key: cord-0806685-e4q19kss
authors: Das Swain, V.; Xie, J.; Madan, M.; Sargolzaei, S.; Cai, J.; De Choudhury, M.; Abowd, G. D.; Steimle, L. N.; Prakash, B. A.
title: WiFi mobility models for COVID-19 enable less burdensome and more localized interventions for university campuses
date: 2021-03-24
journal: nan
DOI: 10.1101/2021.03.16.21253662
sha: 9571b0c66840ad93772b85fa1f0b03f479d52109
doc_id: 806685
cord_uid: e4q19kss

Infectious diseases, like COVID-19, pose serious challenges to university campuses, and they typically adopt closure as a non-pharmaceutical intervention to control spread early and ensure a gradual return to normalcy. These policies, like remote instruction (SQ), reduce potential contact but also have broad side-effects on campus by hampering local economy, students learning outcomes, and community wellbeing. In this paper, we demonstrate that university policymakers can mitigate these tradeoffs by leveraging anonymized data from their WiFi infrastructure to learn community mobility (WF) and in turn explore more granular policies like localized closures (LC). WF can construct contact networks that capture behavior in a variety of spaces, highlighting new potential transmission pathways and temporal variation in contact behavior. Additionally, WF enables us to design LC policies that close super-spreader locations on campus. On simulating disease spread with contact networks from WF, we find that LC maintains the same reduction in cumulative infections as SQ while showing greater reduction in peak infections and internal transmission. Moreover, LC reduces campus burden by closing fewer locations, forcing fewer students into completely online schedules, and requiring no additional isolation. WF can empower universities to conceive and assess a variety of closure policies to prevent future outbreaks.

WiMob mines these logs to characterize mobility as a bipartite graph that describes people (e.g., P 1, P 2) visiting locations (e.g., L1, L2) on campus during di↵erent times (e.g., t 1 , t 2 ). Since people's devices can proxy their presence, we estimate collocation (e.g., P 1 and P 2 were collocated at L1 at t 1 ), and movement (P 2 dwelled at L1 and then at L2). The collocation graph forms the basis of the contact structure for our ABM. WiMob captures contact behavior at a community scale for a variety of campus spaces, describes temporal variations in contact, and provides a better estimate of local context by being aware of occupancy and the non-student population. Leveraging WiMob also reveals that En overestimates the impact of RI on reducing contact on campus. Hence, we propose a less burdensome alternative to RI, by deriving more targeted LC policies based on WiMob ( Figure 1 ) (indeed En is too coarse-grained for designing targeted LC policies). We further exhibit that LC presents better disease control outcomes than RI by constructing and simulating an agent-based model (ABM) over the WiMob contact networks, calibrated with GT on-campus COVID-19 cases from the Fall semester of 2020 [25] and infection rates from Fulton County [40] . To compare the e↵ect of interventions, we describe a counterfactual semester that is unaltered by other policy-induced behaviors of 2020 by leveraging WiFi data from Fall 2019 to determine the contact structure of the simulation. En describes contact based on classes that students are expected to attend based on their enrollment. En assumes 90% of students to be connected in a single component, but WiMob reveals that on given week only 69% of the academic population is in the largest component (those that do not visit campus are isolated and shown in the circumference). In the first week of Fall 2019, both En and WiMob show high connectivity. However, En does not change, but WiMob reveals that density of connections changes over the semester. (b, c) The di↵erence in the two disease models is because of the di↵erence in structural characteristics of networks built with WiMob compared to En. (b) En depicts campus contacts to be connected closely into a "small world". WiMob shows that contacts evolve over time. As mobility captures interactions outside classrooms we observe that for the first 6 weeks the shortest transmission path between people is shorter than what is reported by En. (c) Enrolling into a course does not necessitate physically collocating with the class for extended periods (students can also choose to be entirely absent). WiMob reflects this behavior and highlights a decline in average contacts over time. (d) These structural di↵erences can help policymakers anticipate the e↵ect of closure policies by describing how it fragments the underlying contact network. En shows that remote instruction leads to a 94% reduction in contacts and 50% increase in transmission path length (similar to numbers reported in [59] , shown as En (Ext.)). However, the estimate is significantly lower with WiMob. As a result, WiMob emphasizes the limits of remote instruction policies and in turn motivates new policies that can be designed and evaluated with actual on-campus behavior.

between courses (via instructors) and organic interactions outside classes (e.g., waiting areas, dining, parties, and extra-curricular activities). Therefore, using En can overemphasize the disease-mitigating structural changes to the network by RI interventions. By contrast, WiMob is more grounded in community behavior as it captures multiple scheduled and serendipitous contact situations dynamically over the semester. We compare the features of contact networks constructed with WiMob, against networks constructed with En using data from GT for Fall semester of 2019 (August 19 -December 14), which is before any COVID-19 reported cases in the U.S. En approximates contact based on students enrolling for classes that could potentially collocate them in the same room during lectures. WiMob infers contact when any two individuals actually collocate near the same WiFi access point [13, 55] for extended period (see explanation in SI WiFi Mobility). We find that WiMob renders new insight into contact on campus that is invisible to the En methodology.

Das Swain, V. et al • 5

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  WiMob characterizes temporal variation in proximity Variation in contact over the semester would naturally impact the severity of disease spread. However, En describes a static network that does not capture such dynamics (Figure 2a) . Instead, we find that WiMob shows contacts get sparser as the semester progresses. Figure 2c presents a notable decline in contacts after the first two weeks, which coincides with multiple orientation seminars and the so-called "course shopping" period of Fall 2019. In fact, contact decreases considerably in classrooms, with a steeper slope possibly because of reduction in attendance. WiMob is able to reveal other observable changes, such as drop in contacts during exam period (week 15) and increase after fall recess (week 10). Since, En renders a highly connected static network, which can miscalculate the speed at which a disease spreads. By contrast, the longitudinal behavior represented by WiMob can help universities anticipate disease spread more accurately.

Campuses can assess risk of an outbreak by characterizing the number of individuals that would be at risk of infection through contact. In fact, En indicates that 99% of the individuals on campus are clustered in a single component -if any of them would have been infected in Fall 2019, everyone in the component would be at risk. From the lens of En a virus can exhaust an entire population with infection very early. However, WiMob shows that only 69% of the population is connected in a single component (Table S3 ). This di↵erence is because WiMob can distinguish how many individuals are active on campus. Therefore, WiMob provides a pragmatic estimate of risk by grounding it in local occupancy and helps campuses budget for resources better.

Reports suggest that a key contributor to cases in the pandemic is actually clustering of individuals in non-academic spaces [39] . However, En does not depict a holistic view of campus contact. It is limited to classrooms and, therefore, fixates on contacts in lectures, while ignoring other spaces. In fact, WiMob shows that in the first 6 weeks of Fall 2019, the shortest path among individuals is smaller than that approximated by En (Figure 2b ). With WiMob, we observe new paths in the contact network from situations outside classes. On a given week, WiMob shows the average shortest path with contact is 3.26(±0.5) when only considering lectures, whereas capturing all contexts reduces the average shortest path to 2.67(±0. 28) . Characterizing shorter pathways is crucial for policymakers as closure policies by design aim to disconnect these pathways.

Remote Instruction (RI) The status quo for data-driven policies o↵ers strictly online instruction for large class enrollment, while continuing the other classes in person.

For En we implement this by removing connections between students who are only in contact through courses where size 30. For WiMob we remove connections between students if they are only connected because of collocations during scheduled lectures of such courses.

We evaluate the e↵ectiveness of such a policy if it were applied in Fall 2019, with both WiMob and En. Figure 2d shows that RI with En reduces contact by 94% and increases shortest path by 50%. However, the same intervention with WiMob shows a relatively milder impact (contact reduction 45%; shortest path increase 11%). This reinforces that contact outside courses are significant and remain una↵ected by enrollment-oriented policies like RI. WiMob provides a more encompassing view of the structural e↵ects to a network and motivates design of more impactful closure policies.

As outlined above, En does not capture comprehensively the contact on campus. A campus is composed of many di↵erent spaces and En does not have the flexibility to design closure of such spaces or assess its impact. These drawbacks naturally motivate a new approach to design interventions. Since WiMob mitigates the limitations of En, we leverage it to demonstrate the e↵ectiveness of localized closure or LC. We evaluate the community health outcomes and burdens to campus of closure interventions by simulating COVID-19 with our ABM that uses WiMob to define the contact structure for each day. This is overlayed by a modified SEIR compartmental model for COVID-19. GT also had implemented a robust surveillance program on campus. Hence we calibrate the ABM on the positivity rate for COVID-19 for GT [25] in the first 5 weeks of Fall 2020 also incorporating external seeding from the surrounding Fulton County, GA [40] . We validate our model by predicting future trends for the rest of Fall 2020. For robustness, we perform additional calibrations by varying time windows and university context (details in SI Sensitivity Analyses). We study interventions by applying the ABM over the contact networks produced by WiMob with data from Fall 2019 -a counterfactual to Fall 2020 if no closure had occurred (see SI Simulation Model for further details).

