Global and local associations between urban greenery and travel propensity of older adults in Hong Kong

Abstract

A significant increase in the older population has been observed in numerous cities. Understanding the correlates of older adults’ mobility is, therefore, of critical importance to meet the needs of this growing population. Based on multisource data, including Google Street View street image data and Hong Kong Travel Characteristics Survey 2011 data, we developed a series of multilevel and geographically weighted logistic regression models to scrutinize the global and local associations between urban greenery (eye-level street greenery, the number of parks, and the normalized difference vegetation index (NDVI)) and the travel propensity of older adults. Notably, eye-level street greenery was assessed by using readily available street images and machine learning techniques. The modeling results reveal that eye-level street greenery has a positive effect on older adults’ mobility, but that the number of parks and the NDVI do not significantly affect older adults’ mobility. Robustness checks verified the plausibility of these findings. Furthermore, the effect of street greenery varies over space, larger in suburban areas than in urban areas. This study advances the understanding of relationships between urban greenery and travel behavior.

Full text

1
Global and local associations between urban greenery and travel propensity of
older adults in Hong Kong
Linchuan Yanga, Jixiang Liub, Yi Luc, Yibin Aod,
Yuanyuan Guoe, Wencheng Huangf, Rui Zhaog*, Ruoyu Wangh
aDepartment of Urban and Rural Planning, Southwest Jiaotong University, Chengdu, Sichuan Province, China
yanglc0125@swjtu.edu.cn
bDepartment of Urban Planning and Design, The University of Hong Kong, Hong Kong, China
u3004679@hku.hk
cDepartment of Architecture and Civil Engineering, City University of Hong Kong, Hong Kong, China
yilu24@cityu.edu.hk
dCollege of Environment and Civil Engineering, Chengdu University of Technology, Chengdu, Sichuan Province,
China aoyibin10@mail.cdut.edu.cn
eDepartment of Geography and Resource Management, The Chinese University of Hong Kong, Hong Kong,
China guoyuanyuan@link.cuhk.edu.hk
fSchool of Transportation and Logistics, Southwest Jiaotong University, Chengdu, Sichuan Province, China
1261992248@qq.com
gFaculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu, Sichuan
Province, China ruizhaoswjtu@hotmail.com
hInstitute of Geography, School of GeoSciences, University of Edinburgh, Edinburgh, UK
r.wang-54@sms.ed.ac.uk
To cite: Yang, L., Liu, J., Lu, Y., Ao, Y., Guo, Y., Huang, W., Zhao, R., & Wang, R.
(2020). Global and local associations between urban greenery and travel propensity of
older adults in Hong Kong. Sustainable Cities and Society,63, 102442.
Abstract: A significant increase in the older population has been observed in
numerous cities. Understanding the correlates of older adults’ mobility is, therefore,
of critical importance to meet the needs of this growing population. Based on
multisource data, including Google Street View street images and Hong Kong Travel
Characteristics Survey 2011 data, we developed a series of multilevel and
geographically weighted logistic regression models to scrutinize the global and local
associations between urban greenery (eye-level street greenery, the number of parks,
and the normalized difference vegetation index (NDVI)) and the travel propensity of
older adults. Notably, eye-level street greenery was assessed by using readily
available street images and machine learning techniques. The modeling results reveal
that eye-level street greenery has a positive effect on older adults’ mobility, but the
number of parks and the NDVI do not significantly affect the mobility. Robustness
checks verified the plausibility of these findings. Furthermore, the effect of street
greenery varies over space, larger in suburban areas than in urban areas. This study
advances the understanding of relationships between urban greenery and travel
behavior.
Keywords: Old adult, Travel propensity, Street greenery, Population aging, Mobility
behavior, Google Street View, Geographically weighted logistic regression model

