State v. Ross (358 F. Supp. 3d 965, 2019)

United States District Court for the Northern District of California 2019-03-06

Census Labor The Public Health and Welfare Government Organization and Employees

Citations

Full opinion text

RICHARD SEEBORG, United States District Judge TABLE OF CONTENTS

I. SUMMARY OF DECISION...973

II. BACKGROUND...976

III. STANDING...977

A. Legal Standard...977

B. Findings of Fact Related to Standing...977

C. Conclusions of Law Related to Standing...1002

1. Injury-in-Fact...1002

2. Traceability...1006

3. Redressability...1007

IV. APA CLAIM...1007

A. Legal Standard...1007

B. Scope of Review...1008

C. Findings of Fact Based Exclusively on the Administrative Record...1009

D. Findings of Fact Based on Extra-Record Evidence...1026

E. Conclusions of Law...1037

V. CONSTITUTIONAL CLAIMS...1046

A. Legal Standard...1046

B. Scope of Review...1047

C. Conclusions of Law...1048

VI. REMEDIES...1049

I. SUMMARY OF DECISION

The formal decision by Secretary of Commerce Wilbur L. Ross, Jr. on March 26, 2018 to add a citizenship question to the 2020 Decennial Census violated the Administrative Procedure Act ("APA") and the Enumeration Clause of the United States Constitution. Nearly a year before issuing that decision, on May 2, 2017, Secretary Ross sent an email to Deputy Chief of Staff Earl Comstock stating in part "I am mystified why nothing [has] been done in response to my months old request that we include the citizenship question. Why not?" What ensued was a cynical search to find some reason, any reason, or an agency request to justify that preordained result.

As to the APA, one need look no further than the Administrative Record to conclude that the decision to include the citizenship question was arbitrary and capricious, represented an abuse of discretion, and was otherwise not in accordance with law. In response to Secretary Ross's demand, Comstock began to search for an agency that would be willing to request the inclusion of the citizenship question in the 2020 Census. When initially approached by Comstock about the citizenship question, the Department of Justice ("DOJ") opted not to request its inclusion in the census. Comstock then reached out to the Department of Homeland Security, which similarly declined to request the addition of the question. Only after Secretary Ross personally interceded with then Attorney General Jeff Sessions did the DOJ switch its position and request the inclusion of a citizenship question, ostensibly to assist in the enforcement of Section 2 of the Voting Rights Act ("VRA").

Despite unrefuted evidence produced by the professional staff of the Census Bureau that inclusion of a citizenship question would likely result in a significant differential decline in self-response rates within noncitizen and Latino communities and that the requested data could be obtained by other means, Secretary Ross insisted upon adding the citizenship question to the census. When Census Bureau staff offered to meet with DOJ staff to ascertain if other available data could be used to meet their VRA enforcement needs, DOJ took the unprecedented step of refusing to allow even such an inter-agency meeting to take place.

These facts and other evidence contained in the Administrative Record, along with all reasonable inferences to be drawn therefrom, demonstrate that Secretary Ross's reliance on VRA enforcement to justify inclusion of the citizenship question was mere pretext and the definition of an arbitrary and capricious governmental act. Moreover, Secretary Ross's conclusion that adding the citizenship question would enable the Census Burau to obtain more "complete and accurate data" in response to the DOJ's request is not only unsupported, it is directly contradicted by the scientific analysis contained in the Administrative Record. PTX-26 at 1, 7. While it is of course appropriate for an incoming cabinet member to advocate for different policy directions, to solicit support for such views from other agencies, and to disagree with his or her professional staff, this record reflects a profoundly different scenario: an effort to concoct a rationale bearing no plausible relation to the real reason, whatever that may be, underlying the decision.

Again confining review solely to the Administrative Record, it is evident that the inclusion of the citizenship question on the 2020 Census violated both Sections 6(c) and 141(f)(3) of the 1976 Census Act. Section 6(c) mandates that, to the maximum extent possible, the Secretary use administrative records as opposed to additional census questions to obtain secondary data, such as demographic information. Section 141(f) mandates certain timely reports to Congress regarding the subject and questions to be included on the census and limits the Secretary's ability subsequently to modify the contents of the census absent new circumstances that necessitate a change. Quite simply, Secretary Ross ignored these statutory requirements in issuing his March 26, 2018 decision.

While finding a violation of the APA logically flows from the Administrative Record in this action alone, the facts here satisfy the requisite standard warranting consideration of extra-record evidence. Such evidence includes the absence of any effort to test the impact of the addition of the citizenship question to the census, the deviation from the Census Bureau's usual process for adding new questions to the census, the troubling circumstances under which the DOJ's request letter was drafted and procured, and Sessions' order prohibiting DOJ staff from meeting with Census Bureau officials to discuss alternative sources of data that could meet DOJ's VRA enforcement needs. Going beyond the Administrative Record, in short, confirms that the decision to include a citizenship question runs afoul of the APA.

The analysis of the Enumeration Clause claim similarly involves evidence beyond the four corners of the Administrative Record. As a general proposition, the decision to include a specific question on the census is committed to the discretion of the Commerce Secretary and does not implicate the constitutional command that all persons in each state be counted every ten years. However, if the Secretary's decision to include a question affirmatively interferes with the actual enumeration and fulfills no reasonable governmental purpose, it may form the basis for a cognizable Enumeration Clause challenge.

The evidence admitted in the trial of these actions demonstrates that a significant differential undercount, particularly impacting noncitizen and Latino communities, will result from the inclusion of a citizenship question on the 2020 Census, compounded by macro-environmental factors arising out of the national immigration debate. Efforts to ameliorate these effects through Non-Response Follow-Up ("NRFU"), the evidence showed, would not remediate and could in fact exacerbate the differential undercount of noncitizens and Latino persons. While a citizenship question had been included in the decennial census in 1950 and before, the analysis now must turn on the impact of that question on the prospect of achieving the central constitutional purpose of an actual enumeration in 2020. Viewed through that lens, the inclusion of the question is contrary to the Constitution.

Plaintiffs in each of these actions satisfied their burden of demonstrating standing under Article III of the Constitution. The State of California demonstrated that it will suffer a loss of federal funding and face a substantial risk of losing political representation directly traceable to the inclusion of the citizenship question on the census. California established that the inclusion of this question will also require the expenditure of additional funds to attempt to mitigate the effects of the question and minimize the resulting undercount of California vis-à-vis other states. Similarly, the City of San Jose and the Black Alliance for Just Immigration ("BAJI") each established injury directly flowing from the addition of the citizenship question. In the case of San Jose, it showed the negative impact on federal funding it would receive for various programs dependent on census data and the additional resources that would be required to attempt to mitigate those effects. As to BAJI, the organization demonstrated that it will be obliged to commit additional time and resources to address the specific effects of the citizenship question on its constituents and to encourage them to participate despite the perceived risks.