In addition to RI, we model LC, which we formalize:

Localized Closure (LC) Prior works show a few locations are responsible for majority spread [10] and restricting movement between regions leads to greater control [33] . We intuitively identify rooms-level spaces that are highly central location nodes in the network. We remove contacts between people who are only connected because of collocating at these locations. While, we employ various centrality algorithms to identify Das Swain, V. et al • 7

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Figure 3 : Results of policy interventions with our calibrated ABM on contact networks from Fall 2019, derived from WiMob (a) LC shows improved outcomes (shaded regions) even when constrained to the same restrictions of broader shutdown policies. This holds true especially during the first month after the policy is implemented and is consistent regardless of the behavior scenarios. After the first month, the active infections decline for non-intervention and broader shutdowns appear to be lower, but the cumulative infection count is still higher (inset). The overall decline in infections could be a function of reduced contact (Figure 2b ) as well as fewer external infections from the county. It is important to note that new infections are a proportion of susceptible individuals in the system. Since LC restrict spread early, it leaves more people susceptible (to both internal and external infections) later in the semester. (b) LC not only provided better outcomes (for peak and internal infection reduction) they also tend to be less burdening in terms of certain cost dimensions. For instance, in certain scenarios LC forces fewer students into fully online schedules and therefore keep more students on campus.

such locations, but for the results discussed in this section we use PageRank [42] ). Details in SI Identifying Locations for Closure.

We find that, if COVID-19 spread through Fall 2019 (a regular semester), the cases rise after 7 days (Figure 3a) . Therefore, we apply both RI and LC interventions after the first week.

To make the comparisons between the closure policies, we establish fixed budgets to design LC based on the resource utilization on RI We consider 2 kinds of budgets, (i) mobility reduction -to depict space use on campus, and (ii) risk of exposure -to reflect testing capacity. Also note, response to closure policies can lead to unpredictable side-e↵ects in campus behavior, particularly when a student's schedule is entirely online. Therefore, we design policies within three behavioral scenarios (each with a varying budget):

• S2: Non-Residential Avoidance: Non-residential students stop all visits to campus if they enrolled in at least 3 courses and the policy forces their entire academic schedule online. • S3: Complete Avoidance: Both residential and non-residential students avoid campus if they have at least 3 courses and all move online.

WiMob facilitates the conception of expressive closure interventions. To devise interventions,WiMob estimates how RI uses the budget and then designs LC to match this budget under every scenario. Table 1 describes how the budget for each policy varies. Additional details are present in SI Modeling Policy and Scenarios.

We present di↵erences between LC and RI based on three infection reduction outcomes; peak infections (maximum active cases on a given day), internal transmission (exposure from infected individuals on campus), and total infections (cumulative cases at the end of the semester). Additionally, we measure the burden of policy interventions with the number of locations closed -requires resources to monitor and maintain super-spreader locations, the percentage of students that avoid campus -disruption to learning outcomes [16, 11] , and the percentage of individuals completely isolated -worsens mental wellbeing [48] .

LC cause greater reduction in peak infections, while a↵ecting fewer locations Controlling peak infections relaxes the burden on a university to support positive cases for any given day, and allows resources to be distributed over time. In all scenarios, of our simulation of Fall 2019, we observe that the peak reduction is significantly better in LC ( Figure 3 ) than RI. While RI impacts 58 di↵erent locations (classrooms and lecture halls), in S1 and S2, LC achieves better outcomes by closing fewer locations. For example, in S2, RI achieves a 28.9% peak reduction, but LC shows a reductions of 49.3% (mobility budget) and 48.1% (exposure risk budget). This is attained by closing 38 or 50 locations respectively. Therefore, with such policies, policymakers need to restrict fewer locations to remarkably minimize the pressure of active infections on campus (e.g., diagnoses, treatment, quarantining).

LC lead to comparable reduction in total infections, while keeping more students on campus Universities want to minimize the number of infected cases while ensuring majority of the population remains active on campus to continue successful operation. The total number of infections reduced by both LC and RI are similar. While the di↵erences between policies are statistically significant (Table S3 ) in some scenarios, the magnitude of these di↵erences might not be practically as important. In contrast, the impact the policies have on the student schedules is remarkably di↵erent. RI forces multiple students to adapt to fully online schedules. In the Scenario S2, 9% of students do not visit campus and in S3, 27% of students do not visit campus. On the other hand in LC the number of students expected to avoid campus can be as low as 0 and never exceeds more than 12%. Besides sustaining economic Das Swain, V. et al • 9

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC with RI. We find that LC leads to significantly improved peak infection reduction and internal transmission. In terms of reduction in total infections, the outcomes are comparable in general but can vary by specific scenarios. In addition, every policy also exerts some burden on campus, either in terms of locations a↵ected, students avoiding campus or isolation. We observe that LC policies focus on fewer locations (except in S3). Moreover, these policies a↵ect fewer student's schedules and therefore fewer people avoid campus due to completely remote schedules. Finally, LC does not increase the percentage of people completely isolated on campus (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). loss to the campus, remote instruction can increase anxiety among students and hinder learning outcomes [11, 61] . Compared to RI, LC o↵ers policymakers a way to defend against turnover in the student population, without compromising overall control of disease spread (Table 1) . Limiting the number of students that avoid campus helps preserve on-campus businesses [28, 58] and minimally disrupts the student wellbeing.

LC cause greater reduction in internal transmission without causing further isolation on campus

Universities are responsible for limiting spread on campus, but they must also ensure that aggressive policies do not worsen mental wellbeing of the community. In terms of internal transmission the reduction is significantly larger with LC (Table 1) . However, when LC restricts the infections early in Fall 2019, it leaves more individuals susceptible to external transmission. College student behavior outside campus on weekends and breaks is known to impact local transmission [14] . When policymakers consider LC they should also consider policies on re-entry or required testing based on o↵-campus activities. In terms of isolating individuals on campus, it's notable that LC and RI are similar in S2. Interestingly, in S3, where LC closes more than 100 locations, the percentage of isolated individuals per week is less than that of RI. This finding implies that LC can keep individuals on campus without forcing them into complete isolation. Here "isolation" refers to no form of proximate contact with any individual on campus -extreme social distancing where individuals are not even collocated in the same suite or hall. While social distancing is a recommended coun-Das Swain, V. et al • 10

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 24, 2021. ;  termeasure for COVID-19 [1] , complete isolation can have adverse e↵ects on psychological wellbeing [48, 36, 43] . Staying completely isolated on campus can increase loneliness and limit social connectedness [36] , which are both related to depression [48] . Although the proportions are similar (Table 1) , LC does not necessarily isolate the same sets of individuals. This qualitative di↵erence could also explain the di↵erence in internal transmissions -LC could be isolating individuals who are less likely to spread the virus.

LC identifies a wider variety of auxiliary spaces By using WiMob to design LC we are able to identify locations for closure at the granularity level of rooms, including unbound spaces such as lobbies and work areas. First, in S1, we find that most locations that LC targets are a subset of the auditoriums-like rooms where large classes would take place in Fall 2019. Note, LC needs to restrict only a few such spaces to be under the same budget as RI. This is because, under S1, RI policies only alter visits to lectures, while these spaces are used for other purposes during other times (e.g., club activities and seminars). We also note that LC targets 'high tra c' locations like conference center lobbies which are typically used as waiting areas or for networking events. Next, in Scenario S2, we see that in addition to spaces mentioned earlier, interestingly LC further restricts the use of smaller rooms (occupancy 13 35) which would not be a↵ected by RI (as only classes of size 30 are o↵ered online). LC also targets areas in the recreation center (which includes locker rooms and indoor courts for 4 20 people). This insight indicates that our methodology WiMob is sensitive to other student activities. Moreover, we also find a selection of spaces that would not be frequented by the undergraduate population, such as lab areas and facility buildings like the police station. Lastly, in Scenario S3, LC targets closure of activity in far more spaces than RI. However, the better outcomes can be attributed to the fact that LC diversifies the potential restriction areas. LC now restricts heavily used small study rooms or breakout rooms (for 1 6 people). Furthermore, it restricts use of spaces where multiple small groups of people can organically assemble, such as cafes, dining halls, and reading areas. We also observe that LC restricts activity in about 10 Greek Houses but does not target other housing areas -demonstrating its ability to restrict social behavior that could amplify disease spread. Figure S22 shows the diversity in locations for various LC policies.

The results above use an ABM calibrated on the positivity rate of the first 5 weeks of Fall 2020. At a university, this rate can be influenced by many latent factors (e.g., mask-wearing, hand washing, distancing, and compliance). To study any e↵ect of these variations on our results, we also calibrated on di↵erent time windows throughout the semester. We calibrate on weeks 5 9 and 10 14 in Fall 2020, and validate on the remaining semester. In both cases, compared to RI, we find that LC still exhibits better reduction in peak infections (up to 90%) and internal transmission (up to 77%). In the original calibration, LC also significantly reduces total infections and maintains the same level as RI, but with the new periods we find Das Swain, V. et al • 11

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 24, 2021. ;  total infections are substantially less than RI (Table S9 and Table S10 ).

Another important variable for positivity is the wider context of the campus e.g. urban/rural, the surrounding county, city, etc. To investigate this, we also calibrated our ABM on the positivity rate of di↵erent universities in the US in Fall 2020 (along with information from their county to seed external cases). Consider this as a hypothetical where the mobility of the GT community remains the same but disease outcomes resemble a different campus. We calibrate on data from University of Illinois at Urbana-Champaign and University of California, Berkeley. We find no remarkable di↵erences from our findings with GT (Table S11 and Table S12 ).

