UNITED STATES DEPARTMENT OF COMMERCE
The Secretary of Commerce
Washington, D.C. 20230

To:

Karen Dunn Kelley, Under Secretary for Economic Affairs

From: Secretary Wilbur Ross

U

~~

Date:

March 26,2018

Re:

Reinstatement of a Citizenship Question on the 2020 Decennial Census Questionnaire

Dear Under Secretary Kelley:
As you know, on December 12,2017, the Department of Justice ("DOJ") requested that the
Census Bureau reinstate a citizenship question on the decennial census to provide census block
level citizenship voting age population ("CV AP") data that are not currently available from
government survey data ("DOJ request"). DOJ and the courts use CV AP data for determining
violations of Section 2 of the Voting Rights Act ("VRA"), and having these data at the census
block level will permit more effective enforcement of the Act. Section 2 protects minority
population voting rights.
Following receipt of the DOJ request, I set out to take a hard look at the request and ensure that
I considered all facts and data relevant to the question so that I could make an informed decision
on how to respond. To that end, the Department of Commerce ("Department") immediately
initiated a comprehensive review process led by the Census Bureau.
The Department and Census Bureau's review of the DOJ request - as with all significant Census
assessments - prioritized the goal of obtaining complete and accurate data. The decennial
census is mandated in the Constitution and its data are relied on for a myriad of important
government decisions, including apportionment of Congressional seats among states,
enforcement of voting rights laws, and allocation of federal funds. These are foundational
elements of our democracy, and it is therefore incumbent upon the Department and the Census
Bureau to make every effort to provide a complete and accurate decennial census.
At my direction, the Census Bureau and the Department's Office of the Secretary began a
thorough assessment that included legal, program, and policy considerations. As part of the
process, I also met with Census Bureau leadership on multiple occasions to discuss their process
for reviewing the DOJ request, their data analysis, my questions about accuracy and response
rates, and their recommendations. At present, the Census Bureau leadership are all career civil
servants. In addition, my staff and I reviewed over 50 incoming letters from stakeholders,
interest groups, Members of Congress, and state and local officials regarding reinstatement of a
citizenship question on the 2020 decennial census, and I personally had specific conversations on

1

 the citizenship question with over 24 diverse, well informed and interested parties representing a
broad range of views. My staff and I have also monitored press coverage of this issue.
Congress has delegated to me the authority to determine which questions should be asked on the
. decennial census, and I may exercise my discretion to reinstate the citizenship question on the
2020 decennial census, especially based on DOl's request for improved CV AP data to enforce
the VRA. By law, the list of decennial census questions is to be submitted two years prior to the
decennial census - in this case, no later than March 31, 2018.
Th~ Department's review demonstrated that collection of citizenship data by the Census has been
a long-standing historical practice. Prior decennial census surveys of the entire United States
population consistently asked citizenship questions up until 1950, and Census Bureau surveys of
sample populations continue to ask citizenship questions to this day. In 2000, the decennial
'
census "long form" survey, which was distributed to one in six people in the U.S., included a
question on citizenship. Following the 2000 decennial census, the "long form" sample was
replaced by the American Community Survey ("ACS"), which has included a citizenship
question since 2005. Therefore, the citizenship question has been well tested.
DOJ seeks to obtain CV AP data for census blocks, block groups, counties, towns, and other
locations where potential Section 2 violations are alleged or suspected, and DOJ states that the
current data collected under the ACS are insufficient in scope, detail, and certainty to meet its
purpose under the VRA. The Census Bureau has advised me that the census-block-level
citizenship data requested by DOJ are not available using the annual ACS, which as noted earlier
does ask a citizenship question and is the present method used to provide DOJ and the courts
with data used to enforce Section 2 of the VRA. The ACS is sent on an annual basis to a sample
of approximately 2.6 percent of the population.
To provide the data requested by DOJ, the Census Bureau initially analyzed three alternatives:
Option A was to continue the status quo and use ACS responses; Option B was placing the ACS
citizenship question on the decennial census, which goes to every American household; and
Option C was not placing a question on the decennial census and instead providing DOJ with a
citizenship analysis for the entire populati~n using federal administrative record data that Census
has agreements with other agencies to access for statistical purposes.
Option A contemplates rejection of the DOJ request and represents the statu;s quo baseline.
Under Option A, the 2020 decennial census would not include the question on citizenship that
DOJ requested and therefore would'not provide DOJ with improved CVAP data. Additionally,.
the block-group level CV AP data currently obtained through the ACS has associated margins of
error because the ACS is extrapolated based on sample surveys of the population. Providing
more precise block-level data would require sophisticated statistical modeling, and if Option A'is
selected, the Census Bureau advised that it would need to deploy a team of experts to develop
model-based methods that attempt to better facilitate DOl's request for more specific data. But
the Census Bureau did not assert and could not confirm that such data modeling is possible for
census-block-level data with a sufficient degree of accuracy. Regardless, DOl's request is based
at least in part on the fact that existing ACS citizenship data-sets lack specificity and

