“It’s the perfect storm,” said Ditas Katague, director of the California Complete Count — Census 2020 Office, an organization working to ensure a complete count of the hardest-to-count Californians. “And if you add the shortening of the time, it adds greater potential for inaccuracy.”
Four of the Southeastern states in the path of September’s storm systems — Alabama, Louisiana, Mississippi and South Carolina — also happen to be among the five states that, as of the weekend, had counted fewer than 91 percent of households, the lowest response rates in the nation, according to an analysis by the data science company Civis Analytics and the nonprofit group National Conference on Citizenship (NCoC).
In California, Oregon and Washington state, where enumerators still need to complete the final couple of percentage points of households, fires have ravaged the landscape, directly affecting tens of thousands of people and engulfing most of the states with unhealthy air.
The bureau aims to enumerate at least 99 percent of households; as of Monday, it had counted 95.4 percent, with just over a week to go before the count is set to end. In the states affected by storms, which also include Texas, nearly 2.7 million households remained uncounted as of the weekend. As of Sunday night, about 6.39 million residents were estimated to be affected by Tropical Depression Sally, and on Monday, 13.9 million were estimated to be affected by Tropical Storm Beta, according to an emergency management map maintained by the Census Bureau.
Data from the decennial census is used to determine congressional apportionment, redistricting, and $1.5 trillion each year in federal funding, which includes money to plan for and respond to natural disasters.
Census experts say it has been hard to determine how well the count has been going because the bureau has not shared enough details.
“We’ve asked for information about what does that mean that you’ve completed enumeration” of households left to be counted, Katague said. “What does it mean to close out a case and move on? We recently heard that they were going to places that they classify as easier to count, which means they left those truly harder-to-count places till the end.
“I was kind of shocked to hear that, because [typically] you start with the hardest first, so you can go back there six times. . . . What’s going on now is abnormal. It’s very different from what was going on in 2000 and 2010.”
A Census Bureau spokesperson denied the allegation that enumerators were going to easier places first, adding that at this point all households still uncounted are “among the hardest of the hard to count.”
NCoC senior fellow Denice Ross said she wished the bureau would release information at the county level rather than just the state level, and share “paradata” — information it is collecting on how long each enumerator spends inputting data into their handheld devices, which can help indicate whether the information is being gleaned through asking each person interviewed or by other means such as visual identification.
“The type of transparency that we need would really be unprecedented, but we are in an unprecedented crisis right now,” she said, adding that people don’t need to be displaced from their homes for the count to be affected. “The impact isn’t necessarily that there are flames licking at your feet; it’s the air quality, and then because of the pandemic, the census takers can’t go inside, and so we’re asking people to step outside into terrible air quality and we’re asking enumerators to go around in terrible air quality.”
The census spokesperson said the bureau does not publish paradata during the count because doing so would compromise its ability to evaluate the job performance of individual enumerators.
A final 2020 Census that is considered inaccurate could affect response to future natural disasters, Ross said. “If people don’t believe in this count . . . then the entire country will be flying blind without the benefits of federal statistics,” she said. “You won’t know the right number of buses to help people evacuate from a hurricane or a fire, you won’t know how many seats you need, or how many beds do you need to set up at a shelter.”
The Southeastern states hit by storms have not only some of the country’s lowest response rates but also some of the highest levels of coronavirus cases per capita, which could also make householders more reluctant to open the door to enumerators. All but Florida and Louisiana reported more than 100 coronavirus cases per 100,000 residents in the past week; Florida’s rate was 87, and Louisiana’s was 83. States with more than 100 cases per 100,000 residents are considered “hot spots,” according to the analysis.
The natural disasters have struck at an already uncertain time for the census. The pandemic, which hit just as survey forms were appearing in people’s mailboxes, delayed operations originally scheduled to end in July. Respondents are supposed to fill out the survey based on their whereabouts on April 1, but gathering accurate data becomes harder the more time passes after that date. A Pew Research Center survey in early June found that 3 percent of people in the country said they had been permanently displaced by the pandemic and that 6 percent said someone had moved into their household because of it.
