Applied Geography 154 (2023) 102941

Contents lists available at ScienceDirect

Applied Geography
journal homepage: www.elsevier.com/locate/apgeog

Seeing the forest and the trees: Holistic view of social distancing on the
spread of COVID-19 in China
Danlin Yu a, *, 1, Yaojun Zhang b, **, 1, Jun Meng c, 1, Xiaoxi Wang d, Linfeng He d, Meng Jia d,
Jie Ouyang d, Yu Han d, Ge Zhang e, Yao Lu f
a
Department of Earth and Environmental Studies, Montclair State University, Montclair, NJ, 07043, USA
b
School of Applied Economics, Renmin University of China, Beijing, 100086, China
c
Department of Obs.&Gyn., Beijing Youan Hospital, Capital Medical University, Beijing, 100069, China
d
School of Sociology and Population Studies, Renmin University of China, Beijing, 100086, China
e
School of Management, Minzu University of China, Beijing, 100081, China
f
School of Ethnology and Sociology, Minzu University of China, Beijing, 100081, China

A R T I C L E I N F O A B S T R A C T

Handling Editor: J Peng The human social and behavioral activities play significant roles in the spread of COVID-19. Social-distancing
centered non-pharmaceutical interventions (NPIs) are the best strategies to curb the spread of COVID-19 prior to
Keywords: an effective pharmaceutical or vaccine solution. This study investigates various social-distancing measures’
COVID-19 impact on the spread of COVID-19 using advanced global and novel local geospatial techniques. Social distancing
Global and local spatial analysis
measures are acquired through website analysis, document text analysis, and other big data extraction strategies.
Geographically weighted panel regression
A spatial panel regression model and a newly proposed geographically weighted panel regression model are
Social distancing
China applied to investigate the global and local relationships between the spread of COVID-19 and the various social
distancing measures. Results from the combined global and local analyses confirm the effectiveness of NPI
strategies to curb the spread of COVID-19. While global level strategies allow a nation to implement social
distancing measures immediately at the beginning to minimize the impact of the disease, local level strategies
fine tune such measures based on different times and places to provide targeted implementation to balance
conflicting demands during the pandemic. The local level analysis further suggests that implementing different
NPI strategies in different locations might allow us to battle unknown global pandemic more efficiently.

1. Introduction Law, & Butt, 2022; Redlin, 2022). A recent study on how social
distancing impacts the spread of influenza pandemics (Fong et al., 2020)
The outbreak and rapid spread of COVID-19 in the early days of 2020 found that social distancing, including the isolation of the infected,
in China stricken the world with unprecedented damage. During the active contact tracing and quarantine, prevention of public gathering
initial periods of the outbreak, China faced with the dilemma of con­ including school and workplace closures helped reduce the rate of
taining the spread of the disease with little knowledge of how. Epide­ transmission and flatten the curve of the spread. Social distancing allows
miological practices suggest that effective non-pharmaceutical the disease to spread at a lower rate over a longer period, which enables
intervention (NPI) strategies that center on social distancing and early the health system to respond in a more efficient way and gives time for
quarantine are critical to curb the spread of an unknown, infectious pharmaceutical solutions to be devised to combat the disease. An
disease (Davies et al., 2020; Dayaratna, Gonshorowski, & Kolesar, 2022; investigation of the social distancing measures during the first 50 days of
Fair, Karatayev, Anand, & Bauch, 2022; Flaxman et al., 2020; Nazia, the pandemic in China suggests that intra-city public transportation ban,

* Corresponding author.
** Corresponding author.
E-mail addresses: yud@mail.montclair.edu (D. Yu), zhyaojun@ruc.edu.cn (Y. Zhang), mengjun_98@yahoo.com (J. Meng), wangxiaoxi2016@ruc.edu.cn
(X. Wang), helingfeng@ruc.edu.cn (L. He), jiameng0209@ruc.edu.cn (M. Jia), jie.ouyang@ruc.edu.cn (J. Ouyang), hyzxshyx@ruc.edu.cn (Y. Han), gezhang@
muc.edu.cn (G. Zhang), aria_ly0118@muc.edu.cn (Y. Lu).
1
The authors contribute equally to this work.

https://doi.org/10.1016/j.apgeog.2023.102941
Received 31 March 2022; Received in revised form 21 January 2023; Accepted 18 March 2023
Available online 27 March 2023
0143-6228/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
D. Yu et al. Applied Geography 154 (2023) 102941

large gathering ban successfully delayed the spread of the disease (Tian and the spread of the COVID-19 disease. This “trees” view provides a
et al., 2020). micro perspective of the modeled relationship and might provide a
Still, the real situation, as the pandemic stretches, was more complex critical supplement for the holistic view for policy implementation and
because the spread rate of COVID-19 and what might work better to curb decision-making.
its spread might not be the same everywhere as witnessed in many The current investigation explores the forest view from a spatial
studies in China (Kraemer et al., 2020; Xiong, Wang, Chen, & Zhu, analytical perspective as some have attempted (X. Guo et al., 2021;
2020), UK (Davies et al., 2020), Italy (Ciufolini & Paolozzi, 2020; Xiong et al., 2020), but focuses on the trees view to establish a new
Giordano et al.), Germany (Chae & Park, 2020), the US (Auger et al., modeling strategy and new modeling paradigm that can strike a balance
2020). It is hence critical to consider the effectiveness of different social between better saving lives from the pandemic and meeting the de­
distance based NPI policies at different times and places from the loca­ mands for other socioeconomic needs. We envision the study contributes
tion specific perspective. Different locations have different speed of to the applied geographic research community by proposing a research
responsiveness, different policies regarding a never-before-seen disease, paradigm that starts with a global vision, delves deeper with local var­
different natural and infrastructural conditions, but more importantly, iations, couples advanced global with novel local spatiotemporal
different human social and behavioral responses in dealing with a con­ analytical models to provide a holistic view of relationships between
tagious disease that are often not only related to governmental policies stimulus and outcomes by considering the spatiotemporal effects that
but also individual knowledge and the general cultural environments are embedded in the collected data in the applied geographical com­
(Cevasco et al., 2021; S. Guo et al., 2020; X. Guo, Zhang, & Wu, 2021). munities. This study fills a gap in empirical spatial epidemiological
Based on studies in the early stages of the COVID-19 pandemic (Bertozzi, studies that emphasizes on pragmatic and actionable local strategies. In
Franco, Mohler, Short, & Sledge, 2020; Brauner et al., 2021; Drew et al., addition, this study promotes the spatial holistic research paradigm to
2020; J. Huang et al., 2020; Poletto, Scarpino, & Volz, 2020), accurate explore the effectiveness of various NPI strategies and how such stra­
estimation and prediction of the spread of the disease on different lo­ tegies prevent the spread of COVID-19 at both the macro and micro
cations is critical to estimate necessary medical requirement and ca­ levels. The spatial holistic research paradigm suggests that “seeing both
pacities, enable decision makers to allocate limited medical resources the forest and the trees” contributes to the growing spatial data analysis
during the initial outbreak, reduce the loss of human lives, and flatten knowledge base of spatial epidemiology studies. The research is ex­
the curve of the spread. Many predictive epidemiological models have pected to supply the governments more efficient strategies against
been established prior to or during the outbreak of COVID-19 (Ciufolini future global pandemic, allow the governments to act more responsively
& Paolozzi, 2020; Dehning et al., 2020; Dowd et al., 2020; Enserink & and responsibly, flatten the curve of spread, save lives, and enable the
Kupferschmidt, 2020; S. Guo et al., 2020). The experiences and con­ society to return to normal earlier.
clusions from the models are in consensus: reliable short-term prediction After this introduction section, we will detail our data and method­
is the key to the success of combatting the spread, flattening the curve ology. This is followed by the results from the models in the third sec­
especially in the initial stages of the spread, regardless of where the tion, and discussion of these results in the fourth section. We conclude
prediction was made and how it was made (Baker et al., 2020; Chen our study with a summary of the findings of the study.
et al., 2022; Ciufolini & Paolozzi, 2020; Davies et al., 2020; Dayaratna
et al., 2022). 2. Data and methods
Most such models and predictions, however, investigate the impacts
of NPI on the spread of COVID-19 from a global perspective (we define 2.1. Measuring social distancing via big data: the social, behavior and
this as the forest view). That is to say, the models are built on the premise policy responses
that NPI strategies will work universally, regardless of locations that
have varied degrees of the diseases occurring and spread, and social, Social distancing measures vary from study to study. The essence of
economic, demographic, and natural conditions. Results from those social distancing, however, is to prevent individual physical closeness as
global models are valuable because they provide timely and important would normally be required under usual social interaction scenarios.
predictions of what might work and what might not work from a macro The current study extends previous investigation on social distancing
perspective, which will enable the country to take immediate actions at measures’ effect (Tian et al., 2020) to May and attempts the “forest and
the initial stages of the outbreak as China did to save lives and flatten the trees” research paradigm with data mined from non-traditional sources
curve (Chinazzi et al., 2020; Wei & Wang, 2020; Xiong et al., 2020). In (other than official reporting and records) for China, as the first country
addition, under the premise of providing legitimate short-term predic­ that reported the outbreak and the country that implemented one of the
tion, oftentimes models need to rely on available information. Many strictest social distancing measures to curb the spread of the disease. We
varying conditions regarding the places and times are either unavailable chose the dates from the start of the pandemic to early May (May 6th)
or less reliable to be used for better modeling practices under the global because on April 8th, Wuhan, the epicenter of the pandemic, finally
modeling framework. As to whether similar strategies will work or work decided to relax its strict local quarantine lock-down. We did not stop
to the same degree in different locations, the global level models are not our data collecting points at exactly April 8th, however. Instead,
equipped to answer these questions. Yet answers to these questions considering many studies suggest that COVID-19 has a 5–8 days of in­
could mean not only actual difference between life and death for in­ cubation period (Alene et al., 2021; Y. Y. Cai et al., 2020; Cheng et al.,
dividuals in different places during the spread of COVID-19, but also a 2021; Cimolai, 2021; Guan et al., 2020; Quesada et al., 2021; Wan, Liu,
balance between strict lockdown and meeting the local demands for & Liu, 2021; Wu et al., 2022; L. Zhang et al., 2021), and in clinical
adequate and flexible socioeconomic activities. Without the detailed practices, a 14-day extended quarantine period is often adopted in the
knowledge, all decisions might come down to a choice between strict policy sphere (Charvadeh, Yi, Bian, & He, 2022), we extended our data
national level lock-down, which carries prohibitively high socioeco­ collection period to the early May (May 6th) until one of the data entry,
nomic cost as witnessed in China (Chinazzi et al., 2020), or a the Baidu migration index, is no longer available.
half-hearted, hesitant lock-down that might not work at all as the US With the help of search engine, website text and policy analyses, the
situation pans out in the past year (S. Guo et al., 2020; Manchein, study team identified eight proxies that could roughly represent the
Brugnago, da Silva, Mendes, & Beims, 2020), and we are now registering many different social distancing measures in China during the outbreak
over a million lost lives. For this matter, to combat the spread of the of the pandemic. These proxies include the intensity of within city travel
disease more efficiently and save human lives at various locations, (travelcity), inter-city travel ban (intercity, 1 for ban, 0 not), the intensity
models need to also look at the individual “trees” (individual locations) of daily migration (addmig), the daily net change of the number of in­
that might have different relationships between the mitigation strategies dividuals for each prefecture (netmin), school closures (school, 1 for

