Content uploaded by Boniphace Kutela
Author content
All content in this area was uploaded by Boniphace Kutela on Sep 19, 2022
Content may be subject to copyright.
1
Insights into the Long-term Effects of COVID-19 Responses on Transportation Facilities
Boniphace Kutela, Ph.D., P.E
Assistant Research Scientist
Roadway Safety Program
Texas A&M Transportation Institute
1111 RELLIS Parkway, Bryan, TX 77807
Email: b-kutela@tti.tamu.edu
Tabitha Combs, Ph.D.
Research Associate & Lecturer
Dept. of City & Regional Planning
University of North Carolina at Chapel Hill
Email: tab@unc.edu
Rafael John, Mwekh’iga
Civil Engineer
Ibra Contractors Limited
P.O.BOX 20881, Dar es Salaam, Tanzania
Email: rafaelmwekhiga@gmail.com
Neema Langa, Ph.D.
Assistant Professor
Department of Sociology/African American Studies
University of Houston
3553 Cullen Boulevard
Email: nmlanga@central.uh.edu
2
Abstract
The impacts of COVID-19 on transportation sector have received a substantial research attention,
however, less is known about localized COVID-19 responses that provided safe space for mobility and
other daily activities. We applied logistic regression and text mining approaches on the Shifting Streets
COVID-19 Mobility Dataset to explore the long-term outcomes of the localized responses. We explored
the purpose, affected space, function, and implementation approach. We found that responses instituted for
economic recovery and public health are less likely to be long-term, while responses meant to improve
safety or bicycle/pedestrian mobility are more likely to be long-term. Further, operational or regulatory
responses are less likely to be long-term. Additionally, responses affecting curb space are more likely to be
long-term than those affecting other right-of-way areas. Text-mining of responses’ narratives revealed key
patterns for both short-term and long-term outcomes. Study findings showcase the possible design and
operations changes during post-COVID-19 era
Keywords: COVID-19, long-term impacts, shifting streets
3
1. Background
To date, over 320 million cases of COVID-19 have been reported worldwide since the first
case reported in Wuhan, China on December 8, 2019 (BBC, 2022; Ritchie et al., 2022; WHO,
2020). The United States, India, Brazil, the United Kingdom, and France are the leading countries
in cumulative total cases and total deaths (BBC, 2022; Ritchie et al., 2022).Transportation is
among the sectors most heavily affected by COVID-19. The impact of COVID-19 on
transportation comes in different ways, including large-scale mobility restrictions (e.g., internal
travel restrictions, cross-border travel restrictions, national and regional lockdowns) and localized
responses.
There is a rapidly growing body of literature examining the impacts of these responses on
global and local travel demand (Van Wee & Witlox 2021; Suau-Sanchez et al. 2020), active
mobility patterns (Hunter et al., 2021), individual decision-making (such as residential relocation
and vehicle ownership; Habib & Anik, 2021), energy use, and air quality (Abu-Rayash & Dincer,
2020). Research is also emerging on the wide-ranging impacts of street-level interventions meant
to lower local transmission of COVID-19 and provide safe, physically distanced space to walk,
bike, and conduct business outdoors by altering the allocation, use, and regulation of space in the
roadway right-of-way (Firth et al., 2021; Fischer & Winters, 2021; Mayo, 2021; Vecchio, Tiznado-
Aitken, & Mora-Vega, 2021; Wright & Reardon, 2021). Others have explored factors enabling
street-level responses (e.g., Combs & Pardo, 2021), cities’ motivations and objectives for street-
level responses (e.g., Fischer & Winters, 2021), the distribution of benefits of the responses (e.g.,
Wright & Reardon, 2021), and overall public perceptions about the responses (e.g., Shirgaokar et
al., 2021).
4
Limited literature is available on the long-term impact of COVID-19 in transportation
(Advani, Sharma, & Dhyani, 2021; Habib & Anik, 2021; Marra, Sun, & Corman, 2022; Truong,
2021; Zhang & Zhang, 2021). A study by Truong (2021) examined the medium and long-term
impact of COVID-19 on air transportation using neural network and Monte Carlo simulation. The
study found weekly economic index and travel distance as the key to air travel in the post-pandemic
period, concluding that it would take several years before the air travel recovers to pre-pandemic
levels. Another study by Marra et al., (2022) focused on changes in public transport use in
Switzerland. They reported that the key difference was the perception of costs of transfers and of
travel time in train. A study by Habib & Anik, (2021) in Nova Scotia, Canada investigated the
long-term impact of COVID-19 on transport and land use. The study predicted an increase of
vehicle ownership for suburban areas by up to 74% by 2030. Other studies by Advani et al., (2021)
and Zhang & Zhang, (2021) focused on non-motorists and decarbonization of the transport sector,
respectively.
Little is known yet about the long-term outcomes of the street-level responses. For instance,
it is unknown whether and which responses will outlast the pandemic, or the factors associated
with their longevity and durability. Given long-standing calls around the world for massive
overhauling of the transportation sector in order to address on-going crises of climate change and
deepening inequality, understanding whether and how the pandemic might lead to enduring
changes to how roadway space is allocated, used, and regulated is critical (Combs & Pardo, 2021).
