key: cord-0121046-bd9j1din
authors: Cho, Jungwoo; Shin, Yuyol; Kim, Seyun; Kim, Namwoo; Oh, Soohwan; Cho, Haechan; Yoon, Yoonjin
title: Running the COVID-19 marathon: the behavioral adaptations in mobility and facemask over 27 weeks of pandemic in Seoul, South Korea
date: 2020-09-09
journal: nan
DOI: nan
sha: 48068c5666cddaa88afcc6297161fd56cac2ae93
doc_id: 121046
cord_uid: bd9j1din

Battle with COVID-19 turned out to be a marathon, not a sprint, and behavioral adjustments have been unavoidable to stay viable. In this paper, we employ a data-centric approach to investigate individual mobility adaptations and mask-wearing in Seoul, South Korea. We first identify six epidemic phases and two waves based on COVID-19 case count and its geospatial dispersion. The phase-specific linear models reveal the strong, self-driven mobility reductions in the first escalation and peak with a common focus on public transit use and less-essential weekend/afternoon trips. However, comparable reduction was not present in the second wave, as the shifted focus from mobility to mask-wearing was evident. Although no lockdowns and gentle nudge to wear mask seemed counter-intuitive, simple and persistent communication on personal safety has been effective and sustainable to induce cooperative behavioral adaptations. Our phase-specific analyses and interpretation highlight the importance of targeted response consistent with the fluctuating epidemic risk.

Mobility intervention has been embraced as one of the most effective and immediate control measures since the early days of COVID-19 pandemic, and universal quarantines such as lockdown and shelter-in-place have shown measurable success. Sjödin et al. 1 conducted a modeling study to show that the lockdown duration could be shortened when household size and time spent outside one's home were brief. Since household size could not be easily changed, the study concluded that strict quarantine measures are the key to reduce its duration. Leung et al. 2 showed that reproduction number has dropped below 1 after lockdowns in 14 Chinese provinces, and the fatality risk could also be decreased. Based on the Susceptible-Infectious-Recovered model, the study projected that premature relaxation of lockdown measures would increase the cumulative case count exponentially. López and Rodó 3 used a stochastic Susceptible-Exposed-Infectious-Recovered to project the epidemic progression after lockdown, suggesting that another larger scale infection wave is likely if lockdown measures were alleviated prematurely and abruptly.

Although strong mobility stringency can minimize uncertainties arising from individual differences to achieve prompt and well-coordinated response, high social and economic costs have been inevitable. Suppressing personal mobility was found to deepen economic inequalities 4,5 , generate disconnects among communities 6 , and even cause mental illness 7 . As the pandemic prolonged, concerns over accumulating costs and exit strategies have deepened. Phased exit strategy, prioritizing people at the highest risk, and managing misinformation were suggested to reduce unintended cost in the post-lockdown period 8 .

Another key control measure of massive testing and contact-tracing has shown promising outcomes in several countries including South Korea 9 . However, a recent study found that such approaches are feasible only when case numbers remain at a manageable level, and the success of prevention measures depends on the population's behavioral changes in response 10 . Hellewell et al. 11 stated that contact tracing and isolation without mobility reduction might not be sufficient to contain the outbreak under plausible scenarios.

Even if the testing-contact-tracing has been effective to contain the outbreak, it cannot serve as a direct incentive to induce behavioral adaptations. The main motivation of our study is consistent with the aforementioned studies, and we aim to evaluate the role of mobility when case numbers remain at a traceable level. Several recent studies also emphasized the need to understand behavioral aspects and perception changes regarding policy adherence 12 , and the importance of mobility data analyses for a broader understanding of the efficacy of public communication and social distancing interventions 13 . Despite the urgent need for adaptive mobility policy targeting behavioral adaptations, there are limited academic and technical resources available. It is encouraging, however, that an increasing number of researches utilize large-scale near-real time datasets to study the interrelationship among mobility change, social impact, and policy response 5, 6, 14 .

