Symposium: Evenwel v. Abbott and the Constitution’s big data problem

Nathaniel Persily is the James B. McClatchy Professor of Law at Stanford University Law School. (For a more developed version of the argument presented here, please see Nathaniel Persily, The Law of the Census: How to Count, What to Count, Whom to Count, and Where to Count Them, 32 Cardozo L. Rev. 755, 775-81 (2011).)

The plaintiffs in Evenwel v. Abbott deserve credit for highlighting an unappreciated feature of our system of representation and exposing a gap in the jurisprudence of one person, one vote. Their constitutional argument, however, is academic, at best, and at worst, dangerous and destabilizing for the redistricting process and the U.S. Census. Leaving aside the weighty and contested philosophical arguments concerning the proper basis for legislative representation, a constitutional rule of redistricting based on citizenship or eligible voters presents insurmountable logistical difficulties. These difficulties are of constitutional import because no national database of citizenship exists at the level of granularity necessary to draw legislative districts that comply with one person, one vote.

First some basics on the redistricting process. With each decennial census virtually every level of government in the country – from Congress to state legislatures to school boards and any other manner of local government – must redraw their districts to adjust to population shifts the Census reveals.  The redistricting dataset released by the Census (the so-called PL 94-171 datafile) and used by virtually all jurisdictions in the United States includes information regarding the total population and voting-age population, as broken down by race. The data are derived from the Census form we now receive every ten years, which only asks questions as to race/ethnicity, age, gender, and owner/renter status of the home.   It does not ask anything about citizenship, voter registration, or anything comparable.

From where, then, do all of the data concerning citizenship we hear so much about come? They come from a yearly survey of 2.5% of American households called the American Community Survey (ACS). As one of many surveys conducted by the Census, the ACS contains a host of questions covering everything from a house’s plumbing, internet connectivity, and tax rates, to its occupants’ citizenship, marital, employment, and veteran status. The Census releases results from each year’s ACS, as well as averages for the previous three- or five-year period. Whenever you see citizenship rates reported in the news, the ACS is usually the source. Unlike the decennial census, the ACS is not mandated by the Constitution, and Congress could terminate or defund it at any time, as some representatives have threatened.

The ACS does not produce a dataset, however, that is appropriate for redistricting. First, because they sample such a small share of the population, the yearly surveys do not produce data at sufficient granularity for redistricting, namely the census block or neighborhood level, akin to the decennial census. Just like other public opinion polls, a survey like the ACS can paint, in broad outlines, the profile of a state or large subsections of it. But as you move down to your neighborhood, a 2.5% national survey will often not include enough people from your immediate vicinity in order to create accurate estimates of citizenship rates. Any such estimates, therefore, come with margins of error, just like other public opinion polls.

Because any yearly ACS is too small and inaccurate to be used for redistricting purposes, three- and five-year averages might give better estimates of the citizenship rates at smaller levels of geography. Even those statistics, though, are not released at the census block level, which is the building block for redistricting plans. Even if they were, they do not provide estimates of the citizenship population at the time of redistricting – that is, they are averages of earlier years and give estimates of what the citizenship rates have been, not what they are at the time districts need to be drawn. Unlike the decennial census, which presents a snapshot of the population, the ACS is more like a movie, highlighting general movements in the population data viewable only on the big screen. (To make matters even more complicated, those ACS averages are derived from the earlier census geography – i.e., the census tracts as they were drawn for the census ten years earlier – not the census immediately preceding the redistricting.)

The other population base for redistricting put forward by the Evenwel appellants is registered voters. With this suggestion, they appear on more familiar ground, given that the Supreme Court in Burns v. Richardson (1966), upheld, though clearly did not mandate, the use of registered voter data for redistricting purposes. Burns somewhat reluctantly sanctioned the use of registered voters as the population base, only because in that context, “registered voters was chosen as a reasonable approximation of both citizen and voter population.” Requiring as a constitutional rule redistricting on the basis of equal numbers of registered voters comes with its own set of serious problems, however.

First, as with a “citizen census,” the Constitution does not require voter registration. In fact, North Dakota does not require its voters to register and nothing prevents other states from following suit. Another fourteen states allow for Election Day registration; for them, the number of registered voters in a given jurisdiction can fluctuate considerably depending on whether one uses data before or after the election.

Indeed, the extreme variability in voter registration data for any jurisdiction is a key reason such data should not be used for redistricting purposes. The size and accuracy of a voter registration list will often depend on the proximity to an election. When voter registration drives are in full force, especially for an impending presidential election in a battleground state, the number of registered voters can increase substantially. If the next election is a congressional election or even an off-year local election, the number of registered voters in a jurisdiction might be smaller or out of date. The voter registration list is, in this sense, a moving target. If it is to be used for redistricting, the decision as to which voter list to use, and when, can lead to dramatically different outcomes. (Expressing caution as it blessed the use of registered voters, Burns recognized this problem: “[F]luctuations in the number of registered voters in a given election may be sudden and substantial, caused by such fortuitous factors as a peculiarly controversial election issue, a particularly popular candidate, or even weather conditions.”

The inaccuracy of voter rolls, moreover, has been a continuing source of frustration for election administrators and should caution against using voter registration as a population base for redistricting. As the Report of the Presidential Commission on Election Administration, for which I served as Senior Research Director, detailed, roughly eight percent or sixteen million voter registration records are invalid or significantly inaccurate. But the degree of inaccuracy varies considerably by state and over time. For some states, as many as fifteen percent of the names on their voter registration list constitute “deadwood”: names of people who have likely moved or died since they were registered at the given address. As a result, states conduct periodic purges of their voter rolls, as regulated by the National Voter Registration Act (NVRA). Whether a jurisdiction redistricts on the basis of a list before or after a purge can lead to different results. To guard against disenfranchisement by way of purging, furthermore, states retain different lists of voters: an active voter list of those who have regularly voted and an inactive voter list, which includes people who have missed voting in the last few elections. As with the ACS citizenship data, mandating redistricting on the basis of “registered voters,” as the Evenwel appellants urge, does not settle the question concerning the proper population base for redistricting: It would require further decisions as to which voter registration list, at which time.

These points concerning the inaccuracy and variability of the potential alternative data sources for redistricting will likely receive less attention in Evenwel than familiar constitutional arguments concerning the proper interpretation of the Fourteenth Amendment. But these seemingly technical and logistical issues should foreclose the constitutional debate.   Current data on citizenship or on registered voters is simply too inaccurate or contested to be used in redistricting. Unless the Justices are prepared to mandate a new kind of “citizen census” or to constitutionalize the voter registration process, then they should leave it to the states to draw their districts using the most accurate data available.