Census IT costs close in on $5 billion

Census documents reveal the rising costs of the 2020 census -- and indicate that existing tech can be leveraged to obtain accurate citizenship data without adding a new question to the population survey.

Census 2020 By Maria Dryfhout Stock photo ID: 790714156
 

Technology costs for the 2020 population count will reach almost $5 billion, according to an official Census Bureau estimate released June 8 as part of document production for an ongoing lawsuit.

The estimate covers a 12-year period from 2012, when planning for the 2020 census launched, to 2023, when the data releases from the enumeration are expected to be concluded. The report, included in a tranche of more than 1,300 pages of documents made public in a legal battle over the controversial citizenship question set to be included in the 2020 population survey, puts the total IT spend at $4.97 billion.

The documents include letters to and from Commerce Secretary Wilbur Ross as well as several official-use-only reports that shed new light on both the IT costs of the 2020 population count and the ability to use existing data and technology to obtain block-level counts of citizen and non-citizen U.S. residents.

The documents suggest that the bureau has the technological resources to provide a solid estimate of the size of the pool of non-citizens residing in the United States without putting the question to respondents on the main survey. Such an estimate would draw on existing administrative data, which the bureau has the right to use under inter-agency agreements, at a cost of less than $2 million.

That same report, from John Aboud, the bureau's chief scientist and associate director for research and methodology, indicates that adding the citizenship question to the main survey would increase follow-up response costs by at least $27.5 million.

The IT effort includes a new system of systems called Census Enterprise Data Collection and Processing (CEDCaP) to support the 2020 enumeration as well as future bureau efforts. CEDCaP also includes an internet self-response option, which allows respondents for the first time ever to complete their surveys online.

The readiness of CEDCaP remains an issue, with watchdogs including the Government Accountability Office worried that the program is behind schedule and missing key authority-to-operate certifications.

A risk assessment notes that in February 2018, the Census Bureau obtained a waiver from the federal CIO to run the self-response program in the cloud, outside the "trusted internet connection" policy that usually applies to public-facing government websites. That move is necessary, according to Census documents, because of latency issues arising from communication between far-flung survey respondents and the two bureau sites with trusted internet connections – one at headquarters in Suitland, Md., and the other at a Bowie, Md., data center. Under the bureau's plan, if there's a surge in self-response activity, some traffic will communicate directly with a commercial cloud provider – as yet unnamed. Bureau documents also note that plans for a cloud-based architecture are not yet finalized and still at risk not being finished before the start of the 2020 enumeration.

Other big IT charges include IT support and services for field offices and the decennial device-as-a-service contract to give mobile devices and software to canvassers walking door-to-door to conduct survey follow-up. The largest single contract is to integrator T-Rex Solutions, with an estimated spend of just under $1.3 billion.

Census reports also indicate that cybersecurity remains a top-three risk for the 2020 enumeration, along with denial-of-service attacks and infiltration of bureau systems. The bureau is using "red teams" from the Department of Homeland Security and industry to conduct "slow and under the radar" cyberattacks on census systems to identify vulnerabilities, the documents show. Additionally, the self-response system underwent security tests by contractors in January and by DHS in February.

Overall, the projected cost of the 2020 census is $15.6 billion, which is more than $3 billion over 2012 estimates that the population count could be done for $12 billion. Still, the bureau is promising some cost savings over the 2010 enumeration, which included a major technology program failure involving mobile devices. The estimate projects the 2020 population count will cost $107 per housing unit, down from $120 per unit in 2010. The biggest spending period is just on the horizon, according to bureau documents, with $10.8 billion to be spent in fiscal years 2019 and 2020.