1

Introduction

This proceedings summarizes the presentations and discussions at the Workshop on the 2020 Census Demographic and Housing Characteristics File, held June 21–22, 2022. The workshop was convened by the Committee on National Statistics (CNSTAT) of the National Academies of Sciences, Engineering, and Medicine to assist the U.S. Census Bureau with its new disclosure avoidance system¹ for 2020 Census data products, which implements algorithms providing differential privacy (see Chapter 3). The workshop focused specifically on the Demographic and Housing Characteristics (DHC) File, a major source of data for local governments, particularly those with small populations, and many other data users in the federal, state, academic, and business sectors. The intent was to garner feedback from users on the usability of the privacy-protected data by evaluating DHC “demonstration” files produced with the proposed TopDown Algorithm (TDA) on 2010 Census data.²

BACKGROUND

This workshop was a successor to a December 2019 public workshop³ held by CNSTAT to provide user feedback on the first effort by the Census

___________________

¹ A disclosure avoidance system is a process of protecting confidentiality and limiting ability to link a specific individual to person or housing characteristics in public use data files.

² The 2010 equivalent of the DHC File is referred to as Summary File 1.

³ https://www.nationalacademies.org/event/12-11-2019/workshop-on-2020-census-data-products-data-needs-and-privacy-considerations

Bureau to implement the TDA on 2010 data similar to the DHC File. In response to the first workshop, the Census Bureau made changes⁴ to the TDA for the DHC File and released another DHC demonstration file in May 2020, but then focused on implementing the TDA for the 2020 Public Law 94-171 redistricting file, which was released in August 2021. Following the redistricting file release, the Census Bureau revisited the DHC File and released a 2010 demonstration file with person tables on March 16, 2022, and a second file with housing tables on April 14, 2022 (see Chapter 3). These demonstration files were the focus of this second workshop (see Box 1-1 for the Statement of Task).

The workshop brought together applied demographers; academics; and local, state, and federal government officials to share use cases and provide feedback to the Census Bureau before it planned to finalize the parameters of the TDA for the 2020 DHC File during summer–fall 2022. In addition to speakers and staff, the workshop was attended by 297 people who registered to attend virtually and 646 viewers (as measured by unique Internet protocol addresses) who synchronously watched some or all of the workshop via the webcast.

___________________

⁴ These changes to the algorithm are not public information.

OVERVIEW OF DEMONSTRATION DATA

Throughout this proceedings, the terms published data and demonstration data are used frequently and require explication to understand the different data sets analyzed by the presenters. Published data refers to the release of the 2010 decennial DHC data in what was previously called Summary File 1 (SF1). Many presenters simply referred to SF1 in their presentations. As explained in more detail in Chapter 3, SF1 used a mechanism called swapping for its disclosure avoidance system to protect the confidentiality of respondents. Although these are not the raw data because swapping was already infused in the SF1 data, they serve as the baseline for all comparisons.

The Census Bureau adopted a new disclosure avoidance system for 2020, based on differential privacy, called the TDA, discussed more fully in Chapter 3. Knowing this would be a change from how past Census data products were created, the Census Bureau released three versions of DHC demonstration data for 2010 (depicted in Chapter 3, Figure 3-2) prior to this workshop:

1. October 2019;
2. May 2020; and
3. March 2022 (separate person and housing files; the housing file was reissued in April 2022).

The purpose of the demonstration files created using 2010 data was to enable public feedback by using these data, protected with the new TDA, to compare accuracy with the published (SF1) data as the baseline.⁵ Although most presenters prepared use cases for this workshop using the most recent demonstration data released in 2022, some presenters compared the SF1 data with one or two of the earlier demonstration data sets to illustrate improvements in accuracy with each iteration.

In addition to direct outreach by the planning committee, CNSTAT issued a Call for Input that was circulated broadly across the federal government; state demographers; multiple professional associations; and numerous civic and community-based organizations, including those representing hard-to-reach populations. Box 1-2 provides an excerpt from the Call for Input that delineates the types of use cases sought for the workshop.

___________________

⁵ Unlike the TDA, swapping did not change population totals for geographic areas and, more generally, introduced less noise, making the published SF1 suitable for comparison with the demonstration data.

ORGANIZATION OF PROCEEDINGS

The workshop was held in a hybrid format—some presenters and attendees participated in person in Washington, DC, while others participated virtually through a live-streamed public webinar. Given the limited time available to recruit presenters before the workshop, not all presenters were able to attend in person, which resulted in broader participation but also less flexibility in scheduling given prior commitments. As a result, the agenda listed sessions generically as “Use Cases Parts I–V” and was organized by speaker availability; however, they are described in this proceedings thematically. When presenters offered a use case in more than one area, the relevant portions of their presentations appear in the corresponding thematic sections in this proceedings. In addition to this publication, all

slides and videos of the 2-day workshop are available in their entirety on the project’s website.⁶

Chapter 2 outlines opening remarks and goals for the workshop from the planning committee and the U.S. Census Bureau, focused on data equity and communication strategies. Chapter 3 provides overviews from the Census Bureau on the TDA and DHC feedback and demographic programs. Chapters 4–7 summarize use cases presented at the workshop, grouped thematically by applications using data on age, housing and tenure, small areas and populations, and public health, respectively. Chapter 8 discusses privacy concerns raised by a panel of participants, as well as one use case focused on the risk of re-identification, underpinning why privacy is important to protect. Chapter 9 provides observations on use cases and needs for 2020 decennial data and beyond. Chapter 10 addresses themes raised in presentations over the course of the two-day workshop regarding statutory requirements for decennial data and the impact of the quality and timeliness of these data on local, state, and federal resource allocations. This proceedings concludes with Chapter 11, which is devoted to reflections and ideas for moving forward.

The full meeting agenda⁷ and biographical sketches of the planning committee members and workshop presenters, moderators, and discussants appear in the appendixes. This proceedings was prepared by the workshop rapporteur as a factual summary of what occurred at the workshop. The planning committee’s role was limited to planning and convening the workshop. The views contained in the proceedings are those of individual workshop participants and do not necessarily represent the views of all workshop participants; the planning committee; the U.S. Census Bureau; or the National Academies of Sciences, Engineering, and Medicine.

___________________

⁶ https://www.nationalacademies.org/event/06-21-2022/2020-census-data-products-workshop-on-the-demographic-and-housing-characteristics-files

⁷ Appendix A reflects changes made to the agenda for Day 2 to make time for the Closing Reflections and Ideas final session that was added after Day 1. As a result, some sessions on Day 2 were shortened and eliminated the discussant’s role and/or time originally allocated for Q&A because the intent was to deliver the use cases to the Census Bureau to help refine the TDA, as described in the statement of task.

This page intentionally left blank.