Key Issues > High Risk > 2020 Decennial Census
High Risk Medallion

2020 Decennial Census

This information appears as published in the 2017 High Risk Report.

View the 2017 Report

  1. Share with Facebook 
  2. Share with Twitter 
  3. Share with LinkedIn 
  4. Share with mail 

One of the most important functions of the U.S. Census Bureau (Bureau) is conducting the decennial census of the U.S. population, which is mandated by the Constitution and provides vital data for the nation. This information is used to apportion the seats of the U.S. House of Representatives; realign the boundaries of the legislative districts of each state; allocate billions of dollars in federal financial assistance; and provide social, demographic, and economic profiles of the nation's people to guide policy decisions at each level of government. A complete count of the nation's population is an enormous challenge as the Bureau seeks to control the cost of the census while it implements several new innovations and manages the processes of acquiring and developing new and modified information technology (IT) systems supporting them. Over the past 3 years, we have made 30 recommendations to help the Bureau design and implement a more cost-effective census for 2020; however, only 6 of them had been fully implemented as of January 2017.

The cost of the census, in terms of cost for counting each housing unit, has been escalating over the last several decennials. The 2010 Census was the costliest U.S. Census in history at about $12.3 billion, and was about 31 percent more costly than the $9.4 billion 2000 Census (in 2020 dollars).1 The average cost for counting a housing unit increased from about $16 in 1970 to around $92 in 2010 (in 2020 constant dollars). Meanwhile, the return of census questionnaires by mail (the primary mode of data collection) declined over this period from 78 percent in 1970 to 63 percent in 2010. Declining mail response rates—a key indicator of a cost-effective census—are significant and lead to higher costs. This is because the Bureau sends enumerators to each non-responding household to obtain census data. As a result, non-response follow-up (NRFU) is the Bureau's largest and most costly field operation. In many ways, the Bureau has had to invest substantially more resources each decade to match the results of prior enumerations.


[1] The fiscal year 2020 constant dollar factors the Bureau used are derived from the Chained Price Index from “Gross Domestic Product and Deflators Used in the Historical Tables: 1940–2020” table from the Fiscal Year 2016 Budget of the United States Government.

Challenges Implementing Innovations

The Bureau is planning many previously unused innovations for the 2020 Census: The decennial census is an inherently challenging undertaking, requiring many moving parts to come together in a short time and be completed according to a prescribed schedule. To help control costs and maintain accuracy, the Bureau is introducing significant change to how it conducts the decennial census in 2020. Its planned innovations include (1) making greater use of local data, imagery, and other office procedures to build its address list; (2) improving self-response by encouraging respondents to use the Internet and telephone; (3) using administrative records to reduce field work; and (4) reengineering field operations using technology to reduce manual effort and improve productivity. While the census is under way, the tolerance for any breakdowns is quite small. As a result, given the new four innovation areas for the 2020 Census, it will be imperative that the Bureau have systems and operations in place for the 2018 End-to-End Test.

Using administrative records is promising but introduces challenges: Although administrative records—information already provided to the government as it administers other programs—have been discussed and used for the decennial census since the 1970s, the Bureau plans a more significant role for them to reduce the amount of data collection fieldwork, which has the potential to help significantly limit the cost increases of the 2020 Census. The Bureau has estimated that using these records could save up to $1.4 billion compared to traditional census methods. In 2015, we found that while the Bureau has already demonstrated the feasibility of using administrative records, it still faces challenges with using them for the 2020 Census.1 For example, although the Bureau has no control over the accuracy of data provided to it by other agencies, it is responsible for ensuring that data it uses for 2020 Census are of sufficient quality for their planned uses.

Another challenge we identified is the extent to which the public will accept government agencies sharing personal data for the purposes of the census. Related concerns involve trust in the government and perceptions about burden on respondents as well the social benefits of agencies sharing data. Moreover, in addition to using administrative records to reduce fieldwork, the Bureau is considering several additional opportunities to leverage administrative records to help improve the cost and quality of the 2020 Census. It will be important for the Bureau to set deadlines for deciding which records it will use and for which purpose to help the Bureau monitor its progress and prioritize which activities—or records—to continue pursuing, or to abandon, if time becomes a constraint.

