One City Two Tales Using Mobility Networks to Understand Neighborhood Resilience and Fragility during the COVID-19 Pandemic

2025-05-02 0 0 2.74MB 21 页 10玖币

侵权投诉

One City, Two Tales: Using Mobility Networks to

Understand Neighborhood Resilience and Fragility

during the COVID-19 Pandemic

Hasan Alp Boz1,+, Mohsen Bahrami2,*,+, Selim Balcisoy1, Burcin Bozkaya3, Nina Mazar4,

Aaron Nichols4, and Alex Pentland2

1Sabanci University, Faculty of Engineering and Natural Sciences, Istanbul, 34956, Turkey

2Massachusetts Institute of Technology, Institute for Data, Systems, and Society, Cambridge, MA 02139, USA

New College of Florida, The Graduate Program in Applied Data Science, 5800 Bay Shore Rd, Sarasota, FL 34243,

USA

4Boston University, Questrom School of Business, Boston, MA 02215, USA

*bahrami@mit.edu

+These authors contributed equally to this work. Authors are listed alphabetically.

ABSTRACT

What predicts a neighborhood’s resilience and adaptability to essential public health policies and shelter-in-place regulations

that prevent the harmful spread of COVID-19? To answer this question, in this paper we present a novel application of human

mobility patterns and human behavior in a network setting. We analyze mobility data in New York City over two years, from

January 2019 to December 2020, and create weekly mobility networks between Census Block Groups by aggregating Point

of Interest level visit patterns. Our results suggest that both the socioeconomic and geographic attributes of neighborhoods

signiﬁcantly predict neighborhood adaptability to the shelter-in-place policies active at that time. That is, our ﬁndings and

simulation results reveal that in addition to factors such as race, education, and income, geographical attributes such as access

to amenities in a neighborhood that satisfy community needs were equally important factors for predicting neighborhood

adaptability and the spread of COVID-19. The results of our study provide insights that can enhance urban planning strategies

that contribute to pandemic alleviation efforts, which in turn may help urban areas become more resilient to exogenous shocks

such as the COVID-19 pandemic.

Introduction

Mobility in urban and metropolitan settings is the result of the dynamic interaction, over space and time, of a large number

of agents with diverse goals and characteristics. Understanding mobility patterns is crucial for predicting future social and

economic well-being as well as growth

1,2

, as it has been shown to reﬂect social interactions

3,4

, productivity

5,6

, and economic

prosperity and resilience7–10.

The dynamic interactions in mobility data are complex and are typically represented by network structures. However,

despite the ability for network analyses to yield unique and critically important insights

, researchers have yet to use network

analyses to explore how the COVID-19 pandemic and the shelter-in-place and social distancing policies that it triggered – two of

the most effective non-pharmaceutical interventions (NPI) aimed at reducing the spread of the Corona virus

12,13

(for a differing

conclusion, see a recently published article by Berry et al.

) – inﬂuenced human mobility and interaction patterns. Such

interventions directly impact the human mobility network by reducing levels of mobility and interactions among individuals

and points of interest (POIs) such as restaurants or supermarkets

15,16

. The decline in movement activities during the current

COVID-19 pandemic has negatively impacted economies all around the world, and these economic repercussions are predicted

to continue for years17.

Although network analysis approaches are well suited to investigate the multifaceted impact that policy changes have on

mobility outcomes, early studies have examined the short-term impacts of the COVID-19 pandemic on various unidimensional

mobility metrics such as travel distance, visit patterns, and dwelling times in different countries and cities

18,19

, or used

mobility networks along with epidemiological models to understand the effects of changes in mobility patterns on the spread

of the Corona virus

15,20

. Indeed, there is a lack of research investigating how the COVID-19 pandemic and corresponding

interventions impacted the structure of mobility networks (e.g. the centrality metrics at a node and network level). For example,

investigating the betweenness centrality metric may reveal the role certain neighborhoods played during the pandemic, such as

arXiv:2210.04641v1 [physics.soc-ph] 6 Oct 2022

acting as a bridge among the nodes in a mobility network, thereby producing a spreader effect.