In light of the statutory and constitutional violations outlined above, the issue becomes the appropriate remedy. With respect to the APA claim, consistent with and for the reasons stated in New York v. United States Dep't of Commerce , No. 18-cv-2921 (S.D.N.Y. Jan. 15, 2019) (the "New York matter"), vacatur of the Secretary's decision, remand to the agency, and an injunction against inclusion of the citizenship question on the 2020 census is warranted and will be ordered. As to the Enumeration Clause violation, an injunction is the proper relief. The Department of Commerce urges that any relief should be limited to the particular plaintiffs before the Court. While mindful of the concerns regarding individual district courts issuing orders of national scope, the limitation advanced by Defendants here is simply impractical in light of the nationwide nature of the questionnaire at issue. Accordingly, no such limitation will be included in the injunctive relief ordered by this Court.

II. BACKGROUND

This action, comprising two related cases, arises from the U.S. Census Bureau's decision to include a question regarding citizenship status on the 2020 Census questionnaire. Plaintiffs in Case No. 18-cv-1865 are the State of California, the County of Los Angeles, the City of Los Angeles, the City of Fremont, the City of Long Beach, the City of Oakland, the City of Stockton, and the Los Angeles Unified School District (collectively, "California Plaintiffs"). Plaintiffs in Case No. 18-cv-2279 are the City of San Jose and the Black Alliance for Just Immigration ("BAJI") (collectively, "San Jose Plaintiffs"). Defendants in both matters are Wilbur L. Ross, Jr., in his official capacity as Secretary of the U.S. Department of Commerce; the U.S. Department of Commerce; Stephen Dillingham, in his official capacity as Director of the United States Census Bureau; and the U.S. Census Bureau. During the relevant period Dr. Ron Jarmin served as Acting Director of the United States Census Bureau.

On March 26, 2018, Secretary Ross issued a memorandum (the "Decision Memo") directing the Census Bureau to add a question on citizenship status to the 2020 Census. PTX-1 at 1313-20. Plaintiffs contend the decision to include this question violated the Constitution and the APA. They specifically argue Secretary Ross's decision violated the Enumeration Clause, U.S. Const. art. I, § 2, cl. 3, and was "arbitrary, capricious, [and] otherwise not in accordance with law" under the APA, 5 U.S.C. § 706(2)(A). The San Jose Plaintiffs also allege a violation of the Apportionment Clause of the Constitution. U.S. Const. amend. XIV, § 2.

The process by which the decennial census is taken has changed significantly over the years, as have the questions asked in the census instrument. From 1790 to 1960, the Bureau collected data directly from households through in-person interviews. Undisputed Fact ("UF") 75. Moreover, from 1820 to 1950, with the exception of 1840, respondents were asked a question concerning citizenship or birthplace. UF 67. The Census Bureau subsequently transitioned to a mailed questionnaire, which involved sending a "short form" questionnaire to most residences, and a "long form" questionnaire with significantly more questions to the remaining households. UF 77-78.

The long form questionnaires used in 1970, 1980, 1990, and 2000 included a question about citizenship status, whereas the short form questionnaires did not. UF 80. After the 2000 Decennial Census, the functions performed by the long form questionnaire were replaced by the American Community Survey ("ACS"). UF 83. The ACS is a yearly survey of approximately 2% of households-about 3.5 million-across the United States. UF 85. A question concerning citizenship status currently appears among more than 50 questions on the ACS questionnaire. UF 86.

In keeping with recent practice, the 2020 Census will be "short form only." UF 102. The ACS will continue to be distributed as usual and will continue to include a citizenship question. UF 103. Per Secretary Ross's Decision Memo, the 2020 Census will also include a citizenship question. The text of this newly added question will read, "Is this person a citizen of the United States?," with the answer options "Yes, born in the United States"; "Yes, born in Puerto Rico, Guam, the U.S. Virgin Islands, or Northern Marianas"; "Yes, born abroad of U.S. citizen parent or parents"; "Yes, U.S. citizen by naturalization - Print year of naturalization"; and "No, not a U.S. citizen." UF 104. As in past years, the 2020 Census questionnaire will also pose questions regarding sex, Hispanic origin, race, and relationship status. UF 106.

III. STANDING

A. Legal Standard

In order to establish standing under Article III of the Constitution, a "plaintiff must have (1) suffered an injury in fact, (2) that is fairly traceable to the challenged conduct of the defendant, and (3) that is likely to be redressed by a favorable judicial decision." Spokeo, Inc. v. Robins , --- U.S. ----, 136 S.Ct. 1540, 1547, 194 L.Ed.2d 635 (2016) (citing Lujan v. Defs. of Wildlife , 504 U.S. 555, 560-61, 112 S.Ct. 2130, 119 L.Ed.2d 351 (1992) ). As the party invoking federal jurisdiction, the plaintiff bears the burden of establishing all three requirements by a preponderance of the evidence. Lujan , 504 U.S. at 561, 112 S.Ct. 2130.

B. Findings of Fact Related to Standing

1. Inclusion of the Citizenship Question on the 2020 Census Will Cause a Differential Decline in Self-Response Rates

1. Undisputed evidence in this case shows that adding a citizenship question to the 2020 Census will cause a differential decline in self-response rates for noncitizen and Hispanic households.

2. Defense expert Dr. John Abowd, Chief Scientist and Associate Director for Research and Methodology at the Census Bureau, testified credibly that the Census Bureau has produced quantitative evidence that adding a citizenship question to the 2020 Census will lower self-response rates. Tr. 797:20-25 (Abowd). He specifically endorsed the Census Bureau's finding that the citizenship question will lead to a lower self-response rate in both noncitizen and Hispanic households, New York Tr. 881:19-882:1 (Abowd); id. at 918:3-919:1 (Abowd), and that this lower response rates will harm the quality of census data, id. at 882:2-5 (Abowd). The Plaintiff experts similarly endorsed these conclusions. See Part III.B.1.e, infra .

a. December 22 Census Bureau Memo

3. The analysis that underpins Dr. Abowd's testimony is set forth in three memoranda issued by the Census Bureau: the December 22 Memo, PTX-148; the January 19 Memo, PTX-22; and the Brown, et al. Memo, PTX-160. New York Tr. 896:7-15 (Abowd); Census Bureau 30(b)(6) Dep. Vol. II 353:2-6, 353:19-21.

