The U.S. Census Bureau stopped collecting online responses to this year’s census at 11:59 p.m. Hawaii Time Thursday. Mail-in responses must be postmarked no later than Thursday, Oct. 15 and received no later than Oct. 22 to count.
The bureau announced the adjusted timeline soon after an unsigned Supreme Court brief released Tuesday overrode a lower court order that extended the count’s schedule to Oct. 31. Sonia Sotomayor was the lone dissenting justice.
Between the coronavirus pandemic, concerns around the bureau’s attempt in 2019 to add a question asking for citizenship, and the shifts in timeline — a months-long delay to start, then a deadline extended and shortened multiple times prior to Tuesday’s ruling — it has not been a normal census in any sense.
Census Bureau data indicated that as of Thursday, 99.9% of Colorado households had been counted, whether from residents self-responding or from non-response follow-up operations.
While Colorado looks to be doing alright, state demographer Elizabeth Garner noted there have been questions on the validity of the bureau’s data, given how politicized its operations have become. Statistical experts have posited that this year’s census may end up as one of the worst counts in the bureau’s history.
“Nobody can say how accurate the data is right now, because we have no idea what the actual count is,” Garner said.
Counting a household does not always mean that a census worker logged residents as living there. Some addresses listed in the census’ system don’t have anyone living there, whether because they are second homes, or because that address is for some reason not habitable. Once a census worker determines if an address is vacant, that address is considered counted.
Garner gave the example of an apartment complex which was slated for demolition, and where residents were moving out as late as September — Census Day is April 1 — but the property manager did not keep records on former residents. The address was marked as resolved, even though no actual residents had been counted there and residents may still have lived there on Census Day.
Data on the combined percent counted, from both self-response and non-response follow-ups, is not yet available on a scale more detailed than the state average. However, some counties in Colorado have reported relatively high self-response rates, meaning that census field workers had less to do in those areas to fill the gap. As of Thursday, Douglas County was in the lead with 81.4% of its residents self-responding, better than its final 2010 self-response rate of 77.3%. Other localities have declared a complete count, such as the Ute Mountain Ute Tribe reservation, which overlaps with the southwest corner of the state.
The decennial census gives the proportions for how federal funds are allocated to dozens of programs, many of which serve low-income communities, older residents, and young children. The people that have not been counted, and likely will remain uncounted given the count’s abrupt end, are often in one or more of these categories.
There have been some silver linings to this year’s count. For one, it’s the first time that the Census Bureau has offered an online form, in addition to the usual mail and phone options. Garner said that roughly 60% of the state responded to the count through an online form, with just 9% responding by mail or phone. The rest of responses were captured through follow-up operations, such as going door to door.
Additionally, this is the first year where the bureau has offered data on non-response follow-up as they progress, which offers a better idea on how close the census is to completion than the self-response rates do.
“At least they gave us something,” Garner said, though she lamented that state-level non-response follow-ups data was still not as much detail as experts ideally need to understand the count’s progress in real time.
Also of note: While this year’s wildfires have been unprecedented in scale and intensity, since Census Day is on April 1, Garner said it’s likely that most of the related evacuations in Colorado — which have primarily occurred in the second half of the year — have not affected census counts of those communities. The same may not be true for California’s wildfire evacuations; there are also concerns about undercounting hurricane-displaced residents in some of the southern states, especially Louisiana and Mississippi.
The Census Bureau has until Dec. 31 to clean up the data, and experts, including Garner, question whether less than two months is enough time to do so. This process includes doing a statistical analysis called imputation, as well as consolidating any duplicate responses and using administrative records to fill in gaps when absolutely necessary. The bureau will also double check its work with post-enumeration surveys, which go to certain neighborhoods to assess the accuracy — or inaccuracy — of the final count.
Correction: this article has been corrected to clarify the nature of the census. It is more thorough than a survey, which implies a sampling rather than counting every of a whole.