2
D. Yu et al. Applied Geography 154 (2023) 102941

closure, 0 not), workplace closures (work, 1 for closure, 0 not), the differences in place attractivity due to socioeconomic or natural reasons
number of tourist sites comments (tourists), and quarantine status (Yu, Zhang, & Wu, 2020), daily change of the number of individuals,
(quarantine, 1 for full prefecture lock down, 0 otherwise). All data are however, might reflect a dynamic of spontaneous movement primarily
acquired daily from January 11th, 2020, to May 6th, 2020, the period because of personal reasons (such as visiting friends, doing business,
when it is closest to the mandatory quarantine of Wuhan City on January personal shopping trips, among others, especially with the convenience
23rd, 2020, and the gradual reopening since April 8th, until one of our of high-speed rail system) instead of regional socioeconomic differences.
data entry, the Baidu migration index was no longer available after May The hypothesis is that when more people move in a prefecture, it will be
6th, 2020. more likely for COVID-19 to spread. This social distancing measure is
The data used in this study is acquired at the prefecture adminis­ hypothesized to be positively related with the outcome variable.
trative level. In China, a prefecture is the sub-provincial administrative The school, work, intercity, and quarantine are all binary variables in
unit that is composed of a central district and many counties. Counting that 1 represents a positive action that is designated to curb the spread of
the four provincial level municipalities (Beijing, Tianjin, Shanghai, and COVID-19 (so 1 represents closure of schools and workplaces, shutdown
Chongqing) as such prefectures and excluding Hong Kong, Macau, and of intercity transportation, and enabling prefecture-wide quarantine),
Taiwan where not all data is available, the study counts 359 analytical and 0 otherwise. These four variables are all hypothesized to be nega­
units in China. The study uses the number of daily confirmed newly tively related with the outcome variable.
added cases of COVID-19 for each prefecture as the outcome variable. The tourists data attempts to capture individual behaviors that might
This daily data is readily available from Tencent’s COVID-19 real-time not follow regional or national policies entirely. A heighted number of
reporting website (https://news.qq.com/zt2020/page/feiyan.htm? tourists might suggest the higher probability for the spread of COVID-19.
from=timeline&isappinstalled=0#/?nojump=1). The eight proxies for The relationship is hypothesized to be positive with the outcome
social distancing measures are acquired through website analysis, policy variable.
related text analysis and document analysis. The intensity of within city Because the relationship between these social distancing measures
travel (travelcity), the intensity of daily migration for each prefecture and the spread of COVID-19 is well defined (Tian et al., 2020), for
(addmig), and the daily net change of the number of migrants for each ensuing regression models, the tests of coefficients’ significance are all
prefecture (netmin) are acquired through the Baidu migration website based on one-way test instead of the commonly reported two-way test.
(https://qianxi.baidu.com/#/2020chunyun). A Python code is devised This test applies to both the global and local models (the forest and trees
by one of our team members to acquire and organize the three types of views).
data. A team of researchers also combed through news reports and
published policy documents from January 11th to May 6th, 2020 via 2.2. The spread of COVID-19 and the indirect inversed normal
Baidu web searches using the keywords “school closure,” (Fengxiao) transformation supported regression model
“workplace closure” (Tinggong), “quarantine” (Fengcheng), or “intercity
transportation closure” (Quji Tingyun) and the relevant prefecture’s Daily data of the COVID-19 in China at prefecture administrative
name to acquire school closures (school, 1 for closure, 0 not), workplace level is acquired through Tencent’s COVID-19 website. The current
closures (work, 1 for closure, 0 not), quarantine status (quarantine, 1 for study focuses on the daily newly confirmed cases since this item is
full prefecture lock down, 0 otherwise), and closure of inter-city trans­ representative of the spread of the diseases. Studies of the initial cases in
portation (intercity, 1 for closure, 0 not) for each prefecture, and the time Wuhan suggest that COVID-19 has an average incubation period of 5–6
points those measures were enacted and lifted through these searches. days (Anderson, Heesterbeek, Klinkenberg, & Hollingsworth, 2020;
This process was done through a Python code for intelligent sematic Charvadeh et al., 2022; Guan et al., 2020; Wu et al., 2022). The quar­
analysis that picks up the key words, the prefectures’ names, and sort the antine practice adopted by many countries requires either mandatory or
queries according to the time stamp it was posted during our study pe­ self-quarantine for a period of 14 days (Guan et al., 2020; Wan et al.,
riods’ Baidu Search results. Furthermore, a Python code searching the 2021). To see how the NPI strategies prevent the spread of COVID-19,
Dazhong Dianping (www.dianping.com) website for each prefecture’s we use the 14 days lagged NPI strategies as the intervention input
popular tourism sites’ daily comments was also devised to collect the (meaning for any one day’s daily newly confirmed cases, the NPI stra­
number of comments. This data (tourists) is used as a proxy for potential tegies used in the model are the ones implemented 14 days ago). For this
travels initiated by individuals without utilizing public transportation. matter, our initial analysis date moves from January 11th, 2020, to
The purpose of social distancing is to prevent close contact among in­ January 25th, 2020, which is also one day after the traditional Chinese
dividuals from both global (city) and local (individual) levels. Increased Lunar New Year, when normally the largest inter and inner-city move­
social distancing policies, societal, and individual behaviors that prevent ments are recorded. All the data in their raw format of the 103 days over
close contact will be the efficient ways to curb the spread of COVID-19. the 359 prefectures are summarized in Table 1.
The detailed description of each social distancing measures we have To investigate the potential impacts of social distancing on the
collected data for this study and how it is related with the spread of spread of COVID-19, this study adopts regression models with the daily
COVID-19 are as follows. newly confirmed cases as the outcome, and the social distancing
The travelcity variable measures the percentage of the total number
of people who attempted any types of travel activities in the total pop­ Table 1
ulation of a prefecture. More people who are moving about during the Summary of the raw data.
pandemic will increase the risk of spreading the disease. This social Min 1st Median Mean 3rd Max
distancing measure is hypothesized to be positively related with the Quantile Quantile
outcome variable.
newly − 107 0 0 2.149 0 13,436
The addmig variable represents the total number of people that move added
in or out of a prefecture for any given day. This variable measures the cases
intensity of voluntary movement of individuals even under the strict travelcity 0.300 2.960 4.212 3.939 4.889 8.878
lock-down policy has been instituted. Higher intensity will exacerbate addmig 0.003 0.345 0.795 1.351 1.588 31.228
netmin − 23.115 − 0.087 − 0.013 0.000 0.041 6.533
the spread of COVID-19. This social distancing measure is hypothesized tourist 0 5 12 18.550 25 125
to be positively related with the spread of COVID-19. school 0 1 1 0.839 1 1
The netmin variable measures the difference between the numbers of work 0 0 0 0.267 1 1
people who moves in and out of a prefecture. Longer term net change of quarantine 0 0 1 0.531 1 1
intercity 0 0 0 0.306 1 1
the number of individuals for each prefecture is usually a reflection of

