Memo Final AR Used

August 11, 2021

Administrative Data Used in the 2020
Census

Karen D. Deaver
Decennial Census Programs Directorate

Intentionally blank

Table of Contents
INTRODUCTION ....................................................................................................................................................... 1
BACKGROUND .......................................................................................................................................................... 1
DATA SOURCES ........................................................................................................................................................ 3
TITLE 26 COMPOSITE ......................................................................................................................................... 3
TITLE 13 COMPOSITE ......................................................................................................................................... 4
SOURCES NOT IN A COMPOSITE .................................................................................................................... 5
ADDITIONAL SOURCES ONLY USED FOR CITIZENSHIP ....................................................................... 5
FRAME DEVELOPMENT ....................................................................................................................................... 5
RESPONDENT MOTIVATION .............................................................................................................................. 6
A. INITIAL CONTACT STRATEGY......................................................................................................................... 6
B. ADVERTISING CAMPAIGN ................................................................................................................................ 6
SELF-RESPONSE PHASE ....................................................................................................................................... 7
A. NON-ID ADDRESS ENHANCED MATCHING ............................................................................................... 7
B. PAPER DATA CAPTURE QUALITY ASSURANCE ....................................................................................... 8
C. SPECIAL ENUMERATIONS ................................................................................................................................ 9
NONRESPONSE FOLLOWUP ............................................................................................................................... 9
A. VACANT AND DELETE IDENTIFICATION ................................................................................................ 10
B. ADMINISTRATIVE RECORDS ENUMERATION ...................................................................................... 10
C. BEST TIME TO CONTACT MODELING ....................................................................................................... 12
D. PHONE NUMBER AVAILABILITY ................................................................................................................ 12
RESPONSE DATA VERIFICATION .................................................................................................................. 12
A. SELF-RESPONSE QUALITY ASSURANCE .................................................................................................. 12
B. ENUMERATOR QUALITY CONTROL .......................................................................................................... 13
POST-RESPONSE PROCESSING....................................................................................................................... 13
A. ADDITIONAL UNDUPLICATION OF PEOPLE IN HOUSING UNITS ................................................. 13
B. COUNT IMPUTATION ....................................................................................................................................... 13
C. CHARACTERISTIC IMPUTATION ................................................................................................................. 14
PUBLISHING DATA .............................................................................................................................................. 15
A. CITIZEN VOTING AGE POPULATION (CVAP) TABLES ....................................................................... 15
B. COUNT QUESTION RESOLUTION ................................................................................................................ 15

COVERAGE EVALUATION ................................................................................................................................. 16
A. POST-ENUMERATION SURVEY .................................................................................................................... 16
B. DEMOGRAPHIC ANALYSIS ............................................................................................................................. 16
REFERENCES .......................................................................................................................................................... 16
APPENDIX A –DATA SOURCES USED IN THE 2020 CENSUS .......................................................................... 1
APPENDIX B - OVERWRITING EXAMPLE................................................................................................................ 1

ADMINISTRATIVE DATA USED IN THE 2020 CENSUS
INTRODUCTION
Based on the results of the 2018 End-to-End Census Test, prior census tests, and other research, the
Decennial Census Programs Directorate determined its approach to using administrative data in many
operations of the 2020 Census. Events that occurred during the fielding of the 2020 Census, including
the COVID-19 pandemic and natural disasters, required some adjustments to these plans.
In this context, we are using “administrative data” broadly to include:
•
•

Microdata records contained in files collected and maintained by federal, state, and local
government agencies (traditionally referred to as administrative records).
Microdata records contained in files collected and maintained by commercial entities (often
referred to as third-party data).

As well as:
•
•

Macro- and microdata from Census Bureau data collections for statistical purposes and address
enhancement operations (internal data).
Macrodata from publicly available sources (public data).

The intended uses of administrative data and the expected sources of data evolved throughout the
research and planning for the 2020 Census and their actual use was expanded during 2020 Census
fielding. This report documents the final use of administrate data during the 2020 Census. It updates
and can replace the similar “intended use” memo made public in 2020 (Deaver, May 2020).
BACKGROUND
To meet the strategic goals and objectives of the 2020 Census, the Census Bureau made fundamental
changes to the design, implementation, and management of the decennial census. These changes built
upon the successes and addressed the challenges of previous censuses, while also balancing objectives
of cost containment, quality, flexibility, innovation, and disciplined and transparent acquisition decisions
and processes. The 2020 Census Operational Plan included using administrative records, third-party
data, internal data, and public sources (collectively referred to in this document as “administrative
data”) to avoid cost, maintain quality, and improve efficiency of operations (U.S. Census Bureau,
December 2018).
Several core sources were used to support this effort, including data collected from Census Bureau
operations, which is protected under 13 U.S.C. § 9, and data from the Internal Revenue Service (IRS),
which is referred to as Federal Tax Information (FTI), and is protected by 26 U.S.C. § 6103(b)(8). In
accordance with Title 26 and a 2013 Memorandum of Agreement between the Department of Treasury,
Internal Revenue Service, and the Department of Commerce, U.S. Census Bureau, the scope of work for
which FTI may be used includes frame building, enumeration, imputation, and evaluation. Other
agreements that the Census Bureau maintains with other administrative data providers similarly
delineate allowable uses of their data.
Below are the instances for which the Census Bureau used administrative data in the 2020 Census,
excluding evaluations and experiments in support of early planning and research to inform the transition
and design of the 2030 Census. Several instances used extracts from either a “Title 26 composite” or a
August 11, 2021

1

“Title 13 composite” rather than directly accessing original administrative datasets. The process to
create these composites is described in the next section. More detailed information on each instance is
provided in sections IV- XI. Appendix A summarizes the sources for each instance. For more information
about the 2020 Census operations, see the detailed operational plans (U.S. Census Bureau, various).
Final assessments of each operation will be available on www.census.gov in the coming years.
•

Frame Development: Develop and update the address frame, including group quarters, and
spatial data that serve as the universe for 2020 Census enumeration activities.
• Respondent Motivation:
o Support the initial contact strategy (appropriate delivery of census invitations and
questionnaires).
o Develop and execute the microtargeted advertising campaign.
• Self-Response:
o Augment respondent-provided address data to enhance matching of non-ID responses to
the updated enumeration address list: the addresses for responses that are returned
without the preassigned census identification number and do not easily match to the
Master Address File (MAF) will be compared with administrative record information in order
to obtain missing address information, or correct errors, such as misspelling. By improving
the address, another attempt can be made to associate the response data to a household in
the enumeration universe so that no further effort is required to obtain a response, such as
follow-up mailings or fieldwork.
o Quality assurance in the Paper Data Capture operation: Optical Character Recognition (OCR)
capture of written-in names is compared to existing administrative data to ensure quality of
the OCR process.
o Enumerate or supplement field enumeration of nontraditional or unique living
arrangements, such as group quarters, military installations, and federally affiliated people
overseas.
• Nonresponse Followup (NRFU):
o Reduce contacts for cases in the NRFU workload through identification of vacant housing
units and deletes (units that do not meet the Census Bureau’s definition of a housing unit).
o Enumerate nonresponding, occupied housing units with quality, reliable information.
o Model the “best time to contact” occupied, nonresponding housing units (modeled as
intended, but feature disabled during fielding).
o Provide additional phone contact information to cases in the NRFU workload.
• Response Data Verification:
o Self-Response Quality Assurance: Corroborate respondent-provided information to detect
potentially suspicious responses.
o Enumerator Quality Control: Validate enumerator-provided information to ensure reporting
accuracy and more efficiently target reinterview efforts.
• Post-Response Processing:
o Additional Unduplication of People in Housing Units: Determine in which household to keep
people with resolved duplicate links.