The Bureau needs to identify and analyze root causes of non-interviews during testing: When households do not respond to the census and when the Bureau does not obtain information about the household while knocking on doors during its NRFU operation, the Bureau may have to impute attributes of the household based on the demographic characteristics of surrounding housing units as well as on administrative records. We reported in 2016 that during the Bureau's 2016 Census Site Test, the Bureau experienced about 20 and 30 percent of its test workload as non-interviews at its two test sites in Harris County, Texas, and Los Angeles County, California, respectively.2 According to the Bureau, non-interviews are cases where no data or insufficient data are collected, either because enumerators make six attempted visits without success (the maximum number the Bureau allows), or are not completed due to, for example, language barriers or dangerous situations.3 Identifying root causes of problems is something we look for when determining progress within a high-risk area. Accordingly, while the 2016 Census Test non-interview rate is not necessarily a precursor to the 2020 non-interview rate, because of its relationship to the cost and quality of the census, it will be important for the Bureau to better understand the factors contributing to it.

Bureau cancelled field tests for 2017: The Bureau plans to conduct additional research through 2018 in order to further refine the design of the 2020 Census, but recently had to alter its approach. On October 18, 2016, the Bureau decided to stop two field test operations planned for fiscal year 2017 in order to mitigate risks from funding uncertainty. Specifically, the Bureau said it would stop all planned field activity, including local outreach and hiring, at its test sites in Puerto Rico, North and South Dakota, and Washington State. The Bureau will not carry out planned field tests of its mail-out strategy and follow up for non-response in Puerto Rico or its door-to-door enumeration. The Bureau also cancelled plans to update its address list in the Indian lands and surrounding areas in the three states.

However, the Bureau will continue with other planned testing in fiscal year 2017, such as those focusing on systems readiness and Internet response. Further, the Bureau said it would consider incorporating the cancelled field activities elements within the 2018 End-to-End Test. The Bureau maintains that stopping the 2017 Field Test will help prioritize readiness for the 2018 End-to-End Test, and mitigate risk. Nevertheless, as we reported in November 2016, it also represents a lost opportunity to test, refine, and integrate operations and systems, and it puts more pressure on the 2018 Test to demonstrate that enumeration activities will function as needed for 2020.4

 

Critical IT Uncertainties

The Bureau needs to strengthen the management and oversight of all IT programs, systems, and contractors supporting the decennial: The redesign of the 2020 Census relies on many new and modified IT systems. In addition to those systems that are being managed and developed within the 2020 Census Directorate, the 2020 program is also heavily dependent upon 11 systems that are being delivered by the CEDCaP program—a large and complex modernization program within the IT Directorate. Importantly, as a result of the Bureau's challenges in key IT internal controls and its rapidly approaching deadline, we identified CEDCaP as an IT investment in need of attention in the February 2015 high-risk report.5 In addition, in August 2016, we reported that the 2020 program and CEDCaP program lacked effective processes for managing their schedule, risk, and requirements interdependencies.6

For example, among tens of thousands of schedule activities, the two programs were expected to manually identify activities that are dependent on each other, and rather than establishing one integrated dependency schedule, the programs maintained two separate dependency schedules. We reported that this contributed to misaligning milestones between the programs. We stated that until the two programs establish schedules that are completely aligned, develop an integrated list of all interdependent risks, and finalize processes for managing requirements, both programs are at risk of not delivering their programs as expected.

The Bureau is also relying on contractor support in many key areas, including technically integrating all of the key systems and infrastructure, and developing many of the key data collection systems. Specifically, in August 2016, the Bureau hired a contractor to technically integrate the 2020 Census systems and infrastructure, to include evaluating the systems and infrastructure, developing the infrastructure (e.g., cloud or data center) to meet the Bureau's scalability and performance needs, integrating all of the systems, and supporting testing activities.

