To estimate the spread of COVID-19 inside immigration detention facilities, Vera modeled ICE’s patterns of booking people into custody (“book-ins”), transferring them (“transfers”), and releasing people from custody to the community or through deportation (“book-outs”) within its network of hundreds of detention facilities and local jail beds.^[]ICE reports 137 dedicated detention facilities on its website, but relies on a network of several hundred facilities where it may detain small numbers of people. See https://perma.cc/LF5L-ZQQH. Vera produced these estimates based on ICE detention data from fiscal year 2016, which is the most recent full fiscal year of individual-level data available to the public.^[]This dataset was provided to Vera by David Hausman of Stanford University’s Immigration Policy Lab and was originally obtained through a Freedom of Information Act (FOIA) request. Vera used this data to calculate average daily populations and daily inter-facility transfer probabilities. The researchers then scaled down the number of people booked in to custody to provisionally calibrate to March 2020 ICE-reported initial book-ins and average daily population (ADP), leading to a decrease in overall facility populations throughout the 60-day simulation period. The model Vera designed is an extended version of a susceptible-exposed-infected-removed (SEIR) epidemiological model, adapted from a paper by Eric Lofgren et al.^[]Vera’s model was adapted from Eric Lofgren, Kristian Lum, Aaron Horowitz, Brooke Madubuonwu, Kellen Myers, and Nina H. Fefferman, “The Epidemiological Implications of Incarceration Dynamics in Jails for Community, Corrections Officer, and Incarcerated Population Risks from COVID-19,” https://perma.cc/U9FG-PHH7. Given that available evidence suggests that the majority of people detained by ICE are 45 years old or younger, Vera used the “jail, low-risk adult” component of Lofgren’s model, which most closely aligns with the population demographics of people in ICE detention centers.^[]Based on data that Freedom for Immigrants collected from 5,823 people detained by ICE, Vera determined that roughly 80 percent of people in their sample were 45 years old or younger. See Freedom for Immigrants, “Detention by the Numbers,” accessed May 29, 2020, https://perma.cc/D74D-JU4J. The Transactional Records Access Clearinghouse found a median age of 30 among people deported by ICE in fiscal year 2012 and fiscal year 2013. See Transactional Records Access Clearinghouse, “ICE Deportations: Gender, Age, and Country of Citizenship,” April 9, 2014, https://perma.cc/2UF2-RH5N. Details on how Vera implemented the model, and the derivations and values of the parameters the researchers used, are included in the technical appendix.

Vera’s model presents a simulation covering a 60-day period following a “day 0,” on which there are zero COVID-19 cases among people in detention facilities. In this model, people booked-in to ICE detention facilities are set at a constant probability of having been exposed to COVID-19. Once an infected person is inside an ICE facility, the virus can spread to other people detained in that facility and to people in any other facility that ICE transfers them to.

This simulation is a simplified depiction of how COVID-19 may actually spread. Vera’s model accounts for spread only between people detained and does not simulate possible transmission of the virus between facility staff and detained people or between people detained by ICE and people detained by other agencies at the same facilities. Vera’s model tracks the progression of individual people’s conditions while they remain detained and does not account for risk or the progression of people’s conditions once they are no longer in custody (e.g., due to deportation or release). In these respects, this approach is a conservative one, yet the results still show a substantial likely underreporting of COVID-19 cases by ICE, the need for more testing, and a substantial risk to people in ICE custody.

Vera’s model is based on ICE detention data from fiscal year 2016, provisionally calibrated to March 2020 ICE initial book-ins and average daily population (ADP) to simulate a trajectory of ICE detentions and COVID-19 infections over a 60-day period. A second iteration of the model will extend the calibration beyond the March 2020 ICE book-ins and detention. This update will allow Vera to improve the accuracy of predictions and credibly extend the simulation past the original 60-day mark.

Vera’s model differs from the only other published model of COVID-19 infection in detention, recently released by Michael Irvine and colleagues in the Journal of Urban Health, in that Vera’s model produces a lower total infected share of the cumulatively detained population.^[]For the only other published model of COVID-19 infection in ICE detention, see Michael Irvine, Daniel Coombs, Julianne Skarha, Brandon del Pozo, Josiah Rich, Faye Taxman, and Traci C. Green, “Modeling COVID-19 and Its Impacts on U.S. Immigration and Customs Enforcement (ICE) Detention Facilities, 2020,” Journal of Urban Health (2020), https://perma.cc/7YH3-7447. This difference is due to how ICE’s movement of people is modeled and how the outbreak starts. Irvine et al. model static ICE facility populations that are subject to synchronized outbreaks across detention facilities. In contrast, Vera assumes a dynamic ICE facility population that reflects ICE continuing to book in people who may have COVID-19, accounts for ICE transferring people between facilities, and allows facilities to experience their first infections at randomly distributed times.

Vera’s model uses a transmission rate that is comparable to the one used in the Irvine et al. “pessimistic scenario.” In this comparable scenario, Irvine et al. estimate that 99 percent of their static populations of detained people would be infected by day 60. However, because Vera modeled dynamic—rather than static—ICE facility populations, the researchers arrived at different results: Vera’s simulations project a lower aggregate rise in cumulative infections compared to any of the Irvine et al. scenarios. Vera estimates that 19 percent of people detained at any point during a 60-day period would be infected by day 60.^[]Irvine et al. estimate in their optimistic scenario that 36 percent of detained people will be infected within 60 days and 72 percent will be infected by day 90. Readers may be interested in FiveThirtyEight’s explanation of why COVID-19 models often arrive at different results. See Maggie Koerth, Laura Bronner, and Jasmine Mithani, “Why It‘s So Freaking Hard to Make a Good COVID-19 Model,” FiveThirtyEight, March 31, 2020, https://perma.cc/D7Z7-A4R9. In addition, the Vera model predicts that the COVID-19 outbreak would not have peaked by day 60. Though these two models differ, they both require a certain amount of guesswork given that ICE shares little actual data with the public, and they each provide a useful starting point for understanding what is surely an underreporting of infection on ICE’s part.