2
1. Introduction
Population aging is a conspicuous, pervasive, and enduring global demographic
phenomenon. The proportion of older adults (defined as those aged 60 years or over
here) is observed to be growing in both developed and developing countries,
including Japan (where 33.4% of people were older adults in 2017), Italy (29.4%),
Germany (28.0%), and Portugal (27.9%). Internationally, the proportion of older
adults was 12.7% in 2017 and is predicted to be 21.3% in 2050. The number of the
oldest adults (aged 80 years or over) is also rapidly increasing and will reach 425
million by 2050, more than triple the 2017 figure (137 million) (Population Division,
2017). Like many regions worldwide, Hong Kong is undergoing a dramatic change in
demographic profile. Hong Kong has the second-highest proportion of older adults in
Asia (23.5%), exceeded only by Japan (33.4%) that tops in global rankings. Moreover,
the older population in the region is predicted to grow progressively, with its
proportion being 40.6% by 2050 (Population Division, 2017). While approximately
one in five Hong Kong people were aged 60 years or over in 2017, this figure will rise
to about two in five within around three decades.
It is widely accepted in a huge body of existing literature that transportation
mobility (abbreviated to “mobility” hereafter) is of paramount importance to older
adults and becomes a prerequisite for healthy and active aging (Böcker, van Amen, &
Helbich, 2017; Wong, Szeto, Yang, Li, & Wong, 2017, 2018). For older adults,
mobility is closely related to independence, autonomy, good health, life satisfaction,
social integration/engagement/inclusion, quality of life, and well-being (Giuliano, Hu,
& Lee, 2003; Metz, 2000; Böcker et al., 2017; Tacken, 1998; Nordbakke & Schwanen,
2015). Banister & Bowling (2004) asserted that mobility is one of the three essential
components of older adults’ quality of life on the transport dimension, along with
locality and social networks. Metz (2000) and Tacken (1998) proposed that mobility
is closely connected to quality of life during old age. Giuliano et al. (2003) and
Olawole & Aloba (2014) noted that for older adults, mobility is a necessary but
insufficient requirement of social integration, which may result in physical and
psychological well-being (Wentowski, 1981). Delbosc & Currie (2011) concluded
that mobility is correlated with social inclusion, quality of life, and well-being for the
transport-disadvantaged (e.g., older adults and the disabled). Li (2020) presented that
mobility is part and parcel of well-being for older adults. However, due to
deteriorating health, physical fragility, and visual/hearing/cognitive impairments
(Cirella, Bąk, Kozlak, Pawłowska, & Borkowski, 2019), the mobility of older adults
is arguably lower than that of young adults (Rantakokko, Mänty, & Rantanen, 2013).
Improving the mobility of older adults is, therefore, of paramount importance for a
society that equitably addresses their needs. It should be deemed essential for future
urban planning and policy.
Understanding the mobility characteristics and patterns of older adults is an
essential part of and the first step in reaching the goal of older adults’ mobility
improvement. A branch of studies utilizes descriptive analyses to investigate this issue
in a host of cities or countries, such as the United States (Collia, Sharp, & Giesbrecht,
2003; Rosenbloom, 2001; Buehler & Nobis, 2010), Canada (Newbold, Scott, Spinney,
Kanaroglou, & Páez, 2005), England (Schmöcker, Quddus, Noland, & Bell, 2005),
Germany (Buehler & Nobis, 2010), Australia (Rosenbloom & Morris, 1998),
Netherland (Schwanen, Dijst, & Dieleman, 2001; Tacken, 1998), Hong Kong (Szeto,
Yang, Wong, Li, & Wong, 2017), South Korea (Choi et al., 2014), and Nigeria
(Ipingbemi, 2010; Olawole & Aloba, 2014). Such studies are evidently useful but still
not always adequate for targeted policy prescriptions. Another branch of studies

