African Americans, Hispanics, and other minority populations are in danger of losing representation in Congress as well as their share of more than $400 billion a year in federal funds for health care, education, job training, and community development. That possibility should get anyone’s attention, yet few have noticed that it will be the likely result if Congress cuts the budget for the U.S. Census Bureau to the extent it now threatens to do.

The Constitution requires a decennial census to determine congressional apportionment, and federal law relies on the numbers to allocate funds among states and localities. Historically, the census has missed large numbers of people in poverty and racial and ethnic minorities. By the 2000 and 2010 censuses, however, the national undercount had dropped to less than 2 percent, due primarily to the Census Bureau’s dogged determination to walk America’s streets and knock on the doors of the roughly 100 million U.S. residents who didn’t mail back their forms. Racial and ethnic minorities were still more likely to be missed than whites. But the Census Bureau could not have reduced the disparity in counting minorities without budgetary support.

Now, Congress is insisting that the Census Bureau spend less preparing for and conducting the 2020 census than it did on the 2010 census, even though the U.S. population is expected to have grown by more than 25 million people by 2020. The bureau has chosen not to fight this directive, which census experts call delusional. Instead, the bureau has embarked on a high-risk strategy to save $5 billion by rolling back door-to-door canvassing and conducting a largely electronic, Internet-based census.

Adoption of new information technologies would be good for the census and likely would save money. But the bureau’s all-encompassing overhaul may be too ambitious and complex to pull off in the time remaining, especially given the resource constraints Congress has imposed. Delays in the bureau’s schedule could result in “another HealthCare.gov,” warned Carol Cha, lead technology expert for the Government Accountability Office, in recent Senate testimony.

What’s more, civil-rights advocates worry that the bureau’s plan won’t be as effective in reaching communities of color and immigrant populations even if all the technology works well. About a quarter of all those counted in the 2010 census came from face-to-face non-response follow-up, and this process is critical for reaching people of color who tend to be harder to contact by mail. The bureau is hoping that digital “in-office” canvassing using government and commercial databases will be just as good to prepare for the 2020 census, but that is as unproven as the technology itself.

Success is possible only if Congress makes the necessary upfront investment to develop, test, and refine the bureau’s new systems. As Sen. Barbara Mikulski (D-MD) puts it, “We have to spend millions now to save billions later.” Congress also should be prepared to accept higher costs than expected if tests show savings are coming at the expense of accuracy.

Yet the Census Bureau already has had to delay or cancel field tests because of budget shortfalls, and more cuts could be on the way. In May, the House voted on a mostly party-line vote to cut President Obama’s 2016 budget request for the 2020 census by more than a third. And in June, Republicans on the Senate Appropriations Committee defeated a Mikulski amendment to restore funding for the bureau’s information-technology overhaul by adding back the $360 million her colleagues slashed from the president’s budget. Inadequate funding means that the Census Bureau’s activities for 2016, including overhauling the technology system, creating language assistance materials, and field-testing the use of administrative and commercial records for non-response follow-up, will be delayed or derailed entirely.

“We are reaching a crisis point,” said Terri Ann Lowenthal, co-director of the Census Project, a network of organizations that work to support a full and accurate census. “Without adequate funding next year, we will head into the census with what I think is a high probability of an increased undercount in 2020, particularly affecting people of color, low-income households, and immigrant communities.” Other historically undercounted populations, including poor whites, people with disabilities, and children, also likely would be missed in greater numbers.

Adding to this uncertainty, and on top of the technology overhaul, the Census Bureau is exploring significant changes in the way it asks about race and ethnicity, which also need prior testing. The right changes could improve the quality of race and ethnicity data, but at least one approach under consideration—relying on write-in responses instead of check boxes—would do the opposite, according to civil-rights advocates.

One key vehicle for testing is the American Community Survey (ACS), which in the 2000s replaced the “long form” census and is sent to a random sample of 3.5 million households annually. As part of census appropriations, however, House Republicans have added riders that would effectively make response to the ACS voluntary. They also voted for steep cuts to the ACS budget—20 percent below current-year funding—that could force the Census Bureau to cut the ACS sample size, essentially the same outcome as converting to a voluntary survey.