4. The December 22, 2017 Memo was authored by senior professional staff at the Census Bureau (nicknamed the "SWAT Team"). They found that, based on a comparison of self-response rates to the 2010 Census and the 2010 ACS (which included a citizenship question), noncitizen households were 5.1 percent less likely than all-citizen households to respond to a survey with a citizenship question. PTX-103 at 6-7; PTX-148 at 6-7. This finding is "consistent with citizenship questions being more sensitive for households with noncitizens." PTX-103 at 7; PTX-148 at 7.

b. Dr. Abowd's January 19 Memo

5. Dr. Abowd's January 19, 2018 Memo conveyed the 5.1 percent differential self-response estimate to Secretary Ross. PTX-22 at 4.

6. This finding was the result of just one of the "[t]hree distinct analyses" in the January 19 Memo that "support the conclusion of an adverse impact on self-response and, as a result, on the accuracy and quality of the 2020 Census."Id. The other two analyses focused on indicators that suggest that Hispanic households are disproportionately less likely to respond to a survey with a citizenship question.

7. The first of these anlyses focused on the item nonresponse rates-the rate at which respondents do not answer a particular survey question. New York Tr. 905:10-24 (Abowd). The Census Bureau found that item nonresponse rates for the citizenship question on the ACS were more than twice as high for Hispanics as for non-Hispanic whites from 2013 through 2016, and that the nonresponse rate for Hispanics increased by 2.5 percent relative to non-Hispanic whites over that span. PTX-22 at 4; see also New York Tr. 906:12-908:6 (Abowd); Tr. 156:4-157:19 (O'Muircheartaigh).

8. The second analysis considered breakoff rates-the rate at which respondents stop completing a survey when presented with a particular question. New York Tr. 913:13-24 (Abowd). It found that the breakoff rate for the citizenship question on the 2016 ACS was more than eight times higher for Hispanics than for non-Hispanic whites. PTX-22 at 5; see also New York Tr. 914:5-8 (Abowd); Tr. 158:4-21 (O'Muircheartaigh). Similarly, the breakoff rate for three related questions on immigration status (citizenship, place of birth, and year of entry) on the 2016 ACS was more than three times higher for Hispanics than for non-Hispanic whites. PTX-22 at 5; see also New York Tr. 915:9-13 (Abowd).

9. Based on the Census Bureau's analysis of item nonresponse rates and breakoff rates, Dr. Abowd testified credibly that a citizenship question would be sensitive for Hispanics, and that the sensitivity of the question is increasing for Hispanics (but not for non-Hispanic whites). New York Tr. 917:4-918:2 (Abowd).

c. Brown, et al. Memo

10. The Brown, et al. Memo builds upon and updates the analysis in Dr. Abowd's January 19 Memo. New York Tr. 896:7-12 (Abowd). This memo represents the Census Bureau's best analysis of the consequences of adding a citizenship question to the 2020 Census. Census Bureau 30(b)(6)

Dep. Vol. II 355:15-356:15; New York Tr. 897:4-15 (Abowd).

11. The Brown, et al. Memo summarized its findings as follows:

This paper's examination of several Census Bureau surveys with and without citizenship questions suggests that households that may contain noncitizens are more sensitive to the inclusion of citizenship in the questionnaire than all-citizen households. The implication is that adding a citizenship question to the 2020 Census would lead to lower self-response rates in households potentially containing noncitizens, resulting in more nonresponse follow-up (NRFU) fieldwork, more proxy responses, and a lower-quality population count.

PTX-160 at 54.

12. The memo presented data showing that citizenship-related questions are more sensitive for Hispanics and that, because Hispanics have higher rates of nonresponse for citizenship than for sex or age, they could be disproportionately impacted by adding a citizenship question to the 2020 Census questionnaire. PTX-160 at 7-10.

13. The Census Bureau also updated the estimated 5.1 percent differential decline in the self-response rate of noncitizen households to 5.8 percent. PTX-160 at 39; Census Bureau 30(b)(6) Dep. Vol. II 372:2-12; New York Tr. 897:16-20 (Abowd); Tr. 161:13-21 (O'Muircheartaigh). This revised estimate is the result of a natural experiment that compared response rates on the 2016 ACS, which included a citizenship question, to response rates on the 2010 Census, which did not incorporate a citizenship question, and then compared the change in response rates between all-citizen households and all other households (i.e., households that contain or may contain one or more noncitizens). PTX-160 at 33-34; Census Bureau 30(b)(6) Dep. Vol. II 373:9-15, 374:10-16; New York Tr. 898:2-899:6 (Abowd); Tr. 161:22-164:17 (O'Muircheartaigh).

14. The 5.8 percent figure represents the Census Bureau's best conservative estimate of the differential effect of the citizenship question on noncitizen household self-response. New York Tr. 894:17-895:2, 897:9-12 (Abowd).

15. The Brown, et al. Memo emphasized that the 5.8 percent estimate is "conservative." PTX-160 at 39; New York Tr. 900:21-25 (Abowd); Tr. 164:21-24 (O'Muircheartaigh). The Bureau acknowledged that this figure may underestimate the impact of the citizenship question on census self-response rates for two reasons: (1) the question will be more prominent on the 2020 Census questionnaire, which has just ten other questions, than it was on the ACS questionnaire, which has 75 questions, PTX-160 at 39; New York Tr. 901:22-902:10 (Abowd); Tr. 164:25-165:14 (O'Muircheartaigh), and (2) given "the level of concern about using citizenship data for enforcement purposes," the macro-environment at the time of the 2020 Census may be worse than it was when the ACS data were collected, PTX-160 at 39; see also New York Tr. 902:11-24 (Abowd); Tr. 165:15-21 (O'Muircheartaigh).

16. The 5.8 percent estimate is also conservative because of limitations in the design of the natural experiment. For example, the natural experiment assumed that individuals whose citizenship information was missing from administrative records were citizens, which had the effect of reducing the estimated difference between the response rates of all-citizen households and noncitizen households. Tr. 165:25-166:15 (O'Muircheartaigh).

17. The Brown, et al. Memo also confirmed the findings in Dr. Abowd's January 19 Memo showing that (1) Hispanics were more than twice as likely as non-Hispanic whites to skip the citizenship question on the ACS and that the differential in such item nonresponse rates increased between 2013 and 2016, and (2) the breakoff rate for the citizenship question on the 2016 ACS was more than eight times higher for Hispanics than for non-Hispanic whites. PTX-160 at 8-11. Based on this data, the Census Bureau concluded that Hispanics are more sensitive to survey questions about citizenship than they were a few years ago but that non-Hispanic whites are not. Census Bureau 30(b)(6) Dep. Vol. II 369:1-19. This suggests that nonresponse rates to the citizenship question on the 2020 Census will be higher for Hispanics than for non-Hispanic whites. New York Tr. 910:7-13, 914:9-12 (Abowd).