When facing a pandemic, non-pharmaceutical interventions (NPI) are the first line of defense for universities to respond to contagious diseases like COVID-19 [19, 38] . On a campus, a common form of NPI is closure [29] . Universities consider enrollment data (En) to design remote instruction (RI) for closure to support continued operations safely [59] . However, En can misconstrue contact on campus and RI policies can have broad impacts despite their e↵ects on curbing the disease spread. This paper demonstrates that repurposing logs from a managed WiFi network (WiMob) can help design e↵ective localized closure policies (LC). We show that WiMob uncovers rich contact dynamics and provides policymakers multiple dimensions to design policies like LC. We simulate COVID-19 with an ABM that harnesses WiMob to compare RI and LC. Our results present evidence that LC can lead to improved infection reduction outcomes, while simultaneously relaxing burdens on the campus community caused by coarse-grained broad RI policies.

In practice LC policies should be deployed on campus in conjunction with the other tools as well like testing, tracing, and quarantining. WiMob can complement disease-specific knowledge to identify closure spaces. For example, small indoor spaces with poor ventilation increase the risk of infection for COVID- 19 [51] , while other algorithm-identified locations for closure might not require closure because mask-wearing and testing have high compliance among users of that space. Further, as a pandemic progresses and public health guidance develops [50], with WiMob, campuses can regulate the restriction of LC policies and anticipate the path to 'normal' operations [34, 44] Moreover, WiMob captures various spillover e↵ects that cannot be captured in methods like En. For instance, with WiMob we observe that the mobility in Fall 2020 was 39% of that in Fall 2019 because the on-ground policies lead to certain sta↵ working remotely as well. With additional information, WiMob enables policymakers to model such scenarios and design alternatives like LC with new budgets. Policymakers in universities can use WiMob as a versatile tool to explore dynamic intervention strategies as well. In this paper, LC interventions are non-adaptive fixed policies throughout the semester. Since staggering policy restrictions could have variable impact on campus [65] .

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 24, 2021. However, WiMob could help identify locations from di↵erent mobility phases (e.g., the same week from a prior semester) and assess the e↵ects of closure policies. Additionally, depending on campus priorities and resource limitations, di↵erent campuses can use this same data to model policies di↵erently. The e↵ectiveness of reopening policies is expected to be sensitive to a campus' specific context that includes physical infrastructure, overarching guidelines, and human compliance [5] . For certain campuses policies might not need to be constrained by exposure risk as testing might be frequent, ubiquitous, and voluminous. Other campuses could have limits on quarantining capacity. Policymakers might even consider the cost trade-o↵s by actually forecasting actual financial losses incurred by reduction in mobility [6] , or valuate loss of services based on community needs [53] . We elaborate on these considerations in the SI Implications for Policy Design.

Beyond assessing cost-benefits, universities also need to consider practical methods of obtaining, storing, and processing mobility of the community as WiMob. University can access logs from the managed network internally as it is passively collected. Moreover, it does not require any new form of surveillance sensing but universities must revise terms of use and stay sensitive to community perspectives. While aggregate data on population mobility is valuable for many applications [64] , which includes informing pandemic response [8] , the major privacy challenge with localization data is to avoid accumulation [56] . Instead, operational applications need to conceive approaches that only retain processed insights on locations to shutdown but not individual data. Similarly, any operational use needs to have pre-established access limitations on what stakeholders can learn from the data [4] (e.g., decision-makers can only get a list of candidate locations to close). In the SI Discussion, we further detail approaches to reconcile privacy, ethics and legal considerations.

Lastly, for future investigations of better closure policies, researchers and policymakers need to be cognizant of the limitations of our work. Our analyses do not represent heterogeneity among individuals and therefore our simulation does not account for intrinsic vulnerabilities [31, 43, 22] and di↵erence in mobility behaviors of demographic groups [10] . WiMob can be extended with other streams of data to introduce variability in the population and devise new forms of LC to protect the most vulnerable community members. Additionally, our work explores the extremes of the range of behavioral responses to closure interventions. Henceforth, researchers and policymakers can model more nuanced spillover e↵ects that interpolate between the scenarios we describe, as well as extend them. Further discussion in SI Limitations and Future Work.

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021 

This section summarizes (i) the data used to derive contact networks and policies, and (ii) the dynamics of our simulation model and calibration approach. Additional information for every subsection is present in SI Methods.

The IT management facility at GT accumulates WiFi access point logs over time. This is common in most universities with managed WiFi infrastructure. We actively collaborated with IT management to define safety and security safeguards that allow us to obtain a deidentified version of these raw logs. Before accessing the data we established a data-use agreement and an ethics protocol that was approved by the Institutional Review Board (IRB). For the WiFi data, we were provided access to logs from Fall 2019 and Fall 2020. We process these logs to characterize mobility (WiMob) and it encompasses all 40, 000 unique individuals that connected to the network via 6, 959 di↵erent access points [13] . The logs do not contain any personally identifiable information and locations are also coded. For En we only use aggregate insights for enrollment, which are derived from course registration transcripts. Note, we do not cross-identify any students. We use publicly accessible course schedules to approximate schedules of de-identified nodes and infer if they are students or sta↵, and non-residential or residential. We elaborate on our data in SI Data.

WiMob leverages the logs to create bipartite graphs K t , for each day t, which connect P users to L access point locations ( Figure 1a ). Any edge, {p, l} i indicates the i th instance when a p was dwelling at l. These edges describe the time period of dwelling. Subsequently, by comparing all edges in K t we can infer if di↵erent individuals are collocated near an AP to create a contact network, G t , for each day t -between collocated p 2 P . These networks feed into the ABM at every time-step. Similarly, by inspecting the sequence of dwelling locations for any p in graph K, we compute a mobility network, H t -between locations l 2 L. We provide more details of our approach in SI Data Processing and in SI Modeling Collocation and Movement.

We compare the disease outcomes and burdens of 2 policies, Remote Instruction (RI) and Localized Closure (LC), both of which are modeled with WiMob. For RI we infer enrollment size of each course in Fall 2019 by determining the number of unique individuals that visit lecture locations during scheduled times. After the first week, we apply the RI by removing all visiting edges in K t for any l c 2 L RI if visits were during lecture times of course c with an enrollment 30. This helps create counterfactual contact networks G 0 t . The removal of Das Swain, V. et al • 14

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ; https://doi.org/10.1101/2021.03.16.21253662 doi: medRxiv preprint edges from K describes the mobility budget of RI and the structure of G 0 t indicates the risk of exposure budget. We design LC with these budgets by identifying locations for closure (L LC ) with di↵erent algorithms, such as PageRank, Eigenvector Centrality, Load Centrality, and Betweenness Centrality. When a location is closed, we remove all edges in K t connected to any l x 2 L LC . We aggregate the movement graph H t over a week and apply the algorithms to identify locations. Subsequently, we identify the number of top-ranked locations to remove such that the resultant counterfactual contact network G" t has is within 1% of the budget. The budgets vary for di↵erent behavioral scenarios and we only compare policies within the same scenario. This is further elaborated in SI Modeling Policies and Scenarios.

We construct an agent-based model (ABM) that captures the spread of COVID-19 between individuals active on campus. This ABM leverages the contact networks produced by WiMob. The simulation iterates a time-step each day and the underlying contact networks i.e., G t for no interventions, G 0 t for RI, and G " t for LC. Our ABM follows a modified version of susceptible-exposed-infectious-removed (SEIR) template that disambiguates the infectious compartment into asymptomatic and symptomatic. New infections are introduced to the model either externally or internally. External transmission arises because individuals can contract the virus outside campus and bring the infection back for local spread [24, 39] . We adopt data of positive cases from Fulton county [40] with a scaling factor ↵ to estimate the probability that a susceptible individual, who is active on campus, was infected from interactions outside campus. This is to account for any commute outside campus during the pandemic. Internal transmissions are determined by p, as the probability of susceptible individuals in contact with an infectious one. We calibrate the parameters related to disease transmission by training and validating our models on the positivity rate reported by GT surveillance testing [25] . SI Agent-Based Model details the disease progression and describes the various parameters.

We estimate the ranges of optimal parameters for disease transmission by minimizing the root means square error (r.m.s.e) between the Georgia Tech surveillance testing positive rates [33, 25] and the observed positivity rate of the model every week-percentage of new asymptomatic out of the total testable population. The surveillance testing conducted by Georgia Tech is designed for detecting individuals who contracted Covid-19 without showing Flu-like symptoms within the community [33] . We calibrate the model on the positivity rates on the first 5 weeks of Fall 2020. To attain a point estimation of the optimal parameters, we fit the model to predict trends in the remaining weeks by running a numerical optimization algorithm, Nelder-Mead [25] . To account for quantitative uncertainty, we estimate a range of parameters, within 40% of optimum r.m.s.e. Note, this calibration

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  characterizes latent factors associated with pandemic-related cautious behaviors, including the relationship with external transmission. And these factors could be related to "county characteristics, partisanship, media consumption, and racial and ethnic composition" [1] . Since the e↵ectiveness of shutdown policies can vary by time period and county. In SI Sensitivity we discuss hypothetical variations where the mobility behavior of GT remains constant but disease outcomes change based on time period of calibration and positivity rates from universities at di↵erent counties in the U.S. See SI Calibration for details on the calibration process and results are in Table S4 .

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021 [49] P. Sapiezynski, A. Stopczynski, R. Gatej, and S. Lehmann. Tracking human mobility using wifi signals. PloS one, 10 (7):e0130824, 2015.