Given the close link between changes in mobility patterns and economic outcomes, it is critical that research investigates

how various environmental and demographic factors have inﬂuenced the adaptability of mobility networks during the COVID-19

pandemic. Using network science methodologies, the current research aims to help scientists and policy-makers understand the

factors that contribute to economic resilience and adaptability in the wake of economic shocks. In particular, we extend previous

research about the impact of the COVID-19 pandemic on human behavior by examining changes in mobility patterns using a

dynamic network analysis on one of the most important economic hubs in the world: New York City (NYC). We create weekly

mobility networks, from January 2019 to December 2020, in which nodes represent neighborhoods (i.e. Census Block Groups

or CBGs), and the edges between them correspond to the visitors from the source neighborhood to POIs such as restaurants or

supermarkets in the target neighborhood. For each neighborhood in the weekly networks, we compute node and ego-network

based features21 so the resulting feature vectors not only capture the dynamics of local mobility but also the relationship with

the neighboring CBGs, allowing us to incorporate the complexity of mobility into our analyses. We investigate the dissimilarity

of the resulting feature vectors for each neighborhood, between the same weeks of 2019 and 2020, and break the results down

by different socioeconomic groups. Combining the mobility network metrics with data from various sources, including census

data and COVID-19 test results, we are able to reveal how differences in NYC neighborhood characteristics predict dynamic

structural changes of mobility networks and behavior.

The results of our study indicate that the centrality metrics and geographic attributes signiﬁcantly predict neighborhood

adaptability to shelter-in-place orders. In addition to conﬁrming the results of previous research

17,20,22,23

, our ﬁndings reveal

that not only are race, education, and income important factors in predicting neighborhood adaptability to shelter-in-place

orders, but so are geographical attributes such as access to diverse amenities that satisfy community needs. This indicates that

in the same city, communities with similar socioeconomic and demographic features may have different mobility responses

based on their neighborhoods’ urban structure.

Using the information extracted from the mobility network structure, we study the case of COVID-19 hotspots to investigate

which neighborhoods act as the COVID-19 bridges among the hotspots and other neighborhoods, and uncover the associated

factors. We then utilize the well-known Huff gravity model to perform a hypothetical scenario analysis to show how the higher

levels of access to essential businesses (e.g. grocery stores) could reduce the interaction among the COVID-19 hotspots and

other neighborhoods that could potentially lead to a reduction in the infection rates and save more lives. These novel results

provide signiﬁcant insights and recommend policies that can enhance urban planning strategies that contribute to pandemic

alleviation efforts, which in turn may help urban areas become more resilient to exogenous shocks such as the COVID-19

pandemic that restrict movements and interactions.

Results

To demonstrate the impact of the COVID-19 pandemic on different socioeconomic groups, we ﬁrst analyze the change in

network topologies in a weekly resolution at the neighborhood level. To this end, we extract the node-level feature vectors

summarizing the statistical properties of their respective ego-networks

. The resulting node feature vectors are used to compute

the dissimilarities between paired weekly networks from 2019 and 2020. Then, we illustrate the course of centrality metrics

with respect to different demographic groups and highlight their variability. Next, we analyze the possible COVID-19 bridges,

neighborhoods that frequently interact with COVID-19 hotspots (CBGs with higher numbers of infected residents), by focusing

on the in- and out-going edges between CBGs over two-week periods, which is considered as the virus incubation time

Finally, utilizing the Huff Gravity Model, we analyze the mobility in Staten Island using hypothetical Grocery Store densities in

order to observe the change in visits to hotspot CBGs that frequently appear in the top new COVID-19 cases quartile.

Demographic Disparities: Temporal Changes in Mobility Networks

CBG-level dissimilarity analysis

We compute the node-level dissimilarity scores between paired weekly networks of 2019 (pre-pandemic) and 2020 (pandemic)

using the extracted ego-network feature vectors (the components of these feature vectors and dissimilarity score formulation

are explained in the Methods section). The CBGs are then ranked with respect to their dissimilarity scores at each time step

(week). To demonstrate the differences between the CBGs with distinct behaviors, we focus on the CBGs in the top and bottom

dissimilarity quartiles, and create two cohorts of CBGs that frequently appear in those quartiles during the ﬁrst wave of the

pandemic between March and June 2020. Figure 1shows the spatial and socioeconomic distribution of the resulting cohorts of