18. Recent Census Bureau data show that the differential breakoff is escalating. After the January 19 Memo and the Brown, et al. Memo were issued, the Census Bureau made the 2017 ACS breakoff data publicly available. New York Tr. 915:19-916:3 (Abowd). That data, which was reviewed by the SWAT team, showed that the breakoff rate for the citizenship question on the 2017 ACS is now twelve times higher for Hispanics than for non-Hispanic whites. New York Tr. 916:4-917:3 (Abowd).

d. CSM Memo and CBAMS Results

19. Recent Census Bureau qualitative research suggests that the citizenship question will cause an even greater differential decline in self-response rates than estimated by Brown, et al. The macro-environment, particularly the political environment around immigration, has the potential to amplify the negative effect of the citizenship question on self-response rates. New York Tr. 926:21-927:10 (Abowd).

20. This research includes the Census Bureau's Center for Survey Measurement ("CSM") focus group testing in 2017, which revealed increased concern among immigrants about the confidentiality of their survey responses, PTX-157 at 1, and the Census Barriers, Attitudes, and Motivators Study (CBAMS) conducted in 2018, which revealed concerns among Spanish language respondents about the citizenship question, PTX-153 at 21-22.

i. CSM Findings

21. CSM researchers summarized the respondent confidentiality concerns they observed in a September 20, 2017 memo for the Associate Directorate for Research and Methodology at the Census Bureau, PTX-157, and in presentations of their findings to the American Association of Public Opinion Research (AAPOR), PTX-158, and to the National Advisory Committee on Racial, Ethnic, and Other Populations, PTX-326.

22. During the pretesting studies conducted in 2017, CSM researchers "noticed a recent increase in respondents spontaneously expressing concerns about confidentiality" and "reported that respondents' fears, particularly among immigrant respondents, have increased markedly this year." PTX-157 at 1.

23. For example, CSM researchers observed Spanish-speaking respondents who were "uncomfortable 'registering' other household members," who "left three or four roomers off the roster" and "mentioned being worried because of their '[immigration] status,' " and who stated that "the Latino community will not sign up because they will think that Census will pass their information on and people can come looking for them." Id. at 2.

24. CSM researchers observed that "this level of deliberate falsification of the household roster, and spontaneous mention of concerns regarding negative attitudes toward immigrants, is largely unprecedented in the usability interviews that CSM has been conducting since 2014 in preparation for the 2020 Census." Id. at 3. CSM researchers worried that the concerns expressed by immigrant respondents might be "even more pronounced" during the 2020 Census, because respondents are generally more willing to participate in pretesting surveys "given that they are being paid a cash incentive for their participation and [are] being interviewed by a researcher with whom they have established rapport." Id.

25. During focus group testing, respondents similarly expressed "fear of deportation[ ] [and] concern about how the data are used[ ] and which agencies can see it," specifically asking whether the Department of Homeland Security ("DHS") or Immigration and Customs Enforcement ("ICE") would have access to their data. PTX-326 at 9.

ii. CBAMS Findings

26. The CBAMS is a survey of 50,000 households in a series of 42 focus groups designed to inform the integrated partnership and communications program for the 2020 Census about the macro-environment. Census Bureau 30(b)(6) Dep. Vol. II 437:17-438:6; New York Tr. 927:22-928:6 (Abowd). The Census Bureau finds CBAMS research sufficiently reliable to provide actionable information for the integrated partnership and communications program. Census Bureau 30(b)(6) Dep. Vol. II 438:7-11.

27. After Secretary Ross announced that the 2020 Census would include a citizenship question, Census Bureau researchers began asking for feedback about the question from 30 of the 42 focus groups, including Spanish-language groups. PTX-161 at 6; see also New York Tr. 930:16-19 (Abowd).

28. The CBAMS found that in the Spanish-language (U.S. Mainland) focus groups, the citizenship question was a "determining factor for participation." PTX-153 at 22. Although most participants said that they were not afraid to answer the citizenship question because they are citizens or legal residents, they knew many others who would not participate in the 2020 Census "out of fear." Id. While all participants wanted to participate in the 2020 Census, "fear of deportation outweighs any benefit." Id.

29. The Census Bureau views the results of the Spanish-language focus groups with respect to the citizenship question as "extremely problematic." Census Bureau 30(b)(6) Dep. Vol. II 450:16-451:1; New York Tr. 934:8-12 (Abowd). Other immigrant and non-white groups raised similar concerns. New York Tr. 930:9-24, 938:22-939:17, 940:4-941:14 (Abowd).

30. Census Bureau researchers ultimately concluded that "[t]he citizenship question may be a major barrier" to participation in the 2020 Census because respondents, including citizens and legal residents, believed that the census's purpose "is to find undocumented immigrants" and because "[t]he political discourse is targeting their ethnic group." PTX-465 at 43.

31. The CBAMS results suggest that the citizenship question is sensitive in the current macro-environment and is a "major concern" for the Census Bureau's efforts to encourage participation in the 2020 Census within Hispanic communities. New York Tr. 944:7-24 (Abowd). Moreover, the increased sensitivity to the citizenship question that was observed in the 2018 CBAMS results was likely not captured in Brown, et al.'s 5.8 percent estimate, which was based on 2016 data. Id. at 944:25-945:4 (Abowd).

e. The Plaintiff Experts' Testimony

32. The Plaintiff experts' testimony further supports the conclusion that the citizenship question will cause a greater differential decline in self-response rates than estimated by Brown, et al.

33. Dr. Colm O'Muircheartaigh, professor in the Harris School of Public Policy and senior fellow at the National Opinion Research Center (NORC) at the University of Chicago, testified that he agrees with the Census Bureau research discussed above. Tr. 33:4-17; 145:15-166:15 (O'Muircheartaigh). Dr. O'Muircheartaigh also cited additional factors that will exacerbate the effects of the differential decline in self-response rates caused by the citizenship question on the ultimate enumeration. Id. at 166:16-174:20 (O'Muircheartaigh).

34. First, missing units in the Census Bureau's Master Address File (MAF) contain a disproportionate number of immigrant and noncitizen households. Tr. 166:21-25 (O'Muircheartaigh). The MAF is the "first building block" of census data collection. Id. at 122:4-6. The MAF is constantly updated throughout the census-taking process. Tr. 803:23-805.7 (Abowd). In general, the census is unlikely to count persons whose households do not appear on the MAF. Id. at 46:1-6 (O'Muircheartaigh). Dr. O'Muircheartaigh testified that social science research, including recent research on Mexican immigrants, has observed that the Census Bureau has particular difficulty identifying household addresses for immigrants and noncitizens. Id. at 122:7-123:13, 124:7-17 (O'Muircheartaigh). To the extent that immigrant and noncitizen households are not identified by the Census Bureau and included in the MAF, and the residents of these households choose not to come forward to be counted because of the citizenship question, such households and their residents will not be included in the 2020 Census despite the Census Bureau's NRFU efforts. Id. at 166:21-167:14 (O'Muircheartaigh).

