DIVISION OF BEHAVIORAL AND SOCIAL SCIENCES AND EDUCATION
Committee on National Statistics
Task Force on the 2020 Census
August 7, 2018
Departmental Paperwork Clearance Officer
U.S. Department of Commerce, Room 6616
14th and Constitution Avenue NW
Washington, DC 20230
RE: FR Doc. 2018-12365, Proposed Information Collection; Comment Request; 2020 Census.
Docket number USBC- 2018-0005.
Dear Ms. Jessup:
The National Academies of Sciences, Engineering, and Medicine’s Committee on National Statistics (CNSTAT) established a Task Force on the 2020 Census to consider challenges for the Census Bureau in conducting the 2020 decennial census. CNSTAT, which was established in 1972, has provided assistance to the country on the methods used in the decennial census at the behest of Congress or the U.S. Department of Commerce (DOC) most years since the 1980 census. Our work has included panels that monitored the conduct of the 2000 and 2010 censuses as they occurred and that helped structure this decade’s development and testing work for the 2020 census. CNSTAT studies have also examined the American Community Survey (ACS) since its early pilot-testing days and throughout its full-scale operation as replacement for the “long-form sample” of households in the 2010 census (see Attachment E). In addition, since 1992, CNSTAT, as part of its core mission, has issued regular editions of its Principles and Practices for a Federal Statistical Agency (P&P), identifying and affirming the standards that statistical agencies should meet in order to function effectively as a source of high-quality, objective information to inform policy makers and the public (National Academies of Sciences, Engineering, and Medicine, 2017c; see attachment D). P&P has been used and cited by the U.S. Office of Management and Budget in statistical policy directives and by the U.S. Government Accountability Office in reviews of statistical programs. P&P has also been endorsed by the board of the American Statistical Association.
The conduct of an effective 2020 census is necessary for the functioning of the U.S. government as a whole, as required by the Constitution (Article I, Section 2). High-quality census-based information is essential, not only for reapportionment of the U.S. House of Representatives and redistricting of congressional, state, and local legislative districts, but also for many other governmental functions, including the allocation of federal funds to states and localities. Census statistical information is also widely used by the business community, nongovernmental organizations, researchers, the media, and the general public (U.S. Department of Commerce, 2014). Consequently, careful planning and testing of the methods to be used in
500 Fifth Street, NW, Washington, DC 20001
Phone 202.334.3096 Fax 202.334.3751 www.sites.nationalacademies.org/cnstat
each census is of paramount importance. The proposed 2020 census design incorporates innovations in key areas, including those recommended as priority areas by our Panel to Review the 2010 Census (National Research Council, 2011)—reengineering field operations, making fuller use of administrative records/third party data, optimizing self-response (including Internet response), and reengineering address canvassing. More generally, the design has been developed with the rigor commensurate with recent U.S. censuses.
At this stage in the life cycle of a decennial census, the Census Bureau would ordinarily be focused on fine-tuning systems and operations and not on making significant changes to its plans. The Task Force concluded that the DOC’s recent decision to add a question on citizenship status to the 2020 census is inconsistent with the “proper performance of the functions” of the Census Bureau, which was one of the areas where the DOC requested public comments in the Federal Register notice, as required by the Paperwork Reduction Act (Pub. L. No. 104-13). This conclusion rests on three principal arguments:
- The American Community Survey already meets the stated need for citizenship data: The Secretary’s decision memo of March 26, 2018, discounts the current collection of citizenship data in the ACS—a survey, like its predecessor long-form sample in the decennial census, that is directly intended to address critical information collection needs while mitigating undue burden on the public. ACS data are expressly designed to facilitate the analysis and comparisons of characteristics of specific subpopulations, and they have been used effectively for enforcement of protections in the Voting Rights Act, which is the stated reason for adding the citizenship question to the 2020 decennial census.
- Adding the citizenship question without proper testing will, in our judgment, impair the quality of the 2020 census as a whole: The Secretary’s decision memorandum characterizes the issue as “reinstatement” of a citizenship question. Yet, each census is sufficiently different from prior censuses that a more accurate characterization is that a new question is being added to the census but without the rigorous testing and proper consideration of consequences that are expected in proper survey and statistical practice. According to the Census Bureau’s own analysis, addition of the citizenship question could adversely affect the quality and the cost of the 2020 census.