[50] S. Saul and S. Hubler. Colleges vowed a safer spring. then students, and variants, arrived, 2021. Available at: https://www.nytimes.com/2021/02/09/us/colleges-covid.

html.

[51] L. J. Schoen. Guidance for building operations during the covid-19 pandemic. ASHRAE Journal, 5(3), 2020.

[ 

In this section, first, we describe the primary data source for mobility models (WiMob), the data used for calibrating our simulations, and for comparison of contact networks with methods using enrollment data (En). Next, we describe how we construct counterfactual mobility networks under the two main policies of interest in our study: remote instruction (RI) and localized closures (LC). Finally, we describe an agent-based-model (ABM) of disease transmission, which has a contact structure based on WiMob, and how this model was calibrated.

We use data provided by the IT management facility at Georgia Institute of Technology (GT) which accumulates WiFi access point (AP) logs over time. The primary use of WiFi network logs is for maintenance and security purposes. We mine these logs post-hoc to describe the mobility of individuals on campus, which we refer to as WiMob. Here mobility is expressed by visits to certain locations that are demarcated by a corresponding AP. WiMob can also describe dwelling (duration of visits) and collocation (dwelling in the presence of others around the same AP).

The campus WiFi network spans 6959 APs distributed between 240 buildings (and some outdoor locations). We label APs according to which building they are inside, along with the closest room or space (e.g, hallway, lobby, suite, cafe, etc.). The AP may or may not reside inside the room, however, in most cases, only a single AP is associated with space. For less than 5% of the APs, the AP shared association to space with another AP. This manyto-one mapping is typically in the case of large halls and auditoriums. We resolve such many-to-one associations by using APs as a proxy of the space they are associated with. Therefore, individuals connected to di↵erent APs in the same space will still be identified as collocated. Similarly, an individual could connect to the network with multiple devices. However, less than 1% logs show that a user is connected to multiple APs around the same time. Therefore, WiMob is agnostic to which device connects to the APs to proxy the presence of the individual. For this study, we obtain the WiFi network logs retrospectively for all of Fall 2019, and the data for Fall 2020 was provided on a per-day basis. Each day, approximately 33, 000 di↵erent people connect their devices to the WiFi network on campus. Overall in Fall 2019, approximately 40, 000 di↵erent people connected to the campus network.

Note, that GT's managed WiFi network is not equipped with any Real-Time Location System (RTLS) [9, 27] . RTLS systems use Received Signal Strength Indicator (RSSI) values from multiple neighboring APs to provide high precise localization of individuals in terms of time and space. However, deploying such systems requires surveying the entire network. Additionally, precision localization raises more privacy concerns. These factors together

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  make it challenging for universities to justify the deployment of RTLS, unlike small retail settings that can monetize RTLS insights directly (e.g., insights on footfall can be tied to improving revenue).

We calibrated the ABM using the publicly reported positivity rate on the GT campus as reported through the asymptomatic surveillance and diagnostic testing program [33] . The testing program used pooled saliva sample surveillance with follow-up diagnostic testing. The positivity rate was reported each day, but individuals must wait at least 1 week between tests. We aggregated the positivity rate by week during the Fall 2020 semester.

When calibrating our ABM, we considered the reported confirmed cases in Fulton County [40] , the county in which GT is located. The 'Confirmed COVID-19 Cases' reported in this dataset are cases that have been confirmed with a positive molecular (PCR) test. We considered cases during the Fall 2020 semester to inform external transmissions in the ABM.

We compare structural properties of contact networks constructed with WiMob to contact networks constructed from GT's course enrollment transcripts (En) -the aggregate statistics as reported in [45] . The En network was based on Fall 2019 transcripts for GT's Atlanta campus. These were cleaned to account for cross-listed courses and was used to determine which students were classmates with each other to form a contact network.

WiMob is our approach to describe contact between people and movement of people between locations. The first step requires using WiFi network logs to infer when individuals were at specific locations on campus by determining when devices were connected to the corresponding APs. Our system mines the WiFi network logs that are populated via the Simple Network Management Protocol (SNMP) -a standard and widely used monitoring protocol to organize device association behavior to a WiFi network. Periodic SNMP updates can be caused either by poll requests to the APs that log which devices are associated with it at that time. However, devices can appear invisible to detached from an AP for multiple reasons, for example, when devices are idle. Otherwise, SNMP updates can occur whenever a new device connects, which is typical when individuals move between APs. Our approach exploits this factor to first mine periods when individuals are moving, then identify periods of dwelling between movements, and finally determine collocation when two or more individuals are dwelling near the same AP. This system follows from other studies that mine WiFi

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  logs [18, 57] and the detailed processing pipeline and evaluation is presented in [13] . This system to infer collocations has been tested against lecture attendance and reports a high precision of 0.89, but a relatively lower specificity of 0.79 [13] . While it is not likely to show false-positives, it has a possibility to erroneously mark people absent from a location even though they were there. However, for the purposes of our study, a contact network is made over an entire day and it only needs a single collocation instance for us to consider contact. And therefore we believe this limitation would not significantly a↵ect our models. Figure S4 : In a managed network, SNMP updates the logs by describing device association to an AP at a certain timestamp. WiMob mines these logs to characterize mobility as a bipartite graph. The nodes are partitioned to describe people nodes (e.g., P 1, P 2) connected to locations nodes (e.g., L1, L2). Every edge across the partition describes people visiting locations on campus during di↵erent times (e.g., t 1 , t 2 ). Projecting the bipartite on people nodes helps construct a contact network (e.g., P 1 and P 2 were collocated at L1 at t 1 ), while projecting it on locations helps construct a directed movement graph (P 2 dwelled at L1 and then at L2).

After inferring where an individual is located on campus, we represent the entire community behavior as graphs. We describe a bipartite graph, K, that shows when a user is at a given location on campus ( Figure S4 ). This bipartite graph has edges connecting a set of m people, P , to a set of n locations, L. An individual can have multiple edges connecting to the location if they visited that location multiple times (e.g., t 1 , t 2 ). The edge data contains the start Das Swain, V. et al • 4

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  and end times of these dwelling periods. For these bipartite graphs, we make a projection on set P to describe collocation. This projection graph, G, contains an edge between users if they were visiting the same location during overlapping times. Since we do not use RTLS, our approach can only identify if people were in the vicinity of the same AP, but does not describe the distance between them. However, it can reasonably determine collocation in the same room [13] . Since our study is limited to localizing people indoors, we adapt the definition of proximate contact [27] where people might be "more than 6 feet but in the same room for an extended period". In our work, we use a lower bound threshold of 40 minutes to determine proximate contact. Therefore, individuals are only considered in contact when they are collocated in a room for 40 minutes or more. This threshold was set up to account for typical lecture duration on campus (for standard 3-credit hour courses taught 3 times a week). Every edge between two individuals contains a list of locations where they were possibly in contact. G forms the basis of the contact-network that we use an agent-based model to simulate. Alternatively, we also make a projection on the set L. This projection is a directed graph, H, where an edge from L i to L j represents movement from the first location to the next within a span of 60 minutes. GT's large urban campus with pedestrian pathways and motorized transit services enables direct movement between any two places on campus within the threshold. The 60 minutes threshold helps discount erroneously labeling returning from outside campus (e.g., non-residential students visiting two di↵erent locations between 2 days). H e↵ectively describes how locations are connected and which locations could be more conducive to attracting and disseminating the virus. As a consequence, the H helps inform policy design. We compute the bipartite graph and its projections for each day of the semester.

As a response to COVID-19, prior work has recommended using En to enforce a form of RI-moving classes large to an online remote instruction setup while other classes are o↵ered in-person [26, 7, 59] . While we have access to aggregate insights on En contact networks, our study protocol prohibits us from accessing course-specific information at an individual level. Therefore to infer individual enrollment, we analyze the edges of the bipartite graph K. For this, we first scrape the GT's course roster for Fall 2019 (filtered to only represent the Atlanta campus). This process provides us with a location and weekly schedule for every lecture conducted on campus, including its various sections. With this information, we are able to identify which edges represent visits to lectures, and subsequently, we can account for unique visitors to a lecture. Thus, we can first identify the number of unique individuals on campus who are enrolled in classes. The aggregate data from course enrollment reports that 21, 299 students were enrolled in Fall 2019. In comparison, our inference identifies 22, 248 students. The excess number can be explained by the fact that our method does Das Swain, V. et al • 5

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ; https://doi.org/10.1101/2021.03.16.21253662 doi: medRxiv preprint not distinguish between instructors, TAs, and students. Next, we study the unique visitors to every lecture in the scraped course schedule which gives us an estimate for the size of every class. Given the limitations of our data processing, actual enrollment sizes could be larger, but our process is less likely to count false positives [13] . Finally, to model RI, for the contact network G t , we create a counterfactual network G 0 t for each day t. These exclude collocations that took place at lecture locations during lecture times. If two people were connected solely by proximity during lectures -in a class with large enrollment -they will appear disconnected in the counterfactual network.