3
concentrates on identifying the correlates of older adults’ mobility using statistical
models. Moreover, in existing literature, mobility is assessed in many ways, including
but not limited to trip frequency (the number of trips), travel propensity
(willingness/propensity to travel), travel time/distance, and trip chain frequency.
Travel propensity or trip-making propensity, meaning whether to “travel by any mode
of transport for any purpose” in a predefined amount of time (Beaver, 2012), is a
useful mobility measure, especially for people with reduced mobility (e.g., older
adults and the disabled) (Schwanen et al., 2001). This measure, however, has rarely
been investigated in existing literature (Yang, 2018) in comparison to others (e.g., trip
frequency) (Pa ´ez, Scott, Potoglou, Kanaroglou, & Newbold, 2007; Roorda, Páez,
Morency, Mercado, & Farber, 2010).
Urban greenery has a wide variety of social, economic, ecological, (physical and
physiological) health, and well-being benefits (Hartig, Mitchell, De Vries, & Frumkin,
2014; Xiao, Lu, Guo, & Yuan, 2017; de Keijzer, Bauwelinck, & Dadvand, 2020;
Kaczynski & Henderson, 2008; La Rosa, Takatori, Shimizu, & Privitera, 2018;
Włodarczyk-Marciniak, Sikorska, & Krauze, 2020). Urban greenery may attract or
stimulate people to go out of the door and improve the mobility of people with
preferences toward things in nature (biophilia) (Kellert & Wilson, 1993). This is more
applicable to older adults (Kemperman & Timmermans, 2006; Jorgensen &
Anthopoulou, 2007; La Rosa et al., 2018) compared to young adults, partially because
most trips taken by this population subgroup are not compulsory (commuting or going
to school) but discretionary. Due to the rapid urbanization that diminishes human
contact with nature (Hartig et al., 2014; Zhang & Ramírez, 2019; Zhang, Xia, Chu,
Lin, Li, & Wu, 2019; Bao et al., 2019; Bao & Lu, 2020; Chi, Lu, Ye, Bao, & Zhang,
2020), urban greenery, especially street greenery, has aroused increasing scholarly
attention and come into the limelight in recent years. Still, its effect on mobility has
mostly been studied for the general population rather than for a population subgroup
(e.g., older adults or children) (Hartig et al., 2014; Wen, Albert, & Von Haaren, 2018).
The majority of previous studies have overlooked the role of urban greenery in
shaping older adults’ mobility (Feng, Dijst, Wissink, & Prillwitz, 2013, 2015; Cheng,
Caset, De Vos, Derudder, & Witlox, 2019; Cheng, Chen, et al., 2019; Böcker et al.,
2017; Cheng, De Vos, et al., 2019). A handful of recently emerged studies have
turned their focus to urban greenery (Yang, He, Gou, Wang, Liu, & Lu, 2019; Zang,
Lu, Ma, Xie, Wang, & Liu, 2019).
To address these issues, based on multisource data, including Google Street
View (GSV) street greenery data and Hong Kong Travel Characteristics Survey 2011
(TCS 2011) data, we developed a series of multilevel logistic regression models to
estimate the global association or “average relationship” (Fotheringham, Brunsdon, &
Charlton, 2002, p. 2) that is assumed to be spatially constant/stationary/homogeneous
between urban greenery and the travel propensity of older adults. Three urban
greenery measures were used: eye-level street greenery assessed by GSV images, the
number of parks, and the normalized difference vegetation index (NDVI). Their
effects on older adults’ mobility were then compared, and robustness checks were
conducted to gain confidence in the findings. Moreover, in a departure from existing
literature that only focuses on the global association between street greenery and
mobility indicators, a geographically weighted logistic regression model was
developed to estimate the local association that may be spatially
varying/non-stationary/heterogeneous. The contributions of this paper include: (1)
analyzing the relationship between different urban greenery measures and older
adults' mobility; (2) confirming the role of street greenery in shaping older adults'