CBGs that appear in at least 60% of the time steps in the top and bottom quartiles. This threshold (60%) is the highest frequency

that yields a similar number of CBGs in each group, enabling a better comparison between them. CBGs that ended up changing

their mobility pattern the most (i.e. the top dissimilarity quartile) are primarily located in the Manhattan borough, the ﬁnancial

center of NYC. From all CBGs in this cohort, 63% of them rank in the top quartile for income, 79% in the top education

quartile, 62% in the top white population percentage quartile, and 52% in the bottom quartile for commute time, meaning they

2/13

either do not travel relatively long distances to get to their workplace or are located in areas that have greater access to fast and

frequent transportation. There is no evident socioeconomic proﬁle for the bottom dissimilarity quartile (the CBGs that did not

change their mobility patterns much), although the distributions in quartiles delineate the residents to some extent. However,

there exists a decreasing trend from bottom to top quartiles in income, education, and white population percentage.

Figure 1.

The socioeconomic distribution of the census block groups (CBGs) that changed their mobility patterns the most in

comparison to the previous year in at least 60% of the time steps, versus the least. CBGs that ended up changing their mobility

patterns the most are primarily located in the ﬁnancial center of NYC. Note that there are only signiﬁcant socioeconomic

characteristics for the top quartiles but not for the bottom.

Node Degree and Centrality Metrics

Centrality metrics help us examine a node’s role in the network, such as inﬂuence and information diffusion

24,25

. In the

proposed mobility network, since each node is a CBG, the centrality metrics highlight the CBGs that are noteworthy in terms of

the ﬂow of masses.

In this context, the temporal changes in centrality metrics reveal the interaction patterns between different socioeconomic

communities, which consequently indicate complex mobility behaviors from a network perspective. To this end, we focus on

basic node degree centrality metrics and analyze their course of change in distinct demographic groups (i.e. CBGs in the top

and bottom quartiles). We use the node betweenness to illustrate a CBG’s importance based on its connections and position in

the network. Furthermore, we use degree centrality metrics to reveal the weekly incoming and outgoing visitors among CBGs.

Lastly, we use a custom metric named self-visit ratio to represent the fraction of visits to the POIs inside the home CBG.

Betweenness:

This centrality score measures how frequently a CBG appears along the shortest paths in a network, and

is the only node centrality metric that demonstrates a signiﬁcant difference between the selected demographic groups. As

displayed in Figure 2-A, CBGs in the top income quartile held a higher betweenness value until the beginning of March 2020

(start of the pandemic), meaning that they played a critical role in terms of bridging the ﬂow of masses. However, an abrupt

decrease of betweenness in the top income CBGs took place after the start of the pandemic, while less afﬂuent CBGs gained

higher betweenness scores. That is, less afﬂuent CBGs increasingly acted as connectors among the nodes in the mobility

3/13

network but only until September 2020, when the economic activity revived. The same relationship can also be observed when

focusing on education levels. CBGs with lower education levels had a higher betweenness score in the same time interval

(displayed in the Supplementary Information (SI) Figure 4).

Figure 2. The temporal change in (A) betweenness, (B) total-degree, and (C) self-visit ratio metrics in the top and bottom

income quartiles. The vertical line segments show a 95% conﬁdence interval.

Degree:

Node degree analysis indicates that income and education play a signiﬁcant role in distribution of degree centrality

values as well. Afﬂuent CBGs were more successful at lowering their mobility compared to less afﬂuent neighborhoods as

shown in Figure 2-B.

Self-Visit Ratio:

Another metric we deﬁned to investigate the change in visit patterns is called self-visit ratio. Self-visit

ratio is the fraction of the visits made by the residents of a CBG to their home CBG over all visits paid by them. Figure 2-C

displays the course of the self-visit ratio with respect to the top and bottom income quartiles. From March to June 2020, during

the ﬁrst wave and the most striking decline in mobility, CBGs in the top income quartile had a higher rate of visits to the POIs

inside their home CBGs, while on the contrary, the residents of lower income CBGs displayed a lower self-visit ratio. However,

starting in June 2020, as the re-opening of economic activity commenced, the disparity of self-visit ratio between income

groups is narrowed.