August 11, 2021

2

o

•

•

Count Imputation
▪ Housing Unit: Determine final occupied/vacant/nonexistent status for unresolved
cases and impute the household count for those addresses determined to be
occupied but without a specified number of occupants.
▪ Group Quarters: Impute final count for unresolved, occupied facilities.
o Characteristic Imputation: Impute household and person characteristics where they are
missing from the response.
Publishing Data:
o Create the Citizen Voting Age Population special tabulation (intended but subsequently
cancelled).
o Resolve Count Question Resolution challenges.
Coverage Evaluation:
o Improve matching and characteristic imputation in the Post-Enumeration Survey (PES).
o Create independent estimates through Demographic Analysis.
DATA SOURCES
TITLE 26 COMPOSITE

Initially, administrative data were used to create a robust repository of data and information that was
used for activities that had been approved by data providers (e.g., the Internal Revenue Service and the
Social Security Administration). Data from several federal, state, internal, and commercial primary
sources were combined, standardized, and corroborated to create the Title 26 composite, which
contained variables such as address, householder name, household roster and relationships, and
demographic data about the household inhabitants. FTI such as taxpayer ID and return type were
included in this repository. These data were combined with extracts from the Master Address File
(MAF)/Topologically Integrated Geographic Encoding and Referencing (TIGER) database (MTdb). The
MTdb reflects the latest geographic information for each address. Extracts from the Title 26 composite
were provided for additional processing as described in subsequent sections as appropriate.
The following sources were used to create the Title 26 composite and include historic and current
vintages as available to the Census Bureau:
•

Administrative Records
o Centers for Medicare and Medicaid Services (CMS) Medicare Enrollment Database
(MEDB)
o Housing and Urban Development (HUD) Public and Indian Housing Information Center
(PIC) and Tenant Rental Assistance Certification System (TRACS), now known as the
combined “Longitudinal” File
o HUD Federal Housing Administration (FHA) Integrated Database (IDB), which includes
data from Computerized Homes Underwriting Management System (CHUMS)
o Indian Health Service (IHS) Patient Registration
o Internal Revenue Service (IRS)
▪ 1040 Individual Tax Returns
▪ IRS 1099 Information
o Selective Service System (SSS) Registration

August 11, 2021

3

o

•

•

Social Security Administration
▪ Census Numident (a processed version of the SSA Numeric Identification File,
Numident, which does not contain a Social Security number) as well as the
Census Numident Alternate Name File
o State or local program datasets, examples include:
▪ Homeless Management Information System (HMIS)
▪ Alaska Permanent Fund Dividend (PFD)
▪ Supplemental Nutrition Assistance Program (SNAP)
▪ Temporary Assistance for Needy Families (TANF)
▪ Women, Infants, and Children (WIC)
o U.S. Postal Service (USPS) National Change of Address (NCOA) File
Third-Party Datasets1
o Black Knight
o DAR Partners (Data Advisory Research)
o Targus (Wireless and Federal Consumer)
o Veteran Service Group of Illinois (VSGI)
Census Bureau Data
o American Community Survey (ACS) data
o Census 2000 and 2010 Census edited and unedited datasets
o Census Household Composition Key File (produced using Social Security Administration
data and previous census information linking children 18 years and younger with their
parents)
o Contact Frame file (list of phone numbers) compiled from some sources listed here
(SNAP, WIC, Alaska PFD, ACS Data, DAR Partners, Targus Wireless and Federal
Consumer, and VSGI) as well as the ones below before it was delivered to 2020 Census
systems:
▪ Experian InSource and Experian End-Dated Records
▪ InfoUSA
▪ Melissa Data
▪ National Sample Survey of Registered Nurses
o “Best Race and Ethnicity” information compiled from some sources listed here (MEDB,
HUD Longitudinal, IHS, Census Numident, ACS data, prior census data, and Targus
Federal Consumer) as well as:
▪ CMS Medicaid and Statistical Information System (MSIS)
▪ National-Level Adult and Child TANF Recipient Files
▪ Experian InSource and Experian End-Dated Records
▪ InfoUSA
TITLE 13 COMPOSITE

Similarly, a robust repository of Title 13 data and information was created from the Title 26 composite
and extracts of this composite were provided to various 2020 Census operations that did not need to
use FTI. While FTI was used in the creation of the Title 26 composite, in the creation of the Title 13
composite, the FTI was overwritten such that no FTI appeared in the final Title 13 composite and FTI was

1

Data from other commercial sources (e.g. CoreLogic) were used in census tests but are not considered part of the
production datasets.

August 11, 2021

4

not the sole source of any information. This composite was therefore considered to be Title 13, not Title
26. An example of how the overwriting worked is in Appendix B.
SOURCES NOT IN A COMPOSITE
Some processes used sources, either solely or in part, that were not part of the composites discussed
above. Those sources were:
•
•

•

•

•

•
•
•
•

ACS contact history information
Department of Defense
o List of deployed personnel
o Count of federally affiliated people stationed/assigned overseas, including dependents
Communications-related
o Modeled self-response predictions (Predicted Self-Response Score and Internet
Proportion of Self-Response)
o Audience segmentation information
o Federal Communications Commission (FCC Residential Fixed Internet Access Service
Connections per 1,000 Households by Census Tract [publicly available])
Frame-related
o USPS Delivery Sequence File (DSF)
o Additional Geographic Support Program and 2020 Census program data – address and
spatial data provided by tribal, federal, state and local governments
o Lists of group quarters and transitory locations from federal, state, local, and third-party
sources as well as ongoing operations
Planning Database (PDB) (publicly available datasets at the tract and block group levels that
assemble a range of housing, demographic, socioeconomic, and census operational data;
variables have been extracted from the 2010 Census and ACS databases)
Group quarters administrator lists
Lists of residency information for off-campus college students
Integrated Postsecondary Education Data System (IPEDS)
USPS Undeliverable As Addressed (UAA) Information
ADDITIONAL SOURCES ONLY USED FOR CITIZENSHIP

In addition to some sources mentioned in the sections above, there were several sources that were
expected to be used only to research how to produce, and subsequently produce, citizenship
information in conjunction with the census. This effort was canceled to be in compliance with Executive
Order 13986 issued on January 20, 2021 (Executive Office of the President, January 20, 2021).
FRAME DEVELOPMENT
Throughout the decade, the Census Bureau maintained address and spatial (e.g., roads, boundaries, and
geographic areas) data in the Master Address File (MAF)/Topologically Integrated Geographic Encoding
and Referencing (TIGER) System. The MAF/TIGER System is regularly updated with data from the United
States Postal Service; ongoing geographic partnership efforts with tribal, state, and local governments;
and fieldwork. These efforts were used to update the address frame and reflect changes to the housing
stock that occurred over time.
August 11, 2021

5

For the 2020 Census, additional efforts were employed to finalize the frame. Focused geographic
partnership efforts helped improve the address list. State and local governments provided address
updates during the Local Update of Census Addresses (LUCA) program and New Construction (NC)
efforts. Additional data for group quarters and transitory locations was obtained from multiple sources,
including federal partners, state partners, local governments, and a third-party provider.
Frame development did not use either the T26 composite or the T13 composite. Data within the
MAF/TIGER System―extracts from the MTdb―were used as inputs to several enumeration operations.
RESPONDENT MOTIVATION
Respondent motivation activities did not use T26 or T13 composite data. Most activities used
aggregated data at higher geographic or demographic levels; two activities used person-level data. The
activities are outlined below.
A.