- Adding the citizenship question and using the method described in the Secretary’s memo and the Census Bureau’s review would create a new population register, which has unclear statistical purposes and which could not, under current law, be used for nonstatistical purposes, such as law enforcement against individuals, and still comport with the mission of the Census Bureau: The Secretary’s decision memorandum suggests an intent to use census responses to “correct” or validate citizenship status in administrative records data, and the documentation of the Secretary’s decision directly states that the option ultimately chosen by the Secretary was intended to use 2020 census responses to supplement a “comprehensive statistical reference list of current U.S. citizens,” which would then continue as a regularly updated citizenship registry. The uses of this registry are not detailed, and therefore, the practical utility and need for the collection of information and new active citizenship registry has not been demonstrated. Currently, any “nonstatistical uses” of census data, which include law enforcement, adjudication, and using census
responses in any other manner to directly affect the rights, benefits, and privileges of an individual, are prohibited by federal law (13 U.S.C. § 9), and are contrary to the functions of a statistical agency.
The American Community Survey Already Meets the Stated Need for Citizenship Data
In an attached narrative (see Attachment C), we summarize the history of collection of citizenship information in the census and the ACS. This material underscores a critical point: The citizenship item has never been collected on a complete population basis since the U.S. Department of Justice became responsible for enforcing voting equity challenges under Section 2 of the Voting Rights Act of 1965. In its December 12, 2017, request to the Census Bureau, the Justice Department asserted that its Section 2 enforcement abilities would be improved through the generation of citizenship data at the census block level, hence the need to include the question on a complete population basis in the 2020 census.
Yet, data at that level of precision were not deemed necessary in the 1970, 1980, 1990, 2000, and 2010 censuses; apparently, all parties deemed it sufficient to include citizenship on only the long-form questionnaire in the 1970–2000 censuses (this, after the topic was eliminated completely in the 1960 census except for New York State and Puerto Rico). And it was deemed sufficient for Voting Rights Act adjudication to include citizenship in the ACS when the survey began its operations and when it replaced the long-form sample in 2010. Indeed, we note that the complete “administrative record” of the Secretary of Commerce’s decision on the citizenship question includes several memoranda from the Department of Justice indicating continued support for the collection of citizenship data via the ACS through the last comprehensive ACS content review done in 2014 (U.S. Department of Commerce, 2018).
At present, the ACS collects citizenship voting-age population (CVAP) data that are measured with consistently high quality nationwide and able to be tabulated for small geographic areas. Indeed, the ACS data are of higher quality, measured by nonresponse to individual items, than the long-form sample data from the 2000 census (see National Research Council, 2007, Ch. 2-B). As a result, the ACS has become an essential tool in adjudicating Voting Rights Act disputes of various sorts. Section 2 of the Voting Rights Act (codified at 52 U.S.C. § 10101) prohibits practices or procedures that would impair or dilute the right of a protected minority group to vote. In this context, the U.S. Supreme Court has held (Thornburg v. Gingles, 478 U.S. 30, 1986) that a plaintiff claiming Section 2 deprivation must demonstrate (along with two other tests) that the minority group in question is simultaneously large enough in population yet geographically compact enough to represent the majority in a single-member district. Such determinations have come to rely intensively on ACS CVAP data.