2

 completeness. Any future modeling from these incomplete data would only compound that
problem.
Option A would provide no improved citizenship count, as the existing ACS sampling would
still fail to obtain actual, complete number counts, especially for certain lower population areas
or voting districts, and there is no guarantee that data could be improved using small-area
modeling methods. Therefore, I have concluded that Option A is not a suitable option.
The Census Bureau and many stakeholders expressed concern that Option B, which would add a
citiz~nship question to the decennial census, would negatively impact the response rate for noncitizens. A significantly lower response rate by non-citizens could reduce the accuracy of the
decennial census and increase costs for non-response follow up ("NRFU") operations. However,
neither the Census Bureau nor the concerned stakeholders could document that the response rate
would in fact decline materially. In discussing the question with the national survey agency
Nielsen, it stated that it had added questions from the ACS on sensitive topics such as place of
birth and immigration status to certain short survey forms without any appreciable decrease in
response rates. Further, the former director of the Census Bureau during the last decennial
census told me that, while he wished there were data to answer the question, none existed to his
knowledge. Nielsen's Senior Vice President for Data Science and the former Deputy Director
and Chief Operating Officer of the Census Bureau under President George W. Bush also
confirmed that, to the best of their knowledge, no empirical data existed on the impact of a
citizenship question on responses.
When analyzing Option B, the Census Bureau attempted to assess the impact that reinstatement
of a citizenship question on the decennial census would have on response rates by drawing
comparisons to ACS responses. However, such comparative analysis was challenging, as
response rates generally vary between decennial censuses and other census sample surveys. For
example, ACS self-response rates were 3.1 percentage points less than self-response rates forthe
2010 decennial census. The Bureau attributed this difference to the greater outreach and followup associated with the Constitutionally-mandated decennial census. Further, the decennial
census has differed significantly in nature from the sample surveys. For example, the 2000
decennial census survey contained only eight questions. Conversely, the 2000 "long form"
sample survey contained over 50 questions, and the Census Bureau estimated it took an average
of over 30 minutes to complete. ACS surveys include over 45 questions on numerous topics,
including the number of hours worked, income information, and housing characteristics.
The Census Bureau determined that, for 2013-2016 ACS surveys, nonresponses to the
citizenship question for non-Hispanic whites ranged from 6.0 to 6.3 percent, for non-Hispanic
blacks ranged from 12.0 to 12.6 percent, and for Hispanics ranged from 11.6 to 12.3 percent.
However, these rates were comparable to nonresponse rates for other questions on the 2013 and
2016 ACS. Census Bureau estimates showed similar nonresponse rate ranges occurred for
questions on the ACS asking the number times the respondent was married, 4.7 to 6.9 percent;
educational attainment, 5.6 to 8.5 percent; monthly gas costs, 9.6 to 9.9 percent; weeks worked
in the past 12 months, 6.9 to 10.6 percent; wages/salary income, 8.1 to 13.4 percent; and yearly
property insurance, 23.9 to 25.6 percent.