INITIAL CONTACT STRATEGY

Administrative data were used to determine how the Census Bureau would send the initial invitation to
respondents. Respondents in areas more likely to respond online received the “Internet First” mailing
strategy, where they received invitations to respond online. Those who did not respond online received
reminders to respond and a paper questionnaire before NRFU began. Respondents in areas least likely
to respond online (as determined by using ACS data and Federal Communications Commission internet
connectivity data), received the “Internet Choice” mailing strategy. The Choice strategy consisted of
receiving an invitation to respond online, but with a paper questionnaire in the first mailing.
Respondents then received reminders to respond either online or via the questionnaire they received
earlier. Those who did not respond received another paper questionnaire before NRFU began. The
“Internet First” or “Internet Choice” delineation was determined by geographic area (not by individual
household/address) and used area-based response likelihoods, not person-level data, to identify which
contact strategy the area received.
Addresses in certain geographic areas received bilingual English/Spanish invitations and questionnaires.
These areas were designated where 20% or more of households in a census tract (areas with about
4,000 households) need Spanish language assistance — defined as households that have at least one
person age 15 or older who speaks Spanish and doesn’t speak English “very well” based on ACS data.1
B.

ADVERTISING CAMPAIGN

To increase the effectiveness of advertising and contact strategies, the Census Bureau, through the
communications contract, used demographic and geographic information from various sources to help
target the advertising to specific populations.
•

1

The Census Bureau developed predictive models and used these models to estimate tract-level
self-response propensity for various mode(s). These predictions were then used to develop the
media plan and to aid in campaign optimization. Tract-level self-response rate predictions were
aggregated to larger geographic areas and helped determine the media and messaging
strategies for the campaign.

This use was inadvertently omitted from the previously released memo.

August 11, 2021

6

•

•

Predictions were combined with geographic, demographic, housing, and sentiment information
to create audience segments. The segmentation information was used to design and execute
targeted advertising and communication strategies for geographic segments and audience
groups.
As households completed the census questionnaire, we used campaign analytics to identify the
best messages and modes to reach various segments. Headquarters staff, field teams, and
partnership specialists coordinated to prioritize audiences and align messaging. We also used
internet address and browsing history to customize user experience with the web platform.

In addition to the above general use, two targeted mailings also made use of administrative data.
•

•

Native Hawaiian and Pacific Islanders (NHPI) in the continental United States received a direct
mailing through a partner organization. The partner organization used an address list supplied to
them by NHPI-related organizations.
Residential addresses in targeted USPS carrier routes determined to be at risk for undercounting
young children also received a direct mailer. Tract-level response and demographic data from
the most recent ACS, the 2010 Census, and the planning database, along with address
information from the MAF and route information from the USPS were used to determine the
targeted routes.

An additional effort developed to adapt the communication strategy to a COVID-19 environment, was
engaging the general public in an email marketing campaign that sought to educate recipients about the
2020 Census, motivate them to self-respond, and encourage them to share information with their family
members and friends. Emails for this effort were obtained both from the public subscribing to receive
emails on census.gov, 2020census.gov or through the GovDelivery Network, and from the Census
Bureau Contact Frame. Between July and August, targeted emails were deployed to low responding
regions.
SELF-RESPONSE PHASE
A.

NON-ID ADDRESS ENHANCED MATCHING

A goal in the 2020 Census was to make it easy for people to respond anytime and anywhere to increase
self-response rates. We did this by providing response options that did not require a unique census
identifier (ID).
A “non-ID response” or a “non-ID case” refers to address information and associated person data
provided by respondents without the preassigned, unique identification number. The process of
comparing these non-ID cases to existing address information in the MTdb is referred to as “Non-ID
Processing.” The final MTdb matching and geocoding result for each non-ID case was the outcome of the
process. The Census Bureau used administrative, third-party, and census data to augment respondentprovided address data (referred to as “address enhancement”) to support matching to the MTdb1.
Throughout the entire self-response period, Non-ID Processing matched respondent-provided addresses
from non-ID cases to the MTdb. Census Bureau research suggested that initial address matching efforts
1

The initial census enumeration universe had been updated through Address Canvassing and incorporated into the
MAF that was used during Non-ID processing.

August 11, 2021

7

would pair about 87 percent of all non-ID cases with an existing record in the MTdb. Research also
suggested that an additional 2 to 3 percent could be matched to an existing record in the MTdb as a
result of address enhancement, which focused on the non-ID cases that could not be initially matched to
the MTdb. Information derived from administrative records and third-party data sources was used to
augment respondent-provided address information, creating the most accurate addresses possible.
These “enhanced” addresses then were used in a second attempt to match non-ID addresses to the
MTdb. The remaining non-ID cases that failed the second matching attempt or were not “enhanced”
were validated and geocoded manually.
Only data from the Title 13 composite were used in this operation.
B.

PAPER DATA CAPTURE QUALITY ASSURANCE

Administrative data were used to increase the confidence in the information that was captured from
paper questionnaires. Manual keying or Optical Character Recognition (OCR) was used to capture the
information from returned paper questionnaires, then the information was compared with existing
information as a quality check (i.e., to keep error rates low). Administrative data were not used to
replace any manual or OCR captured data, just to provide a confidence level about it. Greater
confidence increased OCR rates and decreased keying costs.
More specifically, the 2020 Census Paper Data Capture system, Integrated Computer Assisted Data Entry
(iCADE), used OCR to electronically capture response data provided by respondents, thereby eliminating
most data needing to be captured manually. Each character within a response field was assigned a
confidence score by the OCR software for processing.
Results from OCR were also made available to the iCADE application, where low-confidence write-in
fields were manually captured to enter a verification process. This Quality Assurance (QA) process
performed an independent verification process manually of both OCR and keyed entries to ensure the
outgoing error rate was acceptable and met the decennial requirements. Verification was done by
comparing the two data sets looking for a match or not. A sample stratum was defined and used to
determine the sample.
If the error rate within any sample stratum was determined to be unacceptable, those fields were then
sent to a process called Remainder Verification. Remainder Verification manually reprocessed all fields
within the targeted stratum that were not sampled and verified. This process looked to correct any
other errors in the batch.
Having access to supportive demographic information (e.g., age, race, sex, date of birth) from
administrative data allowed iCADE to produce additional matches within a household. This enhanced
the quality related to the confidence levels for “First Names, Last Names, Middle Initial, Race, Age, and
Date of Birth” assisting the OCR and keying data validation processes.
Names and other external data elements integrated into this process were used as a direct replacement
for response data. The data sources, when linked to an MTdb address, were used to increase or
decrease the confidence levels of the data captured. External data elements were integrated to validate
existing OCR and manual keying results.
Only data from the Title 13 composite were used in this operation.
August 11, 2021

8

C.

SPECIAL ENUMERATIONS

The Census Bureau focused additional efforts on nontraditional living quarters, which are traditionally
harder to data capture because of the unique living arrangements they present.
These enumerations did not use T26 composite or T13 composite data.
A group quarters (GQ) is a place where people live or stay in a group living arrangement that is owned or
managed by an entity or organization providing housing and/or services for the residents. This is not a
typical household-type living arrangement. These services may include custodial or medical care as well
as other types of assistance, and residency is commonly restricted to those receiving these services.
People living in GQs are usually not related to each other. GQs include such places as college residence
halls, residential treatment centers, skilled nursing facilities, group homes, correctional facilities,
workers’ dormitories, military barracks, and domestic violence shelters. The Census Bureau worked with
GQ administrators to enumerate the people residing at the GQ. GQ administrators provided 2020
Census response data either electronically to a Census Bureau secured portal or provided a census
worker with a paper listing of response data. These data may have been used as the sole source of
obtaining information or as a supplemental tool to ensure data collection of an entire facility when
other enumeration methods were used. These data are not considered administrative records or thirdparty data, but internally collected data.
The Census Bureau worked with Department of Defense (DOD) to acquire lists of deployed U.S. military
and civilian personnel to ensure a complete enumeration of this population. DOD used its administrative
records to provide the Census Bureau a data file of military and civilian employees who were deployed
outside of the United States (while stationed/assigned in the United States). The Census Bureau used
this file to enumerate these employees at their stateside address that matched to an existing address in
the MTdb.
DOD also provided counts by home state of U.S. military and federal civilian employees stationed or
assigned overseas and their dependent living with them at their overseas duty station as part of the
Federally Affiliated Count Overseas (FACO) operation. In addition, the Census Bureau worked with other
federal departments and agencies to collect counts of their federally employed individuals stationed or
assigned overseas and their dependents living with them.
NONRESPONSE FOLLOWUP
After giving the population multiple opportunities to self-respond to the 2020 Census, addresses for
which the Census Bureau did not receive a self-response formed the initial universe of addresses for the
NRFU operation. The NRFU operation served two purposes: 1) to determine housing unit status for
nonresponding addresses, and 2) to enumerate housing units for which a 2020 Census response was not
achieved.
For the 2020 Census, we used administrative data to reduce the NRFU workload. Data was used to
reduce contacts for vacant and deleted units. The NRFU workload was reduced further by using
administrative data, where feasible, to enumerate occupied households that failed to respond after
several contact attempts. Additionally, administrative data was used to develop a “Best Time to
Contact” model that was intended to reduce NRFU contacts by increasing the likelihood of finding