Using the “Shifting Streets COVID-19 Mobility Dataset,” we seek to identify factors that
predict the likelihood of street-level responses outlasting the pandemic. The Shifting Streets
dataset documents, describes, and catalogs these street-level responses based on information
gathered from 534 cities in over fifty countries around the world (Combs & Pardo, 2021),
5
providing a unique opportunity to assess the short-term and long-term impacts of COVID-19 on
the transportation industry at the local level. We use a mixed methods approach to this research,
combining logistic regression of key variables in the Shifting Streets dataset with text mining of
the dataset’s narrative description of each response.
2. Methods
2.1. Data
The Shifting Streets COVID19 Mobility Dataset contains over 1400 unique responses to changing
demands on public space in 550 cities in 60 countries (as of January 2022). Most of the observations were
collected from North America (56%), Europe & Central Asia (30%), Latin America and Caribbean (5.6%),
East Asia & Pacific (5.2%), while Sub-Saharan Africa had the least proportion of observations (0.8%). Data
have been collected and curated continuously since March 11, 2020, and includes information about local,
state/regional, and national-level responses to changing demands for mobility and on public space that were
initiated between March and September 2020. The dataset is open source and available for public download
at pedbikeinfo.org/shiftingstreets. This paper uses data downloaded from the September 2021 version of
the data.
The Shifting Streets data document and describe the wide range of mobility-related actions cities around
the world took to in response to the pandemic. The dataset was created to highlight innovation and flexibility
in the transport sector, uncover lessons cities learned from their pandemic responses, and support and
inspire research into how the transport sector is evolving based on its experience during the COVID19
pandemic (Combs & Pardo, 2021). Further details about the data collection methods and intended uses are
available on the Shifting Street website (http://pedbikeinfo.org/shiftingstreets) and in Combs & Pardo
(2021).
The variables in the Shifting Streets dataset of interest for this study include:
6
i. Description: a brief narrative describing the response, written by the individual reporting on the
response or gleaned from the news or press releases about the response
ii. Longevity: the anticipated duration of the response, including one-time implementation (e.g., the
response is in place for a single weekend or special event), temporary (the response is intended to
be removed at some point in the future), indefinite (no clearly defined whether long or short term
plans have been established for the response), permanent, and temp-to-perm (the response was
originally intended to be temporary but has or will be converted to a permanent installation). We
collapsed the longevity variable in our analyses “long-term” (including permanent and temp-to-
perm), “short term” (one-time implementation or temporary), and indefinite, and unknown.
iii. Time: the days and/or times of day the response is in effect, which we collapse in this analysis to a
binary variable indicating whether or not the response is in place 24/7
iv. Space: the portion of the roadway directly affected by the response, including the entire roadway,
entire travel lane(s), parking lanes, curb space, sidewalks and other off-street space, and
intersections
v. Purpose: the main purpose behind the response, which includes economic recovery, equity, moving
goods, moving people, public engagement, safety, and public health
vi. Function: the manner in which the response is intended to affect users, including creating street
space for active mobility, other active mobility supports, creating street space for commerce, and
miscellaneous other actions. Given that the focus of the analysis was on changes to the physical
design, use, or regulation of street space, we used the ‘function’ variable to filter out responses that
dealt with changes to transit service, funding streams, access to bicycles, and temporary mobility
restrictions.
vii. Category: implementation approach or general nature of how the response is implemented,
including operational, physical, regulatory, and financial.
7
2.2. Analytical approach
To understand the long-term impact of COVID-19-related street changes, this study employed two
approaches: text mining and logistic regression. Text mining was used to explore the key patterns of short-
term and long-term COVID-19 mobility responses. We used logistic regression to evaluate the likelihood
of responses outlasting the pandemic, i.e., being transformed into permanent installations. The next section
presents the details of each approach.
2.2.1. Text Network
Text mining, while well-established in social science research, is a relatively new method for data
analysis in transportation research. Among text mining methodologies, the text network is a relatively
compact approach that enables visual representation of the language structures in text-based narratives.
Text networks use nodes and links (Figure 1) to present the topology of the narratives.
Figure 1 Skeleton of the text network (B. Kutela, Novat, & Langa, 2021)
In the network, nodes represent keywords, while links represent the co-occurrence of the keywords
(Kim & Jang, 2018; Boniphace Kutela, Das, & Dadashova, 2021; Paranyushkin, 2011). The size of the
Node/keywords
Edge/co-occurrence
Community
8
node represents the frequency of the keyword in the network, while the thickness of the link represents the
frequency of co-occurrence of the keywords. The closer the nodes in the network, the closer the keywords
are in the sentence. Further, keywords of similar themes form a community of keywords.