3

 The Census Bureau also compared the self-response rate differences between citizen and noncitizen households' response rates for the 2000 decennial census short form (which did not
include a citizenship question) and the 2000 decennial census long form survey (the long form
survey, distributed to only one in six households, included a citizenship question in 2000).
Census found the decline in self-response rates for non-citizens to be 3.3 percent greater than for
citizen households. However, Census was not able to isolate what percentage of decline was
caused by the inclusion of a citizenship question rather than some other aspect of the long form
survey (it contained over six times as many questions covering a range of topics). Indeed, the
Census Bureau analysis showed that for the 2000 decennial census there was a significant drop
in self response rates overall between the short and long form; the mail response rate was 66.4
percent for the short form and only 53.9 peicent for the long form survey. So while there is
widespread belief among many parties that adding a citizenship question could reduce response
rates, the Census Bureau's analysis did not provide definitive, empirical support for that belief.
Option C, the use of administrative records rather than placing a citizenship question on the
decennial census, was a potentially appealing solution to the DOJ request. The use of
administrative records is increasingly part of the fabric and design of modem censuses, and the
Census Bureau has been using administrative record data to improve the accuracy and reduce the
cost of censuses since the early 20th century. A Census Bureau analysis matching administrative
records with the 20 1a decennial census and ACS responses over several more recent years
showed that using administrative records could be more accurate than self-responses in the case
of non-citizens. That Census Bureau analysis showed that between 28 and 34 percent of the
citizenship self-responses for persons that administrative records show are non-citizens were
inaccurate. In other words, when non-citizens respond to long form or ACS questions on
citizenship, they inaccurately mark "citizen" about 30 percent of the time. However, the Census
Bureau is still evolving its'use of administrative records, and the Bureau does not yet have a
complete administrative records data set for the entire population. Thus, using administrative
records alone to provide DOJ with CV AP data would provide an incomplete picture. In the 20 1a
decennial census, the Census Bureau was able to match 88.6 percent of the population with what
the Bureau considers credible administrative record data. While impressive, this means that
more than 10 percent of the American population - some 25 million voting age people - would
need to have their citizenship imputed by the Census Bureau. Given the scale of this number, it
was imperative that another option be developed to provide a greater level of accuracy than
either self-response alone or use of administrative records alone would presently provide.
I therefore asked the Census Bureau to develop a fourth alternative, Option D, which would'
combine Options Band C. Under Option D, the ACS citizenship question would be asked on the
decennial census, and the Census Bureau would use the two years remaining until the 2020
decennial census to further enhance its administrative record data sets, protocols, and statistical
models to provide more comple~e and accurate data. This approach would maximize the Census
Bureau's ability to match the decennial census responses with administrative records.
Accordingly, at my direction the Census Bureau is working to obtain as many additional Federal
and state administrative records as possible to provide more comprehensive information for the
population.
"

4

 It is my judgment that Option D will provide DOJ with the most complete and accurate CV AP
data in response to its request. A"sking the citizenship question of 100 percent of the population
gives each respondent the opportunity to provide an answer. This may eliminate the need for the
Census Bureau to have to impute an answer for millions of people. For the approximately 90
percent of the population who are citizens, this question is no additional imposition. And for the
approximately 70 percent of noli-citizens who already answer this question accurately on the
ACS, the question is no additional imposition since census responses by law may only be used
anonymously and for statistical purposes. Finally, placing the question on the decennial census
and directing the Census Bureau to determine the best means to compare the decennial census
responses with administrative records will permit the Census Bureau to determine the inaccurate
response rate for citizens and non-citizens alike using the entire population. This will enable the
Census Bureau to establish, to the best of its ability, the accurate ratio of citizen to non-citizen
responses to impute for that small percentage of cases where it is necessary to do so.
Consideration of Impacts I have carefully considered the argument that the reinstatement of
the citizenship question on the decennial census would depress response rate. Because a lower
response rate would lead to increased non-response follow-up costs and less accurate responses,
this factor was an important consideration in the decision-making process. I find that the need
for accurate citizenship data and the limited burden that the reinstatement of the citizenship
question would impose outweigh fears about potentially lower response rate.

