14.1 Introduction

Over the past several decades the Census Bureau has tried a number of approaches to reduce differential undercounts in the Census, but despite the best efforts of the Census Bureau, many differential undercounts have persisted. In this Chapter, a sample of steps the Census Bureau has taken to improve Census coverage and to reduce differential undercounts are presented. The steps involve things such as improved questionnaire design, field operations, Census promotion and outreach, as well as some changes in data processing.

The Census Bureau is certainly mindful of the differential undercount problem. Awareness of the problem is evidenced by a report following the 2000 Census (U.S. Census Bureau 2004a, p. 1) which states, “Censuses before 2000 have all been plagued by chronic undercount, and particularly by differential undercount of specific minority populations and other subgroups such as renters, males and children.” The report goes on to say, “The pervasive nature of Decennial undercounts has strongly influenced Census design, including adding operations or programs specifically designed to improve coverage.”

The programs initiated by the Census Bureau to address differential undercounts over the past several decades are far too numerous to discuss in detail here. Following the 1970, 1980, 1990 and 2000 Censuses (U.S. Census Bureau 1973, 1974, 1988, 1993, 2004b) the Census Bureau issued reports discussing the coverage improvement efforts for that Census in great detail. Unfortunately, there is no such report following the 2010 Census. These reports provide detailed descriptions of the efforts the Census Bureau has made to reduce differential undercounts. The U.S. General Accountability Office (2010a, b, c, 2017) has also provided a long stream of reports on the Census Bureau’s attempts to improve the Census-taking process and the results.

14.2 Undercount Adjustment

Probably the most visible attempt to remedy the problem of differential undercounts in the Census was the call to adjust Census figures to account for such undercounts (Choldin 1994; Darga 1999). During the late 1960s and early 1970s, there was a significant increase in popular understanding of Census undercounts, particularly in minority communities and among big city mayors. Some of the interest was stimulated by a paper by Hill (1975) which showed high net undercount rates and differentials for large cities. The increased interest in Census undercounts resulted in several changes to the 1980 Census procedures to try and eliminate differential undercounts (U.S. Census Bureau 1988). But there was little impact on the differential undercounts.

The focus during this period was largely on the differential undercount of Blacks and Whites or Non-Blacks. This was partly due to the availability of data on this differential, the large size of the differential, and the fact that Blacks were the largest racial or ethnic minority population at the time. The history of the Black/Non-Black differential is presented in Chap. 8.

The undercount adjustment movement peaked as the country approached the 2000 Census. The Census Bureau plans for the 2000 Census called for the official data to include an adjustment to rectify undercounts. In the late 1990s, as we moved toward the 2000 Census and Census Bureau plans for an undercount adjustment became clear, this issue took on a decidedly partisan political character. Because the Census Bureau’s plan for adjustment involved extensive use of sampling, the term “sampling” mistakenly became synonymous with adjustment in many public and political debates. As Hannah (2001, p. 515) stated, “The recent controversy around the use of sampling methods in the U.S. Census illustrates some important political-geographic dimensions of our decisions regarding whether and how to be counted in surveys.”

Given the fact that a disproportionate share of the people missed in the Census were racial minorities, and racial minorities tended to vote disproportionately for Democrats, adjustment was sometimes seen as a way to give more weight to Democratically leaning constituents. While producing adjusted figures probably would have helped Democrats more than Republicans, it is important to understand that the Census Bureau felt that the properly adjusted data would be more accurate. In other words, the decision to provide adjusted data for the 2000 Census was motivated by an interest in accuracy and not for political reasons.

One of the big impediments to producing official adjusted Census counts is the fact that adjusted figures would need to be produced very quickly after the Census count is completed. The Census Bureau must deliver the data used for political redistricting (Public Law 94-171 files) by March 31st of the year following the Census.

In the 2000 Census, the Census Bureau was prepared to provide adjusted Census data as the official figures but decided against releasing them at the last moment because the initial results appeared flawed and they were not confident in the results (U.S. Census Bureau 2002). A memo from Census Bureau Acting Director William Barron (2001) to Secretary of Commerce Donald Evans stated, “As a member of the Executive Steering Committee for A.C.E. Policy (ESCAP) and as Acting Director, I concur with and approve the Committee’s recommendation that unadjusted Census data be released as the Census Bureau’s official redistricting data.” The memo goes on to say, “The Committee reached this conclusion because it is unable, based on the data and other information currently available, to conclude that the adjusted data are more accurate for use in redistricting.”

The Census Bureau later found an error in their procedures. According to U.S. Census Bureau (2002, p. i), “Evaluations of the March 2001 Accuracy and Coverage Evaluation (A.C.E) coverage estimates indicated the A.C.E. failed to detect a large number of erroneous enumerations.” Following the 2000 Census, very little has been said about adjusting Census figures to account for undercounts.