In addition, the Bureau is relying on contractors to develop a number of key systems and infrastructure; these activities include (1) developing the IT platform that will be used to collect data from a majority of respondents—by using the Internet, telephone, and NRFU activities; (2) procuring the mobile devices and cellular service to be used for NRFU; and (3) developing the IT infrastructure in the field offices. A greater reliance on contractors for these key components of the 2020 Census requires the Bureau to focus on sound management and oversight of the key contracts, projects, and systems.7

Key IT decisions need to be prioritized and made in time for full end-to-end testing in 2017: We have issued a series of reports and testimonies that have discussed the Bureau's challenges in prioritizing and making IT decisions. In April 2014, we reported that the Bureau had not prioritized key IT research and testing needed for its 2020 Census design decisions.8 In February 2015, we reported that the Bureau had not determined how key IT research questions would be answered—such as the expected rate of respondents using its Internet response option or the IT infrastructure that would be needed to support this option.9

Further, we testified, in November 2015, that key IT decisions needed to be made soon because the Bureau was less than 2 years away from preparing for end-to-end testing of all systems and operations, and there was limited time to implement them.10 We emphasized that the Bureau had deferred key IT-related decisions, and that it was running out of time to develop, acquire, and implement the systems it will need to deliver the redesign.

In October 2016, Bureau officials stated that they had 16 IT-related and 32 partially IT-related decisions left to make, including the uses of cloud-based solutions, the tools and test materials to be used during integration testing, and the expected scale of the system workload for those respondents who do not use the Bureau-provided Census ID. It will be important to make these decisions in enough time to develop solutions before the End-to-End Test begins in August 2017.

Information security risks and challenges need to be addressed to secure the Bureau's systems and data: In August 2016, we described the significant challenges that the Bureau faces in securing systems and data, such as developing policies and procedures to minimize the threat of phishing aimed at stealing personal information11 and ensuring that individuals gain only limited and appropriate access to 2020 Census data. Because many of the systems to be used in the 2018 End-to-End Test are not yet fully developed, the Bureau has not finalized all of the controls to be implemented, completed an assessment of those controls, developed plans to remediate any control weaknesses, and determined whether there is time to fully remediate any weaknesses before the system test.

Unreliable 2020 Cost Estimate

Estimation does not conform to best practices: We reviewed the Bureau's October 2015 estimated comprehensive life-cycle cost for the 2020 Census and reported in 2016 that it did not conform to best practices, and, as a result, the estimate was unreliable.12 Cost estimates that appropriately account for risks facing an agency can help an agency manage large, complex activities like the 2020 Census, as well as help Congress make funding decisions and provide oversight. Cost estimates are also necessary to inform decisions to fund one program over another, to develop annual budget requests, to determine what resources are needed, and to develop baselines for measuring performance.

We found that although the Bureau had taken significant steps to improve its capacity to carry out an effective cost estimate, its estimate for the 2020 Census partially met the characteristics of two best practices (comprehensive and accurate) and minimally met the other two (well-documented and credible), where all four need to be substantially met in order for an estimate to be deemed high-quality.

According to best practices, to be comprehensive an estimate has to have enough detail to ensure that cost elements are neither omitted nor double-counted, and all cost-influencing assumptions are detailed in the estimate's documentation, among other things.13 While Bureau officials were able to provide us with several documents that included projections and assumptions that were used in the cost estimate, we found the estimate to be partially comprehensive because it is unclear if all life-cycle costs are included in the estimate or if the cost estimate completely defines the program.

Credible cost estimates clearly identify limitations due to uncertainty or bias surrounding the data or assumptions, according to best practices. We found the estimate minimally met best practices for this characteristic in part because the Bureau carried out its risk and uncertainty analysis only for about $4.6 billion (37 percent) of the $12.5 billion total estimated life-cycle cost, excluding, for example, consideration of uncertainty over what the decennial census's estimated part will be of the total cost of CEDCaP.