Not only would this cutback compromise testing for new census initiatives, but also less-populated rural communities and smaller racial or ethnic sub-communities could statistically vanish. For example, in 2010, Canada’s conservative government decided to shift from a mandatory to a voluntary long-form census, and the overall response rate fell from more than 90 percent to less than 70 percent. Data on many of Canada’s aboriginal communities then had to be withheld because response rates were too low to meet standards for accuracy. There wasn’t even a budgetary benefit to the switch—the Canadian government actually had to spend more money sending forms to a greater number of households just to achieve an adequate sample size.

The Obama administration listed the cuts to the census and ACS first among numerous objections to the House-passed appropriations bill that funds the Commerce Department (which houses the Census Bureau), the Justice Department, and science agencies. Another showdown on the budget between congressional Republicans and Obama could be coming by December. Yet even if the president ultimately secures additional census funding, a prolonged standoff—especially if accompanied by a government shutdown—would cause further delays that put the 2020 census at even greater risk.

(Photo: AP/Bebeto Matthews)

Endangered species: Census workers in 2010 assemble after a training course in New York. Cuts to Census Bureau funding and plans to reduce door-to-door canvassing for the 2020 threatens to undermine accurate counting of communities of color and immigrant populations.

What’s needed now is an intervention. Worries about an undercount in 2010 prompted philanthropic foundations and other funders to give close to $40 million to national civil-rights organizations and community groups as part of an unprecedented and successful communications campaign to boost census participation. The success or failure of the 2020 census, however, will depend on decisions by Congress and by the bureau over the next several years. That suggests the need to build and mobilize a political constituency that can make the case for adequate funding and watch over implementation to ensure the bureau’s new methods accurately count the poor and communities of color, who are historically undercounted.

“What funders need to understand is we don’t have until 2020,” Lowenthal said. “It’s now or never.”

Why an Accurate Census Matters

People with low incomes, especially people of color, are more likely than others to rent their homes, move more frequently, and live in nontraditional or multi-family dwellings, all of which makes it harder for the Census Bureau to locate and count them. Additionally, recent immigrants often do not speak fluent English and, whether or not they immigrated legally, may avoid providing information to the government for fear it will be used against them.

The 2010 census, though one of the most accurate in history according to the Census Bureau, still missed millions of people and double-counted or incorrectly included millions of others. It undercounted African Americans by 2.1 percent, Hispanics by 1.5 percent, and American Indians (as the census calls them) by almost 5 percent, according to the bureau’s own estimates. Whites, meanwhile, were overcounted by almost 1 percent, as affluent whites with more than one home sometimes submitted duplicate forms. (College students living away from home also accounted for some of the overcount among whites.)

These are impressively low rates in the context of a nationwide population of more than 300 million, but the numbers are still large enough to make a difference in congressional redistricting and the annual distribution of more than $400 billion in federal funds to state, local, and tribal governments. The California Department of Finance, for example, estimated that the Census Bureau failed to count 1.5 million California residents in 2010. With an accurate count, the state would have received a funding boost for everything from health care and education to transportation and economic development. Andrew Reamer, a research professor at George Washington University, estimated that for every California resident not counted in the 2010 census, the state has lost around $1,145 in federal funding each year. By that reckoning, California has lost more than $1.7 billion per year because of the 2010 undercount. (According to PricewaterhouseCoopers, Los Angeles County lost $650 million between 2002 and 2012 in eight federal programs because of the undercount.) Even with fewer resources, the state continues to pay for services for every resident whether they were counted or not, putting additional strain across the board on vital social and economic programs.

Adding to this distortion, prison inmates are counted as residents of the location where they are incarcerated, not where they lived previously. Rural white, conservative districts, where prisons are typically located, therefore get a boost in their population, as well as in federal funds, while people in prison, who are both disenfranchised and disproportionately racial and ethnic minorities, lose representation at their true place of residence.

