Winter 2019 Bulletin

The 2020 Census: Unprecedented Challenges & Their Implications

On October 30, 2018, Kenneth Prewitt (Carnegie Professor of Public Affairs and Special Advisor to the President at Columbia University) spoke about the 2020 Census at a gathering of Academy members and guests at the House of the Academy in Cambridge, Massachusetts. The program, which served as the 2018 Distinguished Morton L. Mandel Annual Public Lecture and 2073rd Stated Meeting, was live-streamed to groups of Academy members and other participants gathered at the American Philosophical Society, Georgetown University, and Ohio State University. The following is an edited transcript of Dr. Prewitt’s presentation.

Kenneth Prewitt
Kenneth Prewitt
Kenneth Prewitt is the Carnegie Professor of Public Affairs and Special Advisor to the President at Columbia University. He was serving as Director of the United States Census Bureau in 2000. He was elected a Fellow of the American Academy in 1979.

The task of the census is simple to state yet difficult to execute: count everyone once (no undercount), only once (no overcount), and in the right place (no location errors). Keeping this in mind, I would like to outline three major challenges for the 2020 Census: operational issues, partisan interference, and deliberate disruption. They have the potential, especially the latter two, of creating conditions in 2020 that jeopardize an accurate census.

There have always been operational challenges, starting when federal Marshalls mounted their horses and set forth to find every household in the new thirteen states. George Washington was certain that they had missed some, claiming that not every federal Marshall was up to the task. There have been frequent efforts to secure partisan advantage, again starting early: after the first census, conflicting apportionment formulas were proposed by Hamilton, a New York Federalist, and Jefferson, a Virginia Republican. Washington, also a Virginian, chose Jefferson’s version; Virginia benefited for decades. The third challenge, deliberate disruption, is of more recent vintage, at least in the form it may take in 2020, possibly arriving from foreign sources. Every decade the census is challenged; every decade, it has innovated to overcome the challenges.

Before turning to each of these impending challenges, let me set the stage by describing a late-twentieth-century challenge and a 2000 innovation that dealt with it. Non-response was the challenge; paid advertising and a mobilization campaign were the innovation. In previous decades, the Census Bureau essentially relied on the willing cooperation of Americans to respond, perhaps with a nudge from public-service-minded broadcasters like pbs urging late night listeners to return their census forms. In the latter decades of the twentieth century it became obvious that something more was needed because the non-response rate was steadily growing and the 1990 experience, with 25 percent of America’s households not responding, was an alarm signal. Estimates for the next census indicated that the non-response rate could reach 35 percent. With the full support of Congress, the Census launched an extensive paid advertising campaign, kicked off with a Super Bowl ad. An unprecedented partnership program – basically unpaid advocacy by civic organizations, such as churches, chambers of commerce, and schools, along with mayors and governors, and much more – joined the campaign. It worked. The 2000 Census arrested what had been a decade-by-decade worsening of the non-response. Building on the 2000 experience, a yet more ambitious engagement campaign was mounted in 2010. It was even more effective.

The success of this initiative notwithstanding, every census has to prepare for millions of households not immediately responding. Enumerators following up with these households, often multiple times, is very costly. The 1990 Census budget was about $3.5 billion; the 2000 Census cost nearly twice that sum. Another major budget increase was required for 2010, and yet another was expected for 2020. Faced with what appeared to be runaway costs, Congress put the brakes on, instructing the Census Bureau to design the 2020 Census at a cost no greater than the previous one, despite population growth and all that it implies. This takes us to the operational challenges for 2020.

Operational Challenges and Technical Innovations

When the budget is held constant while the challenges mount, the only way to produce a strong census is to innovate. For example, the address file needs to be updated for every census. Historically that involved a very costly year of fieldwork, involving census workers walking eleven million census blocks. The 2020 address file update did not repeat that traditional procedure. Instead it made extensive use of satellite imagery, third-party data providers, and geographic information systems. This operation succeeded, with significant cost savings.