August 11, 2021

9

respondents at home, but this feature was disabled during fielding. Given the unique circumstances in
2020, phone numbers were made available to enumerators to help them complete their work.
A.

VACANT AND DELETE IDENTIFICATION

Prior to any fieldwork, an initial set of vacant and nonexistent addresses were identified using
administrative data. Pending further information obtained for these addresses, these cases received
either one or multiple contact attempts.
Data from the Title 26 composite were used in this operation.
To determine housing unit status, the Census Bureau used data from internal sources, such as the ACS,
the MTdb (including USPS DSF status), and the 2010 Census, as well as external sources, such as the U.S.
Postal Service’s Undeliverable-As-Addressed (UAA) file and IRS records.
More specifically, a multinomial logistic regression model was used to assign predicted probabilities that
a given housing unit was vacant or nonexistent. FTI from both IRS 1040 Individual Tax Payer Returns and
IRS 1099 Information Returns was used as inputs to the model. The model also included as predictors
administrative record household roster information (i.e., count and composition), U.S. Postal Service
UAA codes, block group characteristics from the ACS 5-year estimates, and housing unit characteristics
from the MTdb. These predicted probabilities were compared with predefined thresholds designed to
maximize the identification of vacant and nonexistent units while minimizing misclassification error. The
household status probabilities and final status designation was determined in a Title 26 environment
and the status designation was sent to a Title 13 environment for use in production. More information
on the details of the modeling and its application during 2020 Census fielding can be found in
Memorandum 2021.10: Administrative Record Modeling in the 2020 Census (Mulry, Mule, Keller, &
Konicki, April 2021).
B.

ADMINISTRATIVE RECORDS ENUMERATION

Models were used to identify units with high quality administrative data indicating they were occupied.
These models used the same predictors as those used in the vacant and delete model. Occupied
households that failed to respond after several contact attempts were eligible for enumeration using
their own administrative data in lieu of multiple personal visits. These addresses were matched to
administrative data that had been combined to provide the “best” person and housing unit information
found throughout the sources.
There were four additional uses of administrative data that were not originally planned in the 2020
Census.
•

Original administrative records enumeration did not include American Indian reservations
because, as historically hard-to-count areas, we did not want to initially reduce contacts. Given
the large number of unresolved cases, we planned to identify Administrative Record (AR)
Occupied and AR Closeout Occupied addresses with AR data that could be used if NRFU
fieldwork was unable to resolve the status of the address. While not originally planned, this AR
processing did result in the AR Occupied and AR Closeout Occupied determinations being sent
to the field control system. Thus, these addresses were eligible to receive a reduction in contact
attempts during the remainder of the NRFU operation.

August 11, 2021

10

•

•
•

The COVID-19 pandemic caused many college students living in off-campus housing to relocate.
To mitigate the risk of undercoverage of off-campus college students, the Census Bureau
contacted universities and colleges and requested that they share residency information for
their off-campus students. This information was used, if necessary, to enumerate students at
the off-campus address.
Four parishes in Louisiana affected by hurricanes that limited NRFU visits were enumerated with
administrative data in lieu of imputation where possible.
The Census Bureau received permission from the IRS to use a single source of administrative
data to determine household size where necessary.

A summary of administrative records enumeration for count and characteristics is below. More
information on the details of the modeling and its application during 2020 Census fielding and the
enumeration of off-campus addresses can be found in Memorandum 2021.10: Administrative Record
Modeling in the 2020 Census (Mulry, Mule, Keller, & Konicki, April 2021). Response records sent to the
National Archives and Records Administration (NARA) will not include enumerations with administrative
records.
1) Count
Data from the Title 26 composite were used in this operation.
FTI from both IRS 1040 Individual Tax Payer Returns and IRS 1099 Information Returns, as well as nonFTI from sources (including many of those listed in Section IV), was used to determine household count.
Where possible, other data sources were used to corroborate information from FTI and unique FTI was
overwritten as described in Appendix B so that the final enumeration data were Title 13. The IRS
approved the use of its data to determine the number of household members at a NRFU address when
the household could not be corroborated with other data sources. This IRS approval applied only to the
household size. IRS did not approve the Title 13 use of the names or characteristics of the household
members when its data were the sole source. The use of only the household size information for sole
source corroboration also was applied to MEDB and IHS data.
2) Characteristics
The non-FTI sources were also used for characteristic enumeration. When possible, we used the 2010
Census and ACS responses before using information from other sources. No FTI was used in
characteristic enumeration. Sole-sourced household size cases were not assigned characteristics during
administrative records enumeration.
Six characteristics were attempted to be assigned for the 2020 Census, five at the person level and one
at the housing unit (HU) level. Only federal sources were used to assign person-level characteristics.
Characteristics Expected to be Enumerated from Administrative Data for the 2020 Census
Person-Level Characteristics
Housing Unit Characteristic
Name
Tenure (own or rent the housing unit)
Age/Date of Birth
Sex
Race and Ethnicity
Relationship to Householder
August 11, 2021

11

C.

BEST TIME TO CONTACT MODELING

For the 2020 Census, the goal of enumerator route optimization was to increase the productivity of the
enumerators in the NRFU operation by decreasing the miles traveled and total hours spent per case.
Assignments were based on factors including enumerator availability, workload location, and the
probability of successfully contacting cases at various hours of the day. The enumerators received an
optimal routing of attempts to minimize travel.
Data from the Title 26 composite were used in this operation.
Logistic regression models were used to develop predicted probabilities associated with contacting
occupied housing units during each hour of the workday (weekdays and weekends). The Census Bureau
used data from internal sources, such as the ACS and the MTdb, as well as external sources, such as IRS
records, to develop input variables for the model. The predicted contact probabilities were determined
in a Title 26 environment and then sent to a Title 13 environment for use in production to optimize case
assignments.
While the modeling was done as planned, the best time to contact feature was disabled early in the
operation. The program expected that respondents may be home more often given the national
response to the COVID-19 pandemic. Disabling the best time to contact component of case assignment
allowed more cases to be assigned early in the operation.
D.

PHONE NUMBER AVAILABILITY

In areas where in-person follow-up was not permitted due to COVID-19 restrictions, wildfires, or
hurricanes, phone numbers associated with an enumerator's workload for the day were displayed. The
source of the phone numbers was the Contact Frame. This was an unplanned effort to increase response
given the unique circumstances arising during 2020.
RESPONSE DATA VERIFICATION
The Census Bureau used administrative data to examine the consistency between the respondentprovided or enumerator-collected information and the administrative data sources to determine if
further analysis and/or field investigation was warranted.
Name information, in combination with address and other information, can confirm a person’s
association with an address as well as the demographic data associated with their household and
whether the person can be found at another address. A level of consistency (i.e., matched, nonmatched)
between the respondent-provided information or the enumerator-provided data and information
available from administrative data was determined.
A.