a

Importantly, the Department's review found that limited empirical evidence exists about whether
adding a citizenship question would decrease response rates materially. Concerns about
decreased response rates generally fell into the following two categories - distrust of government
and increased burden. First, stakeholders, particularly those who represented immigrant
constituencies, noted that members of their respective communities generally distrusted the
government and especially distrusted efforts by government agencies to obtain information about
them. Stakeholders from California referenced the difficulty that government agencies faced
obtaining any information from immigrants as part of the relief efforts after the California
wildfires. These government agencies were not seeking to ascertain the citizenship status of
these wildfire victims. Other stakeholders referenced the political climate generally and fears
that Census responses could be used for law enforcement purposes. But no one provided
evidence that reinstating a citizenship question on the decennial census would materially
decrease response rates among those who generally distrusted government and government
information collection efforts, disliked the current administration, or feared law
enforcement. Rather, stakeholders merely identified residents who made the decision not to
participate regardless of whether the Census includes a citizenship question. The reinstatement
of a citizenship question will not decrease the response rate of residents who already decided not
to respond. And no one provided evidence that there are residents who would respond accurately
to a decennial census that did not contain a citizenship question but would not respond if it did
(although many believed that such residents had to exist). While it is possible this belief is true,
there is no information available to determine the number of people who would in fact not
respond due to a citizenship question being added, and no one has identified any mechanism for
making such a determination.

5

 A second concern that stakeholders advanced is that recipients are generally less likely to
respond to a survey that contained more questions than one that contained fewer. The former
Deputy Director and Chief Operating Officer of the Census Bureau during the George W. Bush
administration described the decennial census as particularly fragile and stated that any effort to
. add questions risked lowering the response rate, especially a question about citizenship in the
current political environment. However, there is limited empirical evidence to support this view.
A former Census Bureau Director during the Obama Administration who oversaw the last
decennial census noted as much. He stated that, even though he believed that the reinstatement
of a citizenship question would decrease response rate, there is limited evidence to support this
conclusion. This same former director noted that, in the years preceding the decennial census,
certain interest groups consistently attack the census and discourage participation. While the
reinstatement of a citizenship question may be a data point on which these interest groups seize
in 2019, past experience demonstrates that it is likely efforts to undermine the decennial census
will occur again regardless of whether the decennial census includes a citizenship
question. There is no evidence that residents who are persuaded by these disruptive efforts are
more or less likely to make their respective decisions about participation b~sed specifically on
the reinstatement of a citizenship question. And there are actions that the Census Bureau and
stakeholder groups are taking to mitigate the impact of these attacks on the decennial census.
Additional empirical evidence about the impact of sensitive questions on survey response rates
came from the SVP of Data Science at Nielsen. When Nielsen added questions on place of birth
and time of arrival in the United States (both of which were taken from the ACS) to a short
survey, the response rate was not materially different than it had been before these two questions
were added. Similarly, the former Deputy Director and COO of the Census during the George
W. Bush Administration shared an example of a citizenship-like question that he believed would
negatively impact response rates but did not. He cited to the Department of Homeland Security's
2004 request to the Census Bureau to provide aggregate data on the number of Arab Americans
by zip code in certain areas of the country. The Census Bureau complied, and Census
employees, including the then-Deputy Director, believed that the resulting political fire storm
would depress response rates for further Census Bureau surveys in the impacted communities.
But the response rate did not change materially.
Two other themes emerged from stakeholder calls that merit discussion. First, several
stakeholders who opposed reinstatement of the citizenship question did not appreciate that the
question had been asked in some form or another for nearly 200 years. Second, other
stakeholders who opposed reinstatement did so based on the assumption that the data on
citizenship that the Census Bureau collects through the ACS are accurate, thereby obviating the
need to ask the question on the decennial census. But as discussed above, the Census Bureau
estimates that between 28 and 34 percent of citizenship self-responses on the ACS for persons
that administrative records show are non-citizens were inaccurate. Because these stakeholder
concerns were based on incorrect premises, they are not sufficient to change my decision.