This is one of several innovations. In 2020, every household likely to have an internet connection will receive mailed instructions on how to respond easily online. If, as expected, between 50 and 60 percent of households respond electronically this will be another major reduction in costs. Households without connectivity, or that hesitate to use the internet, can fill out a form and respond by mail. Because advertising and mobilization initiatives will have reached the majority of these households, we can assume that a significant number will be persuaded to return their form. The remaining non-responding households will require an in-person follow-up, but costly return visits will be limited by bringing administrative records into the census process. This has been an area of substantial advance in the last decade, opening a new chapter in census history. When administrative records can substitute for census forms, census-takers don’t have to knock on doors, find someone at home, and convince them to answer questions. Many of their answers are already on file – in tax-records or housing starts or vital statistics. There are quality control challenges in data linkage, but these are being worked on and there are reasons to be confident about what the Bureau is projecting for 2020. Less easily managed is public suspicion of anything resembling a national registration system – which, in fact, is not that distant from extensive data linkage across multiple administrative records. Not surprising, public concerns about privacy and the ability to re-identify individual data from aggregate data made publicly available have created additional uneasiness. In this arena, the Census Bureau is planning new and more secure ways to protect privacy.

The Census Bureau gets high marks for these and other technical and operational advances, though – largely because of budgetary constraints – thorough testing has been less than optimal. For example, the Bureau initially planned three dress rehearsal sites, which duplicate the varied conditions that are encountered across the country. Funds were available for only one site. Untested or inadequately tested operations, especially if they rely on new technologies, are cause for concern, but the Census Bureau has a remarkable record of coming up with a Plan B when a Plan A doesn’t fully perform.

Overall, I think the 2020 Census will have a difficult time reaching the performance level of its two predecessors. As noted above, in 2000 and 2010, the big innovation was not technical but instead a large, well-funded public outreach initiative. This initiative fit naturally with the basic mail-out/mail-back census that was being conducted. The 2020 Census budget limits require it to introduce a number of high-tech innovations, using tools less familiar to the public and unlikely to work flawlessly on their first large-scale application. This reservation notwithstanding, from the perspective of 2020 operations there is no reason to doubt a successful census overall. However, the census could be seriously compromised or, in the worst case, undone by one or both of the issues I will discuss next.

Partisan Interference

Before the adoption of the Fourteenth Amendment (1868), every census utilized the constitutional three-fifths clause that was designed to give a significant advantage to the Southern states – providing them with about one-third more House seats and Electoral College votes than warranted by the size of their free population. The 1920 Census reported population totals indicating that the country had become more urban than rural; these counts should have shifted eleven congressional seats from Southern and Midwestern states to the urban Northeast. Partisan interests, however, blocked reapportionment altogether, and distribution of the seats remained unchanged until the 1930 Census. A partisan dispute stretched across three decades of census-taking (1980 – 2000) over dual system estimation and sampling for non-response follow-up designed statistically to adjust the differential undercount (racial minorities missed in the census at higher rates than whites). Partisan pressures are not new.

The most recent example is the Trump administration’s directive that the Census Bureau, against its professional judgment, add a citizenship question to the 2020 Census form. To explain the possible consequences of this directive I first distinguish between the use of census statistics and their production. The political use of government statistics enhances a democracy, and is even mandated at times – for example, the decadal reapportionment without which political representation proportionate to population size could not work. More generally, we expect statistical evidence to be used when policy options are debated and then, after X rather than Y option is chosen, we want evidence used in a post-hoc evaluation of whether X worked as intended. Put simply, official statistics are used politically because governing is a political act.

The mischief occurs when explicitly partisan interests intrude in the production phase, indicated by the examples above. Differential overcounts (the three-fifths clause) and differential undercounts (missing the hard-to-count) damage our statistics, thereby weakening our democracy, our economy, and our society. In early 2019, as I write this, there is vigorous partisan debate about the merits of the late addition of a citizenship question to the 2020 Census form. The Census Bureau has warned the administration that this question will increase census costs and produce a sharp increase in the undercount of immigrant and non-citizen households (5 percent, and perhaps much higher in homes where undocumented immigrants are present). The administration claims that citizenship data will help the Department of Justice apply the Voting Rights Act. A federal judge found that claim to be spurious and ruled against the administration; the case is now being appealed and will likely be resolved by the time readers encounter this essay.

I will not describe the statistical and legal details (although they are certainly interesting) but instead will use the citizenship case to pose more fundamental questions: What, exactly, is partisan interference in census-taking? What is at risk when it occurs?

Defining Partisan Interference

For reasons noted, especially because of the positive benefit to democracy when federal statistics are used in the political act of governing, partisan interference is not easily defined. Start with Article 1 of the Constitution, which states that

The actual [Census] Enumeration shall be made within three Years after the first Meeting of the Congress of the United States, and within every subsequent Term of ten Years, in such Manner as they shall by Law direct.