To create a text network, the text data/narratives need to be normalized, transformed from
unstructured to structured and mapped on the network (Kim & Jang, 2018; B. Kutela et al., 2021;
Paranyushkin, 2011; Yoon & Park, 2004). All the texts are converted to lower case and stop words
(connecting words) are removed during normalization. The transformation from unstructured to structured
involves the creation of a matrix of keywords. Mapping of keywords involves assigning each keyword to
the network. In this step, the algorithm maps the keyword from the matrix to the network as a node. If the
pair of the keywords appears for the first time, the frequency associated with that node and the frequency
of the edge between the two keywords are mapped. If the next pair contains one of the keywords from the
previous pair, an additional frequency of the existing keyword is mapped, followed by the new keyword
and the link (Boniphace Kutela, Das, et al., 2021; Boniphace Kutela & Teng, 2021; Paranyushkin, 2011).
Upon completion of the network, the interpretation depends on various metrics, including the keyword
frequency, document frequency, co-occurrence frequency, collocation frequency, and betweenness
centrality (Kim & Jang, 2018; Paranyushkin, 2011). In this study, the topology of the network, keyword
frequency, document frequency, co-occurred keywords, and collocated keywords (Blaheta and Johnson
2011), are used for interpretation. Keyword frequency represents the number of times the keyword appears
in the entire dataset. On the other hand, document frequency represents the number of times the entire
narrative/description contains a keyword of interest appears in the dataset. Collocated and co-occurred
keywords differ by the location of the keywords. They both represent the keywords appearing in the same
sentence, but the collocated keywords are next to each. For instance, in the statement “Arlington has
implemented automatic pedestrian signal phases at all signalized intersections in its densely populated
corridors”, the keywords pedestrian and signal are collocated while pedestrian and phases are co-occurred
keywords. Thus, collocated keywords provide more insights than co-occurred keywords. Furthermore, in
addition to the frequency, the strength of the association and statistical significance of collocated keywords
9
can be measured using lambda and z-value. The large the magnitude of lambda the strong is the association,
and the z-value greater than 1.96 indicates a statistical significance at a 95% confidence interval (Blaheta
& Johnson, 2011; Boniphace Kutela, Langa, et al., 2021). The analysis was performed in R 4.1.1-
environment (R Core Team, 2021), with the help of quanteda and igraph packages (Benoit et al., 2018;
Csárdi, 2020). For the simplicity of the network, only the top 50 keywords were considered.
2.2.2. Logistic Regression
The likelihood of the response being long-term can be represented as the binary outcome, i.e., either
yes (long-term) or no (short-term). With the two possible outcomes, binary regression models such as logit
and probit are the best candidate for this type of data. Previous studies suggest that logit models are
preferred due to the ease of interpretation of the parameters in terms of odds ratios (Boniphace Kutela &
Teng, 2020; Woodridge, 2012).
Logistic regression can be expressed using the Bernoulli probability function. Given that the dependent
variable Yi has two possible outcomes (1 for long-term response, or 0 for short-term response). The ,
which is the probability that the event is long-term, can be expressed as an inverse logistic function of a
vector of explanatory variables as:
(1)
After linearizing equation 1, the can be written as
(2)
The
represent coefficients of variable that are to be estimated, while
is a constant term.
Further, to apply logistic regression, several assumptions were checked. These include the independence of
the observations, multicollinearity among independent variables, linearity of independent variables, and
sample size.
10
3. Results and Discussion
This section presents the results and discussion of the study. It is divided into three sections:
descriptive statistics, text networks results and discussion, and logistic regression results and discussions.
The text network part of the results explores the content of both short-term and long-term responses using
the narrative description of the response. The logistic regression presents the likelihood of the response
becoming a long-term implementation. The implications of the results on roadway operations are also
presented.
3.1. Descriptive statistics
The September 2021 version of the data used in this analysis contains 1,407 observations; however,
a number of observations had missing or unverified data on some of our variables of interest. After
removing these records, a total of 487 observations with complete, verified data on our variables of interest
were available for analysis. Table 1 presents the descriptive statistics of the variables used in our analyses.
It can be observed that the majority (75%) of the responses recorded in the final dataset were short-term
(“temporary” or “one-time implementation”), while 11.5% were listed as long-term (“permanent” or “temp-
to-perm”). Twelve and a half percent were described as “indefinite,” meaning that at the time of recording,
no decisions had been announced as to their ultimate resolution (i.e., whether they would become short-
term or long-term). Thus, while developing the statistical model, the “indefinite” observations will be
assessed on either side. That is, one model will categorize the indefinite responses as a potential long-term
outcome, and another will categorize them as short-term outcomes. The intention of inclusion of the
indefinite observation was to perform a sensitivity analysis, allowing us to examine whether the extent to
which the results are dependent on indefinite actions being removed or converted to permanent
interventions. The remaining distributions of observations is self-explanatory, as presented in Table 1.
11
Table 1 Descriptive Statistics of the Study Variables
Variable
Variable category
Count
Percent
Longevity
Permanent/ Temp-to-perm
56
11.5%
Indefinite
74
15.2%
One-time/temporary
366
75.3%
Time (all day
everyday)
All day, everyday
399
82.1%
Not all day, everyday
87
17.9%
Space affected
Entire roadway
234
48.1%
Intersection
24
4.9%
Curb
38
7.8%
Travel lanes
116
23.9%
Parking lane
42
8.6%
Parks/plazas/sidewalk
32
6.6%
Purpose
Moving people
217
44.7%
Economic recovery
109
22.4%
Public health
136
28.0%
Safety
24
4.9%
Function
Space for bike/ped
294
60.5%
Space for commerce
96
19.8%
Other bike/ped
38
7.8%
Others
58
11.9%
Category
Physical
366
75.3%
Operational
83
17.1%
Regulatory
37
7.6%
3.2. Text Network Results and Discussion
This section presents results and discussion of the text network analysis. It covers three aspects—short-
term response, long-term response, and indefinite response text networks—and associated metrics. The
networks were developed using the description variable and based on the longevity variable.
Figure 2 and Table 2 present the text network for long-term responses and the associated metrics,
respectively. The network is centered on the keywords streets, bicycle, lane, city, and space. Street appears
most frequently in the narrative descriptions for permanent/long-term responses.
12
Figure 2. Text network for Permanent/Long-term Responses
Table 2 Text Network Performance Metrics for Long-term Responses
Keyword Frequency
Co-occurrence Frequency
Collocation Association
Keywords
Association parameters
Keyword
Count
Docfreq
Keywords
Count
Count
Lambda
Z-value
street
145
66
bicycle lane
71
bicycle lane
51
6.16
20.81
city
96
60
physical distancing
36
physical distancing
35
8.37
13.85
bicycle
103
53
street traffic
20
social distancing
17
7.86
9.07
distancing
68
50
open street
19
shared street
15
5.67
9.47
park
74
45
walking cycling
17
new bicycle
13
3.63
10.83
new
54
45
social distancing
17
public transport
10
5.36
12.13
use
61
42
city street
17
walking cycling
9
6.67
11.72
lane
76
41
new bicycle
17
motor vehicle
9
6.76
9.27
service
62
40
pedestrian space
15
more space
8
3.98
9.54
space
53
40
new lane
14
open street
8
3.59
8.15
Key: Docfreq means Document frequency
The keyword street appears 145 times in 66 reports/observations among reported responses. Further, the
keyword bicycle appears more frequently in one observation but in fewer reports overall compared to
streets. Other keywords in the top ten list include park, new, lane, service, and space. The observations
13
imply that bicycle lanes were significantly affected, whereby new bike lanes were constructed, and others
were closed to provide more space for physical, social distancing. For instance, a description of the vehicle
lane that was converted to bike lane in Ciudad Pal, Spain states that “vehicle lane converted to bicycle lane
on Paseo Maritimo to offset expected increases in vehicle traffic from shut-down relaxation”. Moreover,
the co-occurrence between street, social, distancing implies that the streets were modified to increase more
space for social distancing.
Table 2 also presents the co-occurrence and collocation statistics. It can be observed that all the
collocated keywords presented in the table are statistically significant at a 95% confidence interval (z-value
> 1.96). Further, the statistical association of the keywords is strongest for physical distancing collocated
keywords with lambda value of 8.37. Table 2 also shows that bike lane was most likely to be a long-term
response. Further, social distancing was the major focus for long-term responses that added new bicycle
lanes. New bike lanes are clearly expected to be a long-term effect of the COVID-19 responses. In addition,
the responses related to health care workers and public transportation as indicated in the text network are
expected to be long-term.
Figure 3 and
Table 3 present the text network and top keyword co-occurred and collocated keywords for short-
term responses of COVID-19. Most keywords that were in the long-term responses network are also
available in the short-term responses network. Such keywords include street, city, bicycle, space, park, and
lane. This observation implies that regardless of the longevity of the response, the same
infrastructure/attribute are affected. All the collocated keywords are statistically significant at a 95%
confidence interval. The collocated keywords motor vehicle have the strongest association with the lambda
value of 8.11.
14
Figure 3. Text network for Temporary/Short-term Responses
Table 3 Text Network Performance Metrics for Short-term Responses
Keyword Frequency
Co-occurrence Frequency
Collocation Association
Keywords
Association parameters
Keyword
Count
Keywords
Count
Count
Lambda
Z-value
street
540
227
bicycle lane
136
bicycle lane
96
4.73
30.03
city
360
218
traffic street
108
physical distancing
86
7.73
26.47
bicycle
299
178
physical distancing
87
outdoor dining
63
7.31
25.45
space
223
160
outdoor dining
66
public transport
48
6.05
22.28
park
354
157
shared street
65
shared street
36
4.59
15.60
temporary
196
152
city street
64
slow street
34
5.89
11.71
lane
233
131
temporary lane
61
healthcare workers
32
6.80
19.54
close
176
129
public transport
55
motor vehicle
32
8.11
12.20
traffic
186
123
space street
53
more space
31
4.08
17.86
distancing
156
122
parking space
51
social distancing
31
7.02
15.86
Key: Docfreq means Document frequency
For instance, whether the response is short or long-term is expected to affect the streets where the responses
are applied. However, several keywords that emerged in the short-term responses were not in the long-term
responses network. For instance, the keyword temporary signifies that the response was for temporary
purposes. For instance, one observation indicated that a temporary bike lane was installed in Austin, Texas,
15
stating: “Temporary bike lanes installed on Congress Ave.”. Furthermore, collocated and co-occurred
keywords outdoor dining, slow street, healthcare workers clearly show the responses were temporary. For
instance, the outdoor dining involved the removal/closure of the roadway. Such a response cannot be
permanent/ long-term. Also, responses for healthcare workers, such as reduced/free passes for public
transport or bike share, are implemented temporarily.