This article demonstrates the e↵ectiveness of localized closures,LC, which are targeted interventions to seize mobility at di↵erent spaces on campus. For this, we identify important locations on campus by analyzing H. In the main paper, LC uses PageRank [42] as an illustrative algorithm to identify important location nodes. For robustness, we apply various additional algorithms to identify highly authoritative nodes in H -betweenness centrality [15] , eigenvector centrality [5] , and load centrality [29] . In the SI Appendix, we distinguish these di↵erent policies as LC PRank , LC BCen , LC ECen , LC LCen . Since RI captures a weekly schedule to determine enrollment, LC is implemented to find locations based on behavior from the past 7 days of mobility. We apply the weighted version of the algorithms mentioned earlier on the directed graph representing movement, H. The edge weight is based on the number of instances of movement between any L i and L j . After sorting the locations by importance, we determine the number of locations to shut down based on di↵erent budgets induced by RI-mobility and risk of exposure. For this purpose, we take the approach of a greedy algorithm which successively removes highly-ranked locations till the constraint is met (within 1% margin of error). Similar to RI, LC also render counterfactual collocation networks, G" t for each day t. In these networks, we remove instances of collocations that occurred at the shutdown locations. Figure S22 and Figure S23 shows the categories of buildings where di↵erent spaces are closed by LC policies.

We now describe how we compare the RI and LC policies. First, we consider the e↵ects of these policies under three behavioral scenarios. These scenarios express the spillover e↵ects of closure that lead to students avoiding campus entirely because their entire schedule is forced online. This analysis assumes that the motivation to be present on campus is determined primarily by enrollment. We consider that, if a student has a full course load (enrolled in a minimum of 3 classes) and all their classes are o↵ered online, that student might have less incentive to visit campus at all (for any engagement) and thus practice Avoidance. Since LC could end up closing classrooms, it can also lead to academic schedules being a↵ected and elicit Avoidance behavior. As a result, we describe three scenarios. Persistence, is the preliminary, or null scenario, which represents no Avoidance. This counterfactual collocation graph only removes edges directly a↵ected by RI or LC. The second scenario Das Swain, V. et al • 6

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  we model is Non-Residential Avoidance where only non-residential students with full online schedules stop visiting campus entirely. Here the counterfactual graph will remove all edges of non-residential students with fully online schedules. Lastly, the third scenario we model is Complete Avoidance where any student with fully online schedules stops activity on campus entirely (including residential students). Here the counterfactual graph will remove all edges from any student with fully online schedules. Since our study protocol prohibits us from mapping our data to other sources, we heuristically infer which individuals are likely to be residential and which are not. We label individuals as residential when they dwell an average of at least 15 minutes at residential locations between 6pm and 10am, on workdays (Monday-Thursday).

Under each scenario, we limit the number of locations that can be closed under the LC policy to ensure the level of restriction is constrained to be similar to the RI policy. We limit the number of locations under two types of restrictive budgets. The first budget is based on mobility, which is the percentage of edges remaining in the bipartite graph if a policy were to be implemented. The second budget is based on exposure risk, which is the number of unique individuals who would be in the 1-hop collocation neighborhood of positive individuals. We compute this budget by randomly sampling 2.5% of the population as positive, based on the highest 7-day average positivity rate reported by GT [25] in Fall 2019. Note, however, the e↵ect of RI on campus can vary in di↵erent behavioral scenarios, thereby changing the budget available to design a comparable LC policy. For instance, the number of people at exposure risk is much lower in Complete Avoidance. As a result, we build multiple alternate networks representing the e↵ect of policies under counterfactual behavioral scenarios.

The infection reduction outcomes and burdens of di↵erent policy interventions (under various scenarios and budgets) is described in Table S5-Table S8 presents boxplots that compares the distribution of disease control outcomes. Figure S14 - Figure S17 show cumulative plots of disease control outcomes.

We constructed an agent-based model (ABM) that captures the spread of COVID-19 between individuals active within the GT community. The model is used to evaluate the e↵ectiveness of di↵erent policy interventions. We consider a modified version of the SEIR framework for simulating the spread of COVID-19 [39, 10] by using an underlying contact network given by WiMob. Figure S5 shows the compartments of the framework. The susceptible state (S) represents individuals who have not been infected and can contract the disease by having contact with an infectious individual. The exposed state (E) is canonically equivalent to the "incubation period" and is similar to the pre-symptomatic state found in related work [44, 20] . Individuals are considered infectious when they are in either the asymptomatic state (Asym) or symptomatic state (Sym). Individuals in the asymptomatic

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Figure S5 : (a) The schematic of the compartments in our modified SEIR model. By the design of the GT surveillance testing [33, 25] , the total testable population is defined as the summation of susceptible, exposed, and asymptomatic. Infectious persons are in either symptomatic or symptomatic. For every e↵ective edge in the mobility network, a susceptible individual that is exposed to an infectious person becomes infected with probability p. Individuals may also get infected due to an exposure not captured by the WiMob network which occurs with probability I out (t) on day t. account for new infected cases. (b) The mobility behavior represented by WiMob changes every day of the semester (shown weekly here). The contact network constructed from WiMob forms the underlying contact structure of the ABM.

state are assumed to be the major "spreaders" [20] and transmit the infections to susceptible individuals before they are recovered (R) [26] -after 7 days [20] . Since asymptomatic is considered a state of mild severity [37] , individuals in this state do not have a risk of fatality. By contrast, for individuals in the symptomatic state, will be eventually isolated (Iso) (e.g. self-quarantine, or hospitalization on campus). Once in the isolated state, they cannot transmit the disease to individuals in the susceptible state. Unlike the asymptomatic track, the symptomatic state is considered critical severity. Therefore, after moving to the isolated state, individuals have risk of fatality and entering the death state (D) . If the isolated individual survives, they enter the recovered state. We assume immunity is preserved and therefore after recovery the individual is no longer susceptible.

Let t = {0, 1, 2, 3, ..., T } be the index of days in simulations. We denote the sequence of dynamic collocation networks indexed by day t, as {G t (A t , B t )} T t=0 . A t is the set of vertices, i.e. individuals on campus, and B t is the set of edges. The universe set of the population throughout the simulation time period is given by M = S T i=1 A t . For convenience, we use a i 2 M to index every person in the universe population set.

The SEIR model consists of seven compartments. Each of these corresponds to a function of population subsets with respect to day t: susceptible S(t), exposed E(t), asymptomatic Asym(t), symptomatic Sym(t), isolation I(t), recovered R(t), and dead D(t). For example, a i 2 I(t) means a i is in the isolation state at day t. We use N t S!E , N t E!Asym , N t E!Sym , Das Swain, V. et al • 8

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ; https://doi.org/10.1101/2021.03.16.21253662 doi: medRxiv preprint N t Asym!R , N t Sym!I , N t I!R , and N t I!D to denote the transitions between states between day t and day t + 1.

The entire population M is fixed where M = S(t)+E(t)+Asym(t)+Sym(t)+I(t)+R(t)+D(t) for all t. To capture the positivity out of the students coming back to campus at the start of the semester, we initialize the system by setting a subset of M into Asym(0) and the reminder into S(0). The initial percentage of asymptomatic is described by:

where I 0 is a parameter defined as the initial percentage of Asymptomatic at day t = 0.

We consider two ways that an individual in the ABM could be exposed: (i) exposures that occur due to contacts among individuals captured by the mobility network (internal transmission) and (ii) exposures that occur due to contacts that occur outside of the mobility network (external transmission).

Internal transmissions happen exclusively among individuals in the model. On any given day, an edge becomes e↵ective, when one of the susceptible individual comes in contact with the other which is infectious, i.e. asymptomatic or symptomatic, individual. Therefore, for every e↵ective edge between two such people, the probability of the susceptible individual getting exposed is described by the transmission probability p, which is another model parameter. The probability for an susceptible individual a i entering exposed at the end of day t is given by the following function:

Here, e(t, a i ) is the number of e↵ective edges of individual a i at time t. Since (1 p) e(t,a i ) is the probability that a i does not contracted the disease at time t under e(t, a i ) Bernoulli trials, 1 (1 p) e(t,a i ) is the probability that at least one e↵ective edge leading a i to exposed.

In addition to exposure due to internal transmission, we also consider new exposure due to external transmission. We consider external transmission to be exposure resulting from the physical collocations outside the scope of mobility network. For instance, the WiMob does not capture the connections between individuals without access to the campus WiFi or Das Swain, V. et al • 9

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  someone contacting infectious persons outside the campus. To reflect this risk in our model, for any day t, I out (t) describes the probability of infection on day t from a collocation that is external to the mobility network. We assume that the probability an individual is infected due to an external source is proportional to the number of cases in the broader community. Therefore, we model the probability of external infection as a function of confirmed cases in Fulton county, where GT is located [40] . C t represents the confirmed cases reported by Fulton County where C max is the maximum number of the cases over the whole period, I out (t) is given by

where ↵ is a parameter scaling the normalized confirm cases in the surrounding county. The resulting number of external infections on day t is then modeled to be are Binomial with |S(t)| trials with probability of success I out (t).

In summary, for every day t > 0, the overall number of individuals that become newly exposed is represented as N t S!E which is the result of both external and internal transmissions.