Accurate estimates are unbiased and contain few mathematical mistakes. We found the estimate partially met best practices for this characteristic, in part because we could not independently verify the calculations the Bureau used within its cost model, which the Bureau did not have documented or explained outside its limited access cost estimation software.

Finally, the Bureau's cost-estimate was not well-documented. Improving cost estimation practices will increase the reliability of the Bureau's cost estimate, which will, among other things, help improve decision making, budget formulation, progress measurement, and accountability for results.

The Bureau's cost estimate had other shortcomings as well. For example, in 2016 we found that the Bureau's cost estimation team did not record how and why it changed assumptions that were provided to it, and the Bureau lacked written guidance and procedures for the cost estimation team to follow.14 Moreover, key risks were not accounted for in the cost estimate although this is an important best practice.


[1] GAO, 2020 Census: Additional Actions Would Help the Bureau Realize Potential Administrative Records Cost Savings, GAO-16-48 (Washington, D.C.: Oct. 20, 2015).

[2] GAO, Decennial Census: Progress Report on Preparations for 2020, GAO-17-238T (Washington, D.C.: Nov. 16, 2016).

[3] According to the Bureau, it needs to collect a number of predefined specific combinations of data elements during field interviews in order to consider the interview complete.

[5] As part of a new entry into the February 2015 update to our High-Risk Series focused on improving the management of IT acquisitions and operations, CEDCaP was identified as an IT investment—among others across the federal government—in need of the most attention.

[7] GAO, Information Technology: Uncertainty Remains about the Bureau's Readiness for a Key Decennial Census Test, GAO-17-221T (Washington, D.C.: Nov. 16, 2016).

[8] GAO, 2020 Census: Prioritized Information Technology Research and Testing Is Needed for Census Design Decisions, GAO-14-389 (Washington, D.C.: Apr. 3, 2014).

[10] GAO, 2020 Census: Key Information Technology Decisions Must Be Made Soon, GAO-16-205T (Washington, D.C.: Nov. 3, 2015).

[11] Phishing is a digital form of social engineering that uses authentic-looking, but fake, e-mails, websites, or instant messages to get users to download malware, open malicious attachments, or open links that direct them to a website that requests information or executes malicious code.

[13] GAO Cost Estimating and Assessment Guide: Best Practices for Developing and Managing Capital Program Costs (Supersedes GAO-07-1134SP), GAO-09-3SP (Washington, D.C.: Mar. 2, 2009).

[14] GAO-16-628.

2020 Decennial Census

The Bureau plans to implement several new innovations in its design of the 2020 Census. In response to our recommendations regarding past decennial efforts and other assessments, the Bureau has fundamentally reexamined its approach for conducting the 2020 Census. Its plan for 2020 includes four broad innovation areas that it believes will save it over $5 billion (2020 constant dollars) when compared to what it estimates conducting the census with traditional methods would cost. The Bureau's innovations include (1) using the Internet as a self-response option, which the Bureau has never done on a large scale before; (2) verifying most addresses using “in-office” procedures and on-screen imagery rather than street-by-street field canvassing; (3) re-engineering data collection methods such as by relying on an automated case management system; and (4) in certain instances, replacing enumerator collection of data with administrative records (information already provided to federal and state governments as they administer other programs). These innovations show promise for a more cost-effective head count. However, they also introduce new risks, in part, because they include new procedures and technology that have not been used extensively in earlier decennials, if at all.

The Bureau is also managing the acquisition and development of new and modified IT systems, which add complexity to the design of the census. To help control census costs, the Bureau plans to significantly change the methods and technology it uses to count the population, such as offering an option for households to respond to the survey via the Internet or phone, providing mobile devices for field enumerators to collect survey data from households, and automating the management of field operations. This redesign relies on acquiring and developing many new and modified IT systems, which could add complexity to the design.

These cost risks, new innovations, and the acquisition and development of IT systems for the 2020 Census, along with other challenges we have identified in recent years, raise serious concerns about the Bureau's ability to conduct a cost-effective enumeration. Based on these concerns, we have concluded that the 2020 Census is a high-risk area and have added it to the High-Risk List in 2017.