Likewise, Section 203 of the Voting Rights Act requires that voting materials be made available in non-English languages in jurisdictions for which primary use of a foreign language exceeds a particular threshold level. These determinations are arguably more sensitive and difficult to make than a strict tabulation by citizenship status—requiring, as they do, disaggregation of data by citizenship, age, and primary language spoken at home. Yet, since the section’s enactment in 1975 (P.L. 94-73), Congress has explicitly vested authority in the director of the Census Bureau to make such determinations, emphasizing that the director’s determinations “shall be effective upon publication in the Federal Register and shall not be subject to review in any court.” In 2006, the law was amended to explicitly make “the 2010 American Community Survey census data and subsequent American Community Survey data in
5-year increments, or comparable census data” the source of Section 203 determinations (120 Stat. 581). CNSTAT conducted a workshop in June 2012, which produced the volume Benefits, Burdens, and Prospects of the American Community Survey: Summary of a Workshop (National Research Council, 2013), that included expert presentations and case studies of the ACS in Voting Rights Act and related challenges—all noting the advantages of the more timely ACS data and its analytical richness relative to decennial “snapshots” of characteristics like citizenship. As a final point, statistical modeling has been successfully used in the ACS for the 2010 Census round of data collection to improve the precision of estimates for small communities and small geographic areas for purposes of Section 203 of the Voting Rights Act. Similarly, state-of-the-art statistical models using ACS data produced the 2016 Section 203 determinations, successfully implementing these requirements throughout the decade, and there is no reason to expect that modeling could not be successfully used to produce valid estimates for other sections of the act.
Adding the Citizenship Question Without Proper Testing Will Impair the Quality of the 2020 Census as a Whole
That a citizenship question has been included in decennial censuses as early as 1820 or in the ACS for a decade does not constitute sufficient evidence to argue that it is a tested and proven method of measurement for the 2020 census. The attached historical material shows that “citizenship” (or, more commonly in the 19th century censuses, naturalization status) information has been collected over the decade in different ways, including different modes of administration (e.g., personal interview, self-response). One of our consensus study reports (Envisioning the 2020 Census; National Research Council, 2010), a particularly thorough examination of the Census Bureau’s longstanding research and development process, demonstrates that even small changes in question wording and in mode of administration can have significant effects on responses and on decisions whether to respond, and that the Census Bureau has properly subjected such changes to years of repeated testing and evaluation. The proposed addition of the citizenship question to the 2020 census has not been subjected to the same scrutiny, and it has not been assessed under conditions that resemble the actual decennial census to the greatest extent possible. Most notably, it was added too late to be included in the current 2018 End-to-End Census Test, the final trial run in the field for the 2020 census.
Further, the addition of a question to a decennial census is not simply a matter of a few more words on the printed page or an additional question in the Internet form, nor is there evidence that detrimental effects on census participation (as the closing passage of the Secretary’s decision memorandum argues) can be mitigated simply by placing the new question at the end of the questionnaire. Respondents can see all of the questions on the paper form before completing it, and they can also go back to previous questions on the Internet form after seeing the citizenship question. Furthermore, as the listing of dozens of constituent operations and numerous specific information collection documents in the Census Bureau’s Federal Register notice and the Census Bureau’s operational plan for the 2020 census make clear, the 2020 census is much more than a single, simple questionnaire:
- The vast majority of households will receive an invitation in the mail to respond via the Internet, although they may instead request a paper questionnaire to fill in and mail back. Households in areas where broadband Internet connection may be spotty
will receive both Internet login information and a paper questionnaire in a single mailing, and households in predominantly rural areas will have their questionnaire packet dropped off by census enumerators. Both paper and electronic questionnaires would need to add the citizenship item. Until and unless proven by testing, it cannot be assumed that respondents would react the same way to any question, citizenship or other, on a paper as on an electronic form. With respect to a citizenship question, specifically, it is not known how many respondents to either the paper or electronic questionnaire would leave it blank.
- Because many households will not respond to the 2020 Census, either via the Internet or by mail (and the extent of nonresponse could be increased due to publicity about the citizenship item), the citizenship question would also have to be included on the Enumerator Questionnaire used in nonresponse follow-up (NRFU) operations. The 2020 NRFU will be conducted primarily through handheld mobile devices and not paper questionnaires as in previous censuses. (Enumerators will also use paper questionnaires to interview households in remote areas, such as remote Alaska.) Though an objective of the 2020 census is to reduce the NRFU field workload through recourse to administrative records data from other federal government sources, the quality of citizenship information in those administrative data is known (and acknowledged, explicitly, in the Secretary’s decision memorandum) to have issues. And even with the use of administrative records, enumerators will be making millions of NRFU field visits throughout the country. It is not known the extent to which publicity about the citizenship question would induce households to not provide this information or avoid the interview entirely.