SELF-RESPONSE QUALITY ASSURANCE

Data from the Title 26 composite were used in this operation.
FTI and non-FTI sources were used in matching algorithms to develop a set of Title 13 match results. The
match results were then used as inputs to scoring models that help identify suspicious cases for further
investigation. The algorithms were run in a Title 26 environment and the match results sent to a Title 13
environment for use in production.

August 11, 2021

12

B.

ENUMERATOR QUALITY CONTROL

Only data from the Title 13 composite were used in this operation.
Like many of the other processes, quality control efforts used the Title 13 composite described in
Section II above as the comparison data set to verify that enumerators conducted interviews
appropriately and collected accurate data. The data in the composite was never used in lieu of
additional field verification for those housing units with consistency measures below an acceptable
level. Furthermore, the administrative data matching results were never used as a sole indicator of
whether a response is suspicious and required further analyst and/or field follow-up. Instead, the results
from administrative data were used in conjunction with the results of statistical modeling on response
data, paradata, and other data elements to flag responses as suspicious.
POST-RESPONSE PROCESSING
A.

ADDITIONAL UNDUPLICATION OF PEOPLE IN HOUSING UNITS

During post-processing, various steps were planned to remove duplicates (a person counted more than
once). As we tallied census results, we compared these results to other benchmark data and determined
that additional duplicates remained in the 2020 Census, even after taking all the steps we had planned.
Consequently, after data collection was complete, we took more steps to identify and remove additional
duplicated individuals from the census. We did a broad search for duplicates within each state and the
District of Columbia. If we found individuals who were initially counted at two different housing units,
we first checked if there was any indication from census responses that one of the units was vacant or
nonexistent. If so, we removed the duplicated people from the vacant or nonexistent unit and kept
them at the other address.
If neither housing unit appeared vacant or nonexistent, we checked whether the people were only
associated with one of the addresses in administrative records. If so, we kept them at that address. Data
from the Title 26 composite were used in this operation. Review of the administrative data was done in
a T26 environment and only a list of the duplicated addresses to drop from the person records was sent
to a T13 environment for additional processing. Administrative sources corroborated a person’s 2020
Census enumeration at the location. We only used Protected Identification Keys (PIKs) and MAF
Identification Numbers (MAFIDs) from the administrative sources. For more information on
unduplicating responses, see the Census Bureau blog “How We Unduplicated Responses in the 2020
Census (Keller & King, 2021).
B.

COUNT IMPUTATION

After data collection, count imputation 1) assigned final statuses (i.e., occupied, vacant, and
nonexistent) to those cases that were unresolved and 2) assigned population counts to records with an
occupied status, but without a specified number of occupants. Response records sent to NARA will not
include imputations from administrative records.

August 11, 2021

13

1) Housing Units
Data from the Title 26 composite was used in this operation.
We used the USPS Undeliverable As Addressed (UAA) file as well as information from IRS 1040, IRS 1099,
Census Master Address File, 2020 Census paradata, American Community Survey, and other sources
including Medicare and Indian Health Service to place all addresses into groups with similar status and
household size (if occupied)1. Within each group, a hot-deck imputation process was used to fill in the
missing status and count (if occupied) from a census address with a complete status to the address
without a complete status. FTI was never used as a direct source for final status assignment.
•

•
•

The classification of administrative records information for each address was completed in a
Title 26 environment using the overwriting process described in Appendix B. These variables
were transferred to a Title 13 environment to model the final status when it was unknown and
the household roster count for occupied units where the count was unknown.
UAA information was assigned to each CensusID. The National Processing Center downloaded
this information from USPS for the Census Bureau.
We only used PIKs and MAFIDs from the IRS data.
2) Group Quarters

For some group quarters, we did not have response data to determine precisely how many people
should be assigned to an occupied facility. We developed, tested, and applied imputation procedures to
cases that could not be resolved through other means. These imputations did not use T26 composite or
T13 composite data. Instead, we used information already available on the group quarters under
consideration—such as the expected count or the maximum capacity the group quarters reported
during advance contact or from data collected in our current surveys. We used the IPEDS database to
help determine the GQ count for some colleges and universities if the information from advance contact
or current surveys was not available.
The Census Bureau had never before conducted count imputation on unresolved group quarters. This
change resulted from the unique difficulties presented by the COVID-19 pandemic, including higher
numbers of unresolved cases and difficulty obtaining information through the intended avenues, like
personal contact with GQ representatives who may not have been at the GQ location. For more
information, see the Census Bureau blogs on this topic (Cantwell, 2021; Jarmin, 2021).
C.

CHARACTERISTIC IMPUTATION

After all responses are processed and the final population count is established, there are sometimes
missing, inconsistent, or nonvalid characteristic data. Administrative records sources were used as part
of the process to impute missing characteristics. Imputation methods included direct substitution from
prior Title 13 responses or from other administrative record sources. They also included procedures

1

The 2010 Census was incorrectly listed as an intended source in the previously published version of this memo.
We did use 2010 Census counts for the 2018 End-to-End Census Test model, but further research showed that
other sources were more accurate for the 2020 model so we did not use 2010 Census data in the 2020 production.
We also used the Census Numident, which was inadvertently omitted in the previous memo.

August 11, 2021

14

where missing data were imputed from donors with similar characteristics at the person- or housingunit level. Response records sent to NARA will not include imputations from administrative records.
Only data from the Title 13 composite above were used in this operation. Sources of data included 2010
Census data, ACS responses, the Census Numident, the Best Race and Ethnicity file, Census Household
Composition Key File, HUD PIC/TRACS, and Black Knight. When possible, we tried to impute from the
2010 Census and ACS data before imputing from other sources. No FTI was used in characteristic
imputation.
Five characteristics were imputed for the 2020 Census, four at the person level and one at the housing
unit (HU) level. Only federal sources were used to impute person-level characteristics.
Characteristics Expected to be Imputed for the 2020 Census
Person-Level Characteristics
Housing Unit Characteristics
Age/Date of Birth
Tenure (own or rent the housing unit)
Sex
Race and Ethnicity
Relationship to Householder
PUBLISHING DATA
A.

CITIZEN VOTING AGE POPULATION (CVAP) TABLES

The Census Bureau suspended all work on the Post-2020 Census CVAP Special Tabulation on January 12,
2021, following the Executive Order on Ensuring a Lawful and Accurate Enumeration and Apportionment
Pursuant to the Decennial Census. The Census Bureau had planned to produce this tabulation using
2020 Census and administrative records data and publish it to the census block level. On January 21,
2021, the Census Bureau advised that it would reengage with the U.S. Department of Justice (DOJ) to
confirm that the Citizen Voting Age Population (CVAP) data produced from the American Community
Survey (ACS) continue to meet its statistical needs. On February 16, 2021 the Census Bureau received a
letter from DOJ stating that the CVAP data from the ACS, on which it has traditionally relied, are
adequate for its enforcement of Section 2 of the Voting Rights Act. DOJ did not request compilation or
release of additional citizenship or CVAP data beyond this ACS data.
B.

COUNT QUESTION RESOLUTION

The 2020 Count Question Resolution program (CQR) allows tribal, state, and local government elected
officials to request review for corrections to their jurisdiction’s 2020 Census counts. There are
historically a small percentage of cases where an incorrect geographic boundary or coding of a housing
unit was used to produce the official census population and housing count for a local area. There may
also be cases where, because of processing errors, the Census Bureau mistakenly duplicated or deleted
living quarters that were identified during the census. The Census Bureau does not collect any additional
data during the CQR process but will access existing information to research cases. The CQR process will
not research whether respondents or census enumerators incorrectly determined an address’s
occupancy status or household size during field operations, but only resolve boundary disputes, address
geocoding errors, and address coverage errors. Census Bureau records used in the research phase
include the MTdb, records collected or used during frame development activities (see section IV), and

August 11, 2021

15

2020 Census return metadata. Case submission for the CQR operation will take place from January 3,
2022, through June 30, 2023. Case research and final disposition will complete by September 30, 2023.
COVERAGE EVALUATION
A.