“They,” of course, refers to congression­al members – ever eager to stay in power. Might then their exercise of direction over the decennial census be partisan interference? It depends. The best I can offer is that interference is the attempt to gain partisan or regional advantage by shaping the production of statistics against the judgment of a nonpartisan and apolitical statistical agency, which overrules a statistical agency’s responsibility to offer its best expert judgment, which prevents it from using state-of-the-art science, or which insists on preclearance of a statistical product.

On the face of it, these principles seem reasonable, but they come under pressure when partisan interests are threatened. The 2010 Census counted forty million immigrants, citizens and noncitizens, which redistributed eighteen congressional seats. Sixteen of those seats went to states that voted for Barack Obama in 2012. In a period as politically polarized as the present, it is unsurprising that the party disadvantaged would try to minimize the harm it sees as coming from the census. For example,

Because illegal aliens should not even be in the country, and other nonimmigrants such as foreign students and guest workers are here only temporarily, it makes no sense to distribute congressional seats as if these foreign nationals deserve representation the same as American citizens. . . . The U.S. population that logically should be enumerated includes U.S. citizens and legal permanent residents (immigrants). As only the former may vote in federal elections, the apportionment of seats in Congress should be done on the basis of the number of U.S. citizens in each state.”1

Many lawyers, and to date the courts, have rejected this line of reasoning as unconstitutional. Many other lawyers argue to the contrary, as in the quote above. Whatever the eventual outcome of the legal debate, much more than reapportionment depends on a complete count of the entire resident population. Consider Houston, Texas, where the mayor recently convened leaders from business, education, religious organizations, and other sectors to address a “$6 Billion Census Problem: Frightened Immigrants.” A Houston citizen who lives with undocumented residents made it clear to the mayor that she has no intention of cooperating with the census “for fear it will lead to deportations.” There are millions like her, and that is why “Frightened Immigrants” frightened the mayor. Every uncounted person in the census represents a loss of federal funds – among other things, for roads, disaster relief, schools, and health programs, the latter at more than $1,000 lost per person. Multiply this in a region where 37 percent of immigrants are undocumented (the largest percentage in the country) and you quickly get to $6 billion lost to Houston over ten years.

The point is simple. If you are a mayor, reapportionment is not the issue. Your worry is a $6 billion hole in your budget, and not knowing how many new students to prepare for or how many elderly live in neighborhoods vulnerable to flooding. What can the Houston mayor do? Ask churches and schools to help, get buy-in from the Chamber of Commerce, count on free media coverage – anything to ensure that the census count matches the true size of Houston’s population because it is that entire population that is the mayor’s responsibility. A version of Houston’s dilemma is beginning to play out in thousands of cities and towns across the country. Mayors and city councils need accurate statistics to know whether businesses are investing in or departing from their cities, whether recent immigrants are assimilating into their communities or hiding in fear, among a host of other things.

The Houston problem is real and consequential, and I don’t mean to dismiss it as I now emphasize why there is a longer-term harm in the making. This involves a shift in the public’s view of the census – from the very model of American democracy pictured in the iconic Saturday Evening Post image, to a fear that the census is being used as an instrument for government surveillance. For more than a half-century the Census Bureau has produced a statistically sound and widely used measure of citizenship from sample surveys and government records. If we already have the data, the public can reasonably ask, why does the Department of Commerce and the Department of Justice suddenly want a new, more intrusive, and almost certainly less accurate measure? Might surveillance be the purpose? Even posing that question puts census cooperation at risk, and not just for non-citizens and recent immigrants. Many are asking if their census answers could be used against them – when seeking a job, borrowing money, taking out insurance. The Census Bureau will, correctly, insist that answers are protected, but they are fighting against the privacy concerns resulting from social media practices.

Perhaps if media coverage and public anxieties stopped there, the harm to the 2020 Census could be managed. I don’t think it will stop there. The citizenship question is being pulled into the intense political polarization that afflicts the country. The public takes note that a census question is being vigorously attacked by one party and equally vigorously defended by the other. Might the public come to believe that there can be a census tilted toward the Republican Party when it is in power, and then a census favoring the Democratic Party when it is in power? And treat census cooperation as if it is a vote to be cast differently depending on which party is in power? I do not predict this development, but neither do I believe it impossible. Already some large commercial players, which in the past have urged census cooperation, are hesitant to appear to be partnering with the Trump administration.

