Watchdogs concerned as census dress rehearsal approaches
- By Chase Gunter
- Nov 16, 2016
With eight months remaining before the Census Bureau undertakes its dress rehearsal for the 2020 decennial headcount, watchdogs have expressed concerns about the enumeration's large-scale changes, untested IT systems and the recent decision to cancel several tests scheduled for next year.
For the 2020 main event, Census officials plan to employ technology at an unprecedented scale to cut costs considerably.
However, Rep. Mark Meadows (R-N.C.), chairman of the House Oversight and Government Reform Committee's Government Operations Subcommittee, said he has serious concerns about the schedule in the lead-up to the end-to-end test in 2018 -- the last large-scale test for all the new methodologies before the 2020 census.
Census Bureau Director John Thompson assured the committee that the bureau is "on track and on schedule" and testified that 74 percent of the 350 design decisions in the integrated schedule have been made.
However, Meadows cited a Government Accountability Office report that states only about half of the bureau's 50-plus IT systems have been tested to date.
"Any time you have that many moving parts, it can be troubling," Meadows said. "The fact that there is even a suggestion that IT products would go untested is unacceptable, and no system and product can be used to collect and protect the American public's sensitive information."
Census Bureau CIO Kevin Smith attempted to quell concerns about Americans' data security.
"My first and foremost importance is to make sure the systems are ultimately secure, the data's confidential...and the systems are available," he said.
Smith also said that although some 40 percent of the systems might not be delivered until after the end-to-end testing, which is scheduled to begin in August 2017, that does not mean the bureau cannot test them before the decennial headcount.
David Powner, GAO's director of IT management issues, agreed.
"If everything doesn't go smoothly with the end-to-end test, we still do have some time to test, not ideally, leading up to the actual decennial," said Powner, who added that, after the 2010 census, ensuring "transparency on delivery dates and costs" was a priority of GAO's.
Thompson said the biggest takeaways from the 2016 tests included the realization that enumerators need better training and Census officials need to tinker with the interview processes and address closeout procedures for data collection operations.
GAO reported that in the 2016 tests, not enough data was collected in 20 percent to 30 percent of cases, and they were classified as "non-interviews."
"One factor we observed that may have contributed to the non-interview rate was that enumerators did not seem to uniformly understand or follow procedures for completing interviews with proxy respondents," said Robert Goldenkoff, GAO's director of strategic issues.
GAO has agreed with Census officials that the expanded use of technology to conduct the headcount promises to significantly cut costs, though it has disputed the reliability of the bureau's $5 billion estimate. GAO also reported on issues associated with the mobile devices enumerators will use -- problems that did not arise during past paper-based tests. For instance, auditors said the mobile device prevented enumerators from logging visits to a housing unit more than once per day, reopening a closed case or exceeding the maximum of six contact attempts.
Thompson also addressed the bureau's decision to cancel the 2017 field tests and future budget uncertainty head-on and made a pitch for more congressional funding.
"The current House and Senate fiscal year 2017 appropriations marks from the spring of 2016 fund the program at 16 and 9 percent below the president's budget, respectively," he said. "This would be the fifth consecutive year that the program has received appropriations significantly below the request, and we are at a point where there is a significant cost to continuing to defer work."
Chase Gunter is a former FCW staff writer.