In this paper, we employ data-centric approaches to trace the evolution of mobility behavioral adaptation and its relationship to epidemic progression in Seoul, South Korea for the first 189 days (27 weeks). As the 4 th country infected with COVID-19, the nation had been the worst-affected country outside China in the beginning 15 . However, the nation has managed to control the infection rate under 20.99 per million population since January 20, 2020 without lockdown-type mandates.

The nation's capital Seoul also has been able to contain the maximum daily confirmed cases under 52 or 5.34 per million population in the same period with minimal mobility restrictions.

Due to the time-dependent and transient nature of epidemic progression and mobility patterns, modeling their relationship over the entire 27 weeks is meaningless. As the first step, we propose a framework to identify epidemic phases based on the case count as well as the geospatial spread of COVID-19. The mobility patterns are represented with the reductions in subway ridership (public transit) and traffic volume (private mode) compared to the same period of 2019, given the mobility network under full operations. Association between mobility and epidemic progression is then modeled as linear models in six epidemic phases using a changing point detection technique.

Seasonality specific to mobility including the day-of-week and the time-of-day effect is also evaluated to discover common focus in mobility decisions. Lastly, findings from the statistical analyses are assessed with risk perception and facemask use in a collective manner.

The major transitions in epidemic progression were determined by detecting the major structural changes in daily case counts with a changing point detection (CPD) method of breakpoint detection algorithms 16, 17 . The algorithms simultaneously detect the major interventions in sequential data, and have been widely used in multiple disciplines including finance 18-21 , genetics 22,23 , climate change 24,25 , social science 26-28 , ecology 29 , and geology 30,31 . In our analysis, the intercept-only linear model of = 0 + was assumed to detect the day of major structural change, and used in the COVID-19 case count analysis and the mobility seasonality analysis.

Although the breakpoint analysis method detects the structural changes in a robust manner, sequential data in general represent long-term trends in a retrospective manner, and is sensitive to the timespan under investigation. To complement such shortcomings, we propose to measure the geospatial progression of COVID-19 in addition to the confirmed case count. Conceptually, it is intuitive that quantifying spatial dispersion provides useful insights for an infectious disease like COVID-19, as the risk cannot be the same when 100 new cases are located in a single location versus spread evenly across a region. Methodologically, we employed two topological measures of grouped distance ( ) and Hausdorff distance ( ). Since both measures represent the distance between two sets, one can reduce the two-day changes into single metrics. If the daily case sequence represents the long-term trend, the topological measures are designed to capture the short-term momentum.

Given the set of locations exposed by COVID-19 patients, which we call the contact locations, is the geospatial peak. In the analysis, the metric − was used to identify the geospatial peak period compared to the geospatial contraction and expansion.

Once the epidemic phases of trigger, escalation, peak, and de-escalation phases were established using both the count interventions and the topological metric, the relationship between mobility and epidemic progression was modeled as linear models by individual phases. In the bivariate linear model, the response variable is the mobility reduction, and the COVID-19 count is the regressor.

There are four main datasets used in the study -the hourly subway ridership and traffic volume, Regarding risk perception and universal mask adoption rate, we resorted to several independent survey results due to the lack of direct observation data and its highly qualitative nature. Two types of risk perception-overall risk of COVID-19 and risk of personal infectionwere obtained from biweekly surveys conducted by Hankook Research. Overall risk of COVID-19 represents the percentage of people who responded that the spread of the pandemic within the nation was "very" or "somewhat" serious. Risk of personal infection represents the percentage of people who responded that the likelihood of contracting the virus was more than or equal to 50% 32 .

The rate of universal mask adoption was based on five independent survey outcomes 33-37 , which captured the proportion of the surveyed having worn masks whenever necessary. Additional details and availability of data used are included in the supplementary.