Figure 4 presents the text network for indefinite observations. The content of the text networks is
similar to the two previous networks. The keywords street, social, and distancing are the major keywords
in this network as they were in the previous two networks. Other common keywords include bus,
pedestrians, transport, shared, riders, and drivers, among others. The network, however, does not have the
keyword temporary, which was in the short-term network. The remembrance of the indefinite network
signifies that the observation can be either short-term or long-term.
Figure 4. Text network for Indefinite Responses
16
3.3. Logistic Regression Results and Discussion
Table 4 presents logistic regression results. Two logistic regression models were developed considering
the “indefinite” observations. It should be noted that, as discussed earlier, the indefinite longevity outcome
might end up being either permanent or temporary. Thus, the first model considered indefinite longevity as
the long-term, while the second considered them as short term.
Table 4 Logistic Regression Results
Indefinite as long-term
Indefinite as short-term
Estimate
OR
P-value
Estimate
OR
P-value
Time (all day every day)
No
Base
Yes
1.382
3.98
0.007
0.485
1.62
0.545
Space coverage
Entire roadway
Base
Intersection
0.454
1.57
0.642
-0.710
0.49
0.604
Curb
0.358
1.43
0.391
1.670
5.31
0.005
Travel lanes
-0.131
0.88
0.677
1.477
4.38
0.003
Parking lane
0.518
1.68
0.282
0.836
2.31
0.282
Parks/plazas/sidewalk
0.627
1.87
0.455
0.527
1.69
0.682
Purpose
Moving people
Base
Economic recovery
-3.975
0.02
0.004
-4.160
0.02
0.138
Public health
-0.702
0.50
0.031
-1.931
0.14
0.005
Safety
1.176
3.24
0.047
0.868
2.38
0.181
Function
Space for bike/ped
Base
Space for commerce
2.208
9.10
0.110
2.392
10.93
0.398
Other bike/ped
1.563
4.77
0.073
3.468
32.07
0.007
Others
0.550
1.73
0.418
1.041
2.83
0.226
Category
Physical
Base
Operational
-0.166
0.85
0.653
-0.141
0.87
0.837
Regulatory
-2.553
0.08
0.004
-3.040
0.05
0.048
Intercept
-1.957
0.14
0.000
-3.011
0.05
0.000
Model Summary
Number of observations
487
AIC
501.4
303.7
BIC
564.2
366.5
The discussion of the result uses the Odds Ratios (OR) and associated p-values. The ORs are
computed by exponentiating the estimated coefficients. The OR greater than 1 implies that the variable
category is associated with the increased likelihood of long-term effects, on the other hand, OR less than
17
one implies that the variable is not associated with the increased likelihood of long-term effects (Boniphace
Kutela & Teng, 2019; Woodridge, 2012).
According to the results in Table 4, the odds of the COVID-19 response causing a permanent
change of the transportation facilities varies significantly per space, purpose, function, category, and time
that the response covers. Furthermore, the logistic regression results provide the association between
various responses and the likelihood of being long-term or short-term. However, the analysis could not
portray the exact content of such responses. Therefore, the text mining of selected responses of interest was
performed to understand the content of various responses. The section also presents the text mining results
for the following types of responses: responses taking place in travel lanes, responses with function “other
bike/ped,” and responses whose implementation approach was categorized as “regulatory.” The selection
was based on the fact that these values were the only ones that were statistically significant at a 90%
confidence level. The text mining results provide more information on the content of the variables of interest
than the traditional logistic regression model results.
The logistic regression results show a few dissimilarities when the indefinite observations are
considered as long-term compared to short-term. Such dissimilarities involve the changes in the magnitudes
and directions of the estimates as well as changes in the statistical significance level. The next section
presents the discussion of the results focusing on the five variables of interest.
Purpose of the response
According to the results in Table 4, the responses for economic recovery and public health are less
likely to result in permanent changes in the transportation infrastructures. On the other hand, safety-related
responses are more likely to result in permanent changes. All three categories are statistically significant at
a 95% confidence level when the indefinite observations are considered as long-term. On the other hand,
only public health is statistically significant at the same level when the indefinite observations are
considered as short-term. Study results show that the economic recovery responses' odds of permanent
18
changes in transportation infrastructure are 98% lower than that of moving people. Further, the odds of the
responses for public health are about 50% to 86% lower than that of moving people. Lastly, the safety-
related responses' odds are between 2.3 and 3.3 times more likely to result in permanent changes of
transportation infrastructures.
Space Coverage
Results in Table 4 shows that compared to the responses that covered an entire roadway, all other
responses are not statistically significant at a 90% confidence level if the indefinite observations are
considered as long-term. On the other hand, responses that affected curb and travel lanes are statistically
significant at a 95% confidence level. The responses that affected travel lanes and intersections showed the
changes in the direction of the estimate. The odds ratios for responses that affected travel lanes suggest that
responses that such responses are 43% less likely to be permanent but also about four times more likely to
be permanent, depending on how the infinite observations are considered. The less likelihood of being long-
term responses can be explained by the importance and functionality of travel lanes for moving people.
Thus, any modifications/changes of use of travel lanes are likely to be temporary. In fact, Figure 5 shows
that responses applied on the travel lanes mainly focused on the bicycle lanes as were temporary, as
indicated by a large node of the keywords lane, bicycle, and temporary.
19
Figure 5. Text network for Responses affected Travel Lanes
Furthermore, the intersection-related responses are about 57% more likely and 51% less likely to
be long-term, depending on the consideration of the indefinite observations. Further, curb, parking lane,
and parks/plan responses are more likely to be permanent. In fact, the responses that affected parking lanes
are 69% - 87% more likely to be permanent. Likewise, curb and park/plaza responses are 46% and 67%,
respectively, more permanent. This observation can be attributed to the type of responses adopted in the
locations. For instance, at the intersections/curbs, the responses that involved modifying curbs or
automating walk signals are likely to be permanent.
Function
Responses that created space for commerce, provided other bicycle or pedestrian supports, or had
other or miscellaneous functions are likely to be permanent as compared to responses to create space for
walking and cycling. Spaces created for commerce are between 9 and 11 times more likely to be permanent
compared to space for walking and cycling. Similarly, responses for other bike/ped supports are between 4
and 33 times more likely to be permanent compared to space for bike/ped, while responses for others are
20
about 1.7-2.9 times likely to be permanent. However, only responses for “other bike/ped” are statistically
significant at a 90% confidence level. Further, according to the text mining results in Figure 6, the responses
for other bike/pedestrian supports involve changes at the intersections, which may include the signals, walk
signal automation, and reductions in speed limit, which are likely to be permanent. On the other hand, space
for commerce tended to involve blocking entire sections of roads, which is less likely to be a permanent
option.
Figure 6. Text network for Other Bike/Pedestrians Responses
Category
As shown in Table 4, there are three main implementation approaches (category variable):
physical, operational, and regulatory. The OR of the operational and regulatory are less than one, implying
that these responses are less likely to be permanent when compared to be physical. In fact, regulatory
responses are 92%-95% less likely to be permanent when to compared to physical responses. Similarly,
operational responses are 13%-15% less likely to be permanent. However, only regulatory responses are
statically significant at a 95% confidence level. This observation reflects that the regulatory responses were
21
more likely to be simplified than others and thus are likely to be permanent. Conversely, operational
responses involve substantial infrastructures changes that are less likely to be permanent. Figure 7 shows
key patterns for regulatory responses. It can be observed that most of these responses involved parking.
This observation is based on the size of the node for the keyword parking. The co-occurrence of the
keywords in Figure 7 suggests that regulatory responses affected parking meter, parking permit, parking
suspended, on-street parking, free parking, among others. It is clear that the parking-related responses,
whose summary is presented in Figure 7, are likely to be on a short-term basis.
Figure 7. Text network for Regulatory Responses
4. Conclusions and Future Studies
This study seeks to explore the long-term impact of COVID-19 on transportation facilities. It uses data
from the Shifting Streets COVID-19 Mobility Dataset, which documents ways in which cities around the
world modified streets to provide more spacing for social distancing while performing walking, biking or
22
other business. Two analysis methods—text mining and logistic regression—are applied to explore the
long-term outcomes of street-level responses to COVID-19.
This study found that responses occurring in travel lanes, the creation of space for outdoor dining,
measures to improve access to work for healthcare workers were intended for the short term. On the other
hand, new bicycle lanes and full street closures (a.k.a. ‘open streets’) were more commonly intended to
outlast the pandemic. Partial street closures (a.k.a. 'shared streets’), improvements to public transport,
responses meant to curtail disease transmission, and responses taking place inside parks were intended as
both short-term and long-term interventions. The text network provided the low-level details of the
responses.
The logistic regression results intended to explore the likelihood of the response being long-term (i.e.,
outlasting the pandemic). In this analysis, five variables were of interest: the temporal coverage of the
response (time), the street space affected (space), the purpose and function of the response, and the
implementation approach (category). The study found that responses that were applied all day, every day
were more likely to be long-term. Further, responses that covered some, but not all travel lanes only were
less likely to be long-term than responses that covered the entire roadway. However, the findings for this
kind of response do change depending on the consideration of the indefinite observations. Regarding the
purpose, compared to responses for moving people, responses for economic recovery and public health
were less likely to be long-term while those implemented primarily to improve safety were more likely to
be long-term. Furthermore, responses that focused on other (i.e., non-street space) supports for walking and
cycling were more likely to be long term compared to those that created street space for walking and
cycling. Lastly, regulatory responses were less likely to be long-term than to physical responses.