Analysis of COVID-19 Hotspots, Bridge CBGs & the Case of Staten Island

As explained in the Methods section, we deﬁne the COVID-19 hotspots as those CBGs that frequently appear among the CBGs

with highest weekly new cases. Additionally, we refer to the CBGs that have a high level of interactions with hotspots in

the beginning of the virus incubation time as COVID-19 bridge CBGs. Examining the COVID-19 bridge CBGs may reveal

invaluable insights for policy makers and urban planners attempting to prevent the spread of new infections and build cities that

are resilient to future pandemics.

To this end, we ﬁrst obtain the CBGs in the top weekly new cases quartile for each time step

. Then, we create a list of

CBGs that had edges to the previously obtained CBGs with the highest cases in time step

t−2

considering a period of two

weeks as the incubation time for new cases to surface

. Subsequently, we apply a frequency analysis on the possible bridges

to check how often they were connected to CBGs with the highest weekly cases, and ﬁnally, consider the CBGs in the

75th

frequency percentiles as bridges, for further analyses.

As demonstrated in Figure 3, the majority of the resulting CBGs in the

75th

frequency percentile are comprised of those

in the lower quartiles for income and education and higher quartiles for commuting time. Yet the spatial distribution of the

4/13

potential bridges in the

75th

frequency percentile in Figure 3unveils the special case of Staten Island, where 85% of the CBGs

located in Staten Island appeared in the bridges. Moreover, as the threshold value is increased to the

95th

frequency percentile,

the demographic features begin to display Staten Island’s presence. Figure 4shows the box plot of the COVID-19 bridge CBGs

at a borough level. Additionally, the results of our OLS regression analysis considering occurrence in the bridge CBG set as a

dependent variable and borough code as an independent variable, showed the boroughs are signiﬁcant in predicting occurrence

of CBGs in the bridge set (the regression analysis results are provided in SI). This is counter-intuitive and does not align with

the previous observations, because 48% of the CBGs in Staten Island are from high income quartiles and their residents are

mainly white. That is, the CBGs in Staten Island display a distinctive behavior compared to CBGs in other boroughs that are

similarly in high income and high white population quartiles. This observation is particularly noteworthy since Staten Island is

geographically relatively isolated with a low level of connectivity to other boroughs (connected through bridges and ferry; and

not connected with the NYC subway system). Thus, if anything, one may conclude the opposite, that is, Staten Island would

have been more shielded from the pandemic. Therefore, we followed up with additional analysis at the borough level to try to

shed light on this observation.

Figure 3. Spatial and demographic distributions of the CBGs in the 75th (top) and 95th (bottom) frequency percentiles as

COVID-19 bridges. Staten Island clearly stands out.

Borough Level Analyses Results

As depicted in Figure 5from January 2019 to January 2021, residents in Staten Island always had the highest mobility (average

visit count per smartphone user) among all NYC boroughs, regardless of the COVID-19 pandemic and the rise of different

containment policies (e.g. shelter-in-place or physical/social distancing).

Since the Safegraph mobility data has limited coverage on workplaces and ofﬁces, we use Google’s COVID-19 Community

Mobility Reports

to investigate the mobility trends for places of work. As shown in Figure 6Staten Island has the minimum

relative change in mobility trends for workplace among all NYC boroughs, indicating that the residents of Staten Island reduced

their mobility and visits to POIs noticeably less than the residents of other boroughs.

Additionally, the POIs analysis results show that Staten Island has the lowest number and diversity of POIs among all

boroughs of NYC, where the majority of visits to POIs inside the city originating from Staten Island are made to Brooklyn

and Manhattan, which are the neighboring boroughs of Staten Island. This observation is in line with a report by the NYC

government’s planning department

documenting that 24% of workers residing in Staten Island have their workplaces located

in Manhattan.

5/13

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

OneCity,TwoTales:UsingMobilityNetworkstoUnderstandNeighborhoodResilienceandFragilityduringtheCOVID-19PandemicHasanAlpBoz1,+,MohsenBahrami2,*,+,SelimBalcisoy1,BurcinBozkaya3,NinaMazar4,AaronNichols4,andAlexPentland21SabanciUniversity,FacultyofEngineeringandNaturalSciences,Istanbul,34956,Turkey2Massac...

展开>> 收起<<

One City Two Tales Using Mobility Networks to Understand Neighborhood Resilience and Fragility during the COVID-19 Pandemic.pdf

共21页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

One City Two Tales Using Mobility Networks to Understand Neighborhood Resilience and Fragility during the COVID-19 Pandemic

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: