Lessons Learned from the 2020 Census
Published: March 24, 2021
The Census Bureau faces issues with data processing, quality and privacy as it wraps up its latest, and likely most tumultuous, decennial count.
- Late changes to the design of the census due to last year’s unprecedented events led to a reduction of scope and compressed duration in data quality checks. Accordingly, the GAO spent the last several months raising issues in data processing, public trust and data privacy facing the 2020 Census.
- Challenges and lessons learned by the Census Bureau in the 2020 count gives contractors an early indication of the solutions and needs of the agency for the next decadal count.
While the world hails 2020 as an unforgettable year, many at the Census Bureau likely carry the same sentiments towards the latest decadal count. Despite the advanced uses of technology to carry out the 2020 Decennial Census, the national count faced hurdles in its implementation, jarred by natural disasters and a global pandemic that shut down operations for several weeks.
Briefly recapping the count’s chain of events, the Bureau suspended operations in March 2020, slowly resuming implementation in mid-May. Non-Response Follow-Up (NRFU) visits took place in waves between July and August, wrapping up mid-October 2020. On August 3, 2020, the Bureau announced a series of late changes to the design of the census to accelerate delivery of population counts by statutory deadlines in December 2020. After a series of delays in data processing, the Bureau is now promising delivery of the count by April 30, 2021, yet concerns remain in the quality and trustworthiness of data due to the Bureau’s altered plans.
Late changes to the design of the census included:
- Making NRFU phone calls to collect data in areas hit by natural disasters instead of making personal visits
- Fewer than originally planned visits to households self-reported as vacant
- Reduced scope of NRFU re-interviews for quality control and a reduced time frame for internal Bureau experts to review apportionment population counts
- Elimination of a second administrative record to corroborate an initial source of records
- The Master Address File was locked-down on September 27, 2020 prior to NRFU completion
- Accelerated testing for 12 IT systems to support census response processing, completed in October 2015 instead of January 2021, making less time available for thorough integration testing and operational testing
The combination of fewer visits to nonresponding households, reduction of scope, and a compressed time span for quality checks puts the Census Bureau in the hot seat with the GAO. The government’s watchdog issued a flurry of reports with concerns surrounding data processing, public confidence and weakened data privacy protections in the count.
In terms of data processing, Bureau officials shared with the GAO in November 2020 several anomalies in processing responses for the apportionment count. For example, as the data integrates into group quarter counts, mismatches and duplicates are arising in the data. Moreover, Bureau officials found the system aging people incorrectly. As of mid-March 2021, Bureau officials identified and addressed 33 anomalies.
From a quality standpoint, the GAO urged the Census Bureau to examine how late changes to the census design could affect data quality and provide real-time transparency on what is known and yet unknown about the census quality to increase public confidence. In a December 2020 report, the GAO provided additional quality indicators the Bureau should consider with the release of apportionment counts.
Moreover, the GAO found that the Bureau has yet to finalize plans on the protection of respondent data. To protect the confidentiality of data, the Bureau will apply differential privacy to its publicly released statistics. However, census delays and late changes have put implementation of differential privacy at risk. As a remediation, the Bureau plans to release mock redistricting data files by the end of April 2021 to reveal what the final product will look like. The agency will then collect input on the demonstrated data files with key decisions on how to present the data made by June 2021. Nonetheless, as of March 2021, the GAO found that the Bureau has not yet finalized plans for other data products, such as demographics and house characteristics data, nor time frames for disclosure avoidance-related activities for those products.
Looking ahead to the 2030 Census, the GAO advises the Census Bureau to apply changes and challenges from 2020 to the next count, stating, “Actions taken by the Bureau out of necessity to meet compressed deadlines may be considered as options in future censuses. As part of its planning for the 2030 census, it will be important for the Bureau to assess the impact of the 2020 late design changes and the operations’ challenges that arose.” From a contractor point of view, understanding each challenge faced by the Census Bureau and its follow up assessments from the 2020 count is an early indicator of the solutions needed by the agency in the next count.