- The citizenship question would also have to be included in the electronic version of the questionnaire to be used by operators on the Census Questionnaire Assistance (CQA) phone line. CQA is partially meant as a mechanism for persons with limited English proficiency to request a questionnaire or be interviewed, and it is not known the extent to which the presence of a citizenship question may undercut the use and effectiveness of that operation.
- Along with all the other questions, the citizenship question would also have to be included in the foreign-language versions of the paper and electronic questionnaires, as well as translated and described in dozens of language assistance guides—raising the possibility of cultural or linguistic misinterpretations.
- The citizenship question would also have to be included in quality assurance (QA) reinterviewing, a routine check of the veracity of enumerators’ work. Though the Federal Register notice describes a streamlined QA approach based on initial attempts to verify only household roster/name information by telephone, millions of households will still be subject to QA reinterviewing.
- The citizenship question would also have to be included on questionnaires for special populations, such as residents of group quarters.
- Finally, the citizenship question would need to be included on the postenumeration survey (PES) questionnaire, involving in-person interviews with an independent sample in order to generate estimates of net undercount or overcount in the census.
The ACS has included a citizenship question with no apparent problems to date. However, this use is insufficient evidence of how a citizenship question would function in the
2020 census. The ACS contains scores of questions, and the salience of the citizenship question to the overall request for participation in the ACS is minimal. In contrast, the 2020 census with a citizenship question would include only 11 questions in total; moreover, the salience of the citizenship question has risen to the highest level of all other questions, given the publicity surrounding the Secretary’s decision memorandum. Expecting that such publicity would continue into the census period likely means that the decision to participate in the census would be driven by attitudes toward that question. Such a phenomenon could affect the “actual enumeration” of the full population that the Constitution requires.
The Secretary’s memo notes that there is not clear evidence that addition of the citizenship question would depress response. However, because the citizenship question has not been tested in the proper context or with the proper scrutiny, the burden of proof would seem to lie with the DOC to establish that the addition of the question would not degrade the quality and utility of the decennial census.
Given a lack of testing, we are in no better position than anyone else to estimate how much the citizenship question’s addition would affect response in the 2020 census. However, lack of evidence of an effect is not evidence of the lack of any effect. The presence of the question certainly could not improve response, and our decades of census observation suggests numerous predictable, deleterious effects. Indeed, the mere knowledge of the citizenship question’s existence could put a damper on willingness to respond to the census at all, whether online or on paper, or could induce households to leave the citizenship question blank.
The Census Bureau has procedures for imputing values for missing responses and for imputing whole households when absolutely necessary. However, the necessary modifications to these procedures to handle the citizenship question would be untested and could raise doubts about the quality of the data should levels of nonresponse be greater than in past censuses.
Of key importance is that the Census Bureau relies on the cooperation of the public to participate in the decennial census and has effectively employed advertising and outreach through partnerships with localities and local organizations in the 2000 and 2010 censuses. These efforts would be strained and impaired with many constituencies due to the presence and surrounding adverse publicity about the citizenship question. Enumerators could also experience additional difficulties in obtaining cooperation when they visit nonresponding households and would need to be trained for such an eventuality.
The costs of the census are heavily driven by the level of the participation generated by the initial request. If households do not respond initially, human resources are required to seek participation. In the 2010 census, 600,000 enumerators were used for this task. With a lower initial participation rate that might occur if the citizenship question were incorporated, many more enumerators would be needed, and census costs would almost certainly increase, perhaps to a major degree. More important, census quality would be affected, not only because enumerators could encounter difficulties in obtaining answers to the citizenship question, but also because they might not be able to obtain an interview at all, leading to a likely increase in “last-resort” or proxy enumeration for such cases (e.g., interviewing a neighbor or landlord) or to the need for computer-based imputation.