To help the Bureau mitigate the risks associated with its fundamentally new and complex innovations for the 2020 Census, the commitment of top leadership is needed to ensure the Bureau's management, culture, and business practices align with a cost-effective enumeration. For example, the Bureau needs to continue strategic workforce planning efforts to ensure it has the skills and competencies needed to support planning and executing the census. It must also rigorously test individual census-taking activities to provide information on their feasibility and performance, their potential for achieving desired results, and the extent to which they are able to function together under full operational conditions. 1 We have recommended that the Bureau also ensure that its scheduling adheres to leading practices and be able to support a quantitative schedule risk assessment, such as by having all activities associated with the levels of resources and effort needed to complete them. The Bureau has stated that it has begun maturing project schedules to ensure that the logical relationships are in place and plans to conduct a quantitative risk assessment. We will continue to monitor the Bureau's efforts.

The Bureau must also improve its ability to manage, develop, and secure its IT systems. For example, the Bureau needs to prioritize its IT decisions and determine what information it needs in order to make those decisions. In addition, the Bureau needs to make key IT decisions for the 2020 Census in order to ensure they have enough time to have the production systems in place to support the end-to-end system test. To this end, we recommended the Bureau ensure that the methodologies for answering the Internet response rate and IT infrastructure research questions are determined and documented in time to inform key design decisions.2 Further, given the numerous and critical dependencies between the Census Enterprise Data Collection and Processing (CEDCaP) program—a large and complex modernization program within the IT Directorate—and 2020 Census programs, their parallel implementation tracks, and the 2020 Census's immovable deadline, we recommended that the Bureau establish a comprehensive and integrated list of all interdependent risks facing the two programs, and clearly identify roles and responsibilities for managing this list.3 The Bureau stated that it plans to take actions to address our recommendations.

It is also critical for the Bureau to have better oversight and control over its cost estimation process and we have recommended that the Bureau ensure its cost estimate is consistent with our leading practices.4 For example, the Bureau will need to, among other practices, document all cost-influencing assumptions; describe estimating methodologies used for each cost element; ensure that variances between planned and actual cost are documented, explained, and reviewed; and include a comprehensive sensitivity analysis, so that it can better estimate costs. We also recommended that the Bureau implement and institutionalize processes or methods for ensuring control over how risk and uncertainty are accounted for and communicated within its cost estimation process. The Bureau agreed with our recommendations, and we are currently conducting a follow-up audit of the Bureau's most recent cost estimate and will determine whether the Bureau has implemented them.

Sustained congressional oversight will be essential as well. In 2015 and 2016, congressional committees held five hearings focusing on the progress of the Bureau's preparations for the decennial. Going forward, active oversight will be needed to ensure these efforts stay on track, the Bureau has needed resources, and Bureau officials are held accountable for implementing the enumeration as planned.

We will continue monitoring the Bureau's efforts to conduct a cost-effective enumeration. To this end, we have ongoing work focusing on such topics as the Bureau's updated life-cycle cost estimate and the readiness of IT systems for the 2018 End-to-End Test, which is essentially a dress rehearsal for the decennial.


[1] GAO, 2020 Census: Additional Actions Could Strengthen Data Collection Efforts, GAO-17-191 (Washington, D.C.: Jan. 26, 2017).

[2] GAO, 2020 Census: Key Challenges Need to Be Addressed to Successfully Enable Internet Response, GAO-15-225 (Washington, D.C.: Feb. 5, 2015).

[3] GAO, Information Technology: Better Management of Interdependencies between Programs Supporting 2020 Census Is Needed, GAO-16-623 (Washington, D.C.: Aug. 9, 2016).

[4] GAO, 2020 Census: Census Bureau Needs to Improve Its Life-Cycle Cost Estimating Process, GAO-16-628 (Washington, D.C.: June 30, 2016).

Looking for our recommendations? Click on any report to find each associated recommendation and its current implementation status.