35. Second, respondents, especially those that live in households containing noncitizens, may omit certain household members on the census questionnaire because of fears generated by the citizenship question. Id. at 167:15-20 (O'Muircheartaigh). In particular, the 2017 CSM research observed that Spanish-speaking respondents were reluctant to provide a complete roster of household members. Id. at 147:18-148:16 (O'Muircheartaigh) (citing PTX-157). Dr. O'Muircheartaigh testified that such rostering omissions are a particularly problematic form of nonresponse because "[t]he quality of the census is fundamentally dependent on complete rostering of individuals within households," and "the census protocol has no mechanism for remediating such a response." Id. at 147:10-16, 148:8-149:9 (O'Muircheartaigh); Census Bureau 30(b)(6) Dep. Vol. II 397:19-399:2, 459:21-460:7.

36. Dr. O'Muircheartaigh credibly testified to the following conclusions relating to the impact of the citizenship question on self-response: (1) current survey methodology research, primarily by the Census Bureau, has observed that Latinos and immigrants hold considerable fears about participating in the 2020 Census, (2) the citizenship question will increase the Census Bureau's misidentification of households as unoccupied, particularly among Latinos and households with noncitizens, (3) the citizenship question will depress self-response rates, particularly for Latinos and households with noncitizens, and the Census Bureau's conservative estimate is that the self-response rate for households containing a noncitizen will be 5.8 percent lower than for all-citizen households, and (4) factors such as rostering errors will exacerbate the difference in the effective self-response rates of noncitizens versus citizens. Tr. 175:1-19 (O'Muircheartaigh).

37. Dr. Matthew Barreto, a professor of political science and Chicano studies at the University of California, Los Angeles, Tr. 366:13-17 (Barreto), similarly testified that adding a citizenship question to the 2020 Census will reduce self-response rates, particularly among immigrants and Latinos, Id. at 374:7-15 (Barreto). Dr. Barreto's findings were based on a comprehensive literature review of research publications and reports, including those produced by the Census Bureau, related to response rates (as well as NRFU and imputation); an original survey he fielded in which he asked people about whether they intend to participate in the 2020 Census; and his expertise and years of experience implementing surveys in Latino and immigrant communities. Id. at 375:18-376:4, 379:19-380:7 (Barreto) (citing PTX-499).

38. Dr. Barreto identified three interrelated factors that affect survey participation: (1) trust, (2) sensitive questions, and (3) the macro-environment in which the survey is administered. Tr. 380:19-381:7, 383:13-16 (Barreto). Applying the literature on these factors to the citizenship question, Dr. Barreto concluded that the citizenship question will cause a significant decline in self-response rates on the 2020 Census because it is a sensitive question that will exacerbate trust issues in the current macro-environment, particularly for immigrants and immigrant-adjacent communities. Id. at 386:21-25, 411:5-14 (Barreto). Dr. Barreto defined "immigrant-adjacent communities" as communities with mixed-status households, where one family member is a U.S. citizen and another family member is not, and communities in which residents would interact with immigrants daily at work, school, or in other similar environments. Id. at 387:1-14 (Barreto).

39. A consistent finding in the social science research is that "if a potential respondent does not trust the survey taker to keep their information confidential and not put them at risk, then the survey respondent won't participate in the survey at all." Tr. 381:17-23 (Barreto). With regard to census participation specifically, Dr. Barreto observed that the Census Bureau, particularly in Manuel de la Puente's ethnographic studies of the 1990 and 2000 Censuses, found that "immigrant and undocumented populations in particular [ ] don't trust the federal government to fully protect or keep in confidence their information." Id. at 385:3-19, 390:12-395:7 (Barreto) (citing PTX-308 and PTX-309), see also id. at 388:11-389:8 (Barreto) (citing PTX-339). To break down the barriers he observed in his studies, Dr. de la Puente recommended that the Census Bureau work with community groups to assure them that the Census Bureau isn't seeking information about respondents' citizenship status. Id. at 393:25-394:15 (Barreto).

40. Moving on to the second factor, Dr. Barreto testified that "a sensitive question is one that asks a respondent for some very personal information that they may be uncomfortable revealing." Tr. 383:2-6 (Barreto). Social science research suggests that survey takers should "reduce unnecessary sensitive questions because they do create considerable trust issues with respondents." Id. at 383:10-12 (Barreto). Whether a question is sensitive varies in different environments and contexts and across subpopulations. Id. at 383:13-20, 384:19-385:2 (Barreto). Dr. Barreto observed that the citizenship question is likely to be most sensitive to "those who are closer to the immigrant experience or closer to [ ] immigrant communities," particularly "in the Latino community where there have been concerns over immigration-related issues over the past few years." Id. at 387:15-23 (Barreto).

41. The third factor, macro-environment, is "the context in which any survey is being implemented," including "the social and political environment, the atmosphere that is present when the survey is being administered." Id. at 395:11-19 (Barreto). A respondent "may be more willing to participate if the context or the environment seems very agreeable and welcoming, and they may be far less likely to participate if the environment seems threatening or concerning." Id. at 395:20-25 (Barreto). Dr. Barreto observed that social science research has found that "[i]mmigrants and mixed-status households are likely to avoid government contact when they suspect it is unsafe to participate." Id. at 397:19-398:2 (Barreto). This observation holds true for a census with a citizenship question, because the question will be asked in a macro-environment that is perceived by many immigrants to be "threatening or negative." Id. at 396:3-13 (Barreto).

42. To evaluate participation in the 2020 Census, Dr. Barreto conducted a large national survey that inquired about people's attitudes and behaviors. Id. at 411:15-23 (Barreto). Within the scientific community, survey research is considered reliable and has predictive value. Id. at 414:2-7 (Barreto).

43. Dr. Barreto conducted his survey on a sample of 6,309 respondents from across the United States, including oversamples of Latinos nationwide and residents of the State of California, the City of San Jose, and two border counties in Texas. Id. at 424:8-19 (Barreto). Respondents were randomly chosen, and weighting was applied to balance out the demographic characteristics of the sample. Id. at 415:19-418:12, 434:2-435:19 (Barreto). In addition, the survey response rate-28.1 percent-was within the American Association of Public Opinion Research (AAPOR) response rate guidelines (at least 20 to 30 percent) for telephone surveys. Id. at 425:2-23 (Barreto).

