The accurate count of the populations of American Indian reservations is known to be difficult and subject to net undercount error. When a census is conducted using a new response tool (the Internet) while subject to the turmoil of a pandemic and political interference, one wonders how accurate the census count is. As a means of exploring the effect of these issues, I compare the 2020 Census results for the Hopi Reservation, located in Arizona to the results for the Lummi Reservation, located here in Whatcom County. A motivation in comparing these reservations is that in addition to differences in Internet access and pandemic containment strategies, there are other differences between the two reservations that could affect census results, including isolation and the presence of street addressing systems.
New Tools and Turmoil in the 2020 Census
Starting in 1970, the decennial census relied upon self-response via a “mail-out/mail-back” initial contact. In 2020, the mail-out/mail-back means of initial self-response changed in that a new self-response tool was introduced - the internet. As in the case of the earlier post-1960 census counts, enumerators were sent to housing units that did not respond via the internet. Among their many tasks, the enumerators determine whether the unit is real and that it was a dwelling unit on April 1, then determine whether it is occupied or vacant as of April 1st, 2020, and ultimately, determine the number of occupants, if any, and their characteristics.
In addition to implementing a new internet self-response tool, the Covid-19 pandemic caused a number of adjustments that affected Census Bureau’s operational schedule and delayed the release of data. Importantly, the Non-Response Follow-up period was shortened to a three month period by the Trump Administration, October 1st through December 31st. Dr. Howard Hogan, now retired and a former Associate Director of the U.S. Census Bureau, observes that the Census Bureau itself has implicitly admitted that the shortened schedule is insufficient to conduct the Census with its usual statistical standards.
The Trump Administration also attempted to add a “citizenship” question late in the timeline chain leading to the actual census, which created an unnecessary diversion of staff focus and attention because the citizenship question was already in the American Community Survey. Add to these issues, the efforts by the Census Bureau to implement a second new tool in the form of a Disclosure Avoidance System called “Differential Privacy” created consternation among researchers and other stakeholders. These concerns were crystalized in the form of a lawsuit initiated by Alabama (and joined by 16 states) that attempted to stop the use of Differential Privacy that required the attention of the Census Bureau’s senior staff, which again led to a diversion of staff focus and attention.
Comparing the Hopi and Lummi Reservations
What effects, if any, did these factors have on the 2020 census results for the Hopi Reservation in Arizona and the Lummi Reservation in Washington? In setting the stage for a possible answer to this question, I start with the total 2020 population counts for the Hopi Reservation and the Lummi Reservation, which are, respectively 6,377, and 5,748.
Continuing to set the stage for a possible answer to the question of what effect these issues had on the 2020 census count of the Hopi Reservation and the Lummi Reservation, let us now look at the population counts for the Hopi Reservation and the Lummi Reservation as found in the immediately preceding American Community Surveys (ACS). The 2015-19 ACS shows a total population of: 5,641 (with a 90% Margin of Error of +-359) for the Lummi Reservation; and 9,222 (with a 90% Margin of Error of +-645) for the Hopi Reservation and Off Reservation Trust Land.
For the Lummi Reservation, the total population shown in the 2020 census (5,748) is consistent with the preceding ACS total population: The 2020 Census is 107 persons higher than the 2015-19 ACS total population number of 5,641, which is well within the 90% Margin of Error of the 2015-19 ACS of +-359. However, this is not the case for the Hopi Reservation, where the 2020 Census shows a total population (6,377) that is 2,845 less than the 2105-19 ACS total population number of 9,222 and well below the lower 90% Margin of Error boundary of 8,577 (where 8,577 = 9,222 - 645) for the ACS total population number. Is this an indication of an undercount on the Hopi Reservation?
To assist in answering this question, let us first explore Internet subscriptions, which are key to the Census Bureau’s attempt to use the Internet as the initial method of collecting information for the 2020 census. As a benchmark for the Hopi and Lummi reservations, 89% of all households in the United States are estimated by the Census Bureau to have an internet subscription. Compare this benchmark to the Hopi Reservation where only 30% are estimated to have an internet subscription; whereas 85% are estimated to have an internet subscription on the Lummi Reservation
With only 30 percent of its households having an internet subscription, how does the Hopi 2020 Census response rate on the Hopi Reservation compare to the response rate on the Lummi Reservation, with its internet subscription rate of 85%?