After exposure, individuals in the model will progress through other disease states in our model. We update the number of individuals in each state daily to reflect transitions between them. The transitions between the states on day t are summarized according to the following equations: . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ; https://doi.org/10.1101/2021.03.16.21253662 doi: medRxiv preprint

After an individual has been exposed, they will spend S days in an incubation period. At day S after their exposure, individuals will become a symptomatic infection with probability p S . Otherwise the agent will become an asymptomatic infection This process is given by the following two equations:

Individuals who enter the asymptomatic state will recover after Asym!R days since they were first exposed. Thus, we represent the number of transitions from asymptomatic to recovered on day t as:

On the other hand, individuals who enter the symptomatic will eventually enter the isolation state [20] . The time that individuals spend in the symptomatic state before entering the isolated state is normally distributed t I ⇠ Normal( I , 2 I ). We simulate each individual's transition between symptomatic and isolated by using a sampling function (a i , t, t ) and a function ⌧ (a i , t) that returns the days since exposed respectively:

first day of a i entering exposed , a i 2 Sym(t)

Das Swain, V. et al • 11

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. Transmission probability: For any edge between a susceptible and infectious individual in the contact network, p is the probability that the susceptible person will enter into the exposed state. This only dictates internal transmission 0.034 0.007 Calibration ↵ Scaling factor of the normalized confirmed cases in the surrounding county(()). This is the parameter for us to generate I out (t) 0.032 0.0032 Calibration I 0 Proportion of population that is asymptomatic at day 0 0.012 0.0009 Calibration p S Probability of exposed persons becoming symptomatic 0.66 - [20] S Incubation period (days) since the first day of exposure 5 - [20] Asym!R Asymptomatic duration (days); it is the time taken for an asymptomatic person to recover since the first day of exposure The variables p, ↵, and I 0 are estimated by calibrating the simulation model on the first 5 weeks of positivity rates provided by GT surveillance for Fall 2020, while incorporating external cases from Fulton County. These parameters were found by validating the ABM on the remaining weeks of Fall 2020. Figure S6 shows model estimate during the calibration and validation period. The aggregated transitions N t Sym!I between symptomatic and isolated is the sum of the distribution above on each day t.

Individuals who enter the isolated state may end up with one of two states: dead or recovered. We defined N t I!D as following another binomial distribution with parameter p D :

The transitions between isolation and recovered is quite similar to the transitions between symptomatic and isolation except t R ⇠ Normal( R , 2 R ) where R and R are the two parameters standing for the mean and standard deviation of days for an individual in the isolation state entering recovered since the first day of infection. This leads to:

Most of our model parameters can be estimated from previous studies (see Table S2 ). However, three parameters in our study are not easily estimated from previous studies: (1) Das Swain, V. et al • 12

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  the proportion of the agents that begin the semester asymptotically infected, I 0 , (2) the probability of transmission between a given infectious individual and susceptible individual given a contact in the mobility network, p, and (3) the scaling factor ↵ used to determine probability of transmission due to contact outside of WiMob network on day t, I out (t) (see ()). We fit these three parameters to the published weekly positivity rate (percentage of asymptomatic cases) as reported by GT's asymptomatic surveillance testing program [33] . To fit the parameters, we performed calibration to minimize the root mean square of error(r.m.s.e) between the simulation estimates of the weekly positivity rate and the observed weekly positivity rate on GT's campus of the Fall 2020 semester as reported by the surveillance testing program.

To perform the calibration, we used two sets of public data pertaining to 2020 Fall semester at GT: (i) the confirmed cases in Fulton County [40] , and (ii) the aggregated surveillance test positivity rate for each week [33] . The former helps estimate the daily external infection percentage. The latter is the ground truth trajectory we fit our model on. We consider the data aggregated by week because each individual on campus can only get tested once per week. The positivity rate provided by the surveillance testing data can be interpreted as the estimated percentage of new asymptomatic cases out of the total testable population which includes susceptible, exposed, and asymptomatic -with an assumption that every testable population get tested at the same rate.

To formalize the calibration problem, let R w be the surveillance-testing aggregated result at week w. Let S(I 0 , ↵, p, w) be the function of the simulation model which returns the percentage of new asymptomatic in week w out of the total testable population. For every combination of parameters, the predicted result for each week w is estimated by taking the average of N simulation outputs. The objective function is:

The optimization problem is: min

We fit our model to the first 5 weeks of Fall 2020 and validate the results on the remaining weeks. After obtaining the optimal set of parameters, for robust comparison of policies with di↵erent viral variants, we generate a range of parameters by compromising the r.m.s.e within 40% of the minima [10] . First, we implement the Nelder Mead method [25] to discover the optimal set of parameters that minimizes the r.m.s.e. Next, we sample 40 di↵erent combinations of parameters within 40% of the minimum r.m.s.e to estimate the means and standard deviations of these parameters ( Table S2) . Throughout this paper, we pool together all simulation results across those parameters over multiple runs (N = 15) and report the 2.5th and 97.5th percentiles of the simulation outputs for every policy experiment.

Das Swain, V. et al • 13

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021 

In this section, we design complementary experiments to inspect the robustness LC policies under di↵erent setups and calibration approaches. These variations are defined as follows:

• Calibration periods (V1): For the results in the main paper, we discuss results with our ABM calibrated on the first 5 weeks of surveillance testing data. For additional analyses, the model parameters are re-estimated based on the surveillance data from week 5 9 and 10 14 in Fall 2020 at GT. The calibration is validated on the remaining weeks in the semester. Figure S6 shows the calibration and validation. The results of policy comparison with these variations can be found in Table S9 and Table S10 , for weeks 5 9 and 10 14 respectively. Additionally, Figure S12 shows boxplots to compare the distributions of di↵erent policies, while Figure S18 and Figure S19 show cumulative plots of the disease control outcomes, for weeks 5 9 and 10 14 respectively.

• Campuses and counties (V2): For the results in the main paper, the calibration of our ABM reflects certain latent factors inherent to GT that could a↵ect both mobility behavior as well as testing results. To complement this we consider calibrating our data under di↵erent settings informed by surveillance testing from other similar large universities. This analysis is intended to represent the GT community in a di↵erent geographic setting, which is influenced by a di↵erent surrounding community, policies and resources. The new parameters are estimated based on the first 5 weeks of surveillance testing from the University of Illinois at Urbana-Champaign (UIUC) and the University of California, Berkeley (Berkeley) [31, 38] , and the corresponding county data [12, 11] The calibration is validated on the remaining weeks in the semester. Figure S7 and Figure S8 show the calibration and validation for UIUC and Berkeley respectively. The results of policy comparison with these variations can be found in Table S11 and Table S12 . Additionally, Figure S13 shows boxplots to compare the distributions of di↵erent policies, while Figure S20 and Figure S21 show cumulative plots of the disease control outcomes.

The estimated parameters with these calibration variations are described in Table S4 . Both RI and LC are evaluated in the same infection reduction metrics and burden metrics again under scenarios S1, S2, and S3. Since the budgets are structural (mobility, and exposure risk) the LC policies are unchanged among the variants. Moreover, since the burden metrics are structural, those results are invariant.

To evaluate the e cacy of policies, we inspect infection reduction by simulating the disease with contact networks from Fall 2019. Since managed WiFi networks accumulate logs for

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  long periods of time, policymakers can use WiMob to model data from previous semesters and experiment with closure policies like LC. We show that WiMob can provide retrospective disease-mitigating insight into multiple counterfactual scenarios. For instance, policymakers can consider studying seasonal behaviors over multiple semesters for more robustness. Since the underlying data is longitudinal, it provides the flexibility to realistically assess policy interventions at di↵erent time points and also study updating policies. Restricting movement on campus at di↵erent time-points is known to exert varying degrees of control on disease spread [10] . Our data also shows that mobility on campus varies across the semester and therefore, allows policymakers to consider loosening shutdowns depending on the phase of the semester.

Policy design is determined by practical budgets. We model two kinds of budgets, mobility reduction and risk of exposure. The former represents disruptions in space utilization, availing services, and social life. The latter translates to the testing burden on campus. Our analysis determines the budget in di↵erent scenarios by observing the changes to the graph when large classes are moved online. This is to ensure an equitable comparison with targeted policies. However, in real situations these budgets can be relaxed or restricted based on that campus' preparedness to tackle a pandemic. For instance, a hypothetical campus that can test everyone every day might not be constrained by risk of exposure. Alternatively, policymakers can model other tangible budgets such as the capacity in isolation wards or available hospital beds. This can be informed by practical limitations of the campus. Similarly, this paper only assesses limited forms of cost, e.g., students avoiding campus or closing locations. From a financial perspective, university campuses can digitize their core serviceeducation-but still realize losses from other curtailed services [23, 6, 58] . When students avoid campus it can lead to direct losses from meal passes and parking and also quantifiable losses to learning outcomes [2, 16] Policymakers can compute actual costs by complementing this data with information from other sources (e.g., revenue generated by cafes and stores on campus). This can help qualifying WiMob to reflect di↵erent costs and in turn help design policies that optimize for financial losses. Di↵erent campuses have di↵erent priorities and challenges in implementing policies.

We purposefully compare our prototype targeted policies against moving classes online because of practical budgets within the university. Both the WiMob and En based contact networks are derived from archival data accumulated by universities. This does not require instrumenting campus or its community with any new form of surveillance infrastructure. However, its use for a di↵erent purpose demands approval by an IRB. Moreover, acquiring these kinds of data would require collaborating with data-stewards (e.g., the IT department) to establish a data-use agreement. This document must clarify how the data will be deidentified, transferred, and stored.