4
mobility; (3) proposing the spatially varying effect of street greenery on older adults'
mobility; and (4) revealing that street greenery has a larger effect in suburban areas
than in urban areas.
The remainder of this paper is organized as follows. The ensuing section
(Section 2) presents a review of studies on older adults' mobility and the built
environment. Section 3 shows the TCS 2011 data, the measurement of urban greenery,
and variables utilized in this study. Section 4 introduces the standard, multilevel, and
geographically weighted logistic regression model successively. Section 5 shows the
results of global and local modeling and robustness checks. Section 6 discusses the
strengths and limitations of this study and suggests avenues for future research. The
final section (Section 7) concludes the paper.
2. Literature review
It is widely recognized that mobility (or travel) behavior is affected by the
socio-demographic characteristics of the individual (e.g., age, gender, and car
availability) and the living environment (natural environment + built environment +
social environment) (Cervero & Kockelman, 1997; Ewing & Cervero, 2010; Lu, Chen,
Yang, & Gou, 2018; Ao, Yang, Chen, & Wang, 2019a, 2019b; Du, Cheng, Li, &
Yang, 2020; Liu, Kemperman, & Timmermans, 2020). Numerous empirical studies
have investigated the role of built environment attributes (e.g., density and diversity)
in shaping the mobility of the general population (irrespective of age) and produced
fruitful outcomes. However, such findings for the general population may not be
easily extended to older adults, a population subgroup with many distinct
characteristics (e.g., deteriorating health, visual/hearing/cognitive impairments, and
dominance of discretionary trips).
Limited studies, which were mostly conducted in developed regions or countries,
specifically focus on older adults. Table 1 provides a review of selected studies on
built environment correlates of older adults' mobility. As expected, numerous built
environment attributes are considered in these empirical studies, including population
density, land use mix, four-way intersection density, and accessibility to transit (Pa´ez,
Scott, Potoglou, Kanaroglou, & Newbold, 2007; Roorda et al., 2010; Yang, 2018; Su
& Bell, 2009, 2012). They can be well incorporated into the famous “3Ds” or “5Ds”
framework, in which the built environment is categorized into three or five
dimensions, including density, diversity, design, distance to transit, and destination
accessibility (Cervero & Kockelman, 1997; Ewing & Cervero, 2010).
These empirical studies produced mixed, even conflicting results. For example,
transit accessibility is found to have no bearing on older adults' mobility in many
cities of the United States (Evan, 2001; Hess, 2009). By contrast, transit is frequently
identified as a salient mobility tool that significantly influences older adults' mobility
in Hong Kong and Nanjing, China (Feng et al., 2013, 2015; Feng, 2017; Yang, 2018;
Yang & Cui, 2020). The difference in such empirical results is not too difficult to
interpret, which can be explained by the context: in car-dominant settings such as the
United States, Canada, Australia, and New Zealand, people mainly rely on the car
(mobility tool) to travel and participate in social activities, and transit plays a marginal
role in shaping older adults' mobility. By contrast, many East Asian cities, such as
Hong Kong, Singapore, Tokyo, and Beijing, have high transit ridership and can be
classified into transit-dependent cities. In such contexts, transit is heavily used by city
residents, particularly older adults (Szeto et al., 2017), so its accessibility should, in
theory, plays an indispensable part in defining older adults' mobility. A few existing
studies have corroborated this (e.g., Feng, 2017; Yang, 2018; Yang & Cui, 2020).

5
Among the built environment attributes investigated, density has received the
most exceptional scholarly attention, followed by diversity. Urban greenery has
recently come into the limelight, and its associations with older adults' mobility have
insufficiently been examined. Moreover, to our knowledge, existing studies have only
tested the global association between urban greenery and travel outcome measures of
older adults, while they have not yet focused on the local association. These are what
the present study attempts to do.

