat. Utah sought an injunction compelling the Department to change the official census results. North Carolina intervened in the case as well. The District Court for the District of Utah found for the Department. Utah appealed to the Supreme Court.

In July 2002, the Court decided on a 5-4 vote that the Census Bureau’s use of hot-deck imputation did not violate (1) the Census Act and (2) the Constitution’s Enumeration Clause. The majority opinion was written by Justice Breyer and was joined by justices, Stevens, Souter, Ginsburg, and Chief Justice Rehnquist. The majority opinion determined that hot-deck imputation was not like the statistical method known as sampling, which is not permitted for enumeration per earlier Court decisions. The surprise member of the majority was the Chief Justice who had been expected to decide in favor of the respondents based on his position in the House of Representatives v. Department of Commerce case.

The adjective “hot-deck” refers to the fact that imputation uses data from the current census. If the Census Bureau used data from the past censuses it would be referred to as “cold-deck” imputation. The example used by the Court to describe imputation deals with a librarian who wants to count all the books held by the library. The librarian goes about counting all the books on the shelves. In counting the books the librarian sees that some shelves have spaces between the books. The librarian looks at the books on either side of the gap and estimates an approximate average width among those books. Based on this, the librarian infers or imputes (assigns a value by inferring) that the space holds one, or two, or some other number of books.

The majority opinion declared that imputation differed from sampling in three essential characteristics.

• Sampling collects data from a small subset of a population to count the entire population. Imputation is a method of processing data – it gives a value to missing data and does not collect the data.
• Imputation does not rely on the same statistical method used for sample selection
• Its immediate objective is to determine the characteristics of the missing data rather than extrapolating characteristics from the sample to the entire population.

The dissenting justices produced several dissenting opinions. Justice Scalia disagreed that the appellants had standing. He therefore would “vacate the judgment of the District Court and remand with instructions to dismiss for want of jurisdiction.”

Justice Thomas, joined by Justice Kennedy, agreed that imputation did not violate the Census Act. But argued that the Constitution’s Enumeration Clause prohibited the use of imputation for enumerating the population for purposes of apportionment. Citing historical examples of censuses, from Biblical times through to censuses in France and England in the 18th century and colonial America and how dictionaries used at the time of the framing of the constitution, Thomas determined that the framers of the Constitution meant person counting when they settled on the use of the word enumeration. Such a definition was used, argued Thomas, to avoid political manipulation in undertaking the censuses.

In her dissent, Justice O’Connor saw imputation as a form of sampling. She opined that the Census Bureau used hot-deck imputation “to make the same statistical inferences it could not make through more transparent reliance on sampling. In her words, “the Bureau’s use of imputation was a form of sampling” and therefore violated both the Census Act and the Constitution’s enumeration clause.

Comment

The two Court cases illustrated in this post show the political importance of the decennial census. Just about every decennial census provokes one or more legal suits. These two Court cases focus on important constitutional issues. They also show a political party divide.

Republicans steadfastly support maintaining a conservative, traditional view that opposes the use of sampling and advocates the use of actual enumeration in the decennial censuses. They take this position because they believe, accurately or not, that any move away from enumeration will increase minority and urban populations that tend to vote for Democrats.

Democrats advocate for the use of sampling in part because it generally results in a more accurate estimate of the country’s population. In doing so, they expect the population increase due to minimizing the differential undercount will aid the party electorally. To an extent, the positions on these cases can be seen as either supporting or opposing voter suppression.

The politics of the census spills over to the Supreme Court. Conservative justices support strict enumeration and oppose sampling. Liberal justices tend to support sampling and the importance of obtaining more accurate population estimates. Interestingly, a parallel exists between the Utah v. Evans case the conservative Chief Justice Rehnquist sided with the liberal justices to support hot-deck imputation, resulting in a five-vote majority. In the 2020 Census citizenship case, the conservative Chief Justice Roberts sided with the liberal justices to gain a 5-vote majority that determined the only rationale given by the Commerce Department was pretextual, thereby canceling the adoption of the citizenship question.

The next and final post of this series of posts on the census discusses the opinions of the Court regarding the 2020 Census and the issue of including a question on citizenship.