For this form of data, the critical privacy challenge might not be localization itself, but rather the aggregation of data over a period of time [56] . Data spanning a longer period are more susceptible to cross-analyzing and identifying. To mitigate over-accumulation of data,

CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  we suggest an adherence to principles of data minimization [36] . Instead of storing entire mobility graphs, the campus can compute and preserve only high-level insights, such as the importance of locations. This redacts any underlying individual behavior and corresponding identifiable information. Actually, for future purposes campuses can consider a form of di↵erential privacy that authorizes limited forms of data querying depending on the privileges of the stakeholder [4] .

An operational application would require the university to update the terms of use for its managed network. Particularly, the university should disclose how this data can be used in critical circumstances that invoke shared vulnerabilities [7] . On notifying the campus community of this change it o↵ers individuals the choice to refrain from using the university network. Prior work on a sample within the same university campus shows that 90% of students are connected to the network on any given day [13] . Therefore, proposing such an opt-out condition can be viewed as an unfair choice. As a result, the campus needs to develop a contingency plan to accommodate network access to users who do not want their mobility behavior to constitute the aggregated insights.

This work presents evidence that university campuses can repurpose existing data sources to inform the design of LC policies that can control COVID-19. We evaluate these policies as alternatives to other data-driven, but, broad impact policies that universities consider implementing, such as moving large classes online. One of the drawbacks of this analysis, however, is that it assumes all edges to be the same. For example, when constraining by mobility, in real scenarios losing certain visits might be more valuable than others. Decline in mobility around profit-making services, such as shops and cafeterias, versus losing mobility at common rooms have a di↵erent tangible e↵ects on campus. Currently we take an agnostic stance towards the mobility behavior, where all visits at all locations are the same. In reality, implementing policies could have inequitable qualitative impacts despite appearing to have similar network configuration. This can be improved by embedding more qualitative information into the network and conceiving ingenious ways to associate costs to edges.

Similar to the assumption that all visits and locations, the current work also assumes all people to be equal. However, di↵erent people have di↵erent underlying conditions that can make their vulnerabilities more concerning [43] . The privacy safeguards of this study restricted the research team from acquiring any additional demographic or historical information. Further work can attempt to characterize the nodes by randomly seeding the network to reflect the approximate demographic break up of the community. Alternatively, researchers could try to estimate some demographic based on behavior as well. However, to leverage accurate individual information, even for operational use during a public health emergency, policymakers and researchers need to develop new privacy protocols [28] .

Lastly, this paper only studies three rudimentary scenarios, persistence, non-residential avoidance, complete avoidance. These scenarios assume that when a location is shutdown, the individuals who ought to have visited that location do not come into contact with anyone else during the same time. Yet, other substitution behaviors are possible and the richness of Das Swain, V. et al • 16

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021. ;  networks leveraged with WiMob enables the exploration of various new scenarios that can be triggered by policy interventions on campus. For instance, individuals might not even visit transitory spaces, such as lobbies or cafes between classes. Certain collocations could be the consequence of social ties which might never be developed because of a shutdown (e.g., project teams meeting outside of class). Further research can illuminate the e↵ects of policies in more specific scenarios by modeling post-intervention behavior more accurately.

Das Swain, V. et al • 17

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021 We create a contact network of only students with WiMob and compare it with insights from contact networks created with En. On average, we find the contact network constructed with WiMob shows fewer average contacts, lower density and higher average shortest path (between reachable paths). Moreover, within WiMob itself, characterizing all spaces reveals more contacts and shorter paths than only focusing on contacts in lectures. While the proportion of the largest component appears similar, note that with WiMob, on average about only 70% of the students visit campus on a given week. We further inspect the disease-mitigating structural changes of the RI policy on the network. We observe that the changes across all metrics with En appear to be more drastic than compared to WiMob. The results in the main paper use variables p, ↵, and I 0 as estimated by calibrating the simulation model on the first 5 weeks of positivity rates provided by GT surveillance for Fall 2020, while incorporating external cases from Fulton County. For sensitivity analyses, we perform calibrations on GT data for weeks 5 9 and 10 14. Additionally, we perform calibrations on first five weeks of UIUC and Berkeley positivity rate (along with data from their respective county). These parameters were found by validating the ABM on the remaining weeks of Fall 2020. To assess the basic reproductive number (R 0 ) of our ABM we study the first 4 weeks of the disease. We find the e↵ective R 0 to be higher for Fall 2019 than Fall 2020 as the mobility behaviors between the 2 semesters was vastly di↵erent. Note, Fall 2020 exhibits only 39% of the mobility we observe in Fall 2019. In fact, the ABM is calibrated on Fall 2020, where behavior was subject to pandemic related closures, but in Fall 2019 the mobility was not hindered by any interventions. Thus, Fall 2019 reflects a counterfactual of Fall 2020 without any closures. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 24, 2021 Note that this table is the same as Table 1 . We repeat the results here for easier comparison of LC PRank to other algorithms shown in Table S6 , Table S7 and Table S8 . Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC PRank with RI. We find that LC PRank leads to significantly improved peak infection reduction and internal transmission. In terms of reduction in total infections, the outcomes are comparable in general but can vary by specific scenarios. In addition, every policy also exerts some burden on campus, either in terms of locations a↵ected, students avoiding campus or isolation. We observe that LC PRank policies focus on fewer locations (except in S3). Moreover, these policies a↵ect fewer student's schedules and therefore fewer people avoid campus due to completely remote schedules. Finally, LC PRank does not increase the percentage of people completely isolated on campus (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC BCen with RI. We find that LC BCen leads to significantly improved peak infection reduction and internal transmission, when designed with the exposure risk budget, but can be worse with the mobility budget. In terms of reduction in total infections, the outcomes are typically worse. In addition, every policy also exerts some burden on campus, either in terms of locations a↵ected, students avoiding campus or isolation. We observe that LC BCen policies focus on fewer locations (except in S3). Moreover, these policies a↵ect fewer student's schedules and therefore fewer people avoid campus due to completely remote schedules. Finally, LC LCen does not increase the percentage of people completely isolated on campus (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC ECen with RI. We find that LC ECen leads to significantly improved peak infection reduction and internal transmission. In terms of reduction in total infections, the outcomes vary by specific scenarios. In addition, every policy also exerts some burden on campus, either in terms of locations a↵ected, students avoiding campus or isolation. We observe that LC ECen policies focus on fewer locations (except in S3). Moreover, these policies a↵ect fewer student's schedules and therefore fewer people avoid campus due to completely remote schedules. Finally, LC ECen does not increase the percentage of people completely isolated on campus (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC LCen with RI. We find that LC LCen leads to significantly improved peak infection reduction and internal transmission. In terms of reduction in total infections, the outcomes are comparable in some scenarios but can vary in specific scenarios. In addition, every policy also exerts some burden on campus, either in terms of locations a↵ected, students avoiding campus or isolation. We observe that LC LCen policies focus on fewer locations (except in S3). Moreover, these policies a↵ect fewer student's schedules and therefore fewer people avoid campus due to completely remote schedules. Finally, LC LCen does not increase the percentage of people completely isolated on campus (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC PRank with RI. We find that LC PRank leads to significantly improved peak infection reduction and internal transmission. In terms of reduction in total infections, the outcomes are better in general but can be comparable in specific scenarios. The burden exerted on campus is the same as structural impacts of LC PRank (Table S5 ). (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC PRank with RI. We find that LC PRank leads to significantly improved peak infection reduction and internal transmission. In terms of reduction in total infections, the outcomes are better in general but can be comparable in specific scenarios. The burden exerted on campus is the same as structural impacts of LC PRank (Table S5 ). (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC PRank with RI. We find that LC PRank leads to significantly improved peak infection reduction, internal transmission and total infections. The burden exerted on campus is the same as structural impacts of LC PRank (Table S5 ). (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). Within each scenario, we perform the Kruskal-Wallis H-Test [35] to compare outcomes of LC PRank with RI. We find that LC PRank leads to significantly improved peak infection reduction, internal transmission and total infections. The burden exerted on campus is the same as structural impacts of LC PRank (Table S5 ). (p-value: < 0.01: ⇤ , < 0.001: ⇤⇤ ). . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Figure S6 : We calibrate ABM on positivity rates from Fall 2020 at GT. The objective function of the calibration is to minimize the r.m.s.e. with the weeekly average of positivity rate obtained from surveillance testing results at GT [25] . (a) The parameter that determines external transmission of infections on a given day, I out (t), is a function of cases in Fulton county (where GT is located). (b) The models discussed in the main paper are calibrated using the first 5 weeks of data. We illustrate the output for a range of parameters that incorporate quantitative uncertainty, i.e., within 40% of the r.m.s.e. (c, d) illustrate calibration on the second period of 5 weeks and third period of 5 weeks respectively. These only show the optimal parameter output. The shaded region around the lines show the 2.5 th and 97.5 th percentile.

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Figure S7 : We calibrate ABM on positivity rates from first 5 weeks of Fall 2020 at UIUC. The objective function of the calibration is to minimize the r.m.s.e. with the weekly average of positivity rate obtained from surveillance testing results at GT [25] . (a) The parameter that determines external transmission of infections on a given day, I out (t), is a function of cases in Champaign county (where UIUC is located). (b) We illustrate the output for a range of parameters that incorporate quantitative uncertainty, i.e., within 40% of the r.m.s.e. The shaded region around the lines show the 2.5 th and 97.5 th percentile. Figure S9 : We calibrate ABM on positivity rates from first 5 weeks of Fall 2020 at UC Berkeley. The objective function of the calibration is to minimize the r.m.s.e. with the weekly average of positivity rate obtained from surveillance testing results at GT [25] . (a) The parameter that determines external transmission of infections on a given day, I out (t), is a function of cases in Alameda county (where UIUC is located). (b) We illustrate the output for a range of parameters that incorporate quantitative uncertainty, i.e., within 40% of the r.m.s.e. The shaded region around the lines show the 2.5 th and 97.5 th percentile.