6
Table 1
Review of studies on older adults' mobility and the built environment
Reference
Study area
Sample
Mobility
measure(s)
Built environment
attributes
Econometric
model(s)
Findings
Evans (2011)
The United
States
1,573 non-drivers
aged 75 years or
over
Propensity to
travel by car,
walk, and transit
Transit availability,
population density,
employment density,
central city, etc.
Stepwise
discriminant
analysis
Transit availability is not significantly
related to propensity to travel.
Kim (2003)
Puget Sound
region, the
United
States
2,454 daily activity
records of retired
people aged 65
years or over
Non-home activity
time, travel time,
and travel distance
Retail employment density
and population density
Structural equation
model
Urban form does not significantly affect
older adults' mobility.
Hess (2009)
San José
and Buffalo,
the United
States
171 people aged 60
years or over in
Buffalo and 286 in
San José
Transit ridership
frequency and
transit ride
propensity
Transit accessibility and
percentage of
single-family homes
Linear and binary
logistic models
Transit accessibility has a significant
effect on transit ridership frequency in
San José, but not in Buffalo.
Pa’ez et al.
(2007)
Hamilton,
Canada
27,071 people aged
65 years or over
Number of work
and non-work trips
Distance to major activity
centers and coordinates
Mixed
ordered probit
model
Zonal variables are only marginally
significant.
Mercado & Páez
(2009)
Hamilton,
Canada
16,190 people aged
65 years or over
Mean travel
distance
Population density and
land use
Multilevel linear
regression model
Built environment variables are too
weak to shape older adults’ mean travel
distance.
Moniruzzaman,
Páez, Scott, &
Morency (2015).
Montreal,
Canada
13,127 people aged
65 years or over
Trip frequency of
walking, car, and
bus
Population density, job
density, land use mix,
destination accessibility,
Trivariate ordered
probit model
High job density is positively associated
with transit frequency, but negatively
with walking and car frequency.

7
etc.
Schwanen, Dijst,
& Dieleman
(2001)
The
Netherlands
28,419 people aged
50 years or over
living independently
Propensity of
travel for leisure
and mode choice
for leisure trips
City size
Binary or
multinomial
logistic regression
models
City size affects mode choice for leisure
trips, but not the propensity of travel for
leisure.
Su, Schmöcker,
& Bell (2009)
London, the
United
Kingdom
4,186 people aged
65 years or over
Shopping mode
choice
Bus accessibility and rail
accessibility
Multinomial logit
and nested logit
models
Bus accessibility significantly affects
the transit mode choice, but rail
accessibility not.
Truong &
Somenahalli
(2015)
Adelaide,
Australia
117 people aged 70
years or over
Frequency of
public transport
use
Distance to CBD and
transit stop density
Multinomial
logistic regression
model
Distance to CBD and transit stop
density are not significant in shaping
public transport travel frequency.
Pettersson &
Schmöcker(2010
)
Manila, the
Philippines
10,680 peopled aged
60 years or over
Trip frequency
and tour
complexity
Population density
Ordered probit
model
Population density is significantly
related to trip frequency and tour
complexity.
Koh, Leow,
Wong (2015)
Singapore
168 people aged 65
years or over
Walking time
Presence of shops,
eateries, and sheltered
social interaction areas,
directional sign,
stairs/slope, etc.
Linear regression
model
The presence of shops, eateries, and
sheltered social interaction areas and
road crossing delays are significantly
correlated with walking time.
Feng et al.
(2013)
Nanjing,
China
547 people who are
over 50 years old
and retired
Trip generation
and distance
Population density and
distance to the nearest
metro station
Ordered and linear
regression models
Population density is negatively related
to both trip generation and distance.
Distance to the nearest metro station
adversely affects trip generation but
does not influence trip distance.

8
Feng et al.
(2015)
Nanjing,
China
2,090 people aged
60 years or over
Commuting/shopp
ing frequency and
distance
Population density and
distance to the nearest
metro station
Ordered and linear
regression models
Population density and distance to the
metro station mostly affect
commuting/shopping frequency and
distance.
Feng (2017)
Nanjing,
China
969 people who are
over 50 years old
and retired
Daily activity
participation and
travel distance
Population density, land
use mix, transit
accessibility, intersection
number, distance to the
nearest vegetable market,
etc.
Ordered probit and
linear regression
models
"[P]ublic transportation accessibility
instead of auto transportation
accessibility, vegetable markets instead
of supermarkets and convenience stores,
open spaces and parks along with chess
and card rooms instead of gyms and
sports centers" play a role in shaping
older adults' mobility.
Yang (2018)
Hong Kong,
China
19,703 people aged
60 years or over
Propensity to
travel by any
means
Bus accessibility
Mixed binary logit
model
Transit accessibility significantly affects
older adults’ propensity to travel.
Yang et al.
(2019)
Hong Kong,
China
10,700/1,083 people
aged 65 years or
over
Walking
propensity and
walking time
Population density,
land-use mix, intersection
density, number of retail
shops, urban greenery
exposure, etc.
Multilevel logistic
and linear
regression models
Urban greenery exposure positively
affects both the walking propensity and
walking time of older adults.
Yang & Cui
(2020)
Hong Kong,
China
19,703 people aged
60 years or over
Trip frequency
Bus accessibility and road
density
Poisson and
negative binomial
models
Bus accessibility and road density are
positively and negatively associated
with older adults’ trip frequency,
respectively.
Cheng, Chen,
Yang, Cao, De
Nanjing,
China
702 people aged 60
years or over
Daily active travel
frequency and
Population density, land
use mix, distance to the
Zero-inflated
ordered probit and
Distance to market, park/square, and
chess/card room significantly affects

