The Administration is Lying About the Census

Joseph Fishkin

The Commerce Department has announced that it is adding a question about citizenship to the 2020 Census, for the first time in seventy years.  There has been a lot of speculation about possible political motivations for this action.  It is difficult to know exactly what motivates government actors whose deliberations are not public.  But it is possible to know one thing: the government’s sole stated reason for adding the question—improving enforcement of Section 2 of the Voting Rights Act (VRA)—is false.  It is not the real reason.

 “Lying” is a strong word.  Many falsehoods are based on mistakes, confusion, carelessness, wishful thinking, and so on.  But today’s situation puts me in mind of an evidentiary pattern that arises constantly in employment discrimination law: The plaintiff thinks something nefarious has occurred, but has no direct proof; the defendant offers a legitimate, non-discriminatory reason for the disputed action; all the plaintiff can prove is that the proffered reason is false.  It’s then, ordinarily, up to the jury to sort out what really happened.  But the fact that the defendant offered up a reason that was definitely false is significant.  It raises a potential (contestable) inference that it may have been a lie, covering up something the defendant could not admit.

My sole aim in this post is to explain why the government’s stated reason for adding the citizenship question is false.  To understand this, you need to understand something about the role Census data plays in redistricting.  It plays two completely different roles.



At the bottom of our law of redistricting, since the reapportionment revolution of the 1960s, we have the foundational principle of one-person-one-vote: Each district must have the same population.  The U.S. Constitution requires that “Representatives shall be apportioned among the several States according to their respective numbers, counting the whole number of persons in each State.”  Crucially, it says “whole number of persons,” not citizens.  To make sure each state gets the right number of representatives, and then after that, to draw district lines within states, we need a count of all persons, and it needs to be exact.  For this, we use the Census.


The Voting Rights Act, Section 2, prohibits racial vote dilution, which is drawing maps where members of one racial group have less of an opportunity than others to “elect representatives of their choice.”  This is a much richer and messier concept than equalizing district populations.  To some extent it involves predicting election outcomes.  We must ask questions such as: Is voting racially polarized, so that white voters vote as a bloc to defeat the representatives minority voters choose?  (Section 2 is triggered only when the answer is yes.)  We need to analyze degrees of racial polarization, rates of voter eligibility, registration, turnout, and other factors, in order to determine whether a given district is one where a given minority group of voters will, in fact, be able to “elect representatives of their choice.”  No single data source will ever provide all of the information about all of those factors.  Many of them—turnout, for instance—are epistemically impossible to predict with anything like the degree of precision that is involved in one-person-one-vote.  That is ok.  The goal is not to nail a perfect prediction of an election outcome, which would be impossible.  It’s just to determine whether or not a district gives minority voters an opportunity to elect their candidates.

We use citizenship data as part of the Section 2 analysis.  Why?  Because it’s one piece of data that helps predict the future outcome of an election in the district.  Suppose a district is racially polarized, with Hispanic voters voting one way and white voters a different way.  Now suppose the Hispanic part of the district has more children, more non-citizens, a lower rate of registration among the citizens, and lower turnout among the registered voters.  In that case, you might need to draw a district that was well over 50% Hispanic, in total population, in order to make it truly a district where Hispanic voters could elect a representative of their choice.  We use citizenship data from the annual American Communities Survey (ACS) to calculate the citizen voting age population (CVAP), as an important middle step in this calculus.  Just how high a CVAP do we need?  This is inexact; it depends on the predicted size of gaps in registration and turnout between the Hispanic voters and the white voters.

The American Communities Survey (ACS) data is perfectly adequate for Section 2 analysis.  It’s true that this data is not as granular as census data.  It’s not as precise.  But it doesn’t need to be.  When we are doing Section 2 analysis, what we need is to predict whether a district gives a minority group a good chance of winning—a question that is necessarily approximate.  ACS citizenship data is not perfect, but it’s far more accurate than our ability to predict differences in voter registration and turnout in the future over a decade.  Thus, tightening up the precision of the citizenship data, which is what the DOJ argues will be gained by moving the citizenship question to the Census, is not in fact going to materially improve our ability to tell which districts are ability-to-elect districts and which are not.

That is why no DOJ official, including political appointees of both parties, from the inception of the enforcement of Section 2, had ever called for adding a citizenship question to the basic Census form, before the current Justice Department called for it.  This step simply is not going to materially improve our ability to enforce Section 2.

What introducing the citizenship question will do is screw up the basic count in predictable ways.  This will, on net, make Section 2 calculations less accurate.  It will increase the nonresponse rate in an uneven way, producing a less accurate picture of where the people are, not only non-citizens but also citizens.  This is likely to have real effects on he distribution of representation and political power.*  The administration knows that a lot of mixed-status families (households with some citizens and some non-citizens), for instance, may refuse to answer the Census at all, and although Wilbur Ross makes much of the fact that the magnitude of this problem is unknowable, from a partisan Republican perspective, it’s all gravy.  The more immigrants refuse to answer the door, the more political power flows toward places with fewer immigrants.  I strongly suspect (and have argued on this blog before) that much of the point of this process may be to set up the next iteration of the Evenwel case—the effort by some Republicans to change the basis of representation, at least at the state level, so that only citizens count, not all persons.

As I discuss in this essay, the U.S. Constitution casts a long shadow over that argument.  I am not at all sure that a government that does not even purport to represent all the people is consistent with our constitutional design.  But that is the next fight.  For today, without knowing the precise combination of reasons why the government is doing what it is doing, I simply want to make it as clear as possible, for the record, for journalists, and for anyone trying to understand this somewhat intricate issue, that the government’s claim that adding a citizenship question will aid Section 2 enforcement is false.  The claim does not hold up.  It is a pretext.


*I edited the first sentences of this paragraph in response to an email from Justin Levitt.
Posted 10:30 PM by Joseph Fishkin [link]