POST-ENUMERATION SURVEY

The Post-Enumeration Survey (PES) is one of two primary evaluation tools used to produce estimates of
census coverage. For the PES, addresses in the selected sample blocks were listed and enumerated in
operations that were independent of the 2020 Census. The Census Bureau identifies matches and
nonmatches and discrepancies between the 2020 Census and the PES, for both housing units and people
in the sample area. Both computer and clerical components of matching are conducted. The system that
conducted computer matching used telephone numbers from administrative data (that is, the Contact
Frame) for census records in the sample areas when no telephone number was reported in the census.
As a result, the use of the updated telephone numbers could improve computer match rates, thereby
improving the efficiency of clerical matching and potential followup operations. Additionally, during PES
data processing, the same characteristic imputation methodology that was used for the 2020 Census will
be used for the PES. Final PES work will be completed in the summer of 2022.
B.

DEMOGRAPHIC ANALYSIS

Demographic Analysis (DA) is the other approach for measuring census coverage. DA refers to a set of
methods that have historically been used to develop national-level estimates of the population for
comparison with decennial census counts. 2020 DA estimates were developed from current and
historical vital statistics from the National Center for Health Statistics (NCHS), data on international
migration from ACS, and Medicare records. The data were independent of the census being evaluated.
The results will be compared with the census counts by age, sex, limited race groups, and Hispanic origin
to produce estimates of net coverage error in the census. The DA program initially planned to use
administrative records on Legal Permanent Residents (LPR) from the Office of Immigration Statistics
(OIS), which is part of the Department of Homeland Security. Ultimately, these records were not used
because of changes to the method for estimating the foreign-born population. For more information,
see the technical documentation on the Census Bureau’s website (Jensen, et al., 2020). IRS tax returns
and the Census Numident file will also be used to develop state and county DA estimates for young
children age 0 to 4; however, this is an experimental set of estimates and not one of the official series of
2020 DA estimates. Final DA work will be completed by the end of 2022.
REFERENCES
Cantwell, P. (2021, April). How We Complete the Census When Households or Group Quarters Don’t
Respond. Retrieved from https://www.census.gov/newsroom/blogs/randomsamplings/2021/04/imputation-when-households-or-group-quarters-dont-respond.html
Deaver, K. D. (May 2020). Memorandum 2020.06: Intended Administrative Data Use in the 2020 Census.
Washington, DC: U.S. Census Bureau. Retrieved from https://www.census.gov/programssurveys/decennial-census/decade/2020/planning-management/plan/memo-series/2020-memo2020_06.html

August 11, 2021

16

Executive Office of the President. (January 20, 2021). Ensuring a Lawful and Aaccurate Enumeration and
Apportionment Pursuant to the Decennial Census. Retrieved from
https://www.federalregister.gov/documents/2021/01/25/2021-01755/ensuring-a-lawful-andaccurate-enumeration-and-apportionment-pursuant-to-the-decennial-census
Jarmin, D. R. (2021, February). 2020 Census Processing Updates. Retrieved from
https://www.census.gov/newsroom/blogs/director/2021/02/2020-census-processingupdates.html
Jensen, E. B., Knapp, A., King, H., Armstrong, D., Johnson, S. L., Sink, L., & Miller, E. (2020). Methodology
for the 2020 Demographic Analysis Estimates. U.S. Census Bureau. Retrieved from
https://www2.census.gov/programs-surveys/popest/technicaldocumentation/methodology/2020da_methodology.pdf
Keller, A., & King, R. (2021, April). How We Unduplicated Responses in the 2020 Census. Retrieved from
https://www.census.gov/newsroom/blogs/randomsamplings/2021/04/how_we_unduplicated.html
Mulry, M. H., Mule, T., Keller, A., & Konicki, S. (April 2021). Memorandum 2021.10: Administrative
Record Modeling in the 2020 Census. Washington, DC: U.S. Census Bureau. Retrieved from
https://www.census.gov/programs-surveys/decennial-census/decade/2020/planningmanagement/plan/memo-series/2020-memo-2021_10.html
U.S. Census Bureau. (December 2018). 2020 Census Operational Plan v4.0. Washington, DC: U.S. Census
Bureau. Retrieved from https://www.census.gov/programs-surveys/decennialcensus/decade/2020/planning-management/plan/planning-docs/operational-plan.html
U.S. Census Bureau. (various). 2020 Census Operational Plan and Detailed Operational Plans. Retrieved
from https://www.census.gov/programs-surveys/decennial-census/decade/2020/planningmanagement/plan/op-plans.html

August 11, 2021

17

APPENDIX A –DATA SOURCES USED IN THE 2020 CENSUS
Use
Frame
Resp
Non-ID
Paper Data
Special
Source
Development Motivation Address Validation
Capture QA
Enumerations.
USPS DSF
XXX
Ongoing MAF/TIGER System updates from
XXX
partnerships and fieldwork
2020 Census partnership data (LUCA, NC)
XXX
Lists of GQs and TLs from federal sources
XXX
(Medicare, BOP, ICE, Bureau of Indian
Affairs, U.S. Marshals, Maritime Agencies,
Military), state & local partners
STR Inc. list of hotels/motels
XXX
MTdb Extracts
XXX
XXX
XXX1
Modeled self-response predictions
XXX
Audience segmentation information
XXX
FCC Internet Connectivity
XXX
GQ administrator lists
XXX
Off-campus student records
IPEDS
DOD Deployed data and FACO inputs *
XXX
CMS MEDB
XXX
XXX
HUD FHA IDB (CHUMS)
XXX
XXX
HUD Longitudinal (PIC/TRACS)
XXX
XXX
IHS Patient Registration
XXX
XXX
IRS 1040
IRS 1099
NCHS Vital Statistics (births and deaths)
SSS Registration
XXX
XXX
State/Local Program Data
XXX
USPS NCOA
XXX
XXX
1
ACS Data
XXX
ACS Contact History
2000 Census Data
XXX
XXX
XXX
2010 Census Data
XXX
XXX
XXX
Census Numident
XXX
XXX
Census Numident Alternate Names
XXX
Census HH Composition Key
Contact Frame
XXX
XXX
XXX
Best Race and Ethnicity
Black Knight
DAR Partners
XXX
XXX
Targus Federal Consumer and Wireless
XXX
XXX
VSGI
XXX
XXX
USPS UAA
Planning Database
XXX
* Other federal agencies provided their FACO numbers directly into the system, not as a file to 2020 production systems and are
therefore not listed here.

1

This source was inadvertently omitted from the previously released memo.

August 11, 2021

Appendix A – 1

Use
Source

NRFU
Vacant/Delete
Identification

NRFU AR Enumeration
Occupied / AR
Count Characteristics
Enum. Eligible

NRFU
Phone
Avail.

