The Supreme Court is expected to decide in coming weeks whether next year’s decennial census will include a new and controversial question on citizenship added by the Trump administration. The question is already being asked each year of a small fraction of the nation’s population in the Census Bureau’s American Community Survey.

Hispanics with no citizenship answer on the 2017 American Community Survey

Mexican, Central American

Other Hispanics

High density area with both

1 dot=500 people

Mexican, Central American

Other Hispanics

High density area with both

1 dot=500 people

New York

Chicago

D.C.

Denver

Los Angeles

Phoenix

Atlanta

Houston

Miami

Mexican, Central American

Other Hispanics

High density area with both

1 dot=500 people

Seattle

Minneapolis

Boston

Detroit

New York

Chicago

San

Francisco

Salt

Lake

City

D.C.

Denver

St. Louis

Los Angeles

Phoenix

Atlanta

Dallas

Houston

Miami

In the latest ACS, in 2017, the citizenship question went unanswered by about 1 in 12 Hispanics, a far higher rate than that of whites. Where these non-responses occurred can tell us where millions of Hispanics would probably be missed if the same question were added to the 2020 Census.

Hispanics from Mexico and Central America, areas targeted by the Trump administration’s immigration policies, were most likely to skip the citizenship question, so their neighborhoods would be most affected by the undercount.

The question is simple: “Is this person a citizen of the United States?”

But those words would lead to a 2020 undercount of 6 million Hispanics, or about 12 percent of the Hispanic population, according to a study published this spring by Harvard’s Shorenstein Center on Media, Politics and Public Policy.

“It’s just a really simple experiment where individuals are receiving a census form that looks remarkably similar to what people actually receive in 2020,” said Bryce Dietrich, a Shorenstein research fellow. “And half the time that form includes a question about household members’ citizenship, and in the other half of the time the question is not present.”

The Washington Post worked with Dietrich to estimate where the Hispanic undercount, as a result of the citizenship question, would have the biggest impact, distributing it by state using the citizenship non-response data for Hispanics in the ACS. For example, California accounted for 28 percent of the nation’s unanswered ACS citizenship questions, so it was assigned that share of Shorenstein’s estimate of the overall Hispanic 2020 undercount.

Hispanic undercount and its share of the total state population

1%

2

3

4%

WA

OR

ID

NY

NV

NE

IN

IL

CO

KS

CA

NC

OK

AZ

NM

TX

FL

1%

2

3

4%

Wash.

Ore.

N.Y.

Idaho

Mass.

2.3%

R.I.

Conn.

Nev.

Neb.

N.J.

2.7%

Ind.

Ill.

Colo.

Md.

Calif.

3.4%

Kan.

4.6%

N.C.

Okla.

Arizona

N.M.

4.9%

3.6%

Texas

3.9%

Fla.

2.7%

1%

2

3

4%

Wash.

Ore.

N.Y.

Idaho

Mass.

2.3%

R.I.

Conn.

Nev.

Neb.

N.J.

2.7%

Ind.

Ill.

Colo.

Md.

Calif.

3.4%

Kan.

4.6%

N.C.

Okla.

Arizona

N.M.

4.9%

3.6%

Texas

3.9%

Fla.

2.7%

Undercount Share of pop.
Arizona 362,496 4.9%
California 1,843,035 4.6%
Texas 1,147,650 3.9%
New Mexico 76,333 3.6%
Colorado 201,467 3.4%
Nevada 85,698 2.7%
Florida 601,803 2.7%
New York 454,095 2.3%
New Jersey 182,101 2.0%

The Trump administration has argued that the citizenship question is needed to enforce voting rights laws, even as it has acknowledged that it would depress census response rates. Last week, reports emerged that the proposal was designed expressly to benefit non-Hispanic whites and Republicans.

Opponents of the question argue that it would make the census less accurate, because minorities, especially Hispanics, will be less likely to answer some or all questions because of concerns about increasing immigration enforcement.

“It would add to a severe undercount of the population in Arizona,” said state Rep. Diego Espinoza (D). Arizona’s undercount could approach 5 percent of the total state population. “And I stress that, because for every person that is not counted, that is federal dollars that will not come to support Arizonans on a day-in-and-day-out basis.”

Lower Hispanic response rates related to citizenship also would drive up census costs by hundreds of millions of dollars, according to the Shorenstein study. Costs increase because when no one answers for a home or answers incompletely, the Census Bureau attempts a series of follow-up contacts, including sending census takers to attempt in-person interviews.

The impact of an undercount

The 2020 Census is a unique American project, aiming to count every single person in the United States at the address where they live. The public data will go down to the block level, significantly more geographic detail than the neighborhood data from of the American Community Survey, an annual sample of about 2 percent.

The census is the only data with that geographic detail, and it’s used exclusively to draw political districts. A significant Hispanic undercount would change how many congressional seats go to some states, starting with the 2022 elections.

California, with a projected Hispanic undercount of 1.8 million people, would lose two of its 53 House seats under this scenario. Texas, with an undercount of 1.1 million, would gain one rather than two seats from its overall population growth. The undercount would erase Arizona’s projected gain of one seat. On the other hand, Montana could gain a second congressional seat, and three states could avoid losing a seat: Alabama, Minnesota and Ohio.