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Figure S10 : Disease control outcomes in Fall 2019 for di↵erent algorithms of LC with the ABM is calibrated on weeks 0 4 of Fall 2020 at GT. (a c) Comparison of RI with LC PRank . Under all scenarios, for peak infection reduction (b) and internal transmission reduction (c), LC PRank shows better disease control outcomes than RI. For total infection reduction (b), LC PRank is better in S1, worse in S3 when designed within an exposure risk budget, and comparable in others. (d f ) Comparison of RI with LC BCen . Under all scenarios, for peak infection reduction (d) and internal transmission reduction (f ) LC BCen is better when designed within an exposure risk budget. For total infection reduction (e), LC BCen is always worse than RI . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Figure S11 : Disease control outcomes in Fall 2019 for di↵erent algorithms of LC with the ABM is calibrated on weeks 0 4 of Fall 2020 at GT. (a c) Comparison of RI with LC ECen . Under all scenarios, for peak infection reduction (b) and internal transmission reduction (c), LC ECen shows better disease control outcomes than RI. For total infection reduction (b), LC ECen is better in S1 and worse in S3 when designed within an exposure risk budget. (d f ) Comparison of RI with LC ECen . Under all scenarios, for peak infection reduction (d) and internal transmission reduction (f ), LC ECen shows better disease control outcomes than RI. For total infection reduction (e), LC ECen is better in S1 and worse in S3 when designed within an exposure risk budget.

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) (a) S1 (ABM calibrated on UIUC data) (b) S2 (ABM calibrated on UIUC data) (c) S3 (ABM calibrated on UIUC data) (d) S1 (ABM calibrated on UIUC data) (e) S2 (ABM calibrated on UIUC data) (f) S3 (ABM calibrated on UIUC data) (g) S1 (ABM calibrated on UIUC data) (h) S2 (ABM calibrated on UIUC data) (i) S3 (ABM calibrated on UIUC data) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Figure S22 : The locations shutdown by each policy are grouped into the the general building category. The distribution of locations is di↵erent between policies, for example, in S1 (a) and S2 (b), LC closes fewer locations that RI. Even when targeting spaces in similar buildings, the locations are qualitatively di↵erent -RI only a↵ects classrooms, whereas LC also closes smaller spaces like breakout rooms, reading areas and cafes. LC In S3 (c) we find LC to target locations in a greater variety of buildings, but it also targets more locations to utilize the budget.

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Figure S23 : The locations shutdown by each policy are grouped into the the general building category. The distribution of locations is di↵erent between policies, for example, in S1 (a) and S2 (b), LC closes fewer locations that RI. Even when targeting spaces in similar buildings, the locations are qualitatively di↵erent -RI only a↵ects classrooms, whereas LC also closes smaller spaces like breakout rooms, reading areas and cafes. LC In S3 (c) we find LC to target locations in a greater variety of buildings, but it also targets more locations to utilize the budget.

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 24, 2021. ; https://doi.org/10.1101/2021.03.16.21253662 doi: medRxiv preprint

Early evidence on social distancing in response to covid-19 in the united states

Simulating the potential impacts of COVID-19 school closures on schooling and learning outcomes: A set of global estimates

Association between mobility patterns and covid-19 transmission in the usa: a mathematical modelling study. The Lancet Infectious Diseases

Ancile: Enhancing privacy for ubiquitous computing with use-based privacy

Potential community and campus covid-19 outcomes under university and college reopening scenarios. medRxiv

Rationing social contact during the covid-19 pandemic: Transmission risk and social benefits of us locations

Controlling the spread of covid-19 on college campuses

Aggregated mobility data could help fight covid-19

Estimating the impact of school closure on influenza transmission from sentinel data

Mobility network models of covid-19 explain inequities and inform reopening

Undergraduate and graduate students' mental health during the covid-19 pandemic

Characterizing social networks using the sociometer

Leveraging wifi network logs to infer social interactions: A case study of academic performance and student behavior

Jue insight: Were urban cowboys enough to control covid-19? local shelter-in-place orders and coronavirus case growth

Here's a look at the impact of coronavirus

Covid-19 and student learning in the united states: The hurt could last a lifetime

Reality mining: sensing complex social systems

Strategies for mitigating an influenza pandemic

Disease Control and Prevention. Community npis: Flu prevention in community settings

Disease Control and Prevention. People at increased risk and other people who need to take extra precautions

Covid-19 impact on higher education

Covid-19: the conjunction of events leading to the coronavirus pandemic and lessons to learn for future threats

Surveillanceto-diagnostic testing program for asymptomatic sars-cov-2 infections on a large, urban campus -georgia institute of technology

Simulating covid-19 in a university environment

Strategies to support the covid-19 response in lmics

Universities as anchor institutions: Economic and social potential for urban development

Ranking the e↵ectiveness of worldwide covid-19 government interventions

Covid-19: risk factors for severe disease and death

Projecting the transmission dynamics of sars-cov-2 through the postpandemic period

Simulating the impacts of interregional mobility restriction on the spatial spread of covid-19 in japan. medRxiv

Colleges begin mapping out a more normal fall-with caveats

Use of ranks in one-criterion variance analysis

Rapid systematic review: the impact of social isolation and loneliness on the mental health of children and adolescents in the context of covid-19

High resolution proximity statistics as early warning for us universities reopening during covid-19. medRxiv

Next flu pandemic: what to do until the vaccine arrives?

Schools briefing: University outbreaks and parental angst

of Public Health. Georgia department of public health daily status report

Critical preparedness, readiness and response actions for covid-19: interim guidance

The pagerank citation ranking: Bringing order to the web

Privacy and ethical issues in location-based tracking systems

Large-scale automatic depression screening using meta-data from wifi infrastructure

A new front in america's pandemic: College towns

The small-world network of college classes: implications for epidemic spread on a university campus

Quantifying seasonal population fluxes driving rubella transmission dynamics using mobile phone data

Signs of depression and anxiety soar among us graduate students during pandemic

Dynamic modelling of costs and health consequences of school closure during an influenza pandemic

Mounting peril for public higher education during the coronavirus pandemic

Learning individual behavior using sensor data: The case of gps traces and taxi drivers. Forthcoming in Information Systems Research

Staggered release policies for covid-19 control: Costs and benefits of relaxing restrictions by age and risk

Early evidence on social distancing in response to covid-19 in the united states

Simulating the potential impacts of COVID-19 school closures on schooling and learning outcomes: A set of global estimates

Ancile: Enhancing privacy for ubiquitous computing with use-based privacy

Rationing social contact during the covid-19 pandemic: Transmission risk and social benefits of us locations

Some unique properties of eigenvector centrality

Controlling the spread of covid-19 on college campuses

The ethical imperatives of the covid 19 pandemic: A review from data ethics. Veritas: Revista de Filosofía y Teología

Mobility network models of covid-19 explain inequities and inform reopening

Wi-fi location-based services 4.1 design guide white paper

Leveraging wifi network logs to infer social interactions: A case study of academic performance and student behavior

Real-time data of the impact of covid-19

Champaign-urbana covid-19 coronavirus information

Covid-19 and student learning in the united states: The hurt could last a lifetime

Presence analytics: making sense of human social presence within a learning environment

A set of measures of centrality based on betweenness

Surveillanceto-diagnostic testing program for asymptomatic sars-cov-2 infections on a large, urban campus -georgia institute of technology

Simulating covid-19 in a university environment

Strategies to support the covid-19 response in lmics

The impact of social distancing on covid19 spread: State of georgia case study

Use of ranks in one-criterion variance analysis

A model of covid-19 transmission and control on university campuses. medRxiv

The cost of covid-19: A rough estimate of the 2020 us gdp impact

Jue insight: College student travel contributed to local covid-19 spread

Convergence of the nelder-mead simplex method to a nonstationary point

The implications of silent transmission for the control of covid-19 outbreaks

Wifi location monitor accuracy

Big data in public health: terminology, machine learning, and privacy. Annual review of public health

Scientific collaboration networks. ii. shortest paths, weighted networks, and centrality

Schools briefing: University outbreaks and parental angst

On-campus covid-19 testing

of Public Health. Georgia department of public health daily status report

Georgia tech launches campus coronavirus testing

The pagerank citation ranking: Bringing order to the web

Mental health and the covid-19 pandemic

A terminology for talking about privacy by data minimization: Anonymity, unlinkability, undetectability, unobservability, pseudonymity, and identity management

E cacy of tocilizumab in patients hospitalized with covid-19

Forecasting influenza activity using machine-learned mobility map

Privacy and ethical issues in location-based tracking systems

Large-scale automatic depression screening using meta-data from wifi infrastructure

A new front in america's pandemic: College towns

The small-world network of college classes: implications for epidemic spread on a university campus

Modeling between-population variation in covid-19 dynamics in hubei, lombardy, and new york city

Revisiting the small world of college classes: The trade-o↵s between risk and burden of partial remote instruction