NRFU Best
Time to
Contact

USPS DSF
Ongoing MAF/TIGER System updates from
partnerships and fieldwork
2020 Census partnership data (LUCA, NC)
Lists of GQs and TLs from federal sources
(Medicare, BOP, ICE, Bureau of Indian
Affairs, U.S. Marshals, Maritime Agencies,
Military), state & local partners
STR Inc. list of hotels/motels
MTdb Extracts
XXX
XXX
XXX
Modeled self-response predictions
Audience segmentation information
FCC Internet Connectivity
GQ administrator lists
Off-campus student records
XXX
XXX
IPEDS
DOD Deployed data and FACO inputs *
CMS MEDB
XXX
XXX
XXX
HUD FHA IDB (CHUMS)
HUD Longitudinal (PIC/TRACS)
XXX
IHS Patient Registration
XXX
XXX
XXX
IRS 1040
XXX
XXX
XXX
XXX
IRS 1099
XXX
XXX
XXX
XXX
NCHS Vital Statistics (births and deaths)
SSS Registration
XXX
XXX
State/Local Program Data
USPS NCOA
XXX
XXX
ACS Data
XXX
XXX
XXX
XXX
ACS Contact History
XXX
2000 Census Data
2010 Census Data
XXX
XXX
XXX
Census Numident
XXX
XXX
XXX
XXX
XXX
Census Numident Alternate Names
Census HH Composition Key
XXX
XXX
XXX
XXX
XXX
Contact Frame
XXX
Best Race and Ethnicity
XXX
XXX
XXX
XXX
Black Knight
XXX
DAR Partners
XXX
XXX
Targus Federal Consumer and Wireless
VSGI
XXX
XXX
XXX
USPS UAA
XXX
XXX
Planning Database
XXX
XXX
XXX
* Other federal agencies provided their FACO numbers directly into the system, not as a file to 2020 production systems and are
therefore not listed here.

August 11, 2021

Appendix A – 2

Use

SRQA

Source
USPS DSF
Ongoing MAF/TIGER System updates from
partnerships and fieldwork
2020 Census partnership data (LUCA, NC)
Lists of GQs and TLs from federal sources
(Medicare, BOP, ICE, Bureau of Indian
Affairs, U.S. Marshals, Maritime Agencies,
Military), state & local partners
STR Inc. list of hotels/motels
MTdb Extracts

Enum.
Additional Count Imputation Character. CQR
QC Unduplication
HU
GQ Imputation

PES

DA

XXX
XXX
XXX

XXX

XXX

XXX

XXX

XXX*
*

XXX
XXX

Modeled self-response predictions
Audience segmentation information
FCC Internet Connectivity
GQ administrator lists
Off-campus student records
IPEDS
XXX
DOD Deployed data and FACO inputs *
CMS MEDB
XXX
XXX
XXX
XXX
XXX
HUD FHA IDB (CHUMS)
XXX
HUD Longitudinal (PIC/TRACS)
XXX
XXX
XXX
XXX
IHS Patient Registration
XXX
XXX
XXX
XXX
IRS 1040
XXX
XXX
XXX
XXX
IRS 1099
XXX
XXX
XXX
NCHS Vital Statistics (births and deaths)
XXX
SSS Registration
XXX
XXX
State/Local Program Data
USPS NCOA
XXX
XXX
ACS Data
XXX
XXX
XXX
XXX XXX
ACS Contact History
2000 Census Data
XXX
XXX
XXX
2010 Census Data
XXX
XXX
XXX
XXX XXX
Census Numident
XXX
XXX
XXX
XXX
XXX
XXX XXX
Census Numident Alternate Names
XXX
XXX
Census HH Composition Key
XXX
XXX
XXX
XXX
Contact Frame
XXX
XXX
XXX
Best Race and Ethnicity
XXX
XXX
Black Knight
XXX
XXX
DAR Partners
XXX
XXX
Targus Federal Consumer and Wireless
XXX
XXX
VSGI
XXX
XXX
USPS UAA
XXX
Planning Database
* Other federal agencies provided their FACO numbers directly into the system, not as a file to 2020 production systems and
are therefore not listed here
** The count at the GQ from the last ACS or current surveys visit is contained in this source.

August 11, 2021

Appendix A – 3

APPENDIX B - OVERWRITING EXAMPLE
This is an example of how the U.S. Census Bureau can use FTI and information from multiple sources in the 2020 Census to inform census
fieldwork decisions and conduct administrative record enumeration. This approach adheres to procedures provided by IRS Office of Safeguards
regarding validation and overwriting FTI. This example was presented to IRS Statistics of Income staff in July 2016 during the review and approval
process for the 2020 Census Production Predominant Purpose Statement.
This example presents a hypothetical Mars family who filed an IRS 1040 return at 101 Main Street. The source, validation, and overwriting
involving different variables and values are presented. Data variables and values that are FTI are shown in red shaded background with white
letters. Data variables and values that have been overwritten in accordance with the IRS regulations are shown in white background with green
letters.
1.

IRS 1040 and 1099 Information

Each month, the IRS delivers the IRS 1040 Individual tax returns. Table 1 shows the two hypothetical families at 101 and 102 Main Street. The
Mars family has four people and lives at 101 Main Street. John Smith lives by himself at 102 Main Street. Since this is the 1040 information
delivered to the Census Bureau, all of the variables are FTI and thus colored in red. Among the variables, the IRS provides the Census Bureau
with the address, tax identification number (TIN) of filers and dependents, filing status, and other FTI variables. For 101 Main Street, primary filer
Michael Mars had a full name, SSN present but the other people have only last name and SSNs provided. For 102 Main Street, single primary filer
John Smith did not provide an SSN when filing. While not shown in the table, the IRS also delivers a similar file containing the 1099 Information
return file as well.

Depende
nt 3 TIN

Depende
nt 3
Name

Other FTI
Variables
…

Mars

Depende
nt 2
Name

Secondar
y
Name

111-111112

Depende
nt 2 TIN

Secondar
y Filer TIN

Michael
Mars

Depende
nt 1
Name

Primary
Name

111111111
blank

Depende
nt 1 TIN

Primary
Filer TIN

101
Main
St.
102
Main
St.

Filing
Status

Address

Table 1: IRS 1040 FTI Record of the Mars Family at 101 Main St.

Joint
Married

111-111113

Mars

111-111114

Mars

None

None

…

John
Smith

None

None

Single

None

None

None

None

None

None

…

30 November 2019

Appendix B - 1

2.

Census Person Validation System (PVS) Processing of IRS 1040 and 1099 Delivered File

After receipt of the IRS 1040 file, the Census Bureau processes the IRS 1040 file through the Person Validation System. This processing uses the
address provided by the IRS and links this address to an address already to the Census Master Address File (MAF). This assigns the Census
Bureau’s MAF Identification Number (MAFID) to the record. The PVS processing was able to assign the Census Bureau’s Protected Identification
Key (PIK) to replace the IRS TIN based on combinations of SSN, Name and Address fields. For John Smith, the PVS processing was unable to
assign a PIK. This is an example of a file that has a combination of FTI and non-FTI variables. The address and PIKs are colored in green since they
have been overwritten in accordance with IRS regulations. IRS provided name information has been removed. The Census Bureau does similar
processing of the 1099 Information file as well.

Census
MAFID

Address

Table 2: IRS 1040 Record for Mars Family after Census PVS Processing
Census PIK 1
Census PIK 2
Filing Status
Census PIK 3
Census PIK 4

123412341
123412342

101 Main St.
102 Main St.

999-99-9991
Blank

999-99-9992
Blank

Joint Married
Single

999-99-9993
Blank

999-99-9994
Blank

Other FTI
Variables
…
…

Since 102 Main St. has no PIKs, we are not able to build a roster using IRS 1040 data for this address. The remainder of this document will focus
on using the IRS data to see if we can reduce contacts for 101 Main St. For the 2020 Census, the Census Bureau will be building rosters for
households using multiple sources determined by September 2018. The remainder of this document will assume that an additional source
besides IRS information can corroborate that the Mars family lives at 101 Main St.
3.