Projected change in Congressional seats from ...
Would lose seatsHispanic undercountPop. changeNet
California-2-2
Arizona-1+1
Texas-1+3+2
Would gain seats
Alabama+1-1
Minnesota+1-1
Ohio+1-1
Montana+1+1

Similar changes would hit state and local districts across the country when they are redrawn using the same data.

Another set of historic changes could come if governments won the right, probably involving another Supreme Court fight, to use any new detailed census citizen data to draw districts based on citizen population only. Across the nation, total population data is used now to draw districts.

The census plays a role in hundreds of billions of dollars in spending each year across more than 320 federal programs. While the census is short — fewer than 10 questions about each person — its expansive reach makes it the basis of population and housing estimates for the next decade and informs the design of more wide-ranging government household surveys.

Fifty-five of the largest federal programs distributed over $880 billion in the 2016 fiscal year, according to research by Andrew Reamer, a research professor at George Washington University. Half of these federal programs use a state or region’s population or per-capita income to allocate funds, making them particularly vulnerable to a census undercount. The largest of these programs, Medicaid, distributed $361 billion to states in 2016.

55 large census-reliant federal programs distributed $883 billion in 2016

Programs especially sensitive to an undercount

SNAP

$66B

Medicaid

$361 billion

Fed. Direct

Student

Loans

$94B

Medicare

Part B

$60B

Pell

Grants

$26B

Special education

grants

Title I grants

to school districts

CHIP

Programs especially sensitive to an undercount

Medicaid

$361 billion

Federal Direct

Student Loans

$94B

SNAP

$66B

Medicare

Part B

$60B

Pell

Grants

$26B

Special education

grants

Title I grants

to school districts

CHIP

$14B

$14B

$12B

Medicaid

$361 billion

Federal Direct

Student Loans

$94B

SNAP

$66B

Medicare

Part B

$60B

Pell

Grants

$26B

Special education

grants

Title I grants

to school districts

CHIP

$14B

$14B

$12B

Programs especially

sensitive to an

undercount

Medicaid

$361 billion

Federal Direct

Student Loans

$94B

SNAP

$66B

Medicare

Part B

$60B

Pell

Grants

$26B

Special education

grants

Title I grants

to school districts

CHIP

$14B

$14B

$12B

Programs especially

sensitive to an

undercount

An undercount changes the distribution of funds, not the amount of funds, creating winners and losers among states. Reamer estimates that a 5.8 percent undercount of noncitizens and Hispanics would cause California to lose an estimated $10.6 million annually in WIC funds, a supplemental nutrition program for women, infants and children. Maryland would gain nearly $300,000.

A citizenship question would hit Texas state funds especially hard, Reamer found. The state would lose an estimated $378 million annually in Medicaid funds alone. Conversely, Illinois would gain over $9 million in annual Medicaid funds.

“The data that forms the census are the foundation for the relative functioning of the U.S. economy and government at all levels,” says Reamer. Census and its derived data provides the most accurate and reliable demographic, housing and economic data.

The data is a tool for local governments in decisions including budgeting, disaster response, land-use planning, and measuring economic or environmental impacts. Researchers rely on it to study topics as divergent as the spread of diseases and gentrification. For businesses, the data helps decide where to set up shop, who their prospective customers are, what products to launch and how to market them.

Members within all of those groups have voiced concern over the inclusion of a citizenship question and the potential undercount. The current Supreme Court case was brought, in part, by New York state, 16 other states, seven cities and the U.S. Conference of Mayors. Friend-of-the-court briefings have been filed by research and business groups alike, including the American Statistical Association, polling firm Nielsen and ride-hailing company Uber.

Maps by Laris Karklis.

How we did this story

While the U.S. census conducted each decade typically undercounts minorities by a few percent, this story is focused on the Hispanic undercounts that would probably occur solely because of the addition of a question about citizenship. The citizenship question hasn’t appeared on the complete count census since 1950. It and other questions were later removed to improve accuracy and efficiency. Citizenship has remained on other surveys of population samples.

The Shorenstein study that estimated a national Hispanic undercount of 6 million, or 12 percent, was based on a survey of more than 9,000 people. The survey mimicked a 2020 Census form. Authors Matthew Baum, Bryce Dietrich, Rebecca Goldstein and Maya Sen compared results from forms with and without the citizenship question.

The state undercount estimates were produced using allocation rates for Hispanics on the citizenship question derived from the 2017 American Community Survey public-use microdata. Allocation is a Census Bureau process for filling in missing survey responses, and published data reflects when question responses are allocated.

To explore the impact on congressional reapportionment, The Washington Post projected 2020 populations for the states, based on Census Bureau estimates of population change since 2010, and compared how seats would be apportioned in 2020 with and without state adjustment for the estimated Hispanic undercount. How seats are allocated among states can be affected by small differences in relative population change.

Data on federal distribution of funds reliant on census and the estimated cost to select states of an undercount is from Counting for Dollars 2020, including report No. 5, “Distribution of Funding from 55 Large Census-guided Programs by State,” and expert reports to the Southern District of New York and Maryland district courts.