Should enumerators encounter resistance, they may well back off from asking about citizenship. Indeed, in the 2000 census, there is evidence that enumerators, while generally successful in obtaining answers to questions on household relationship, race/ethnicity, and gender, were less successful in obtaining answers to other questions on the census long form, like citizenship (National Research Council 2004, App. H). As a consequence, the quality of
results for the census as a whole—including responses to other questions besides citizenship—would likely decrease because of the increased levels of imputation and proxy response. Moreover, the decrease would almost certainly affect some demographic subgroups, including minorities, more than others, based on the decades of evidence about differential coverage errors in the census (National Research Council, 2004, Ch. 5; 2010, Ch. 2-B). The quality of the core tool by which we learn how well a census has done in coverage—the postenumeration survey—would likely be compromised as well, for the same reasons. In short, then, our decades of census observation strongly suggests that the addition of the citizenship question at this stage, without proper testing and consideration of all its likely effects, would be an unwise attempt to increase the perceived precision of one data item to the detriment of the quality of the census as a whole.
Adding the Citizenship Question and Using the Method Described in the Secretary’s Memo and the Census Bureau’s Review Would Create a New Population Register, Which Has Unclear Statistical Purposes and Which Could Not, Under Current Law, Be Used for Nonstatistical Purposes, Such as Law Enforcement Against Individuals, and Still Comport with the Mission of the Census Bureau
The Secretary’s decision memorandum rejects the option of relying solely on administrative or third-party data for citizenship, citing then-unpublished Census Bureau analysis of matched 2010 census/ACS and administrative records data showing (in the Secretary’s words) that “when non-citizens respond to long form or ACS questions on citizenship, they inaccurately mark ‘citizen’ about 30 percent of the time” and noting that use of the records alone would still require imputation of citizenship status for about 10 percent of the population. What the Secretary settled on was a hybrid approach, called “Alternative D,” of adding citizenship to the 2020 census questionnaire while also directing the Census Bureau to “use the two years remaining” before the census to improve its administrative-records resources for measuring citizenship. In particular, the Secretary directed the Bureau “to determine the best means to compare the decennial census responses with administrative records” in order to “determine the inaccurate response rate for citizens and non-citizens” alike.
The Secretary’s memorandum does not provide a full description of Alternative D. However, at page 001309 of the complete administrative record related to the Secretary’s decision (U.S. Department of Commerce 2018), the Census Bureau’s analysis describes Alternative D as follows:
Administrative data from the Social Security Administration (SSA), Internal Revenue Service (IRS), U.S. Citizenship and Immigration Services (USCIS), and the State Department would be used to create a comprehensive statistical reference list of current U.S. citizens. Nevertheless, there will be some persons for whom no administrative data are available. To obtain citizenship information for this sub-population, a citizenship question would be added to the 2020 Census questionnaire. The combined administrative record and 2020 Census data would be used to produce baseline citizenship statistics by 2021. Any U.S. citizens appearing in administrative data after the version created for the 2020 Census would be added to the comprehensive statistical reference list. There would be no plan to include a citizenship question on future Decennial Censuses or American Community Surveys. The comprehensive
statistical reference list, built from administrative records and augmented by the 2020 Census answers would be used instead. The comprehensive statistical reference list would be kept current, gradually replacing almost all respondent-provided data with verified citizenship status data.
So described, the chosen Alternative D is not a “reinstatement” of a citizenship question to the decennial census for statistical purposes but rather the intended use of census responses as seed data to construct an ongoing citizenship status registry, something never before proposed as a task for the Census Bureau, an agency solely devoted to statistical uses of data.
Although some European countries have population registries, the United States has resisted the creation of an individual-level population register for statistical or other purposes. Such a register would be a new system of records under the Privacy Act (5 U.S.C. §552a). Its merits, limitations, uses, costs, and privacy implications would need to be thoroughly and publicly discussed, including whether the administrative records are sufficiently accurate to identify citizens (and by implication noncitizens) and assign them a place of residence, given the potential for adverse consequences in cases of misclassification.