44. Dr. Barreto set forth the results of his survey in a number of tables. PTX-499A, PTX-863 through PTX-890. He estimates that, because of the citizenship question, census response rates are likely to decline between 6.3 and 8.0 percent nationally and between 10.5 and 14.1 percent in the State of California. PTX-870; PTX-871; see also Tr. 457:17-458:3 (Barreto) (explaining PTX 870); id. at 461:9-20 (Barreto) (explaining PTX-871). The nonresponse rate attributed to the citizenship question in California is statistically higher than the nationwide average. Tr. 463:8-464:15 (Barreto) (explaining PTX-873).

45. Based on the Census Bureau's most current data, the average Latino household is larger than the average non-Latino household. Id. at 1036:12-1037:6 (Abowd). By factoring in the difference in average household size between Latino households and other households, Dr. Barreto estimated that Latinos would constitute approximately 35 percent (over 10 million) of the total number of persons (approximately 28 million) that would not self-respond to the 2020 Census because of the citizenship question, far surpassing the rate of Latinos in the national population (18 percent). Id. at 478:1-481:6 (Barreto) (explaining PTX-880 and PTX-881). This evidence further supports the conclusion that Latinos will be disproportionately affected by the citizenship question. Id. at 480:9-14 (Barreto).

46. Although Dr. Barreto's study provides credible evidence that the inclusion of the citizenship question on the 2020 Census is likely to cause a decline in self-response rates among certain demographic groups relative to the rest of the population, some aspects of the survey design and methodology limit the weight the Court affords to this evidence. In particular, Dr. Barreto asks in Question 2 whether respondents would participate in the 2020 census if the federal government were to include a citizenship question on the questionnaire. Tr. 576:4-13. By contrast, Question 1 specifically referred to the Census Bureau as the agency responsible for the census and asked respondents, without mentioning the citizenship question, whether they would participate in the 2020 census. Id. at 440:13-441:2. It is plausible that respondents who are more distrustful of the federal government writ large than they are of the Census Bureau in particular may have responded negatively to Question 2 in part because of this difference in terminology. While this hardly represents a glaring flaw, it does diminish somewhat the weight to be afforded to the drop in willingness to respond between Question 1 and Question 2.

47. Defendants' remaining arguments that the decline in self-response rates between Question 1 and Question 2 should not be credited are unpersuasive. In particular, the fact that a randomized controlled trial ("RCT") may produce more accurate results than a survey does not automatically render Dr. Barreto's survey unreliable. See Tr. 874:10-19.

48. Ultimately, respondents' increased reluctance to participate in the census between Question 1 and Question 2 of Dr. Barreto's survey provides credible evidence that the addition of the citizenship question is likely to result in a significant decline in self-response rates in California and within the Latino population relative to the public at large.

2. NRFU Will Not Remediate the Differential Decline in Self-Response Rates

49. In keeping with recent practice, the Census Bureau will implement a series of NRFU operations to attempt to count the significant number of persons who do not self-respond to the 2020 Census, UF 39-47, including the millions who will not self-respond because of the citizenship question, PTX-22 at 6; PTX-160 at 42; New York Tr. 894:1-16 (Abowd). All available evidence indicates that at every NRFU stage, including the imputation phase, the Census Bureau will be differentially less effective at counting noncitizens and Latinos-the very subpopulations most likely not to respond to the 2020 Census because of the citizenship question. Tr. 175:20-218:6 (O'Muircheartaigh).

a. Hard-to-Count Populations

50. The Census Bureau has always struggled to count hard-to-count subpopulations, including noncitizens and Latinos, even when the census count for the national population has been fairly accurate. Tr. 57:17-60:8 (O'Muircheartaigh). For example, as measured in the Census Bureau's post-enumeration surveys, Hispanics have been differentially undercounted compared to non-Hispanic whites in each of the last three censuses. Id. at 55:2-15, 56:11-57:5 (O'Muircheartaigh); UF 61-62. In the 2010 Census, Hispanics were undercounted by 1.54 percent and non-Hispanic whites were overcounted by .84 percent, resulting in a net differential undercount of Hispanics of 2.38 percent. Tr. 56:11-24; PTX-211 at 18. In the 2000 Census, Hispanics were undercounted by .71 percent and non-Hispanic whites were overcounted by 1.13 percent, resulting in a net differential undercount of Hispanics of 1.84 percent. PTX-211 at 18. In the 1990 Census, Hispanics were undercounted by 4.99 percent and non-Hispanic whites were undercounted by .68 percent, resulting in a net differential undercount of Hispanics of 4.31 percent. Id.

51. Hard-to-count subgroups include low-income persons, persons who do not live in traditional housing, persons who do not speak English fluently or have limited English proficiency, persons who have distrust in the government, racial and ethnic minorities, renters, undocumented immigrants or recent immigrants, and young children. Tr. 1021:19-1023:2 (Abowd); UF 59-60. Census Bureau research shows that there is "substantial overlap" between these hard-to count subgroups and those households most likely not to respond to the 2020 Census because of the citizenship question. Tr. 1023:3-7 (Abowd).

52. The Census Bureau has identified four primary obstacles to counting hard-to-count subpopulations: that they are hard to locate, hard to contact, hard to persuade, and hard to interview. Id. at 1023:8-24 (Abowd). For some hard-to-count subgroups, more than one of these obstacles applies. Id. at 1024:7-13 (Abowd). Census Bureau research acknowledges that these obstacles apply to those households most likely not to respond to the 2020 Census because of the citizenship question. Id. at 1023:25-1024:6 (Abowd).

b. The Census Bureau's Partnership and Communications Program

53. The Census Bureau has developed a range of strategies to address the net differential undercount of "hard-to-count" populations-including targeted marketing and outreach efforts, partnerships with community organizations, deployment of field staff to follow up with individuals who do not respond, and retention of staff with foreign language skills. UF 64.

54. In the 2000 and 2010 Censuses, the Census Bureau designed and implemented public advertising campaigns to reach hard-to-count immigrant communities, including using paid media in over a dozen different languages to improve responsiveness, and partnered with local businesses, faith-based groups, community organizations, elected officials, and ethnic organizations to reach these communities and improve the accuracy of the count. UF 65-66.

55. Defendants believe that a similar integrated partnership and communications campaign, in tandem with the Census Bureau's NRFU efforts, may mitigate the decline in self-response rates in the 2020 Census. Tr. 798:6-12, 799:21-800:14 (Abowd). Yet there is no evidence in the Administrative Record that Defendants' planned integrated partnership and communications campaign for the 2020 Census will significantly mitigate such a differential decline in self-response rates. Dr. Abowd agreed that it is "highly unlikely" that the integrated partnership and communications campaign can eliminate the negative effects of adding a citizenship question. Id. at 980:3-11 (Abowd).