The study findings have several implications in the post-COVID-19 era. Transportation researchers,
practitioners, and the public are looking forward to a “new normal.” Our analyses suggest that new normal
will include several modifications to the use and function of public roadways. Responses that require
limited resources and/or involve limited changes to traffic patterns, such as automation of walk signal
23
automation and speed limit reductions, are expected to remain beyond the pandemic. More substantial
changes, such as reallocations of travel lanes to new uses and new intersection configurations, may also be
continued in a post-COVID-19 era. Use of on-street parking spaces for other uses may also be a common
post-pandemic feature. More data is needed to support analysis into the role of other factors, such as the
planning processes underlying cities’ responses and the distribution of costs and benefits of those responses,
to further improve our understanding of the long-term impacts of COVID-19 on the transportation sector.
Furthermore, future studies may focus on the heterogenous results across regions, continents, and high vs.
low-income settings.
5. Acknowledgements
Authors would like to thank all the contributors to the Shifting Streets COVID19 Mobility Dataset.
This includes the contributors for Local actions to support walking and cycling during social distancing
dataset, COVID-19 Livable Streets Response Strategies dataset, COVID Mobility Works dataset.
6. Funding Sources
This research did not receive any specific grant from funding agencies in the public, commercial, or
not-for-profit sectors.
7. Declaration of interest: None.
8. Author Contributions
Boniphace Kutela and Neema Langa Conceptualization; Tabitha Combs Data curation;
Boniphace Kutela and Rafael John, Mwekh’iga Formal analysis; Boniphace Kutela and Rafael John,
Mwekh’iga Methodology; Boniphace Kutela and Neema Langa Roles/Writing - original draft;
Boniphace Kutela and Tabitha Combs Writing - review & editing.
24
9. References
Abu-Rayash, A., & Dincer, I. (2020). Analysis of mobility trends during the COVID-19 coronavirus
pandemic: Exploring the impacts on global aviation and travel in selected cities. Energy Research &
Social Science, 68, 101693. https://doi.org/10.1016/J.ERSS.2020.101693
Advani, M., Sharma, N., & Dhyani, R. (2021). Mobility change in Delhi due to COVID and its’
immediate and long-term impact on demand with intervened non motorized transport friendly
infrastructural policies. Transport Policy, 111, 28–37.
https://doi.org/10.1016/J.TRANPOL.2021.07.008
BBC. (2022). Coronavirus pandemic: Tracking the global outbreak - BBC News. Retrieved September 5,
2020, from https://www.bbc.com/news/world-51235105
Benoit, K., Watanabe, K., Wang, H., Nulty, P., Obeng, A., Müller, S., & Matsuo, A. (2018). quanteda: An
R package for the quantitative analysis of textual data. Journal of Open Source Software, 3(30).
https://doi.org/10.21105/joss.00774
Blaheta, D., & Johnson, M. (2011). Unsupervised learning of multi-word verbs *. Proceedings of the ACL
Workshop on Collocations, 54–60.
Combs, T. S., & Pardo, C. F. (2021). Shifting streets COVID-19 mobility data: Findings from a global
dataset and a research agenda for transport planning and policy. Transportation Research
Interdisciplinary Perspectives, 9, 100322. https://doi.org/10.1016/J.TRIP.2021.100322
Csárdi, G. (2020). “igraph”; Network Analysis and Visualization. Retrieved from https://igraph.org/r/
Firth, C. L., Baquero, B., Berney, R., Hoerster, K. D., Mooney, S. J., & Winters, M. (2021). Not quite a
block party: COVID-19 street reallocation programs in Seattle, WA and Vancouver, BC. SSM -
Population Health, 14, 100769. https://doi.org/10.1016/J.SSMPH.2021.100769
Fischer, J., & Winters, M. (2021). COVID-19 street reallocation in mid-sized Canadian cities: socio-
25
spatial equity patterns. Canadian Journal of Public Health, 112(3), 376–390.
https://doi.org/10.17269/S41997-020-00467-3/FIGURES/8
Habib, M. A., & Anik, M. A. H. (2021). Examining the long-term impacts of COVID-19 using an
integrated transport and land-use modelling system.
Https://Doi.Org/10.1080/12265934.2021.1951821, 25(3), 323–346.
https://doi.org/10.1080/12265934.2021.1951821
Hunter, R. F., Garcia, L., de Sa, T. H., Zapata-Diomedi, B., Millett, C., Woodcock, J., … Moro, E.