The Politics of the Undercount

The potential for using budget cuts to produce an undercount of racial and ethnic minorities uncomfortably brings to mind the constitutional compromise between southern and northern states to count slaves as three-fifths of their actual numbers for purposes of congressional apportionment. Before the Civil War, southern states wanted the census count to include all slaves, who of course were not allowed to vote, so as to increase southern seats in Congress and protect the interests of slaveholders. During Jim Crow, southern whites also gained disproportionate representation in Congress through the counting of largely disenfranchised African Americans.

Today’s congressional Republicans, however, seem to have less interest in counting all African Americans or other minority populations. In fact, House Republicans sued the Census Bureau over its plan for the 2000 census to use statistical sampling techniques to effectively eliminate the undercount of racial and ethnic minorities. The Supreme Court, in a 5–4 decision in Department of Commerce v. U.S. House, ultimately blocked the bureau from using these techniques. (Notably, the court’s decision rested on its interpretation of the Census Act, not the Constitution’s census clause, which requires that everyone be counted.)

Republicans could now be rewarded with additional seats if their census budget cuts lead to a higher undercount of racial and ethnic minorities in 2020, as seems increasingly possible. They also might have a freer hand to manipulate electoral maps at every level of government, a strategy they perfected in 2012 with the REDistricting MAjority Project (REDMAP) initiative, which helped consolidate GOP power through legislative and congressional redistricting. That’s because accurate census data on voters’ race and ethnicity are critical for enforcing the Voting Rights Act, to ensure that lawmakers draw fair, representative maps for state legislatures, counties, cities, and school boards as well as the U.S. House of Representatives.

Race and ethnicity data from the decennial census, combined with ACS data, are also used to determine which jurisdictions are required to provide language assistance to non–English speaking voters and to prevent and prosecute discrimination generally, from education to employment to housing. To put it simply, data from the census and ACS are typically the only reliable data on marginalized and minority communities. Without those data, the people who go uncounted become virtually invisible to policymakers. It is far more challenging to allocate resources and provide services to populations that are left out of the census and ACS counts.

Support for an Accurate Census

Fortunately, state and local government leaders are raising their concerns about threats to the census and ACS. Shortly before the House vote on this year’s census appropriations, the so-called Big 7 national organizations representing state and local governments, including the National Governors Association, the National Conference of State Legislatures, and the U.S. Conference of Mayors, underscored the importance of census and ACS data to policymaking in a letter to Congress: “State and local policymakers depend upon this detailed, high-quality data in order to make the right decisions to benefit the citizens and communities they serve.”

Businesses have different concerns than government does, but they rely just as heavily on census and ACS data in deciding where to locate, whether to invest in expansion, and how to meet the needs of their customers and workforce. Twenty-one national business organizations, including traditional Republican allies such as the U.S. Chamber of Commerce and the National Association of Home Builders, wrote to Congress in April that “the broad range of Census Bureau data—including from the decennial census, ACS, and economic surveys—are the foundation for sound decisions that strengthen the private sector’s role as an economic engine vital to thriving communities.”

Another letter from state and local chambers of commerce likewise explained how they regularly assemble and report census and ACS data to attract businesses. ACS data on Tulsa’s labor force, for example, convinced Macy’s to locate a $180 million distribution center in the city, creating 3,500 year-round and seasonal jobs. Businesses in Canada, by contrast, are struggling to determine where to set up shop because they lack reliable information after the country’s long-form census was made voluntary.

Businesses might not locate in undercounted communities if they cannot reliably evaluate the workforce and economic potential, costing those communities jobs as well as stores, restaurants, and services that meet community needs. “Full funding for 2020 census planning is important to ensure an accurate enumeration at the neighborhood level for our increasingly diverse communities,” the regional chambers of commerce wrote.

A Risky Plan

To meet congressional budget demands, the Census Bureau’s technology overhaul takes aim at the costliest aspects of the census: paper processing and printing costs and door-to-door canvassing. By encouraging responses through the Internet and arming census enumerators with electronic devices, the bureau can save data-entry costs and won’t have to print as many forms and other materials. By automating field operations and using government and commercial databases to build the bureau’s mailing list and count non-respondents, enumerators won’t have to knock on as many doors. The Census Bureau anticipates an Internet-based response rate of 47 percent, with projected savings of up to $400 million.