Several lawsuits are challenging the Trump administration’s sudden announcement that it would end the count on Sept. 30 rather than Oct. 31 as previously planned. A California judge this month temporarily stopped the bureau from winding down operations, and it is unclear whether the count will continue past the end of September; a hearing in the California case is scheduled Tuesday, and a three-judge panel in Maryland is considering a request for a similar injunction this week.
When the pandemic hit, the administration originally requested approval from Congress to move the deadline to report census data for House apportionment from Dec. 31 to April 30.
But on Aug. 3, the administration reversed course, saying the count would end Sept. 30 to deliver the data by the end of the year, while President Trump is still in office.
In testimony in the Maryland lawsuit on Sept. 11, Albert E. Fontenot Jr., the bureau’s associate director for decennial census programs, acknowledged that the bureau is “facing significant risks to complete all states by [Sept. 30], due to factors beyond the Census Bureau’s control, such as wildfires in the western part of our country, major storms, resurgence of COVID-19 restrictions and other similar disruptions.”
In a call Thursday with the federal government’s Census Scientific Advisory Committee (CSAC), which includes statisticians, demographers and others, Fontenot said that the bureau is setting up mobile outposts and that enumerators are contacting households by phone, but he also acknowledged that natural disasters may prevent the bureau from reaching 99 percent in all states by the end of the month.
Inaccuracies in the count can stem not only from the shortened data collection time but also from a truncated timeline for quality checks and analysis. The government’s accelerated “replan,” announced in August, reduced the post-count operations from six months to three and eliminated reviews with state demographers designed to help identify people who were either missed or counted more than once.
On Friday, the CSAC unanimously recommended that the bureau keep the count going through October and publish daily response rates at the census-tract level.
“When the weather isn’t right, we postpone the launching of rockets into space,” the committee wrote. “The same should be true of the decennial enumeration.”
Internal documents released this weekend as part of a lawsuit show Census Bureau career officials warning that the count would be compromised under a shortened schedule. And the Commerce Department’s inspector general released a report Monday finding that the decision to accelerate the schedule was not made by the Census Bureau and that the truncated schedule “increases the risks” to obtaining a complete and accurate count.
The Census Bureau did not answer questions about what post-enumeration operations were being eliminated or changed under the accelerated replan, and a publicly released description of the plan did not include details. But a bureau spokesperson said Sunday that at least some reviews that had previously been canceled to meet “statutory completion deadlines” would now be “discussed” this week. The bureau did not say whether such discussions meant that the checks would be reinstated.
Civil rights groups have accused the administration of trying to skew the count in ways that would benefit Republicans and non-Hispanic whites, through actions such as trying to add a citizenship question to the survey or seeking to exclude undocumented immigrants from apportionment, both of which sparked legal challenges. The administration has added an unprecedented number of political appointees to senior bureau staff this year.
But inaccuracies in the count could be politically beneficial or harmful to both Democrats and Republicans. An analysis released Thursday by the American Statistical Association’s Office of Science Policy found that under a Sept. 30 deadline, California, Idaho or Ohio could gain House seats they may not have gained, while Florida or Montana could lose seats they may not have lost, if the count continued until Oct. 31.
The ASA analysis suggests that Arizona, Florida, Georgia, North Carolina and Texas could collectively forfeit as much as $500 million in federal Medicaid funding each year if the count ends on the earlier date. And it estimated that 95.82 percent of the population will have been counted if the count ends Sept. 30, vs. 99.19 percent if it ends Oct. 31.
The Civis analysis estimated that if the count ends Sept. 30, nine states and the District of Columbia will not reach 99 percent, including Oregon and four states in the Southeastern hurricane area.
Testifying during the summer before the House Oversight Committee, former Census Bureau director Robert Groves urged the bureau to release more information, including “process indicators” such as initial response rates, proxy reporting rates, and percentages of people with usable data; comparisons with population estimates from demographic analysis; and initial field data from the post-enumeration survey.
“Credibility requires transparency,” he said. “The sooner the country can see multiple indicators of 2020 Census quality, the sooner the use case for the census can be made.”