14.3 Enhanced Outreach to Promote Participation in the Census

Efforts to educate the public about the importance of the Census and the need for everyone to participate have increased in recent decades. The Census Bureau (2016a, p. 1) states “for Censuses especially, publicity campaigns are a key component for success. An effective communication strategy delivers tailored messaging to audience segments using media and trusted voices.”

Outreach efforts became bigger and more sophisticated after the 1990 Census. Bates (2017, p. 875) lists six interconnected elements of the Census Bureau’s Social Media campaign of the past few decades;

  1. 1.

    Paid advertising

  2. 2.

    Earned media

  3. 3.

    Local and national partnerships

  4. 4.

    The Census website

  5. 5.

    Public relations

  6. 6.

    Census in schools.

A few of the most important programs listed above are discussed below.

14.3.1 Paid Advertising

One important intervention in the 2000 Census was the initiation of a paid advertising campaign. Prior to the 2000 Census, the Census Bureau had relied on free media like Public Services Announcements (PSAs). One problem with PSAs is that they were often run late at night or other times when viewership was low.

The paid advertising program was initiated in the 2000 Census and continued through the 2010 Census and is planned for the 2020 Census (U.S. Census Bureau 2017). In 2000 and 2010, the Census Bureau contracted with a private firm to oversee the paid advertising program. The paid advertising programs focused heavily on hard-to-count groups. The contract for the 2020 Census paid advertising program was awarded to the firm of Young and Rubican in August 2017 for $415 million (Media Post Agency Daily 2017). Unfortunately, some of the earliest activities under the contract had to be postponed because of the budget situation described in the next Chapter. Consequently, the paid advertising program has not gotten underway as quickly as the Bureau had hoped.

Paid ads appear to be effective in increasing response rates. Bates (2017, p. 876) states,

In 1990, (the last Census to depend on a pro bono outreach campaign), the final mail response rate was projected at 70% but achieved only 65%. The 2000 Census (the first to use paid advertising), budgeted for a 61% response rate but achieved 67%. The 2010 Census (also with a paid campaign), also achieved a higher-than-projected mail response rate (projected was 64% with actual at 67%).

While it is likely that the paid advertising campaign increased the self-response rates in the 2000 and 2010 Census, it did not solve the differential undercount problem. For example, data in Chap. 5 indicate the net undercount of young children increased from 1990 to 2000 and again from 2000 to 2010, while the Census coverage rates for adults improved during this period.

14.3.2 Census Bureau Partnership Program

The Census Bureau Partnership Program, an attempt to get organizations outside of the Census Bureau involved promoting participation in the Census, began in the 2000 Census and expanded in the 2010 Census. The Partnership Program was an attempt to get “trusted voices” from hard-to-count communities to help deliver the message about the Census being important and safe (Olson et al. 2014). In the 2000 Census, there were about 140,000 Census partners and in 2010 about 255,000 (U.S. General Accountability Office (2010a, p. 13). Some of the Census partners were deeply involved in promoting the Census while others were partners in name only.

According to an evaluation of the 2010 Census National Partnership Program (U.S. Census Bureau 2012b, p. 9) here is a list of the types of groups and organizations that were sought for Partnerships with the Census Bureau,

  1. 1.

    Faith-based organizations

  2. 2.

    Labor unions

  3. 3.

    African-American population

  4. 4.

    Hispanic population

  5. 5.

    Asian/Pacific Islander population

  6. 6.

    American Indian or Alaskan Native population

  7. 7.

    Educational Institutions

  8. 8.

    Migrant organization

  9. 9.

    Recent Immigrant/Emerging populations

  10. 10.

    Government organizations

  11. 11.

    Federal government agencies

  12. 12.

    Disabled populations

  13. 13.

    Congressional organizations

  14. 14.

    Nonprofits organizations

  15. 15.

    Gay and lesbian populations.

With respect to the Partnership Program, the U.S. Census Bureau (2012b, p. 89) concluded, “The overall program has a measurable effect on increasing mailback rates in Hard-to-Count areas.”

Census Complete Count Committees were another attempt started in the 2000 Census to expand outreach efforts around the Census and they are closely aligned with the Partnership Program. States and localities were encouraged to put together complete count committees to promote the Census in their jurisdiction. According to the Census Bureau (2018, p. 4), “Complete Count Committees (CCC) are volunteer committees established by tribal, state and local governments and community leaders or organizations to increase awareness and motivate residents to respond to the 2020 Census.”

14.3.3 Census in Schools

The Census in School (CIS) program was started to reach households with children in the school system. The Census Bureau provides basic material on the Census to schools who agree to participate. There are two main goals in the Census in Schools Program. First, students learn about the importance of the Census in our system of government and second, they take home material that increase the chances their household will respond to the Census.

An evaluation of the Census in Schools program following the 2010 Census (U.S. Census Bureau 2012c, p. 107) concluded “Overall, teachers looking at the CIS materials during the focus group had a positive response to them and seemed glad to know about them for their future teaching.” The Census in Schools program has been turned into an “evergreen” program called Statistics in Schools (U.S. Census Bureau 2016b).