Although this plan has promise at least in theory, the technology is at high risk of failure, according to the GAO. The new electronic methods, while cheaper, may not provide the same level of accuracy as the old system, which emphasized face-to-face contact. The big risk is a system breakdown. The projected daily volume would be about eight times that experienced by HealthCare.gov, which crashed its first day when about a million people tried to sign on. Early tests in local areas have been successful, but no test can replicate a nationwide deluge. And unlike Healthcare.gov, where people without insurance were strongly motivated to try again, the census would likely lose responses from millions of people who wouldn’t come back a second time to enter information if the system initially broke down. Plus, as Lowenthal stresses: “Healthcare.gov had a few months to work out the kinks. The census has six weeks for self-response to get it right. It can’t stop.”

Aside from the technological pitfalls, civil-rights advocates worry that the digital divide could skew the census count if the bureau relies primarily on the Internet in its communications campaign. Arturo Vargas, executive director of the National Association of Latino Elected and Appointed Officials, worries that “printing a postcard and directing people to go to the Web” is likely to bias census results: “The severely undercounted populations are not sitting in front of a laptop waiting to fill out their census forms. You’ve got to go reach those people.”

The bureau expects its biggest savings, at $2.5 billion, to come from a new sophisticated operational control system that will use real-time data to automatically and efficiently route census workers, linked by smartphones and tablets, to targeted households and communities. GAO has questioned whether the bureau has the capacity to build such a complex system, pointing to the bureau’s failure in 2010 to develop handheld computers for door-to-door canvassing. That failure cost up to $3 billion, about as much as the bureau predicted it would save.

Then there’s the question of whether targeted canvassing, both in compiling the bureau’s mailing address before the census and in following up with non-responders after the census, is enough to reach hard-to-count populations.

In years past, enumerators walked every street in America before each census to compile the bureau’s address list. Though costly and labor intensive, this practice also proved effective in identifying new housing units in fast-changing communities and hidden or nontraditional dwellings such as basement units or unmarked homes on Indian reservations.

For 2020, the Census Bureau plans to validate addresses using U.S. Postal Service data, aerial imagery, and other sources instead of walking every street, with estimated savings of about $1 billion. This “in-office” canvassing will be used for “stable” areas while targeted door-to-door canvassing will be done in changing areas. The bureau still is working out the methodology for distinguishing stable from changing areas and under current budget limits may be unable to assess whether savings would come at the expense of accuracy.

“If the address list isn’t good, the census isn’t good, because operationally the census doesn’t count people, it counts households,” Lowenthal said. “If your household is not on the list, they won’t know they missed you. There won’t be any follow-up because they won’t know you didn’t respond.”

The bureau is also planning to scale back its door-to-door follow-up for households who don’t respond. Instead, the bureau intends to rely on Social Security, tax, and other administrative records as well as data purchased from commercial vendors to count non-responders and eliminate vacant housing units from the follow-up list, with estimated savings of $1.4 billion.

At this point, however, it’s uncertain whether the bureau will gain access to all the information sources it wants. Privacy protections often prevent other government agencies or state and local governments from sharing their records. The bureau will have to obtain waivers or win legislative changes to gain access to records from the IRS, Medicaid, the Supplemental Nutrition Assistance Program (SNAP), the Social Security Administration, and other agencies.

Overcoming these barriers could be politically difficult though, as the aggregation of personal information—including commercial data—on almost every U.S. resident is bound to set off alarm bells. According to the GAO, the Census Bureau currently lacks adequate information security systems to protect against data breaches. That is especially troublesome in the wake of the disclosure last June of a hack of federal government personnel records, which compromised the personal data of more than 20 million Americans.

If the Census Bureau is unable to obtain the records it wants and is forced to rely on incomplete information, the 2020 census will be more likely to miss non-respondents. Assuming all records are obtained, however, there is still the question of how to count the roughly 11 million undocumented immigrants living in the United States. “If you’re undocumented, you don’t qualify for federal benefits, you don’t have a social security number,” Vargas said. “And these are the [people] most likely not to be counted. They’re not going to be captured by administrative records.” Yet the Constitution requires that the census count everyone, including non-citizens.