3
D. Yu et al. Applied Geography 154 (2023) 102941

measures as explanatory variables. In addition to these time variant [ ]

rank(ui ) − k
daily COVID-19 cases and social distancing measures, it is important to INT(ui ) = ф− 1

note that these daily variant data (both cases and responses) are influ­ n − 2k + 1
enced by and associated with the geographically variant but temporally Here ф− 1 is the normal density function, k ∈ (0, ½) is an adjustable
invariant background information. The WHO developed a “social de­ offset, and n is the sample size. By default, the Blom offset of k = 3/8 is
terminants of health” (SDOH) framework (S. Guo et al., 2020; WHO adopted and the transformed variables tend to be very close to normally
Commission on Social Determinants of Health W. H. O, 2008) that em­ distributed (S. Guo et al., 2020). Replacing ui with εit , we obtain the INT
phasizes the strong relationship between societal background and health transformed outcome variable, which is now continous and normally
outcomes, and relates social and behavioral responses towards potential distributed.
health risks. The societal backgrounds often include socioeconomic In the third step, because the INT transformation of the residuals εit is
status, health system capability, and infrastructure status. This study not a linear transformation, the transformed outcome variable will be
adds a demographic status to the SDOH framework since the amount of again related with the daily invariant background information (S. Guo
population in China also plays a crucial role in the spreading of the et al., 2020; McCaw et al., 2019). To mitigate this issue, again separately
disease. Based on these four categories of social determinants of health for each time point, regress the time-variant NPI social distancing
and demographic background information, we collected data of the total measures xikt on the time-invariant covariates wi and obtain the re­
population (demographic category), local economic status (prefecture siduals for each social distancing measures and treat those residuals as
GDP per capita, portion of the secondary industry in GDP, portion of the the new proxies for the social distancing measures, ̃ xikt , (k is the number
tertiary industry in GDP, local financial income per capita, local finan­ of social distancing measures). In so doing, both time variant and
cial expenditure per capita, socioeconomic category), health care system invariant information is incorporated, and the zero-inflated, highly
status (the number of hospitals per 10,000 individuals, the number of skewed outcome variable (14 day lagged newly confirmed case data) is
doctors per 10,000 individuals, the number of hospital beds per 10,000 also transformed to a normally distributed variable. The I-INT trans­
individuals, health system category), and infrastructure status (the formed variables of the 103 days over the 359 prefectures is summarized
density of roads, and the average road network travel time to the nearest in Table 2. As can be seen from Table 2, the outcome variable is now
high-speed rail station, infrastructure category). All data are collected perfectly normally distributed.
from the 2019 China’s statistical yearbook or calculated based on the After the data is preprocessed through these three stages of indirect-
2019 open street map road network (openstreetmap.org) and high-speed INT approach, we further generated smoothed scatterplots between the
rail information, the most recent such data (Yu, Murakami, et al., 2020; outcome (daily newly confirmed cases) and the five NPI strategies.
Yu, Zhang, Wu, Li, & Li, 2021). Approximate linear relationships can be reasonably assumed (straight
To evaluate the impact of social distancing on the daily spread of lines for all scatter plots generated through smoothing across all data
COVID-19, a three-steps indirect inverse-normal transformation (I-INT) points) from the smoothed scatterplots. We can then proceed to apply
strategy (McCaw, Lane, Saxena, Redline, & Lin, 2019) is adopted to the spatial panel and geographically weighted panel regression models
preprocess the data. The preprocessed data is then used to model how to investigate both the forest and the trees views of how social-distancing
social distancing measures curb the spread of COVID-19 with both measures curb the spread of COVID-19.
global and local models. Detailed methodology discussions follow.
2.2.2. The spatial panel regression method
2.2.1. The indirect inversed normal transformation (INT) method Regional studies suggest that data collected over geographic space
The I-INT method is recently proposed in a Genome-Wide Associa­ often exhibits strong spatial effects (Anselin, 1988b; Elhorst, 2014) that
tion Studies (McCaw et al., 2019) that shows efficient and unbiased prevent regular regression estimators from producing consistent and
properties when handling many tie-values and skewed variables because unbiased estimation because the residuals of the regular regression es­
of zero-inflation. Our COVID-19 caseload data (lagged daily newly timators are spatially dependent on one another (spatial autocorrela­
added cases) has many tie-values, and is zero-inflated. The raw data by tion) (Elhorst, 2014). Spatial panel regression model is developed in the
no means follows a normal distribution and using the raw data for early 2000s and fully explained in LeSage and Pace (2009) and Elhorst
regression estimation will not be tenable. In addition, while we have (2014). The fundamental premise is that as data is collected over
collected daily changing caseload information and NPI strategies data geographic units, the First Law of Geography “that everything is related
through web searches and mining, we have also collected some socio­ with everything else, but closer things are more related” (Tobler, 1970)
economic background data for the prefectures as outlined in the data dominates the data generating process, which causes the nonspatial
section. To investigate the relationship between the spread of COVID-19 regression analysis to produce spatially autocorrelated residuals, lead­
and the social distancing measures, we intend to not only consider the ing to biased, inefficient and/or misleading results when using the or­
association between the NPI social distancing measures and the spread dinary least squares estimator (Yu & Wei, 2008). The maximum
of COVID-19, but also the socioeconomic background information. likelihood based alternative estimator is the valid choice for estimation.
While the spread of COVID-19 and the social distancing measures
changed daily (temporally variant), the socioeconomic background in­
formation stayed constant on a daily interval (temporally invariant). To Table 2
incorporate both the temporally variant and invariant information to the Summary of I-INT transformed data.
modeling scheme, the indirect-INT method takes three steps. Min 1st Median Mean 3rd Max
In the first step, separately for each time point t ∈ {1, …, T}, regress Quantile Quantile
the outcome variable yit (lagged daily newly added cases) on the time-
newly − 2.922 − 0.608 0.000 0.000 0.608 2.922
invariant covariates wi (the socioeconomic background information) added
to obtain the residuals εit . cases
In the second step, conduct INT on the residuals zit ≡ INT(εit ) to travelcity − 4.042 − 0.390 0.028 0.000 0.447 4.047
obtain the Z-scores, again separately for each time point t. Detailed for addmig − 9.108 − 0.330 − 0.032 0.000 0.245 15.257
netmin − 13.676 − 0.118 0.002 0.000 0.111 5.982
the INT procedure follow. Suppose u is a skewedly distributed variable tourists − 83.229 − 7.364 − 1.167 0.000 6.077 112.207
with many tie-values (0s in our study). Let rank(ui) denote the sample school − 0.954 0.000 0.000 0.000 0.036 1.013
rank of ui when the measurements are placed in ascending order. The work − 1.278 − 0.033 0.000 0.000 0.006 0.993
inverse normal transformation (INT) for each ui is then defined as: quarantine − 1.150 − 0.050 0.000 0.000 0.023 1.074
intercity − 1.001 − 0.082 0.000 0.000 0.061 1.041

4
D. Yu et al. Applied Geography 154 (2023) 102941

Anselin (2002) and Elhorst (2014) detailed the estimation and testing 2.2.3. The geographically weighted panel regression analysis
procedures through Lagrange Multiplier tests. According to Elhorst To investigate the “trees” view of how social distancing measures will
(2014), three types of interaction effects exist and two of them might curb the spread of COVID-19 at individual prefectures, this study at­
contribute to spatial autocorrelation of the residuals. The first is the tempts a newly developed exploratory spatial data analytical strategy,
spatial autocorrelation among the dependent/outcome variables the geographically weighted panel regression (GWPR) analysis to
(endogenous interaction). The second is the spatial autocorrelation investigate potentially varying relationships between the mitigation
among the errors (error interaction) because of omitted spatially auto­ strategies and the spread of COVID-19. The GWPR analytical procedure
correlated predictors (independent variables). The third type of inter­ (R. H. Cai, Yu, & Oppenheimer, 2014; Yu, 2010, 2014) is a newly
action is the exogenous interaction in which the dependent variable of a developed exploratory spatial data analysis approach extended from
spatial unit depends on the independent variables of neighboring spatial cross-sectional geographically weighted regression (GWR) analysis
units. This interaction, however, will not cause the regression residuals (Fotheringham, Brunsdon, & Charlton, 2002). The fundamental premise
to be spatially autocorrelated. Depending on which interaction causes for GWR is that regressed relationships are not likely the same from
the residuals’ spatial autocorrelation, two types of spatial panel place to place as suggested by conventional regression analysis because
regression model are often adopted: the spatial lag model that assumes of different geographic backgrounds (including socioeconomic, cultural,
the residuals’ spatial autocorrelation is the result of the outcome vari­ demographic, and natural conditions). Through introducing a small
able’s spatial autocorrelation (endogenous interaction), and the spatial bias, cross-sectional GWR analysis often reduces the variance of esti­
error model that assumes the residuals’ spatial autocorrelation is the mated coefficient quite significantly hence the analysis provides better
result of missing spatially autocorrelated explanatory variables (error confidence of the estimation (Fotheringham et al., 2002). The approach
interaction). has seen wide application in many disciplines. Extending the analysis
Per Elhorst (2014), a full panel model considering all three types of from cross-section to including temporal dimension has been pioneered
interactions can be written as: by Yu (2014), Yu (2010), R. H. Cai et al. (2014), B. Huang, Wu, and
Barry (2010), Fotheringham, Crespo, and Yao (2015), among others.
Y t = ρWY t + αιN + Xt β + WXt θ + μ + ξt ιN + μt
Situating GW approach with the panel setting, however, is only explored
in (R. H. Cai et al., 2014; Yu, 2010, 2014).
μt = λW μt + εt
In general, the static panel model is formulated as follows:
Where Yt is the outcome variable at time t. WYt is the endogenous ( )
interaction, also called the spatial lag. W is a spatial weight matrix that y1:T = X1:T β + G1:T γG + H1:T γH + ε1:T , ε1:T ∼ N 01:T , σ2 I1:T
defines the neighboring relationship among states. For contiguous areas,
where T represents time and y1:T is an NT × 1 vector that stacks the
either the rook or queen adjacency rule works fine (Anselin, 1988b,
outcome variables vectors at time 1 to T. N is the number of individuals.
1992; Anselin, Bera, Florax, & Yoon, 1996). For non-contiguous areas, a
Other matrices and vectors with the subscript 1 : T are defined similarly.
graphic based sphere of influence (SOI) approach is often employed
X is the matrix of explanatory variables. G1:T γG captures the individual
(Bivand, Pebesma, Gomez-Rubio, & Pebesma, 2008). The SOI approach
or group-specific effects where G1:T is an NT × KG matrix of KG indi­
defines neighborhood relationships in a set of geographic objects (go) as
vidual dummy variables indicating district, race, sex, and so on (Greene,
such: for any geographical object in the set, goi, let ri be the distance from
2003) and γG is a KG × 1 coefficient vector. H1:T γH captures temporal
goi to its nearest neighbor in the set, and Ci is the circle centered on goi
effects where H1:T is an NT × KH matrix indicating time, such as day,
with the radius of ri, then i and j are SOI neighbors if and only if Ci and Cj
month, and year. γH is a KH × 1 coefficient vector. The static panel model
intersect in at least 2 places. ρ is the coefficient for the spatial lag. ιN is a
is often estimated as an individual one-way model (namely, the term H1:
vector of 1s, Xt is the matrix of predictor variables, β is the vector of
coefficients of the predictors. WXt is the exogenous interaction, and θ the TγH is often dropped).
Noted that the coefficient vector β in equation (1) does not change
vector of its coefficients. μ is the unobservable, individual specific ef­
from location to location, hence the static model. The geographically
fects, and ξt is the time-specific effects. μt is the error term. Wμt is the
weighted extension of the static panel model allows the coefficient
error interaction, and λ its coefficient. εt is the independent and identi­
vector β to change over locations (but remain constant over temporal
cally distributed (i.i.d.) random noise. Because of nesting all three
periods for panel analysis). The geographically weighted panel regres­
interaction effects, this model is called General Nesting Spatial (GNS)
sion model can then be written as:
model (Elhorst, 2014).
( )
Although it is tempting to estimate the GNS model with defined y1:T = X1:T β(ui ,vi ) + G1:T γG + H1:T γH + ε1:T , ε1:T ∼ N 01:T , σ 2 I1:T
spatial weight matrix because the sources of spatial autocorrelation in
the regression residuals are likely from both endogenous and error in­ (ui, vi) is the coordinate pairs of location i. Everything else remains
teractions, the sources usually can only be weakly identified (Elhorst, the same.
2014). For this argument, in practice, either the endogenous (called the The introduction of the spatially varying coefficient immediately
spatial autoregressive model, SAR) or error interaction (called the introduces either a parsimonious problem that we have more unknowns
spatial error model, SEM) will be considered to control for spatial than data if the number of explanatory variables is more than the
autocorrelation in the regression residuals. The choice of either SAR or number of temporal periods, or the collapse of the panel data to a
SEM needs to consider carefully depending on which source might be collection of individual time series estimation if we have longer time
more likely to introduce spatial autocorrelation to the residuals from a periods as in the current study. In the parsimonious scenario, restrictions
theoretical perspective. A robust Lagrange Multiplier (LM) test (Anselin, of the spatially varying mechanisms must be introduced for the co­
1988a) that is based on the loglikelihood of the alternatives and null efficients to be estimable (the geographical weighting). In the collapse
models can provide guidance from an empirical perspective. Technique scenario, though the coefficients are now varying from place to place,
details can be found in Anselin (1988a) and Croissant and Millo (2019) the variation is not really “spatial” in the way a geographically weighted
and will not be repeated here. In addition, depending on how the indi­ approach intends hence estimation that expands from individual loca­
vidual effects are assumed to be fixed or generated from a random dis­ tions will be required. However, if the geographical weighting is not
tribution, panel model can be estimated with either fixed or random assumed to be temporal invariant, we can then avoid the collapse sce­
effects. A Hausman’s test is often used to determine a better alternative nario with geographically weighted panel regression as well. Details
(Baltagi, 2005). follow.

5
D. Yu et al. Applied Geography 154 (2023) 102941

2.2.4. Algorithm for implementing geographically weighted panel regression 1. For each temporal period, the cross-sectional data is extracted.
In a nutshell, geographically weighted approaches assume the Conventional geographically weighted regression approach is
observed data is generated by many overlapping and smooth spatial applied to this cross-section dataset to determine the geographically
processes that follow a distance-decaying mechanism (Fotheringham weighted subsample for each location at this temporal period.
et al., 2002). This mechanism is generally referred to as the First Law of Specifically:
Geography, which states that everything is related, but closer things are 1.1 A distance-decaying spatial kernel function of the form k(dij) = f
more related than distant ones (Tobler, 1970) and is often mathemati­ ((dij/b)-α) or k(dij) = f(-(dij/b)α) is chosen to decide the local
cally represented by a kernel function and graphically a symmetric region around a location and weigh the observations that fall
bell-shaped curve (Fotheringham et al., 2002). The observed data on within this local region (where dij is the distance between loca­
location i is hence the result of overlapping a smooth distance-decaying tions i and j; b is called the bandwidth that determines the flat­
process that centered on location i, and many other smooth ness of the bell-shaped symmetric kernel (lager the b, flatter the
distance-decaying processes that centered on other locations but kernel); α is a parameter that controls how quickly the values on
neighbor location i in various degrees depending on the distance be­ the kernel curve drops from 1, often α takes the value of 2 (bi-
tween those locations and location i. square) or 3 (tri-cube). For a location i, any other location j will
Based on this argument, once a functional form for the distance- be assigned the weight of k(dij) and will be used to weigh the
decaying process is identified, it is possible to find all the spatial pro­ observations on location j; the weighted observations are then
cesses that project influences on location i. So that we can create a assembled with the observation on location i to form the sub­
unique sample for location i that only pertains to location i hence will sample for location i).
not involve spatial effects. Since the distance-decaying process is sym­ 1.2 With the chosen spatial kernel function, a starting bandwidth b is
metric (distance-decaying from location i to location j is equivalent to arbitrarily selected to determine the local region for each loca­
distance-decaying from location j to location i), finding all the spatial tion. The weights are assigned for all the observations fall within
processes that project influences on location i is equivalent to singling the local region using the kernel function with the arbitrarily
out the one spatial process that centered on i and finding out all the other chosen bandwidth b. A local subsample for each location is
locations that this spatial process reaches. Once that spatial process is created.
singled out (represented as a symmetric bell-shaped curve), one can then 1.3 When all the locations have their own subsamples, ordinary least
extract the part of information that belongs only to this spatial process at squares regression analysis at each location is conducted to
location i, and the parts of observed information in any other locations produce the spatially varying coefficients of each explanatory
that this spatial process reaches. Once this is done, for each location i, we variable.
could then construct a subsample of observations that only pertains to 1.4 An “optimal” bandwidth b will be determined through optimi­
location i. Because this subsample does not involve spatial effects, esti­ zation strategies that either maximize the model fit (such as the
mation based on this subsample could utilize the regular ordinary least leave-one-out cross-validation approach) or minimize the in­
squares model. In addition, also because the information for this sub­ formation loss (such as the Akaike Information Criterion
sample will be extracted from existing data, the distance-decaying approach) by repeating steps 1.2 and 1.3 with different b.
mechanism will be used to weigh the original data at every location 1.5 Once an “optimal” b is determined, the local region and weights
the ith spatial process reaches, hence the name “geographically for observations fall within this local region for all locations can
weighted.” The First Law of Geography and its approximation, the be determined (the geographically weighted subsamples) for the
distance-decaying kernel function are the keys to this “geographical specific temporal period.
weighting.” 2. For each location, the geographically weighted subsamples from all
The “geographical weighting” can solve the problem of the parsi­ temporal periods are combined to be a (likely unbalanced) panel
monious scenario well because now for each location, there will be a dataset.
weighted subsample that will have enough data to estimate the co­ 3. For each location, regular panel regression analysis will be applied to
efficients. “Geographical weighting” might also provide a more tenable estimate the coefficients of the explanatory variables.
way for estimation than simply rely on estimating non-related individual 4. The estimation will be repeated for all locations and the coefficients
time series data on each location in the collapse scenario. estimated in such way will be spatially varying.
How to weigh existing data on all locations, however, is often 5. For each local panel regression, the regular statistical tests for the
considered slightly differently in different studies. Time is added as an significance of the coefficients will also be produced. The signifi­
additional dimension as in (Yu, 2014), (B. Huang et al., 2010), and cance test results will be used for mapping purposes. Specifically,
(Fotheringham et al., 2015), or the geographical weighting is applied to only the statistically significant local coefficients’ spatial variation
every temporal period as time invariant mechanism to generate a sub­ will be mapped. In such way, the maps answer directly the question
sample of panel data for each location as in (R. H. Cai et al., 2014). In where the social distancing measures work and how well these measures
this study, however, we do not add time as an additional dimension, nor work to curb the spread of COVID-19.
do we intend to treat the geographical weighting as time invariant
mechanism. Instead, we argue that the geographical weighting should 3. Results: seeing the forest and the trees with advanced
be time variant and should be based on each cross-sectional dataset. We geospatial analyses
argue that although the geographical arrangement of observations does
not change abruptly over the temporal period, assuming the data 3.1. Seeing the forest
generating spatial processes remain the same for different temporal
periods sounds less tenable. As an exploratory approach, a time varying Although our data covers the entire period from January 25th, 2020,
spatial process might be closer to the true data generating mechanism to May 6th, 2020, preliminary data exploration suggests that the newly
(DGM). It is very likely, however, that the time variant geographical added cases for the entire nation flattens on around March 5th, 2020.
weighting will generate different sizes of samples for different temporal Following March 5th, 2020, most of the prefectures in China experi­
period. The subsample for each location is likely an unbalanced panel enced rather sporadic occurring of newly added COVID-19 cases. The
dataset. Still, such treatment could effectively avoid the collapse sce­ large number of cases appear only in or around Wuhan, the original
nario when estimating geographically weighted panel regression. epicenter of the outbreak. For this matter, the study presents results
The entire estimation procedure for the proposed geographically from two models with different temporal span: model 1 reports the
weighted panel regression follows these steps. period before the flattening on March 5th, 2020 (Table 3). Model 2

6
D. Yu et al. Applied Geography 154 (2023) 102941

Table 3 Table 4
Influence of the social distancing measures on the daily newly confirmed cases in Influence of the social distancing measures on the daily newly confirmed cases in
China at prefecture level, January 25th – March 5th, 2020, fixed effect non- China at prefecture level, January 25th – May 6th, 2020, fixed effect non-spatial
spatial and spatial lag panel regression models. and spatial lag panel regression models.
Fixed effect panel regression model results Fixed effect panel regression model results

Residuals Residuals
Min. 1st Qu. Median 3rd Qu. Max. Min. 1st Qu. Median 3rd Qu. Max.
− 4.092 − 0.169 − 0.009 0.149 4.447 − 3.520 − 0.380 − 0.057 0.319 4.455

Social distancing measures Estimate Std. t-value Pr(>|t|) Social distancing measures Estimate Std. t-value Pr(>|t|)
Error Error

travelcity 0.114 0.009 12.393 0.000 travelcity 0.106 0.009 11.425 0.000
*** ***
addmig 0.028 0.006 4.288 0.000 addmig 0.024 0.008 3.010 0.003 **
*** netmin − 0.004 0.008 − 0.539 0.590
netmin 0.005 0.006 0.823 0.411 school − 0.058 0.018 − 3.229 0.001 **
work − 0.285 0.020 − 14.566 0.000 work − 0.087 0.022 − 3.929 0.000
*** ***
tourists − 0.001 0.000 − 2.023 0.043 tourists − 0.001 0.000 − 2.236 0.025
quarantine 0.120 0.051 2.342 0.019 quarantine − 0.069 0.017 − 4.147 0.000
intercity 0.067 0.017 3.941 0.000 ***
intercity − 0.003 0.016 − 0.194 0.847
R-Squared: 0.03268
Adj. R-Squared: 0.0080809 R-Squared: 0.0074346
F-statistic: 69.2719 on 7 and 14,353 DF, p-value: <2.22e-16 Adj. R-Squared: − 0.0026868
F-statistic: 33.6053 on 8 and 35,892 DF, p-value: <2.22e-16
Robust Lagrange multiplier test: RLMlag = 32.68, RLMerr = 5.87
Robust Lagrange Multiplier test: RLMlag = 176.03, RLMerr = 67.91
Fixed effect spatial lag panel regression model results
Fixed effect spatial lag panel regression model results
Residuals
Residuals
Min. 1st Qu. Median 3rd Qu. Max.
Min. 1st Qu. Median 3rd Qu. Max.
− 3.930 − 0.160 − 0.010 0.142 4.434
− 3.360 − 0.330 − 0.049 0.287 4.787
Social distancing measures Estimate Std. t-value Pr(>|t|)
Social distancing measures Estimate Std. t-value Pr(>|t|)
Error
Error
travelcity 0.098 0.009 11.009 0.000
*** travelcity 0.077 0.009 8.945 0.000
addmig 0.030 0.006 4.731 0.000 ***
*** addmig 0.019 0.007 2.521 0.012 **
netmin 0.008 0.006 1.398 0.162 netmin − 0.004 0.007 − 0.569 0.569
work − 0.223 0.019 − 11.742 0.000 school − 0.024 0.016 − 1.491 0.136
*** work − 0.022 0.020 − 1.089 0.276
tourists − 0.001 0.000 − 1.548 0.122 tourists 0.000 0.000 0.190 0.849
quarantine 0.100 0.049 2.024 0.043 quarantine − 0.033 0.015 − 2.172 0.030 *
intercity 0.065 0.016 3.941 0.000 intercity 0.005 0.014 0.328 0.743

Spatial autoregressive coefficient 0.234 0.011 21.363 0.000 Spatial autoregressive coefficient 0.418 0.006 68.197 0.000
Lambda *** Lambda ***

Significance codes: ***: 0.001; **: 0.01; *: 0.05; .: 0.1 (one-tailed test). Significance codes: ***: 0.001; **: 0.01; *: 0.05; .: 0.1 (one-tailed test).

reports the entire period from January 25th to May 6th, 2020 (Table 4). 4. Discussion
The significant test is based on p < 0.05, one-tailed test, and marked
with the “*” symbol in the tables. We do not report the period after the 4.1. The forest view
flattening because the daily newly confirmed cases show only small
amount of variation throughout China. The non-spatial individual fixed 4.1.1. The overall impression
effect panel regression results are also reported for comparison When reading Tables 3 and 4, we found that the non-spatial panel
purposes. models’ (for both periods) adjusted R2s are very low. This is to be ex­
pected, however, because this study focuses on how social distancing
3.2. Seeing the trees based NPI strategies could curb the spread of COVID-19 on a daily basis.
Undoubtedly, other than the NPI strategies, there are many other
The geographically weighted panel regression (GWPR) was first important contributors that also work to curb the spread of the disease.
proposed in (Yu, 2010) as one of the alternative local analytical ap­ The most important ones are medical infrastructure related factors such
proaches that deals with geo-panel dataset. In addition to allowing the as numbers of health providers, capacity of the health care systems.
distance-decaying processes to be singled out for each location, the Unfortunately, those factors are not readily collected at daily interval so
GWPR also allows the processes to be different for each temporal period that they cannot be included in the model. While the I-INT pre-process
(each day in the current study). This is critical for long panels as in the attempted to include the background information, the procedure in­
current study comparing to short panels. The results of the GWPR cludes the information in a rather coarse way. Because we do not have
approach for the period before the flattening is reported in Fig. 1, and for sufficient number of contributing factors other than the mined NPI
the entire period is reported in Fig. 2. Only prefectures whose co­ strategies at the required daily interval, those missing important vari­
efficients are pseudo-significant at 95% confidence level (one-tailed, the ables are all captured by the residuals. This is also a problem that the
absolute cut value for one-tailed test is 1.68) are grey scaled. The vari­ nonspatial global level model cannot address adequately. Still, when
able netmin does not show statistical significance for any prefectures for spatial effects are introduced, the missing variables’ information is
both periods, hence its maps are not reported. indirectly modeled through addressing the spatially autocorrelated re­
siduals in the global spatial panel model, and directly modeled in the

7
D. Yu et al. Applied Geography 154 (2023) 102941

Fig. 1. Spatially varying influence of the social distancing measures on the daily newly confirmed cases in China at prefecture level, January 25th – March 5th, 2020.

GWPR model, albeit not at the same statistic power as when actual robust Lagrange Multiplier tests support the spatial lag panel model. For
variables are included in the model. the period prior to the flattening, however, the spatial and non-spatial
models agree well. The primary contributing factors for the daily
4.1.2. The detailed interpretations spread of COVID-19 are the intensity of inner-city travel, the intensity of
By reading Tables 3 and 4, we can draw some interesting conclusions daily migration, and workplace closure. These results agree well with
of the investigated relationships between the spread of COVID-19 and previous studies found in China (X. Guo et al., 2021; Xiong et al., 2020;
the NPI strategies as well. First, the modeling tests suggest that a non- Y. H. Zhang, Zhang, & Wang, 2020), in UK (Davies et al., 2020), Italy
spatial panel model could potentially produce misleading results due (Giordano et al., 2020), and Germany (Chae & Park, 2020). The results
to significant spatial autocorrelation in the regression residuals. Both suggest that the rigorous lock-down policy implemented in Wuhan after

8
D. Yu et al. Applied Geography 154 (2023) 102941

Fig. 2. Spatially varying influence of the social distancing measures on the daily newly confirmed cases in China at prefecture level, January 25th – May 4th, 2020.

January 23rd, 2020, and later to the entire nation contributes signifi­ Second, when looking at the results from the entire periods, the non-
cantly to curb the overall spread of the COVID-19 in China. This result spatial model suggests that other than the above three social distancing
also suggests that social-distancing policies/strategies are working polices/strategies, school closure (which does not appear in the first
during the initial outbreak of a pandemic, though it might not be im­ period because schools remained closed during that period), and
mediate (Vogel, 2020), but ignoring it could prove to be detrimental whether the prefecture is under mandatory quarantine, also work to
(Mahase, 2020; Manchein et al., 2020). curb the spread. After controlling for the spatial effects, however, the

9
D. Yu et al. Applied Geography 154 (2023) 102941

workplace closure and school closure strategies do not seem to be sig­ 4.2. The trees view
nificant factors. Whether a prefecture is under strict quarantine still
serves as a significant contributor to curb the spread. Quarantine and 4.2.1. The overall impression
lockdowns are still effective means when curbing a contagious disease as When we turn our attention to the results produced by the GWPR
commonly practiced in epidemiological studies (Guner, Hasanoglu, & models in Figs. 1 and 2, a more detailed picture emerges. It is evident
Aktas, 2020; Manchein et al., 2020). that other than the net daily movement of people (netmin, which is also
The discrepancy regarding the closure of the working places and not significant at the global level), all the included social-distancing
quarantine policy suggests that during the initial period of the outbreak, measures show significant contribution to curb the spread of COVID-
the severity of the pandemic was only gradually being realized by the 19 at the 95% confidence level in some prefectures in China, but not
society. Some working places remained open in the hope that the all prefectures. Comparing with the forest view, this result suggests that
pandemic will end soon. While in the meantime, the quarantine policy the global model might mask out important details that merit close
was strict and almost universal except for some relatively remote pre­ attention, especially in the time of a global pandemic from an unknown
fectures in the western parts of China. The model of the first period infectious disease. The variation of the coefficients’ values suggests that
captures the significant workplace closure effect and insignificant estimates of the forest view are averaged results from the trees view. While
quarantine policy effect. However, when the entire period is considered, the forest view is valuable to tell at a national/regional level what NPI
as more workplaces were closed because of clear correlation between strategies shall be implemented immediately, the trees view takes to the
opening of workplaces and the spread of the disease as witnessed in next level to suggest at when and where, what work most effectively.
ensuing studies (Mendez-Brito, El Bcheraoui, & Pozo-Martin, 2021), the Seeing both the forest and the trees provides strategical as well as
effect of closing workplaces seems to become less important in curbing detailed information for both effective and efficient decision-making.
the spread. Quarantine policy, on the contrary, was relaxed in many
relatively remote prefectures after the flattening of the curve but 4.2.2. The detailed patterns
remained a strict policy measure in densely populated eastern and Detailed patterns appear when we look at the figures individually.
southern parts of China. Relaxation of quarantine policy likely facilitates The grey scaled patterns for all the social distancing measures produced
the spread of the disease, like discussed in Cimolai (2021), our model is by GWPR reveal rich amount of information. First, for the period prior to
able to capture this effect. the flattening of the curve (March 5th, 2020), Fig. 1 suggests that in
Third, it is also worth noting that the absolute values of the signifi­ prefectures around the epicenter, Wuhan, extending to the East and
cant coefficients are larger for the period prior to the flattening of the Southeastern coastal regions, the Southwestern regions and some pre­
curve than for the entire period. This agrees with the common knowl­ fectures in Shaanxi and Gansu of the Northwestern China, the inner-city
edge that early implementation of social distancing strategies tends to be travel intensity is the primary factor for the spread of the disease. On the
more effective because the NPI strategies could block the transmission other hand, the number of daily inter-prefecture migration is signifi­
pathways of the virus (Baker et al., 2020; Brauner et al., 2021; Men­ cantly contributing to the spread of COVID-19 in Northeastern and
dez-Brito et al., 2021; Redlin, 2022). In addition, the “forest” view Northern China prefectures. Inter-city travel bans and workplace
suggests that except for the above three measures, all other closure, however, worked primarily in prefectures immediately around
social-distancing measures do not seem to play significant roles in the epicenter and the Peral River Delta Urban Agglomeration, where the
curbing the spread of COVID-19 in China at the prefecture level. A largest temporary migrant destinations in China locate. The quarantine
decision-maker or modeler who works with the global model might policy does not seem to have a distinctive pattern, though where it
suggest those social distancing strategies are less effective hence might worked, it seems to work the best to curb the spread since quarantined
not give them adequate consideration. This is more evident when we prefectures will have a maximum of 0.973 cases less than non-
model the entire study period. Understandably, when the disease was quarantined ones daily, this observation, even at a varying local level,
gradually under control, with both significantly decreased cases and agrees with mainstream epidemiological evidence regarding the general
gradually relaxed policies in China, the relationship between the spread effectiveness of quarantine (Charvadeh et al., 2022; Chinazzi et al.,
of the disease and the social distancing measures will be diluted and 2020; Guner et al., 2020). The tourism activities’ relationship with the
eventually become nonexistent. The results suggest that China’s quick spread of the disease shows an inconsistent pattern. This could be the
and decisive social distancing campaign and quarantine policies worked result of the way the data is collected (the number of comments on
well and quickly as discussed in detail in Tian et al. (2020). The problem popular tourism sites) and it might not be representative of happened
is, however, that such campaign and policies are hardly replicable trips.
elsewhere no matter it is due to the cultural and political system dif­ Second, the patterns in Fig. 2 do not change much since newly
ferences or economic resistance. The recent strict tracking and local confirmed cases after the flattening changed very little, though pre­
lock-down approaches adopted by the Chinese government due to the fectures with significant relationships between the various social
resurgence of sporadic newly added cases suggests the potential cost distancing measures and the spread of COVID-19 expanded from the
could be prohibitively high elsewhere. original clusters. The strengths of such significant relationships as
In a sense, the forest view provides a general, global perspective as to measured by the varying coefficients are also smaller than the previous
what might work and what might not. It is a standard and effective way period (the absolute values of all maximum significant coefficients are
in epidemiology to curb the spread of a contagious disease at early stages smaller than in the previous period). This result suggests that the quick
of the disease when data was scarce, and many remained unknown and strict social distancing measures taken by the Chinese government
(Fong et al., 2020). It is very likely that some social distancing poli­ in the early stages of COVID-19 development was working and
cies/strategies work in certain places and certain times, but not neces­ continued to work even after the flattening of the curve until the end of
sarily in all places and all times (Davalgi, Undi, Annadani, & Nawaz, our study period.
2020). The models established under the forest view tend to average out Third, the patterns in both Figs. 1 and 2 suggest that for the entire
and mask these effects. Considering the startling cost of implementing country, mass movement of population, either within the city (inner city
social distancing measures in nations as vast as China, a global, forest travel), or across cities (intercity travel and migration), are the most
view might not always produce the most ideal results for efficient stra­ important factors that facilitate the spread of COVID-19. The global
tegies to curb the spread of COVID-19 elsewhere, albeit effective. model failed to pick out the intercity travel ban as an important factor,
primarily because the factor only works in areas around the epicenter
and large temporary migrations centers as we now see in Figs. 1 and 2.
Consequently, the global model averages out the local effectiveness of

10
D. Yu et al. Applied Geography 154 (2023) 102941

this factor. Reducing the amount of mass movement of population is the as the pandemic progresses. Knowing where those measures are effec­
priority for the entire nation to curb the spread of COVID-19 (Chinazzi tive is critical for efficient decision making especially when facing un­
et al., 2020; Kraemer et al., 2020). For workplace closure and city-wide known epidemiological events like the COVID-19. The outbreak of
strict quarantine, these measures worked the best in the epicenters and COVID-19 caught the entire human race off-guard. Medical equipment
cities that are either physically or socioeconomically close to the and personnel were limited at the beginning of the outbreak. All coun­
epicenter (Wuhan). tries had to devise and implement various social distancing based NPI
measures ad hoc. China has implemented one of the strictest social
4.2.3. Significant policy implications of these patterns revealed by the distancing strategies that are effective but can hardly be repeated else­
GWPR where. The introduction of local models with advanced geospatial
These specific spatial patterns are of critical importance for devising analysis has the potential to provide a priori strategies guiding more
targeted social distancing policies for different prefectures in China. efficient implementation for social distancing measures during unknown
While it is well known that reduced mass movement of population is the pandemic. Both the forest and trees views suggest that restriction of mass
key to slow down the spread of an infectious disease (Kraemer et al., movement of population and strict quarantine work the best in
2020), reducing mass movement of population can be implemented restricting the spread of COVID-19. The trees view further suggests that
differently in different regions. In China, for the prefectures that are measures such as closure of workplace and strict quarantine policy work
either physically or socioeconomically close to the epicenter, the the best around the epicenter and large migration origins and destina­
strictest social distancing policy, namely, restricting daily outings, tions. Knowing what strategies work at when and where could poten­
closure of working places, and shutting down inter-city travels are tially put the limited medical equipment and personnel into the most
necessary. In the Northeastern and Northern China centered on the efficient use, which could save lives and reduce social disturbance and
capital city, Beijing, a strict temporary migration ban might prove to be economic downturn resulted from the strict national lock-down policy
effective to slow down the spread of COVID-19 because Beijing is one of as implemented in China during the February–May 2020 period. The
the busiest hubs for temporary migrants from neighboring regions. practice proposed in this work is of great potential to facilitate both the
When the gravity of the COVID-19 disease weighed heavily on understanding of how social distancing based NPI measures work (when
China’s society in late January to February, the Chinese government and where), and the implementation of targeted social distancing
took decisive and strict social distancing and quarantine measures ad campaign in certain areas to achieve maximum benefits of curbing the
hoc, which was effective as supported by our models and other studies spread of the COVID-19. Considering during this early period of the
(Kraemer et al., 2020; Tian et al., 2020). Such strict measures, however, pandemic, neither effective COVID-19 vaccines nor other pharmaceu­
can hardly be repeated elsewhere. Our models, especially the local tical solutions were available at the decision-makers’ disposal, such in­
models that provide the trees view can be applied a priori as data accu­ formation is essential to flatten the curve of the COVID-19 diseases, to
mulates and might provide a more efficient strategy when implementing prevent the collapse of the health system, and to allocate the limited
social distancing measures, alongside with the global model. health care resources to places where they are most needed, with min­
The impact of school closure was not measurable prior to the flat­ imal cost of both human lives and economic performance. For this very
tening of the curve because that was during the traditional Chinese New practical reason, the proposal practice in the current study merits further
Year, no school opened before the flattening. For the entire study period, attention of the scientific community.
however, the impact of school closure seems to only have significant
impact in the relatively remote cities in Qinghai, West Gansu, Xinjiang Author statement
and Tibet, and not significant at the global level (Table 4), despite
findings suggesting that school closure has significant impact on curbing Danlin Yu: Conceptualization, Methodology, Software, Validation,
the spread (Auger et al., 2020). Considering that these remote Western Formal analysis, Investigation, Writing – Original Draft, Writing – Re­
cities often have relatively weaker health care system than the Central view & Editing, Visualization, Supervision, Project administration
and Eastern China, including tests and trace intervention, reopening of Yaojun Zhao: Conceptualization, Validation, Investigation, Resources,
schools there was more likely associated with the spread of the disease as Writing – Original Draft, Supervision, Project administration Jun Meng:
modeled in a recent UK study (Panovska-Griffiths et al., 2020). conceptualization, Validation, Investigation, Writing – Original Draft,
The practice of implementing the GWPR model coupled with the Supervision Xiaoxi Wang: Software, Investigation, Data Curation Lin­
global spatial panel model suggests a holistic spatial analysis of feng He: Software, Formal analysis, Data Curation Meng Jia: Data
geographical phenomena (spatial epidemiology in our study) is of Curation Jie Ouyang: Data Curation Yu Han: Data Curation Ge Zhang:
pragmatical significance. Data collected over geographic space shall Data Curation Yao Lu: Data Curation.
always to scrutinized more carefully to make sure spatial effects, if
present, are taken care of (Anselin, 2007). More importantly, however, Acknowledgement
the current study further suggests that it is of pragmatical importance to
always implement a holistic spatial analytical approach that includes This research is funded by the IDC fund (CC10502 Earth IDC) of the
both global and local methods to provide a better understanding of the Department of Earth and Environmental Studies, College of Science and
geographic process and data generating mechanisms. Mathematics, Montclair State University.