9
Vos, & Witlox
(2019)
time
nearest market, distance to
the nearest park/square,
distance to the nearest
chess/card room, etc.
Cox proportional
hazards models
older adults’ active travel patterns.

10
3. Materials
3.1. Travel data
Travel behavior studies generally require a large amount of data. Therefore,
official travel surveys that often include thousands of observations are extensively
used for travel behavior analysis. In this study, the TCS 2011 data are used to measure
the travel propensity of older adults (the dependent variable) and socio-demographic
variables (some control variables).
From September 2011 to January 2012, a large-scale travel survey was carried
out by the Hong Kong Transport Department. The survey primarily documents trips
made by the respondents in a randomly assigned survey day (e.g., trip departure time,
duration, purpose, origin, and destination). The data collection process was suspended
during the Christmas and New Year holiday period. The socio-demographic attributes
(e.g., gender and education) of the respondents were also recorded. A total of 101,385
citizens in 35,401 households were randomly sampled (a sampling rate of 1.5%).
The widely used age threshold (or cut-off value) for the definition of “older
adults” is 60 or 65 years. TCS 2011 contains an attitudinal survey for older adults, in
which they are defined as those aged 60 years or over. This study follows that
definition. Among the 101,385 respondents, 20,062 persons were identified as older
adults; 19,703 older adults living in 1,815 street blocks had complete travel propensity
information, and 16,333 (82.9%) persons made at least one trip or went out of the
door on the reference day.
3.2. Measurement of urban greenery
Three types of urban greenery were objectively measured, including eye-level
street greenery, the NDVI, and the number of parks in a given neighborhood.
Following Handy, Cao, & Mokhtarian (2006) and Lu, Yang, Sun, & Gou (2019), the
neighborhood is defined as a 1600m network distance-based buffer around a dwelling
location. The network distance has a distinct advantage over the Euclidean distance in
representing spatial separation (e.g., natural or artificial obstacles).
The number of parks is a cumulative opportunity accessibility measure
(Vickerman, 1974), which counts the number of opportunities within a pre-specified
geographical area and overlooks the opportunities outside the area. The park data are
from the Land Department of Hong Kong. The parks have a minimal area of 500 m2
(Planning Department, 2015).Fig. 1 shows the geographic distribution of a total of
156 parks in Hong Kong.