Processing of Administrative Records to Determine Census Fieldwork

The Administrative Records Modeling team has been researching under the DMS 863 project the possibility of using administrative records and
third-party data to determine if the number of contacts can be reduced. The 2020 Production DMS 987 project will implement this usage in the
2020 Census. In this example, we will show parts of the processing. Step 2 showed that the Census Bureau had a file that had a combination of
FTI and non-FTI variables. Our research has shown that other variables created using FTI are powerful predictors in our determination. This
example highlights two of them. Table 3 shows an example of where we create additional variables based on FTI information to be used in the
30 November 2019

Appendix B - 2

determination about how many contacts to conduct during fieldwork. There is a history of using IRS information combined with other sources to
determine fieldwork for the decennial census. For the 2010 Census, part of the workload for the Coverage Followup operation was based on the
comparison of census rosters to administrative records households that included IRS 1040, IRS 1099, and other non-FTI sources.
An example of two variables created using FTI information is:
•

Was any person in the Mars household found on another 1040 return filed this year? Our research has found that if any person in the
household is associated with another 1040 return, then there are possible questions about the administrative record roster assembled.

•

Was any person in the Mars household found on the 1040 return filed at this address last year? Our research has seen that finding at
least one person on 1040 returns at the same address in consecutive years is more highly correlated with count and household
composition agreement.

Census
MAFID
123412341
123412341
123412341
123412341

Table 3: Example of Creating Two Additional FTI Variables to be used in Determination
Census PIK
Filing Status
Processing
Found on
Found on last year
Week
another 1040
1040 at same
return
address
999-99-9991
Joint Married
12
No
Yes
999-99-9992
Joint Married
12
No
Yes
999-99-9993
Joint Married
12
No
Yes
999-99-9994
Joint Married
12
No
Yes

The Administrative Record determination processing uses predictive models. These predictive models are based or “trained” on comparison of
the rosters built for 2010 NRFU addresses using people from 2010 versions of the administrative record sources to the person enumerated at
those addresses in the 2010 Census. Based on building a roster using the most recent versions of the administrative record files, we can use
predictive models can allow us to make determinations if the roster from administrative record sources is similar enough to a census fieldwork
enumeration that we can reduce the number of contacts. In our example, we have built a roster of the Mars family.

30 November 2019

Appendix B - 3

For every 2020 household roster built from administrative record sources, we can predict:
•

How likely is it that all of the Mars family should be counted at the address? From a scale of 0 to 1, we want this to be as close to 1 as
possible.

•

How likely is it that the household composition observed for the Mars family from IRS 1040 and other sources would match census
fieldwork6? From a scale of 0 to 1, we want this to be as close as possible to 1 as well.

Figure 1 shows a graphical description of the number of contacts during fieldwork determination. Each axis represents one of the two
predictions listed above. The results are shown in three colors of green, yellow, and red.
For the 2020 Census, we are implementing the following three classifications. In our example, the Mars family will have high enough results for
the two predictions that we will reduce the contacts so that this address is classified as administrative record occupied and will have reduced
contacts during the NRFU operation.

6

•

Administrative Record Occupied in NRFU (Green): If the two probabilities land the address in the green area then the address will be
administrative record occupied. We will be reducing contacts for these addresses.

•

Administrative Record Occupied for Imputation (Yellow): The Census Bureau has done research on using administrative records after all
of the contacts in the NRFU as an alternative to count imputation. For the 2016 Census Test, these addresses will receive six contacts. If
the two predicted probabilities land the address in the yellow area, then the address we will use the roster of administrative records as
an alternative to imputation

•

Not Use (Red) – If the predicted probabilities are in red, then we would implement full fieldwork. We would not want to use these
rosters built from administrative records as an alternative to count imputation.

In this example, we would be saying how likely it would be that the census would enumerate a household with two adults and one or more children present.

30 November 2019

Appendix B - 4

Figure 1: Graphical Representation of Hypothetical Administrative Record Occupied Determination

30 November 2019

Appendix B - 5

4.

Overwriting for Enumeration Purposes

Based on the result for the Mars family, we have determined that we will reduce contacts during NRFU. While we used FTI in determining how
to conduct census fieldwork, the following will use the Mars family as an example to show how we will not be using FTI for enumeration.
For enumeration, the only information originally sourced from FTI that we use is the address and the PIK. Based on the IRS Office of Safeguards
response to the Census Bureau’s January 2016 overwriting inquiry, the Census MAFID and PIK have overwritten the FTI. In accordance with the
guidance from IRS, we will document the overwriting and the date that it has occurred. This file is no longer FTI. Table 4 provides an example.
Table 4: Example of Overwriting Documentation File for Future IRS Office of Safeguards Review
Census MAFID
Unique Person Number
Census PIK
Overwriting Date
123412341
1
999-99-9991
May 2, 2016
123412341
2
999-99-9992
May 2, 2016
123412341
3
999-99-9993
May 2, 2016
123412341
4
999-99-9994
May 2, 2016

30 November 2019

Appendix B - 6

5.

Obtain Short Form Characteristics from non-FTI sources

For census enumeration purposes, we need to obtain the name, age, date of birth sex, relationship, race, Hispanic origin, and tenure information
for the people at this address. For these characteristics, this information will be obtained from non-IRS Title 26 sources. This includes past
Census Bureau responses or other administrative record information sources.
Table 5 shows the administrative record enumeration for 101 Main Street. All of the fields are colored green to indicate that no FTI information
was used in these variables. Name was obtained from the Census Numident file or past census responses. To further, emphasize in this example
that no FTI information is being used for enumeration purposes, the relationship for Mars family is set to missing. From the IRS 1040 return, we
did see that the Mars family filed a Joint Married return, but we do want to point out that we are not using that information. Since IRS is the sole
source of the relationship, we are unable to use this information as a direct assignment. Our characteristic imputation will have to assign the
relationship status for this household.
Table 5: Example of Administrative Record Enumeration of Mars Family Using non-FTI information
Census Address
Unique
Name
Age
Date
Sex
Relationship
Person
Of
Number
Birth
101 Main St.
1
Michael Mars
43
1/1/1973 Male
Missing
101 Main St.
2
Mary Mars
40
2/2/1976 Female Missing
101 Main St.
3
Hershey Mars
17
3/3/1999 Male
Missing
101 Main St.
4
Lucy Mars
14
3/4/2002 Male
Missing

Race

White
White
White
White

Hispanic
Origin
Non-Hispanic
Non-Hispanic
Non-Hispanic
Non-Hispanic

Note: Census Day is April 1, 2016 for this example.

Step 3 showed that for some addresses it may be determined administrative records would only be used as an alternative to imputation. For
these addresses, they will receive the full amount of contacts during the NRFU operation. If at the end of the census fieldwork, their occupancy
and/or population status are still unresolved then administrative records will be used in a similar way as laid out in this section.

30 November 2019

Appendix B - 7

6. Creating a Census Response File
Based on this processing, this information can then be delivered to our Title 13 server and be combined with the other Census Bureau
responses. This information can be used for all Title 13 purposes of the 2020 Census including archiving to be made available 72 years after
Census Day. Table 6 shows a simple example of variables and values that would be included.
Table 6: Census Response File
Address
Enumeration
Operation
101 Main St.
101 Main St.
101 Main St.
101 Main St.
102 Main St.
102 Main St.
103 Main St.

Administrative Records
Administrative Records
Administrative Records
Administrative Records
Internet Self-Response
Internet Self-Response
NRFU

Unique
Person
Number
1
2
3
4
1
2
1

Name

Michael Mars
Mary Mars
Hershey Mars
Lucy Mars
Thomas Lindt
Sally Lindt
John Haribu

Age

43
40
17
14
28
26
33

Sex

Relationship

Race

Male
Female
Male
Female
Male
Female
Male

Missing
Missing
Missing
Missing
Householder
Spouse
Householder

White
White
White
White
Black
Black
SOR

Hispanic Origin

Non-Hispanic
Non-Hispanic
Non-Hispanic
Non-Hispanic
Non-Hispanic
Non-Hispanic
Hispanic

Summary
This document shows how FTI will and will not be used in relation to administrative record enumeration in the 2020 Census. While FTI will be
used during the processing to determine the number of contacts for census fieldwork purposes, the example shows that FTI will not be used in
the direct enumeration.

30 November 2019

Appendix B - 8