(2021). Effect of COVID-19 response policies on walking behavior in US cities. Nature
Communications, 12(1), 1–9. https://doi.org/10.1038/s41467-021-23937-9
Kim, Y., & Jang, S.-N. (2018). Mapping the knowledge structure of frailty in journal articles by text
network analysis. https://doi.org/10.1371/journal.pone.0196104
Kutela, B., Novat, N., & Langa, N. (2021). Exploring geographical distribution of transportation research
themes related to COVID-19 using text network approach. Sustainable Cities and Society, 67.
https://doi.org/10.1016/j.scs.2021.102729
Kutela, Boniphace, Das, S., & Dadashova, B. (2021). Mining patterns of autonomous vehicle crashes
involving vulnerable road users to understand the associated factors. Accident Analysis &
Prevention, 106473. https://doi.org/10.1016/J.AAP.2021.106473
Kutela, Boniphace, Langa, N., Mwende, S., Kidando, E., Kitali, A. E., & Bansal, P. (2021). A text mining
approach to elicit public perception of bike-sharing systems. Travel Behaviour and Society, 24, 113–
123. https://doi.org/10.1016/j.tbs.2021.03.002
Kutela, Boniphace, & Teng, H. (2019). Exploring the associated factors for multiple-threats and near-miss
incidents at signalized midblock crosswalks. Journal of Transportation Safety & Security, 1–22.
https://doi.org/10.1080/19439962.2019.1638476
26
Kutela, Boniphace, & Teng, H. (2020). Evaluating the influential factors for pushbutton utilization at
signalized midblock crosswalks. Safety Science, 122, 104533.
https://doi.org/10.1016/j.ssci.2019.104533
Kutela, Boniphace, & Teng, H. (2021). The Use of Dynamic Message Signs (DMSs) on the Freeways: An
Empirical Analysis of DMSs Logs and Survey Data. Journal of Transportation Technologies,
11(01), 90–107. https://doi.org/10.4236/jtts.2021.111006
Marra, A. D., Sun, L., & Corman, F. (2022). The impact of COVID-19 pandemic on public transport
usage and route choice: Evidences from a long-term tracking study in urban area. Transport Policy,
116, 258–268. https://doi.org/10.1016/J.TRANPOL.2021.12.009
Mayo, J. (2021). Lane Reallocations During COVID: A Comparison of Interventions and Decision-
Making Process. https://doi.org/10.17615/1V4T-1D46
Paranyushkin, D. (2011). Identifying the Pathways for Meaning Circulation using Text Network
Analysis. Venture Fiction Practices, 2(4). Retrieved from www.noduslabs.com
R Core Team. (2021). R: A Language and Environment for Statistical Computing. R Foundation for
Statistical Computing. Retrieved August 17, 2021, from https://www.r-project.org/
Ritchie, H., Mathieu, E., Rodes-Guirao, L., Appel, C., Giattino, C., Ortiz-Ospina, E., … Rose, M. (2022).
Coronavirus (COVID-19) Cases - Statistics and Research - Our World in Data. Retrieved January
16, 2022, from https://ourworldindata.org/covid-cases
Shirgaokar, M., Reynard, D., & Collins, D. (2021). Using twitter to investigate responses to street
reallocation during COVID-19: Findings from the U.S. and Canada. Transportation Research Part
A: Policy and Practice, 154, 300–312. https://doi.org/10.1016/J.TRA.2021.10.013
Suau-Sanchez, P., Voltes-Dorta, A., & Cugueró-Escofet, N. (2020). An early assessment of the impact of
COVID-19 on air transport: Just another crisis or the end of aviation as we know it? Journal of
27
Transport Geography, 86, 102749. https://doi.org/10.1016/J.JTRANGEO.2020.102749
Truong, D. (2021). Estimating the impact of COVID-19 on air travel in the medium and long-term using
neural network and Monte Carlo simulation. Journal of Air Transport Management, 96, 102126.
https://doi.org/10.1016/J.JAIRTRAMAN.2021.102126
Van Wee, B., & Witlox, F. (2021). COVID-19 and its long-term effects on activity participation and
travel behaviour: A multiperspective view. Journal of Transport Geography, 95, 103144.
https://doi.org/https://doi.org/10.1016/j.jtrangeo.2021.103144
Vecchio, G., Tiznado-Aitken, I., & Mora-Vega, R. (2021). Pandemic-related streets transformations:
Accelerating sustainable mobility transitions in Latin America. Case Studies on Transport Policy,
9(4), 1825–1835. https://doi.org/10.1016/J.CSTP.2021.10.002
WHO. (2020). WHO | Novel Coronavirus – China. WHO.
Woodridge, J. M. (2012). Introductory Economics A modern Approach (5th ed.). Retrieved from
http://economics.ut.ac.ir/documents/3030266/14100645/Jeffrey_M._Wooldridge_Introductory_Eco
nometrics_A_Modern_Approach__2012.pdf
Wright, H., & Reardon, M. (2021). COVID-19: a chance to reallocate street space to the benefit of
children’s health? https://doi.org/10.1080/23748834.2021.1912571
Yoon, B., & Park, Y. (2004). A text-mining-based patent network: Analytical tool for high-technology
trend. Journal of High Technology Management Research, 15(1), 37–50.
https://doi.org/10.1016/j.hitech.2003.09.003
Zhang, R., & Zhang, J. (2021). Long-term pathways to deep decarbonization of the transport sector in the
post-COVID world. Transport Policy, 110, 28–36.
https://doi.org/10.1016/J.TRANPOL.2021.05.018