56. The Census Bureau also acknowledges that the "trusted partners" that it relies on to convey the importance of participating in the census will have additional challenges communicating that message if the 2020 Census includes the citizenship question. Census Bureau 30(b)(6) Dep. Vol. II 451:21-452:4, 453:2-17; New York Tr. 937:16-23 (Abowd). The CBAMS focus groups of Spanish-speaking respondents found that, "while there were suggestions of trusted voices, there does not seem to be a single trusted voice that could mitigate [respondents'] distrust of the government to uphold the promise of confidentiality." PTX-153 at 22. Dr. O'Muircheartaigh persuasively testified that this observation shows the citizenship question will "reduce[ ] the potential impact of the positive input of constituency, community, and association leaders" as these trusted voices attempt to convince their constituents to participate in the 2020 Census. Tr. 153:1-154:9 (O'Muircheartaigh).

57. Census Bureau research has noted one messaging strategy that is reassuring to Spanish-speaking respondents is to convey that "[n]one of the questions in this survey will ask about immigration status" and that "[b]y law, [the respondent's] answers cannot be shared with Immigrations and Customs Enforcement." PTX-158 at 16. Dr. Barreto similarly observed that, consistent with the findings in Dr. de la Puente's ethnographic studies, the most effective way- indeed, perhaps the "only way"-to address confidentiality concerns related to the citizenship question is "to assure respondents that no citizenship information is being gathered" in the 2020 Census. Tr. 500:17-501:5 (Barreto). Neither the Census Bureau nor trusted partners can offer such assurances because the citizenship question will be on the 2020 Census, unless the Census Bureau is instructed to remove it. Tr. 1052:8-12 (Abowd).

58. Moreover, despite the barriers to participation in the 2020 Census associated with the citizenship question, the Census Bureau has not significantly increased its spending on 2020 census outreach relative to that expended in 2010. Tr. 1024:18-1025:9 (Abowd).

c. The Census Bureau's NRFU Operations

59. The Census Bureau's NRFU workload includes all households that do not initially self-respond to the census. Tr. 851:16-852:2 (Abowd). In the 2010 Census, over 27 percent of the persons enumerated were in the NRFU workload. PTX-211 at 32-33 (subtracting from the U.S. total population (300,703,000) those persons not in any NRFU universe (219,207,000) and dividing by the total population). The NRFU workload for the 2020 Census is expected to rise to between 34.5 and 44.5 percent of the total population. PTX-1 at 172.

60. The Census Bureau's best conservative estimate is that adding a citizenship question to the 2020 Census will increase the NRFU workload by 2.09 million households and 6.5 million persons. PTX-160 at 42.

61. Based on his survey data, Dr. Barreto estimated that adding a citizenship question to the 2020 Census will increase the NRFU workload by at least 28 million persons, and that Latinos will be disproportionately represented in that workload. PTX-880; Tr. 480:5-14 (Barreto).

62. The Bureau's NRFU operations are designed to obtain an accurate count-and thus, to prevent an undercount-at the national level. Tr. 918:11-16 (Abowd). In recent censuses, however, the Bureau's NRFU operations have been less effective at counting some subpopulations than others. Tr. 178:7-23 (O'Muircheartaigh).

63. Dr. Abowd testified that he is unaware of any "credible quantitative evidence" that adding a citizenship question will increase the net differential undercount of any subpopulation, after accounting for NRFU operations. Tr. 918:21-24 (Abowd). Dr. Abowd admitted, however, that it is "highly unlikely" that the Census Bureau's NRFU operations will eliminate a differential undercount in the 2020 Census. Id. at 980:12-981:2.

64. The Census Bureau's NRFU operations for the 2020 Census include in-person follow-up enumeration, proxy enumeration, administrative record enumeration, and imputation by other methods. UF 39-46; Tr. 176:13-177:20 (O'Muircheartaigh). The Census Bureau's NRFU operations for the 2010 Census included these same processes, with the exception of administrative record enumeration, which was used only on an experimental basis in 2010. Census Bureau 30(b)(6) Dep. Vol. II 400:19-401:21.

65. The weight of the evidence ultimately shows that these NRFU efforts are unlikely to mitigate significantly the differential decline in self-response caused by the citizenship question and may in fact exacerbate the problem. See Tr. 217:4-218:5 (O'Muircheartaigh).

i. In-Person Follow-Up Enumeration

66. The Census Bureau has repeatedly acknowledged that "[t]hose refusing to self-respond due to the citizenship question are particularly likely to refuse to respond in NRFU as well." PTX-25 at 4; see also PTX-160 at 41, 42 n.59 ("Households deciding not to self-respond because of the citizenship question are likely to refuse to cooperate with enumerators coming to their door....").

67. Although in-person follow-up enumeration is typically more effective than mail solicitation, "in this case for this population, the level of threat embodied by a federal agent arriving at your residence to collect the information is far greater than the threat that might be implied by a piece of paper [ ] that arrives at your residence." Tr. 190:2-10 (O'Muircheartaigh).

68. Given these conditions, the enumeration errors that will result "may not be avoidable simply by spending more money on fieldwork. Once a household decides not to cooperate, it may not be possible to obtain an accurate enumeration no matter how many times an enumerator knocks on their door." PTX-160 at 43 n.60; see also Tr. 190:20-191:21 (O'Muircheartaigh).

69. Recent data from ACS in-person follow-up enumeration efforts, specifically the Computer-Assisted Personal Interviewing (CAPI) operation, underscores the challenges that enumerators will face in the 2020 Census if, like the ACS, the census includes a citizenship question. Census Bureau 30(b)(6) Dep. Vol. I 124:19-133:17; Tr. 178:24-185:19 (O'Muircheartaigh) (describing PTX-138). The data, which was collected between 2010 through 2016, is consistent with the notion that questions on citizenship have become more sensitive since 2010. Census Bureau 30(b)(6) Dep. Vol. I 131:4-11.

70. The CAPI data exhibit the following trends: (1) in-person follow-up enumeration has been less effective over time in all census tracts, (2) in-person follow-up enumeration has been differentially less effective in census tracts with a higher proportion of households containing a noncitizen, and (3) the differential between census tracts with a higher proportion of households containing a noncitizen and census tracts with a lower proportion of households containing a noncitizen has grown over time. Census Bureau 30(b)(6) Dep. Vol. I 129:22-130:4, 131:4-18, 133:8-17, Tr. 180:17-181:3 (O'Muircheartaigh).

71. The most recent CAPI data-from 2016-for the half of the population with a higher proportion of households containing a noncitizen indicate that in-person follow-up enumeration was 86.63 percent successful. Tr. 183:21-185:9 (O'Muircheartaigh). This rate "is an approximate representation of how...such households might behave in the context of the census." Id. at 185:10-19 (O'Muircheartaigh). Indeed, the success rate was lower (and conversely, the non-interview rate was higher) for in-person follow-up enumeration in the 2016 End-to-End Test and the 2018 End-to-End Test. Id. at 186:19-187:14 (describing PTX-482 at 26).