The U.S. Census Bureau shows that as of January 28th, 2021 the total self-response rate was:
(1) 67% for the U.S. as a whole and that 53.5% of households for the U.S. as a whole responded by the internet;
(2) 19% on the Hopi Reservation and that only 3.9% of the households on the reservation responded by the internet; and
(3) 57.7% on the Lummi Reservation and that 47.9% of the households on the reservation responded by the internet.
Adding to the lack of internet subscriptions and the shortened census 2020 follow-up schedule, the Hopi Tribal Council issued a stay-at-home order on March 23rd, 2020 in response to increasing cases of COVID-19 in surrounding areas. This order closed all except essential businesses and directed residents to stay home except for the conduct of essential activities such as obtaining food or medical care. It was not until September 11th, 2020 that the Hopi Tribal Council decided to reopen government facilities and to allow phased and restricted reopening of other businesses as well as a phased return to work subject to mandatory mask usage and a nightly curfew. This order became effective on September 30th, 2020.
The Lummi Tribal Council did not institute a lock-down order. Instead, it quickly introduced mitigation and prevention measures such as social distancing, drive-through testing, telemedicine clinics, and a home delivery service for the elderly. Importantly, with its roads, Internet access, close proximity to Bellingham and other towns in Whatcom County, as well as to Interstate 5, the Lummi Reservation is not even close to being as isolated as the Hopi Reservation.
With the combination of a low level of internet access, the Covid-19 pandemic, and a compressed data collection schedule, the 2020 Census of the Hopi Reservation ran into the “Perfect Storm.” The low level of internet access meant that many on the reservation could not respond to the initial census inquiry, the closure of the Reservation made it very difficult if not impossible for census enumerators to conduct non-response follow-up during the summer of 2020 and the compressed data collection schedule led to a non-response follow-up that was only three months in duration, October 1st to December 31st, 2021.
Add to these problems the fact that the non-response follow-up task is very difficult on the Hopi Reservation, which, not only is far larger than the Lummi Reservation, but also, unlike the Lummi Reservation, has a substantial number of residences that lack traditional mailing addresses, many of which are in remote locations. The Indian Legal Clinic reports that one-third of Coconino County (where a substantial portion of the Hopi Reservation is located) voters do not have situs descriptions and the on-reservation voters lack street addresses.
A Substantial Undercount of the Hopi Reservation?
It is likely that the counts of both the Hopi and Lummi Reservations were affected by the general challenges known to affect the census on reservations as well as the use of a new response tool (the Internet) and the turmoil of a pandemic and political interference. However, what is the additional effect of conducting an internet census for a community that has a low rate of internet subscriptions, was closed during the pandemic, and subject to both a shortened non-response follow-up period and the lack of a street addressing system? It is, perhaps, not surprising that the 2020 Census total population of 6,377 for the Hopi Reservation is not only 31% less than the 2105-19 ACS total population number of 9,222 but also far below the lower 90% Margin of Error boundary of 8,577 for the ACS total population number. Compare this to the Lummi Reservation, where the 2020 Census is only 107 persons higher than the 2015-19 ACS total population number of 5,641, which is well within the 90% Margin of Error of the 2015-19 ACS of +-359.
These results strongly suggest that it is very highly likely that the Hopi Reservation was substantially undercounted in the 2020 Census. In fact, it could be reasonably argued that the total population number found in the 2015-19 ACS (9,222) for the Hopi Reservation is closer to the true but unknown 2020 total population number than is the 2020 Census total population number of 6,377. Unlike the 2020 Census, the 2015-19 ACS data for the Hopi Reservation were collected under circumstances that allowed for a full period of non-response follow-up and not subject to the errors introduced by “Differential Privacy.” Moreover, the 2015-19 ACS is based on data and methods with which the Census Bureau has a great deal of experience, something that cannot be said in regard to the Bureau’s introduction of two new tools for the 2020 Census, the Internet Response System and Differential Privacy.
Being undercounted in a decennial census is not a trivial matter. A great deal of funding is distributed to areas on the basis of decennial census counts. And once the census is done, the count is in effect for the full ten years until the next census is completed. One wonders how many other areas there are with a low level of internet access and how accurate their 2020 census counts were. Rural areas in Alaska would likely be on this list as well as rural areas in the Pacific Northwest and elsewhere in the country.