In Figure 1 -a, the breakpoint analysis results of COVID-19 case count trend are shown with four interventions, yielding five phases divided on day 50, 82, 128, 156. In Figure 1 it is unlikely that the city had entered a de-escalation phase with − being mostly positive.

As results, we defined six epidemic phases as the trigger (1-29), escalation-1 (30-50), peak-1 (51-82), de-escalation (83-106), escalation-2 (107-128), and peak-2 (129-189). 

Linear model results are shown in Figure 2 . In subway ridership, an addition of a single case resulted in an additional 4.1% reduction in escalation-1, which converted to a 0.5% recovery in escalation-2. In traffic volume, the additional reduction was 1.3% in escalation-1, and a 0.1% recovery in escalation-2. Between two peaks, the constant 40.8% subway ridership reduction in peak-1 was decreased to 20.9% in peak-2. Traffic volume shows recovery from an 8.8% reduction to 4.2% with a 0.1% additional recovery per additional new case.

The results provide several useful insights. The reduction ratio between public transit (subway) to private trip mode (traffic volume) was approximately 4 to 1, manifesting the strong shift in mode preference. One can also infer that the reductions of approximately 40% in public transit use and 

Since mobility patterns are embedded with strong seasonality, mobility patterns were evaluated for its time-of-day and day-of-week effects. As shown in Figure 3 , the overall trends of two trip modes are similar except in magnitude, and public transit demand continues to recover as early as day 63 (week 9). The subway ridership shows reductions at least three times larger than traffic volume, which agrees with our findings in section 3.2. It is also worth noting that the subway has continued operations during the epidemic, serving more than 50% of the usual demand. Despite the strong preference shift to private mode, public transit has remained a vital urban mobility option, especially to those with limited mode choices.

It is notable that the first major interventions of two modes agree on day 34 (week 5), which is before any major government response is put in place. Moreover, weekends, nighttime, and afternoon trips show larger reductions compared to weekdays and commute hours. Under low policy stringency, such outcomes strongly suggest that early mobility reduction had been driven by proactive individual adaptations with a common focus on reducing non-essential trips while preserving the essential commute trips in private mode. Continuous recovery after the deescalation signifies changes in public perception regarding the role of mobility, which agrees with our findings in section 3.2. 

In this section, subway ridership reduction is assessed with the public risk perception, as it is the main trigger of the behavioral adjustment. In Figure the key characteristics of epidemic phases, as it increased during two escalation phases, remained over 70% during two peaks, and decreased during the de-escalation.

It is notable that the universal facemask wearing had already reached over 60% in escalation-1. Considering the official promotion on universal face covering had started and remained as a recommendation from day 63 (peak-1) to day 217 (peak-2), the public's awareness on the benefit of facemask had been in place much earlier than the government recommendation. As some studies suggested 38 , the high compliance level is likely the result of the past infectious outbreak experiences amplified by the current epidemic risk. Since escalation-2, mask use exceeded both the personal infection risk perception and the subway ridership reduction. The mask-wearing experiences in the earlier period seemed to have led the public to embrace the facemask as the longer-term personal safety measure, especially in returning to one's mobility routines.

It is also worth noting that the personal risk perception remained higher than the overall epidemic risk perception in the second wave. The government's social distancing level also stayed at the lowest Level 1 1 during peak-2, since its level is solely determined by the case count trend.

When the geospatial momentum is considered, it is evident that Seoul had missed the opportunity to contain the second peak, and the inopportune timing of national economic stimulus to promote in-person shopping and dining since day 106 39 . However, the high level of public's risk perception remained as the main force to prevent rapid resurgence despite the reactive government response.

COVID-19 is the percentage who responded that the spread of the pandemic within the nation was "very" or "somewhat" serious 32 . Risk of personal infection is the percentage who responded that the likelihood of contracting the virus was more than or equal to 50% 32 . The rate of universal mask use 33-37 is the proportion who responded they have worn masks whenever necessary such as in public space.