5. Conclusion: implications for future research paradigm References

Through integrating a global spatial panel regression model and a Alene, M., Yismaw, L., Assemie, M. A., Ketema, D. B., Gietaneh, W., & Birhan, T. Y.
(2021). Serial interval and incubation period of COVID-19: A systematic review and
local geographically weighted panel regression model, this study at­ meta-analysis. BMC Infectious Diseases, 21(1).
tempts to provide a holistic view of investigating how NPI strategies Anderson, R. M., Heesterbeek, H., Klinkenberg, D., & Hollingsworth, T. D. (2020). How
influence the spread of COVID-19 diseases in China during the first three will country-based mitigation measures influence the course of the COVID-19
epidemic? The Lancet, 395(10228), 931–934.
months of the outbreak. The results suggest that NPI strategies are in Anselin, L. (1988a). Lagrange multiplier test diagnostics for spatial dependence and
general effective to prevent the spread of the disease, but the effec­ spatial heterogeneity. Geographical Analysis, 20(1), 1–17.
tiveness varies from place to place. Anselin, L. (1988b). Spatial econometrics: Methods and models. Dordrecht, The Netherland:
Kluwer Academic Publisher.
The take-home message from this study is that while social Anselin, L. (1992). Space and applied econometrics - introduction. Regional Science and
distancing based NPI measures work to curb the spread of COVID-19, Urban Economics, 22(3), 307–316.
these measures are usually effective at certain places and certain times