Administrative and commercial records also typically don’t provide census-required data on the location of a person’s residence as opposed to a post office box where a person receives mail. But addresses are needed for congressional redistricting and other purposes.

Data from administrative and commercial records are equally as unreliable for data on race and ethnicity, which the census also requires. “Right now the way the census questionnaire is set up, Asian respondents have an option to check Chinese, Filipino, Asian Indian, etc.,” providing detailed information about Asian residents, said Terry Ao Minnis, director of census and voting programs for Asian Americans Advancing Justice | AAJC. Most administrative records, however, will not even have accurate information about whether the person is Asian. “At best, government [administrative] records may indicate that the respondent is Asian. But even if you could get Asian [information in these records], you wouldn’t be able to get that detailed data, which is really critical.”

Resolving Confusion about Race and Ethnicity

The census might be the best source of data on race and ethnicity, but it is by no means perfect, and respondents often are confused about how to identify themselves. As currently designed, the survey first asks whether the respondent is of Hispanic, Latino, or Spanish origin, and then offers a series of check boxes for Mexican, Puerto Rican, Cuban, or other Hispanic origin, with a write-in box. The next question asks for the respondent’s race, with check boxes for white, black, American Indian or Alaska Native, seven Asian nationalities, four Pacific Islander groups, or “some other race,” followed by a write-in box.

About 20 million people in 2010 checked the “some other race” box—making it the third most selected race category behind white and black—and the vast majority of those were Hispanic. Vargas, who serves on the Census Bureau’s advisory committee examining the race and ethnicity question, summed up the challenge: “Once you’ve asked, are you Hispanic, yes or no, and they answer yes, I’m Mexican American, they go to the next question and are asked, so what’s your race. And people are like, wait a minute, you just asked me that. I just told you I’m Mexican. And the bureau would say, no, being Hispanic is an ethnicity. It’s not a racial category. But they don’t see themselves in the white, black, Asian, [or] Native American categories.”

The same problem exists for people of Middle East or North African origin. The Census Bureau considers them, like most Hispanics, to be white, but they often check the “some other race” box instead.

Because such large numbers check this box, there are comparability problems with other federal race and ethnicity data that don’t include a “some other race” category. The bureau, in response, has developed procedures for assigning a specific race—either white, black, American Indian or Alaska Native, Asian, Native Hawaiian or other Pacific Islander—to those who self-identify as “some other race.”

“The bureau wants to get out of that business,” Vargas said. “They want to redesign the [race and ethnicity] question so that people can easily see and self-identify themselves in the different options.”

The Census Bureau has tested different approaches to accomplish this objective and plans to conduct more tests through 2017. (The bureau will have to submit the final wording for the race and ethnicity questions for congressional approval in April 2018.) Among the possibilities being weighed is whether to add more detailed check boxes for race and ethnicity or whether to take away check boxes and rely on write-ins instead. Some test questionnaires, for example, eliminate all of the boxes for Asian nationalities (Asian Indian, Chinese, Filipino, Japanese, Korean, Vietnamese, as well as “other Asian”) in favor of one check box for Asian with a write-in for national origin. Ao Minnis worries that the quality of national-origin data would decline if the bureau chooses this approach. “Check boxes have been shown to be more effective in capturing data than write-ins,” she said.

“This is what the bureau needs funding for now,” Vargas said. “So they can work out the kinks and bugs in these questions.”

Now or Never

The 2010 census promised to be uniquely challenging. Hurricane Katrina had displaced millions of people who would need to be located. Millions more lost their homes following the 2008 financial crisis. And the post–September 11 security crackdown as well as the polarized debate over immigration reform had left many immigrant communities fearful of providing information to the government.

The Census Bureau, however, was not well-positioned to meet these challenges. After giving up on its effort to develop a custom handheld mobile device for census workers, the bureau had to scramble to secure $3 billion in emergency funds to revert to a paper-and-pencil census. States also had less money to chip in for outreach because of the Great Recession. California, for one, allocated just $2 million for statewide outreach for the 2010 census after providing $24.7 million in 2000. Privately, bureau staff confessed fears that they would be unable to match response rates achieved in 2000.