72. None of the testing that has been used to plan NRFU staffing levels, the number of field offices, enumerator training, NRFU protocols, or census questionnaire assistance has accounted for a citizenship question on the 2020 Census. Census Bureau 30(b)(6) Dep. Vol. I 198:2-10, 200:9-201:10. Although the Census Bureau's NRFU operations were used in the 2018 End-to End Test, Tr. 819:15-820:9 (Abowd), it did not include a citizenship question, Census Bureau 30(b)(6) Dep. Vol. I 225:13-16; Tr. 820:14-15 (Abowd).

73. The Census Bureau considers the NRFU operations to have been a success in the 2018 End-to-End Test. Tr. 820:19-23 (Abowd). But a U.S. Government Accountability ("GAO") report on NRFU implementation in connection with that test "raises some serious concerns." Tr. 98:3-8 (O'Muircheartaigh) (describing PTX-482).

74. That the Census Bureau did not determine the procedures for late-NRFU data collection until after it started work, for example, "seriously undermines the potential of the activity to be successful." Id. at 98:9-99:7. (O'Muircheartaigh) (describing PTX-482 at 11). This finding, in combination with similar findings that the field workforce was unprepared for certain enumeration challenges, id. at 99:8-100:15 (O'Muircheartaigh), and lacked adequate training, id. at 186:9-18 (O'Muircheartaigh) (describing PTX-482), led Dr. O'Muircheartaigh to conclude that the report was "a little disturbing." Id. at 101:9-12 (O'Muircheartaigh). These findings "cast[ ] doubt on...any projections that the Census Bureau has about how successfully it will operate in 2020, compared, for example, to 2010." Id. at 101:16-102:4 (O'Muircheartaigh).

ii. Proxy Enumeration

75. Locating a proxy respondent-a neighbor, landlord, postal worker, or other knowledgeable person who will provide information about another household-is generally not easy. Tr. 195:2-10 (O'Muircheartaigh). The Census Bureau expects that, just as with in-person follow-up enumeration, in census tracts with a higher proportion of households containing a noncitizen, the proxy enumeration rate will be lower than in other tracts. Census Bureau 30(b)(6) Dep. Vol. II 386:2-15; Tr. 196:25-197:6 (O'Muircheartaigh).

76. In other words, the challenge of finding willing proxy respondents will be greater in neighborhoods with households that are "fearful of the Administration and fearful of Census." Tr. 195:13-25 (O'Muircheartaigh). Potential proxy respondents will be "less likely to want to cooperate" if they are concerned about reporting undocumented immigrants. Id. at 521:15-522:2 (Barreto). Given that "reference persons are much less likely to answer the citizenship question for nonrelatives in the household than for themselves...they may be even less likely to answer it for neighbors." PTX-160 at 43; Census Bureau 30(b)(6) Dep. Vol. II 386:16-387:10; Tr. 523:3-17 (Barreto).

77. Even if located and willing to provide a response, proxy respondents generally provide lower quality enumeration data than self-responses. Census Bureau 30(b)(6) Vol. II 382:17-21; PTX-22 at 6; Tr. 931:14-24, 951:11-14 (Abowd). For example, in the 2010 Census, 97.3 percent of self-responses resulted in a correct enumeration, but the correct enumeration rate for proxy responses was just 70.2 percent. PTX-160 at 42 (citing PTX-211 at 33); Tr. 197:14-198:5 (O'Muircheartaigh).

78. Proxy responses are particularly inaccurate for persons in tenuous residential arrangements-a subpopulation that is disproportionately made up of Latinos and immigrants. Tr. 198:6-200:4 (O'Muircheartaigh). Because of the nature of these living arrangements-which include, for example, converted garages-proxy respondents "may not actually know how many people live there." Id. at 522:3-8 (Barreto).

79. Census Bureau research has also found that "proxies supply poor quality individual demographic and socioeconomic information about the person on behalf of whom they are responding." PTX-160 at 41-42; Tr. 200:23-201:8 (O'Muircheartaigh); id. at 937:6-19 (Abowd). Dr. Abowd conceded that the increased use of proxy responses "does impact data quality," including the quality of characteristic data. Id. at 887:13-24 (Abowd).

iii. Administrative Record Enumeration

80. Census Bureau research has observed that the quality of administrative records varies depending on the subpopulation. Tr. 204:18-205:3 (O'Muircheartiagh) (describing PTX-288). More specifically, the Bureau is less likely to be able to use administrative records to enumerate hard-to-count subpopulations, including noncitizens and Hispanics. Jarmin Dep. 285:1-286:20; Tr. 948:7-949:12 (Abowd), 205:4-12 (O'Muircheartaigh). Undocumented immigrants are particularly unlikely to be found in administrative records and will be harder to enumerate using such records. Census Bureau 30(b)(6) Dep. Vol. II 391:4-19; Tr. 205:13-17 (O'Muircheartaigh). Accordingly, the Census Bureau does not expect administrative record enumeration to be as successful with noncitizens as with citizens. Census Bureau 30(b)(6) Dep. Vol. II 391:21-392:4.

81. Similarly, the Census Bureau will be unable to link Hispanics to administrative records at as high a rate as it can link non-Hispanic whites. Census Bureau 30(b)(6) Dep. Vol. II 389:12-390:5.

82. Given the inability of the Census Bureau to use administrative records to count the very subpopulations most likely not respond to the 2020 Census because of the citizenship question, administrative record enumeration will not remediate the differential decline in self-response rates and may indeed exacerbate any differential undercount of noncitizens and Latinos. Tr. 206:4-19 (O'Muircheartaigh).

iv. Imputation

83. If the Census Bureau is unable to enumerate a household through other NRFU operations, it will impute, or model, the number of persons in the household and their characteristics. Tr. 942:17-20 (Abowd). In the decennial census, the Bureau uses "count imputation" to impute the size of the household, and "whole-person imputation" to impute both the size of the household and the characteristics of the people in the household. Id. at 892:10-15 (Abowd); PTX-22 at 5.

84. The Census Bureau concedes that whole-person imputations "are not very accurate." Census Bureau 30(b)(6) Dep. Vol. I 253:7-15.

85. The Census Bureau anticipates that there will be 1.477 million more whole-person imputations in the 2020 Census because of the citizenship question. PTX-160 at 42-43.

86. The Census Bureau has not finalized the algorithms it will use for count imputation in the 2020 Census. Tr. 892:16-19 (Abowd). The accuracy of the Census Bureau's imputation model "is unknown at this time." PTX-160 at 44. The Census Bureau has recognize