11
D. Yu et al. Applied Geography 154 (2023) 102941

Anselin, L. (2002). Under the hood - issues in the specification and interpretation of Fong, M. W., Gao, H. Z., Wong, J. Y., Xiao, J. Y., Shiu, E. Y. C., Ryu, S., et al. (2020).
spatial regression models. Agricultural Economics, 27(3), 247–267. Nonpharmaceutical measures for pandemic influenza in nonhealthcare settings-
Anselin, L. (2007). Spatial econometrics in RSUE: Retrospect and prospect. Regional social distancing measures. Emerging Infectious Diseases, 26(5), 976–984.
Science and Urban Economics, 37(4), 450–456. Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2002). Geographically weighted
Anselin, L., Bera, A. K., Florax, R., & Yoon, M. J. (1996). Simple diagnostic tests for regression: The analysis of spatially varying relationship. West Essex, England: John
spatial dependence. Regional Science and Urban Economics, 26(1), 77–104. Wiley & Sons.
Auger, K. A., Shah, S. S., Richardson, T., Hartley, D., Hall, M., Warniment, A., et al. Fotheringham, A. S., Crespo, R., & Yao, J. (2015). Geographical and temporal weighted
(2020). Association between statewide school closure and COVID-19 incidence and regression (GTWR). Geographical Analysis, 47(4), 431–452.
mortality in the US. JAMA, 324(9), 859–870. Giordano, G., Blanchini, F., Bruno, R., Colaneri, P., Di Filippo, A., Di Matteo, A., &
Baker, R. E., Park, S. W., Yang, W. C., Vecchi, G. A., Metcalf, C. J. E., & Grenfell, B. T. Colaneri, M. Modelling the COVID-19 epidemic and implementation of population-
(2020). The impact of COVID-19 nonpharmaceutical interventions on the future wide interventions in Italy. Nature Medicine, 32.
dynamics of endemic infections. Proceedings of the National Academy of Sciences of the Giordano, G., Blanchini, F., Bruno, R., Colaneri, P., Di Filippo, A., Di Matteo, A., et al.
United States of America, 117(48), 30547–30553. (2020). Modelling the COVID-19 epidemic and implementation of population-wide
Baltagi, B. H. (2005). Econometric analysis of panel data (3rd ed.). Chichester, West interventions in Italy. Nature Medicine, 32.
Sussex, UK: John Wiley & Sons. Greene, W. H. (2003). Econometric analysis. India: Pearson Education.
Bertozzi, A. L., Franco, E., Mohler, G., Short, M. B., & Sledge, D. (2020). The challenges of Guan, W., Ni, Z., Hu, Y., Liang, W., Ou, C., He, J., et al. (2020). Clinical characteristics of
modeling and forecasting the spread of COVID-19. Proceedings of the National coronavirus disease 2019 in China. New England Journal of Medicine, 382(18),
Academy of Sciences, 117(29), 16732–16738. 1708–1720.
Bivand, R. S., Pebesma, E. J., Gomez-Rubio, V., & Pebesma, E. J. (2008). Applied spatial Guner, R., Hasanoglu, I., & Aktas, F. (2020). COVID-19: Prevention and control measures
data analysis with R, 747248717. Springer. in community. Turkish Journal of Medical Sciences, 50(3), 571–577.
Brauner, J. M., Mindermann, S., Sharma, M., Johnston, D., Salvatier, J., Gavenciak, T., Guo, S., An, R., McBride, T. D., Yu, D., Fu, L., & Yang, Y. (2020). Mitigation interventions in
et al. (2021). Inferring the effectiveness of government interventions against COVID- the United States: An exploratory investigation of determinants and impacts,
19. Science, 371(6531), 802–+. 1049731520957415. Research on Social Work Practice.
Cai, Y. Y., Liu, J. Y., Yang, H. T., Wang, M., Guo, Q. P., Huang, D., et al. (2020). Guo, X., Zhang, J., & Wu, X. (2021). Spatio-temporal characteristics of the novel
Association between incubation period and clinical characteristics of patients with coronavirus attention network and its influencing factors in China. PLoS One, 16(9),
COVID-19. Journal of International Medical Research, 48(9). Article e0257291.
Cai, R. H., Yu, D. L., & Oppenheimer, M. (2014). Estimating the spatially varying Huang, B., Wu, B., & Barry, M. (2010). Geographically and temporally weighted
responses of corn yields to weather variations using geographically weighted panel regression for modeling spatio-temporal variation in house prices. International
regression. Journal of Agricultural and Resource Economics, 39(2), 230–252. Journal of Geographical Information Science, 24(3), 383–401.
Cevasco, K. E., Roess, A. A., North, H. M., Zeitoun, S. A., Wofford, R. N., Matulis, G. A., Huang, J., Zhang, L., Liu, X., Wei, Y., Liu, C., Lian, X., et al. (2020). Global prediction
et al. (2021). Survival analysis of factors affecting the timing of COVID-19 non- system for COVID-19 pandemic. Science Bulletin, 65(22), 1884–1887.
pharmaceutical interventions by US universities. BMC Public Health, 21(1). Kraemer, M. U. G., Yang, C. H., Gutierrez, B., Wu, C. H., Klein, B., Pigott, D. M., et al.
Chae, S. H., & Park, H. J. (2020). Effectiveness of penalties for lockdown violations (2020). The effect of human mobility and control measures on the COVID-19
during the COVID-19 pandemic in Germany. American Journal of Public Health, 110 epidemic in China. Science, 368(6490), 493–+.
(12), 1844–1849. LeSage, J. P., & Pace, R. K. (2009). Introduction to spatial econometrics. Boca Raton: Taylor
Charvadeh, Y. K., Yi, G. Y., Bian, Y., & He, W. Q. (2022). Is 14-days a sensible quarantine & Francis/CRC Press.
length for COVID-19? Examinations of some associated issues with a case study of Mahase, E. (2020). Covid-19: UK starts social distancing after new model points to 260
COVID-19 incubation times. Statistics in Biosciences, 14(1), 175–190. 000 potential deaths. BMJ British Medical Journal, 368, 2.
Chen, P. Y., Guo, X. D., Jiao, Z. T., Liang, S. H., Li, L. F., Yan, J., et al. (2022). Manchein, C., Brugnago, E. L., da Silva, R. M., Mendes, C. F. O., & Beims, M. W. (2020).
A multilayer network model for studying the impact of non-pharmaceutical Strong correlations between power-law growth of COVID-19 in four continents and
interventions implemented in response to COVID-19. Frontiers in Physics, 10. the inefficiency of soft quarantine strategies. Chaos, 30(4), 9.
Cheng, C., Zhang, D. D., Dang, D. J., Geng, J., Zhu, P. Y., Yuan, M. Z., et al. (2021). The McCaw, Z. R., Lane, J. M., Saxena, R., Redline, S., & Lin, X. H. (2019). Operating
incubation period of COVID-19: A global meta-analysis of 53 studies and a Chinese characteristics of the rank-based inverse normal transformation for quantitative trait
observation study of 11 545 patients. Infectious Diseases of Poverty, 10(1). analysis in genome-wide association studies. Biometrics, 11.
Chinazzi, M., Davis, J. T., Ajelli, M., Gioannini, C., Litvinova, M., Merler, S., et al. (2020). Mendez-Brito, A., El Bcheraoui, C., & Pozo-Martin, F. (2021). Systematic review of
The effect of travel restrictions on the spread of the 2019 novel coronavirus (COVID- empirical studies comparing the effectiveness of non-pharmaceutical interventions
19) outbreak. Science, 368(6489), 395–+. against COVID-19. Journal of Infection, 83(3), 281–293.
Cimolai, N. (2021). In pursuit of the right tail for the COVID-19 incubation period. Public Nazia, N., Law, J., & Butt, Z. A. (2022). Identifying spatiotemporal patterns of COVID-19
Health, 194, 149–155. transmissions and the drivers of the patterns in toronto: A bayesian hierarchical
Ciufolini, I., & Paolozzi, A. (2020). Mathematical prediction of the time evolution of the spatiotemporal modelling. Scientific Reports, 12(1), 1–13.
COVID-19 pandemic in Italy by a Gauss error function and Monte Carlo simulations. Panovska-Griffiths, J., Kerr, C. C., Stuart, R. M., Mistry, D., Klein, D. J., Viner, R. M., et al.
European Physical Journal Plus, 135(4), 8. (2020). Determining the optimal strategy for reopening schools, the impact of test and
Croissant, Y., & Millo, G. (2019). Panel data econometrics with R. Hoboken, NJ: John trace interventions, and the risk of occurrence of a second COVID-19 epidemic wave in the
Wiley & Sons. UK: A modelling study. The Lancet Child & Adolescent Health.
Davalgi, S., Undi, M., Annadani, R., & Nawaz, A. S. (2020). Comparison of Measures Poletto, C., Scarpino, S. V., & Volz, E. M. (2020). Applications of predictive modelling
adopted to combat COVID 19 Pandemic by different countries in WHO regions. early in the COVID-19 epidemic. The Lancet Digital Health, 2(10), e498–e499.
Indian Journal of Community Health, 32(2), 288–299. Quesada, J. A., Lopez-Pineda, A., Gil-Guillen, V. F., Arriero-Marin, J. M., Gutierrez, F., &
Davies, N. G., Kucharski, A. J., Eggo, R. M., Gimma, A., Edmunds, W. J., Jombart, T., Carratala-Munuera, C. (2021). Incubation period of COVID-19: A systematic review
et al. (2020). Effects of non-pharmaceutical interventions on COVID-19 cases, deaths, and meta-analysis. Revista Clínica Española, 221(2), 109–117.
and demand for hospital services in the UK: A modelling study. The Lancet Public Redlin, M. (2022). Differences in NPI strategies against COVID-19. Journal of Regulatory
Health. Economics, 62(1–3), 1–23.
Dayaratna, K. D., Gonshorowski, D., & Kolesar, M. (2022). Hierarchical Bayesian spatio- Tian, H. Y., Liu, Y. H., Li, Y. D., Wu, C. H., Chen, B., Kraemer, M. U. G., et al. (2020). An
temporal modeling of COVID-19 in the United States. Journal of Applied Statistics, investigation of transmission control measures during the first 50 days of the COVID-
1–18. 19 epidemic in China. Science, 368(6491), 638–+.
Dehning, J., Zierenberg, J., Spitzner, F. P., Wibral, M., Neto, J. P., Wilczek, M., et al. Tobler, W. R. (1970). A computer movie simulating urban growth in the detroit region.
(2020). Inferring change points in the spread of COVID-19 reveals the effectiveness Economic Geography, 46, 234–240.
of interventions. Science, 369(6500), Article eabb9789. Vogel, L. (2020). How long will social distancing take to work? Experts weigh in on
Dowd, J. B., Andriano, L., Brazel, D. M., Rotondi, V., Block, P., Ding, X. J., et al. (2020). Canada’s COVID-19 response. Canadian Medical Association Journal, 192(14),
Demographic science aids in understanding the spread and fatality rates of COVID- E382–E383.
19. Proceedings of the National Academy of Sciences of the United States of America, 117 Wan, S. Y., Liu, J., & Liu, M. (2021). Progress on the incubation period of COVID-19.
(18), 9696–9698. Chinese Science Bulletin-Chinese, 66(15), 1802–1811.
Drew, D. A., Nguyen, L. H., Steves, C. J., Menni, C., Freydin, M., Varsavsky, T., et al. Wei, S., & Wang, L. (2020). Examining the population flow network in China and its
(2020). Rapid implementation of mobile technology for real-time epidemiology of implications for epidemic control based on Baidu migration data. Humanities & Social
COVID-19. Science, 368(6497), 1362–+. Sciences Communications, 7(1).
Elhorst, J. P. (2014). Spatial econometrics: From cross-sectional data to spatial panels. New WHO Commission on Social Determinants of Health, W. H. O. (2008). Closing the gap in a
York, NY: Springer. generation: Health equity through action on the social determinants of health: Commission
Enserink, M., & Kupferschmidt, K. (2020). With COVID-19, modeling takes on life and on social determinants of health final report. World Health Organization.
death importance. Science, 367(6485), 1414–1415. Wu, Y., Kang, L. Y., Guo, Z. R., Liu, J., Liu, M., & Liang, W. N. (2022). Incubation Period
Fair, K. R., Karatayev, V. A., Anand, M., & Bauch, C. T. (2022). Estimating COVID-19 of COVID-19 caused by unique SARS-CoV-2 strains A systematic review and meta-
cases and deaths prevented by non-pharmaceutical interventions, and the impact of analysis. JAMA Network Open, 5(8).
individual actions: A retrospective model-based analysis. Epidemics, 39. Xiong, Y., Wang, Y., Chen, F., & Zhu, M. (2020). Spatial statistics and influencing factors
Flaxman, S., Mishra, S., Gandy, A., Unwin, H. J. T., Mellan, T. A., Coupland, H., et al. of the COVID-19 epidemic at both prefecture and county levels in Hubei Province,
(2020). Estimating the effects of non-pharmaceutical interventions on COVID-19 in China. International Journal of Environmental Research and Public Health, 17(11),
Europe. Nature, 1–5. 3903.

12
D. Yu et al. Applied Geography 154 (2023) 102941

Yu, D. L. (2010). Exploring spatiotemporally varying regressed relationships: The Yu, D. L., Zhang, Y. J., & Wu, X. W. (2020). How socioeconomic and environmental
geographically weighted panel regression analysis. The International Archives of the factors impact the migration destination choices of different population groups in
Photogrammetry, Remote Sensing and Spatial Information Sciences, 38, 134–139. China: An eigenfunction-based spatial filtering analysis (vol 15, pg 129, 2020).
Yu, D. L. (2014). Understanding regional development mechanisms in Greater Beijing Population and Environment, 41(3), 396, 396.
Area, China, 1995–2001, from a spatial–temporal perspective. Geojournal, 79(2), Yu, D. L., Zhang, Y. J., Wu, X. W., Li, D., & Li, G. D. (2021). The varying effects of
195–207. accessing high-speed rail system on China’s county development: A geographically
Yu, D. L., Murakami, D., Zhang, Y. J., Wu, X. W., Li, D., Wang, X. X., et al. (2020). weighted panel regression analysis. Land Use Policy, 100, 11.
Investigating high-speed rail construction’s support to county level regional Zhang, Y. H., Zhang, A. M., & Wang, J. E. (2020). Exploring the roles of high-speed train,
development in China: An eigenvector based spatial filtering panel data analysis. air and coach services in the spread of COVID-19 in China. Transport Policy, 94,
Transportation Research Part B: Methodological, 133, 21–37. 34–42.
Yu, D. L., & Wei, Y. D. (2008). Spatial data analysis of regional development in Greater Zhang, L., Zhu, J. H., Wang, X. Y., Yang, J., Liu, X. F., & Xu, X. K. (2021). Characterizing
Beijing, China, in a GIS environment. Papers in Regional Science, 87(1), 97–117. COVID-19 transmission: Incubation period, reproduction rate, and multiple-
generation spreading. Frontiers in Physics, 8.

13