6

 Finally, I have considered whether reinstating the citizenship question on the 2020 Census will
lead to any significant monetary costs, programmatic or otherwise. The Census Bureau staff
have advised that the costs of preparing and adding the question would be minimal due in large
part to the fact that the citizenship question is already included on the ACS, and thus the
citizenship question has already undergone the cognitive research and questionnaire testing
required for new questions. Additionally, changes to the Internet Self-Response instrument,
revising the Census Questionnaire Assistance, and redesigning of the printed questionnaire can
be easily implemented for questions that are finalized prior to the submission of the list of
questions to Congress.
.
The Census Bureau also considered whether non-response follow-up increases resulting from
inclusion of the citizenship question would lead to increased costs. As noted above, this estimate
was difficult to assess given the Census Bureau and Department's inability to determine what
impact there will be on decennial census survey responses. The Bureau provided a rough
estimate that postulated that up to 630,000 additional households may require NRFU operations
if a citizenship question is added to the 2020 decennial census. However, even assuming that
estimate is correct, this additional Y2 percent increase in NRFU operations falls well within the
margin of error that the Department, with the support of the Census Bureau, provided to
Congress in the revised Lifecycle Cost Estimate ("LCE") this past fall. That LCE assumed that
NRFU operations might increase by 3 percent due to numerous factors, including a greater
increase in citizen mistrust of government, difficulties in accessing the Internet to respond, and
other factors.
Inclusion of a citizenship question on this country's decennial census is not new - the decision to
collect citizenship information from Americans through the decennial census was first made
centuries ago. The decision to include a citizenship question on a national census is also not
uncommon. The United Nations recommends that its member countries ask census questions
identifying both an individual's country of birth and the country of citizenship. Principals. and
Recommendations/or Population and Housing Censuses (Revision 3), UNITED NATIONS 121
(2017). Additionally, for countries in which the population may include a large portion of
naturalized citizens, the United Nations notes that, "it may be important to collect information on
the method of acquisition of citizenship." Id. at 123. And it is important to note that other major
democracies inquire about citizenship on their census, including Australia, Canada, France,
Germany, Indonesia, Ireland, Mexico, Spain, and the United Kingdom, to name a few.
The Department of Commerce is not able to determine definitively how inclusion of a citizenship
question on the decennial census will impact responsiveness. However, even iftliere is some
impact on responses, the value of more complete and accurate data derived from surveying the
entire population outweighs such concerns. Completing and returning decennial census
questionnaires is required by Federal law, those responses are protected by law, and inclusion of
a citizenship question on the 2020 decennial census will provide more complete information for
those who respond. The citizenship data provided to DOJ will be more accurate with the
question than without it, which is of greater importance than any adverse effect that may result
from people violating their legal duty to respond.

7

 To conclude, after a thorough review of the legal, program, and policy considerations, as well as
numerous discussions with the Census Bureau leadership and interested stakeholders, I have
determined that reinstatement of a citizenship question on the 2020 decennial census is necessary
to provide complete and accurate data in response to the DOl request. To minimize any impact
on decennial census response rates, I am directing the Census Bureau to place the citizenship
question last on the decennial census form.
Please make my decision known to Census Bureau personnel and Members of Congress prior to
March 31, 2018. I look forward to continuing to work with the Census Bureau as we strive for a
complete and accurate 2020 decennial census.

CC:

Ron larmin, performing the nonexclusive functions and duties of the Director of the
Census Bureau
Enrique Lamas, performing the nonexclusive functions and duties of the Deputy Director
of the Census Bureau

8