Part of my concern springs from the fact that the 2020 Census will be in the field simultaneously with a very contested presidential primary season. If “partisan interference in the census” is bandied about, the effort by the Census Bureau and its defenders insisting that census-taking is politically neutral will get lost in the noise, and the noise is nasty. I fear that we are watching the stage being set for an intense partisan debate when the census results are presented to the president in December 2020, with the losing side claiming that the counts are illegitimate and the winning side, as has now become common, shouting “poor loser.”

What is at stake, then, is whether the Census Bureau’s operations and innovations in 2020 will be sufficiently robust to prevent the polarization from damaging its long-standing reputation for providing an indispensable contribution to the nation. The next challenge, I regret, is not cause for optimism on this question.


Census Disinformation and Disruption

Although there is no baseline from which to predict attacks on the 2020 Census, it is prudent to assume that the threat is greater than zero. Hacking, of course, has been a threat for decades. The Census Bureau is experienced and, to date, has been successful in designing firewalls against hacking individual data records. I will skip over that issue, and only address efforts to destabilize the census with disinformation or other forms of disruption by adversarial actors–domestic, foreign, or both. 

What we expect from the Census Bureau and supportive third-party actors is vigilance against efforts to seed discord, intimidate, or shape media coverage. While the Census did very little digital advertising in 2010 or turn to social media to get the census message out, we can expect to see concerted effort from the Bureau and its partners to counter misinformation through a drumbeat of messages in 2020. We can expect accurate, precise – and perhaps even witty – content shared online, but the likelihood that this effort will tamp down partisan or polarizing narratives is low. Furthermore, journalists, civil society groups, and technology companies may be put in an impossible position if their efforts to amplify messages about the census are flipped, and used to undermine the count. 

Assigning a probability to malicious attacks on the census should be guided by two considerations. First, we need to recognize that the census is a target. The Census Bureau and its thousands of volunteer partners will pull out all the stops to announce “the census is here,” an effort that will start exactly one year from when I write this sentence (January 21, 2019), well ahead of the official April 1 “census day.” Twenty years ago, I (and the media) went to Alaska, where, on a dogsled, I went forth to count the first household in the 2000 Census–it made front-page news across the country, as it did ten years later when then Director Robert Groves took his turn on the dogsled. It is likely to be repeated in 2020. The advertising campaign will quickly follow. From that day, well into April, the census will be highly visible. 

Second, probability should be guided by an understanding of the scope of the possible damage. Readers know that the census (which includes the American Community Survey) is the benchmark against which critical economic, social, demographic, and housing information for every community in the United States is calibrated. The ripple effects of distorting its count are not only substantial, they are compounding, and last until the next census. To take one example: The Bureau of Labor Statistics’ price and cost-of-living programs – including the Consumer Price Index (CPI), the Producer Price Index, the Consumer Expenditure Survey, and related data – have an estimated budget authority of $250 million annually (much of which is paid to the Census Bureau for data collection). The CPI component of the program is used for annual cost-of-living adjustments for retiree payments and other beneficiaries under Social Security ($941 billion in 2017). A one percentage point difference in the CPI estimate moves approximately $10 billion, in an increase or decrease in SSA payments in the subsequent year. Annual changes in the CPI also affect commercial and residential rents, public- and private-sector wages, and components of the federal income tax code. A very long list of examples could be provided, all indicating that in tampering with the Census Bureau’s statistics, you tamper with the American economy. 


In its entire history the continuous challenges facing the census, by and large, have been met by census-taking with the full support of the then current government and with general good will from the public. Neither is assured for 2020. This is the most disturbing challenge imaginable. If the administration flirts with the idea that a full count is not to its advantage but selective undercounting is, and if the public, for a variety of reasons, is suspicious of the census to the point that trust erodes, an accurate 2020 Census falls out of reach.

Author’s Note: Following the presentation of this talk, I was asked by the Academy if I would prepare my remarks for publication. I was already scheduled to speak several weeks later at the American Philosophical Society on similar issues, with the understanding that I would publish that presentation in the APS’s Proceedings. Readers will appreciate that I was caught in a difficult situation. I am pleased to report that the APS quickly and graciously agreed that dual publication was fine. For the benefit of readers, I note that the versions differ (they are separated in time, and the census is a moving target), but not to the extent that reading both would be rewarding. 

© 2019 by Kenneth Prewitt

For more information