The nonprofit and philanthropic sectors responded by launching a series of outreach efforts focused on hard-to-count communities. Philanthropic foundations contributed about $40 million for census outreach, led by the Funders’ Committee for Civic Participation, which started the Funders Census Initiative to track and encourage grantmaking. In California alone, a group of funders joined together to distribute around $10 million, five times what the state contributed.

As a leading recipient of funds, the Leadership Conference Education Fund (sister to The Leadership Conference on Civil and Human Rights) led and coordinated a campaign targeting hard-to-count populations in 15 cities. Four national partners—Asian Americans Advancing Justice | AAJC, the National Association for the Advancement of Colored People, the National Association of Latino Elected and Appointed Officials Education Fund, and the National Congress of American Indians—conducted outreach and public education campaigns for their own constituencies. Many community-based organizations provided trusted and familiar voices on the ground.

This campaign made a significant difference, those involved say. Nationally, 74 percent of households mailed back their census forms, equaling the response rate of the 2000 census. Mail response rates equaled or exceeded 2000 rates in eight of the 15 cities targeted by the philanthropic community, and the rate dropped by just a percentage point in another three. Only a few especially challenging cities, including Detroit, which lost a quarter of its population from 2000 to 2010, as well as New Orleans and Biloxi, which suffered massive upheaval following Hurricane Katrina, saw more significant declines in mail-back rates.

This sort of independent education and outreach would be needed even under the best of circumstances, according to civil-rights advocates. “Because of what we’re asking the American public to do, there is a fundamental need for messengers other than the federal government to encourage, convince, and mobilize people to fill out their census forms,” Vargas said. “So even if the Census Bureau had all the money in the world, if the neighborhood advocate is not assuring people, ‘Hey, don’t worry, this information isn’t going to the landlord or to the police or to the immigration service,’ people aren’t going to be convinced to participate.” Vargas and others learned this the hard way in 2010 when several Latino evangelical pastors misguidedly called for immigrants nationwide to boycott the census, forcing several organizations that represent Latinos to double-down on their outreach efforts. Even without these kinds of surprise challenges, the 2020 census is shaping up to be even more challenging than 2010.

If funding shortfalls cause delays in the bureau’s technology and questionnaire reforms, work also could be pushed back on the bureau’s communications campaign, including development of promotional materials, instructions for filling out the form, and translations for immigrants covering 60 different languages. The latter might be particularly challenging due to delays and budget cuts, as testing materials in just five non-English languages started three years before the 2010 census. “Those are components of the census that are most needed to improve the count in historically undercounted communities,” Lowenthal said. “It takes a while to develop those things—it’s not easy. You can only put them off so long before you end up with a not-very-robust communications campaign.”

The challenge now is avoiding this scenario. The 2010 campaign covered just the final two years leading up to the census and focused almost entirely on encouraging people to mail back their forms by the April 1 deadline. Such an approach this time would neglect the Census Bureau’s funding shortfalls, the technology overhaul and related decisions, attacks on the ACS, and revisions to the race and ethnicity question, as well as non-response follow-up, which, under the bureau’s plan, will largely be conducted electronically without the typical boots on the ground. How those issues play out will be at least as important as the communications campaign in determining whether the 2020 census is accurate and fully counts racial and ethnic minorities and other hard-to-count populations.

Adding to the challenge is that so few know how important and widely used the census data are. That is true even among those who use the data on a daily basis. While several organizations such as those mentioned in this article have already been working closely with Congress and the Census Bureau to ensure that these crucial policy and budget decisions will lead to an accurate count in 2020, their success depends in part on gaining more allies. Civil-rights and community organizations, business leaders, state and local elected officials, journalists, researchers and academics, and philanthropy all have a stake in making sure the enumeration process works and picking up the slack from a cash-strapped Census Bureau.

With Congress set to slash the Census Bureau’s budget and the bureau set to lock in changes with big consequences, it’s now or never to make sure everyone counts.