In the 2000 Census, The Census in Schools program included a program for pre-schools but that was not included in the 2010 Census program. This is important because preschoolers (age 0–4) had a higher net undercount than any age group in the 2010 Census while the school age population (5–17) had a very low net undercount. As this book is being written, the plans for the Census in Schools program in the 2020 Census are not finalized.

14.4 Changes to the Census-Taking Process

Another approach to reduce differential undercounts relates to changes in the Census-taking operations. Several recent efforts are reviewed below.

Following the 2000 Census, the Decennial Census was changed to a short-form only Census in an effort to save money and increase participation. Prior to the 2010 Census, the Census involved a short form with only a handful of questions that were asked of everyone, and a long form which contained all the short-form questions as well as several dozen questions that were asked of a very large sample of households.

According to the Census Bureau (2003, p. 17) the mail return rate for the 2000 Census short form (76%) was 13 percentage points higher than the mail return rate for the long form (63%). In addition, the differential response rates between Blacks and Whites was a little higher for the long form than the short form. Among other things, it was felt that the change to a short-form only would increase participation and decrease undercount differentials. What was formerly the long-form Census has become the American Community Survey.

In the 2010 Census, the Census Bureau distributed bilingual questionnaires to areas with a concentration of Hispanic families for the first time. This measure seemed to increase response rates for Hispanics. After evaluating the 2010 Census experience with bilingual questionnaires, the Census Bureau (2011, p. v) concluded, “Further results suggest that the bilingual questionnaire provides substantial benefit to the areas that were targeted…”.

In the 2010 Census, the Census Bureau sent replacement questionnaires to household in areas that had a low response rate in the 2000 Census in an effort to increase participation rates (U.S. Census Bureau 2012a). This was done two different ways. For areas that had the lowest response rates in the 2000 Census every household in the area received a replacement questionnaire. In other words, a few weeks after they received their first Census questionnaire, they received a second one even if they had already responded. This was referred to as “Blanketed areas” and there were about 53.7 million people in these areas. For areas that had a mid-range response rate in the 2000 Census, only nonresponding households received a replacement questionnaire. These were referred to as “targeted areas” and there were about 66 million people living these areas.

In the 2010 Census, several probes were added to the Census questionnaire to help ensure everyone in the households was included on the questionnaire. After listing all the people in the household, respondents were asked a series of questions about the types of people often left off census questionnaires, such as newborns, to make sure someone who should have been on the household roster was not left off accidently.

14.5 Census Costs and Coverage Differentials

The expanded efforts by the Census Bureau to try and eliminate Census undercounts and differential undercounts are reflected in the rising costs of the Census over time. The per household expenditures for the Decennial Census has steadily increased from $16.89 in 1960 to $114.93 in 2010 cost in constant 2009 dollars (National Research Council 2010, p. 32).

The increased cost seems to be related to improvement in the coverage of some groups but not others. Figure 14.1 shows the per household costs along with the net undercount for two hard-to-count groups, namely Blacks and young children. Data for the per household costs is expressed as a ratio of the 1960 value. Data for the undercounts of Blacks and young children were taken from earlier Chapters of this book.

Fig. 14.1
figure 1

Census cost per household and net undercount rates of young children and blacks: 1960–2010

Full size image

As Census costs rose steeply after 1990, the net undercount of Blacks declined from 5.5% in 1990 to 2.5% in 2010, but the net undercount of young children increased from 3.7% in 1990 to 4.6% in 2010. This suggests that the increased money for outreach has been more focused and/or more effective on some hard-to-count groups than others.

14.6 The Emergence of Philanthropy

One part of the changing landscape regarding Census promotion is increased involvement of philanthropic partners working with the Census Bureau to reach hard-to-count populations. Over the past 20 years, philanthropic foundations and their grantees have become much more involved in the Decennial Censuses. This has evolved into an informal public-private partnership. The foundation/non-profit world mounted efforts in 2000 and 2010 to raise awareness about the importance of the Census and foundations provided funding for many of the “trusted voices” to help increase responsiveness in hard-to-count communities (Michaels 2010; Goldstein 2011; O’Hare 2009). Crews et al. (2011) describe activities of philanthropic foundations in the 2010 Census and set out plans for the 2020 Census. Crews (2011) reports that foundations contributed at least $33 million dollars in Census promotion related to the 2010 Census.

In the 2020 Census, the work of foundations in Census promotion has increased (Daniels 2018; Gupta 2017). In the 2020 Census, more foundations are involved, the work started earlier, and the amount of money supplied by foundations is higher than ten years ago. The work of the philanthropic community has been focused on hard-to-count communities and decreasing the persistent Census undercount differential.

14.7 Summary

Over the past several decades, the Census Bureau has tried many different approaches to reduce or eliminate differential undercounts. Some differential undercounts have decreased in recent decades while others have increased. Some of the efforts have good evaluations to show they improved the count while data for other efforts are less convincing.