In this paper, behavioral changes during COVID-19 outbreak were evaluated in Seoul, South COVID-19 is a pandemic we have never experienced in our modern history, but it might not be the last. Learning from the evolving evidence on behavioral adaptations and the resulting public cooperation will have a long-lasting impact to fight this pandemic, and to prepare for the next.

However, understanding behavioral changes amid the unprepared pandemics such as COVID-19 is a tall order. As future research, we suggest the following topics for the broader academic communities to explore.

In mobility, our study specifically focused on trip modes in mobility pattern mining, since they are direct observations and have been in service the whole period. From the transportation economics perspective, however, not only the trip mode but also the trip purposes need to be evaluated to fully capture the behavioral aspects, as the mode is a mere means to achieve one's trip purpose. Therefore, expansive data analytics on trip purposes are imperative for adaptive and targeted mobility control. Moreover, analyses on trip purpose may validate the efficacy of various mobility controls, and to identify the high-risk sectors. Incorporating demographic groups such as age, gender and occupations may also broaden our understanding.

In epidemiology, a broader set of metrics might be helpful to monitor and track the progression closer to real-time. In the prolonged battle with COVID-19, the importance of timely response empowered by fast data tracking is no less than searching for the ultimate medical interventions.

The changing point detection method and geospatial progression measures effectively captured the progression of the outbreak on multiple ends. Further geometric pattern mining methodology such as topological data analysis 43 may reveal the spatial characteristics even further, which has critical implications in the infectious outbreaks as COVID-19. The necessary condition of data availability and methodological validation must be met, and requires active and collaborative contributions of broader academic disciplines.

In policy, there needs to be a framework to evaluate the efficacy and relevance of key control measures in a systematic and collective manner. Although the testing-contact tracing-isolating has proved successful in South Korea 44 , the process of contact tracing had unintended social impacts, especially on the underrepresented populations 45-46 . Moreover, adherence to contact tracing has put growing pressure on the nation's universal healthcare system, and required an enormous sacrifice of the healthcare professionals 47 . Our study outcome strongly suggests that the public is willing to adapt to a reasonable level of mobility restriction, which can alleviate the enduring pressure on the national healthcare infrastructure. Contribution of facemasks and social distancing needs to be considered as a set of policies than of competing interests.

The data that support the findings of this study except contact tracing information are available at https://doi.org/10.6084/m9.figshare.c.5015093 48 .

Custom code that supports the findings of this study is available from the corresponding author contributed equally.

The authors declare no competing interests.

Supplementary Information is available for this paper.

Correspondence and requests for materials should be addressed to Y. Y. The second edition of "the disclosure of itinerary and information of confirmed patients" was published on April 12 (day 84). This edition stated that contact information should be disclosed only for 14 days after the patient had the last contact. The time range was extended to two days before the date of symptom onset or testing. This edition specified that each government should notify the public about the sterilization status of places visited by patients.

The third edition was announced on June 30 (day 163). This edition emphasized that each patient's information should be deleted 14 days after the patient had the last contact. The guideline for protecting personal information has also been changed so that gender, age, nationality, detailed address (only provide Gu-wise information) of residence, and names of workplaces are not disclosed. Since this edition, information to be released has been in the form of a list of visited places, including region, type, name, and address of the facility, exposed date, and status of sterilization, instead of the itinerary form. In addition, if all contacts in a particular place have been identified, the information of the place visited by a patient "should not" be disclosed to the public. The information of cluster cases is to be announced by KCDC, not by municipalities. Supplementary Table 3 * Contacts indicate people that a COVID-19 confirmed patient may have been in contact with.

The scope of contacts is determined considering symptoms of the patient, mask-wearing status, duration of the stay, and exposed situation. All contacts, both symptomatic and asymptomatic, are subject to self-quarantine for 14 days with daily monitoring by public health staff.