The Census Bureau has the technical capabilities to construct such an administrative records system from the many existing data sources that are described in a predecisional Census Bureau memorandum (U.S. Census Bureau, 2018). Indeed, the Census Bureau has for many years obtained and used administrative records from other federal program agencies to produce and enhance statistics, and we have argued in several reports that it should make greater use of administrative records in conjunction with survey data and other sources for improving the quality, timeliness, and cost-effectiveness of many of its statistical programs (National Academies of Sciences, Engineering, and Medicine, 2017a, 2017b, 2017c). However, in none of this work does the Census Bureau create or maintain or provide access to administrative records systems for anything other than statistical purposes.
A core tenet of U.S. census law—paraphrased by the Census Bureau in this Federal Register notice—is that information collected by the Census Bureau must be used only for “the statistical purposes for which it is supplied” (13 U.S.C. § 9). The sixth edition of Principles and Practices for a Federal Statistical Agency defines statistical purposes to include “description, evaluation, analysis, or inference for groups of individuals or other units . . . and not interest in or identification of an individual person or economic unit. A statistical agency . . . does not use [the data it collects] for nonstatistical purposes, such as regulation or law enforcement” (National Academies of Sciences, Engineering, and Medicine, 2017c, p. 12).
To prevent such nonstatistical use and to protect the confidentiality of individual census responses as mandated in Section 9 of Title 13 of the U.S. Code, the Census Bureau has subjected the block-level data that it has provided to the states for redistricting since the 1970 census to “data swapping” and other methods to protect against disclosure. Augmentation of these block-level data with citizenship information would require the Census Bureau to use additional disclosure protection methods that could impair the accuracy of the block-level information, thereby undercutting the avowed purpose of collecting citizenship on the census in the first place. Although the Census Bureau is investigating modern methods that may make it possible to report useable block-level information, these methods require considerable evaluation and testing before they could be deployed.
Relatedly, it is not clear why the Census Bureau would need to create a “comprehensive statistical reference list” based on administrative data and census data simply to produce
citizenship statistics for the Department of Justice when it does not use such an approach for any other population characteristics or for any other federal agency. Because there is no apparent statistical justification for the Census Bureau to create this citizenship registry, legitimate concerns arise that this information could somehow be used for law enforcement, adjudicatory, or other nonstatistical purposes in some manner, which would undermine the mission of the Census Bureau (as well as violate Title 13, Section 9).
Principles and Practices for a Federal Statistical Agency (National Academies of Sciences, Engineering, and Medicine, 2017c), identifies four principles for effective operation of a statistical agency like the Census Bureau:
- Principle 1: A federal statistical agency must be in a position to provide objective, accurate, and timely information that is relevant to issues of public policy.
- Principle 2: A federal statistical agency must have credibility with those who use its data and information.
- Principle 3: A federal statistical agency must have the trust of those whose information it obtains.
- Principle 4: A federal statistical agency must be independent from political and other undue external influence in developing, producing, and disseminating statistics.
Proceeding with the inclusion of the citizenship question to the 2020 census as currently planned, without proper preliminary testing and analysis, without a clear demonstration of the need for the information beyond what is already collected through the ACS, and without a sound justification for creating a “comprehensive statistical reference list,” goes against these principles. This endeavor risks undermining the credibility of the Census Bureau and the decennial census, the trust of its respondents, and the independence of the Census Bureau’s professional staff to develop, produce, and disseminate objective information while protecting confidentiality of respondents.
In conclusion, citizenship is an important public policy topic and worthy of high-quality data collection, and this is already being accomplished through the ACS. The addition of the citizenship question to the 2020 decennial census questionnaire, as described in the Federal Register notice and the Secretary’s decision memo, does not demonstrate practical utility as required by the Paperwork Reduction Act, and the use of the census data to seed a “statistical reference list” raises concerns about whether the intended uses comport with the mission of a federal statistical agency. The late-stage insertion of a new and untested question in the 2020 census would almost certainly have damaging effects on the 2020 decennial census.
Committee on National Statistics’ Task Force on the 2020 Census
Attachment A Membership of the CNSTAT Task Force on the 2020 Census
Attachment B Reviewer Acknowledgments
Attachment C Historical Treatment of Citizenship in the Decennial Census and the American Community Survey
Attachment E Committee on National Statistics’ Reports on the Decennial Census and the American Community Survey