The 

Mobility restriction. To back up our statement that the nation has had marginal mobility restriction, we summarize government responses adopted throughout COVID-19 (see Supplementary Table 5 ). We also quantify the stringency of policy responses, by adopting scoring criteria and method of Oxford COVID-19 Government Response Tracker (OxCGRT) 53 .

Specifically, we adopted two indices-government response index and mobility restriction index-to show that mobility restriction has been relatively marginal compared to overall government responses (see figure S1 ). Government response index, which was first proposed in the working paper of OxCGRT, is based on 13 policy indicators including workplace closure and stay-at-home requirement 9 . Mobility restriction index is calculated based on only 3 indicators that directly constrain mobility: public transport closure, stay-athome requirement, and internal movement restriction. The types of indicators as well as the scoring method used in this study are solely based on OxCGRT 9 . Scoring codebooks and relevant sources are detailed in Supplementary Table 5 .

In summary, the government has not mandated stay-at-home for the public, as of Aug 13.

The official advice has been a recommendation on not-leaving-house, even throughout strict social distancing period (day 63-91). Public transport has not been closed during COVID-19.

Internal movement restriction was also on recommending to refrain from traveling from/to special disaster zones. Please see Supplementary Table 5 for a detailed description and references.

Facemask policy responses. KCDC's official advice for mask-wearing before day 63 was that mask-wearing is only recommended for those showing symptoms, those caring for people with symptoms, those visiting medical facilities, those working in environments with a high risk of infection, and those with underlying illness 54 . Only since strict social distancing (day 63), KCDC has recommended the general public to wear masks in crowded indoor spaces 55 .

Mask wearing has been also mandatory in public transport in Seoul from day 115 and in the nation from day 128 56 . 

Only strict quarantine measures can curb the coronavirus disease (COVID-19) outbreak in Italy

First-wave COVID-19 transmissibility and severity in China outside Hubei after control measures, and second-wave scenario planning: a modelling impact assessment

The end of social confinement and COVID-19 re-emergence risk

Financial markets under the global pandemic of COVID-19

Economic and social consequences of human mobility restrictions under COVID-19

Population mobility reductions during COVID-19 epidemic in France under lockdown

Safe but lonely? Loneliness, mental health symptoms and COVID-19

Ten considerations for effectively managing the COVID-19 transition

Effectiveness of isolation, testing, contact tracing, and physical distancing on reducing transmission of SARS-CoV-2 in different settings: a mathematical modelling study

Impact of non-pharmaceutical interventions (NIs) to reduce COVID 19 mortality and healthcare demand

Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts

How behavioural science data helps mitigate the COVID-19 crisis

Aggregated mobility data could help fight COVID-19

Lockdown timing and efficacy in controlling COVID-19 using mobile phone tracking. EClinicalMedicine, 100457

What South Korea Teaches The World About Fighting COVID

Computation and analysis of multiple structural change models

Testing and dating of structural changes in practice

References (supplementary)

Coronavirus Disease-19. Ministry of Health and Welfare

Data Science for COVID-19 in South Korea

Travel Log of Confirmed Patients in Seoul

The Government of the Republic of Korea. Flattening the curve on COVID-19: How Korea responded to a pandemic using ICT

Subway Ridership Data

Seoul Traffic Volume Count Data

COVID-19 Mobility Data Collection of Seoul

Variation in Government Responses to COVID-19 Version 6.0. Blavatnik School of Government Working Paper

Guideline on recommended facemask use

Stronger Social Distancing for 15 Days, Starting with the Government!

Face masks mandatory on public transport

This work was supported by the National Research Foundation of Korea (NRF) grant (No. 2020R1A2C2010200). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

-The need to postpone or cancel the collective event is low, and it is recommended to carry out various events with sufficient quarantine measures 34~ It is advised to refrain from crowded events in a narrow indoor space, including religious events or outdoor events CDSCHQ