11
Fig. 1. Geographic distribution of parks in Hong Kong
The NDVI, a typical vegetation index, estimates vegetative density and evaluates
the gross urban greenery level based on a satellite-based multispectral imagery dataset
(https://earthexplorer.usgs.gov/) (spatial resolution: 10 m). It is calculated as
NDVI = (NIR−Red) / (NIR+Red),
where NIR is the high reflectivity in the near-infrared band (central wavelength: 832.8
nm; bandwidth: 106 nm), and Red is the chlorophyll pigment absorptions of plants in
the red band (central wavelength: 664.6 nm; bandwidth: 31 nm). The NDVI ranges
from -1 to 1.
Previous studies have shown that eye-level street greenery are more associated
with people’s active travel (i.e., walking and cycling) than other greenery measures
since it reflects pedestrians’ actual perceived greenery on the streets (Lu et al., 2019).
Therefore, eye-level street greenery was also included in this study based on data
from GSV images. Its calculation process is as follows (Lu, Sarkar, & Xiao, 2018; Lu,
Yang, Sun, Gou, 2019; Helbich, Yao, Liu, Zhang, Liu, & Wang, 2019): First, all
street segments within the neighborhood of a dwelling location were extracted in
ArcGIS. Second, GSV-generating points were automatically created along the street
network at a constant spacing of 50 m, and their coordinates were recorded. Note that
the choice of the distance is a tradeoff between detail and computation time, and the
distance of 50 m can capture the details at an acceptable level (Helbich et al., 2019).
We selected sampling points along the streets for two reasons: (1). GSV images are
collected by dedicated GSV cars along the streets, so GSV images are available along
most, though not all, of the streets. (2). Constructing sampling points along the streets
(not elsewhere) can better measure pedestrians’ eye-level street greenery exposure
since pedestrians usually travel along the streets. Third, by putting the coordinate
information into a Python script developed by the authors, four GSV images that
collectively reveal the panorama were downloaded for each GSV-generating point,
with a 90° field of view and headings of north, east, south, and west. Four images

12
were collected for each point since they can well reflect pedestrians’ panoramic view
of the greenery exposure (Lu et al., 2019).
Previous studies typically extract greenery from street view images based on
RGB color difference (Lu, Sarkar, & Xiao, 2018). However, this may overestimate
the level of greenness since green objects such as green cars may be erroneously
included. Therefore, we used a machine-learning approach to extract vegetation
objects and calculate greenery from street view images. The automated
machine-learning vegetation extraction method, which is based on a fully
convolutional neural network (FCN-8s), was adopted to calculate the amount of
eye-level street greenery in each GSV image (Long, Shelhamer, & Darrell, 2015). The
authors wrote a Python script for the implementation of such a method. The ratio of
greenery pixels to total pixels in the images was used to evaluate the eye-level street
greenness for that point (Wang, Helbich, Yao, Zhang, Liu, Yuan, & Liu, 2019; Wang,
Lu, Wu, Liu, & Yao, 2020; Ye, Richards, et al., 2019; Ye, Zeng, Shen, Zhang, & Lu,
2019; Zhou, He, Cai, Wang, & Su, 2019). Fig. 2 offers an example of using GSV
images to objectively assess eye-level street greenery. The GSV-based calculation
method of eye-level street greenery for a GSV-generating point can be expressed as
= =1
4
=1
4
.
The arithmetic mean of “street greenness” for all GSV-generating points within
the neighborhood buffer zone is calculated to evaluate the eye-level street greenery
level of a dwelling location.
The applicability of the automated machine-learning vegetation extraction
method was further validated using manual extraction based on Adobe Photoshop
software. In a pilot study, we applied the two methods to 30 randomly selected GSV
images and compared their performance. Observe that the greenness values calculated
by the two methods were highly correlated (r = 0.91; p < 0.01). Notably, the
automated machine-learning vegetation extraction method's effectiveness has been
extensively confirmed in existing literature (Long et al., 2015; Helbich et al., 2019).
Fig. 2. Example of using GSV images to objectively assess eye-level street greenery
3.3. Variables

13
Table 2 shows the definitions and summary statistics of the variables. The
dependent variable, travel propensity, is binary. We also extracted socio-demographic
and geographic location variables considered in previous studies (He, Cheung, & Tao,
2018; Paez et al., 2007; Yang, 2018; Pettersson & Schmöcker, 2010; Roorda et al.,
2010) from TCS 2011. To measure built environment variables used for subsequent
analysis, street block data (Fig. 1) are necessary, which were downloaded from the
Hong Kong Planning Department's official website. The territory of Hong Kong was
officially divided into 4815 street blocks by the government. We accurately measured
five built environment characteristics to control the effects of confounders on older
adults' mobility and single out the impacts of urban greenery variables. The
measurement of the built environment follows the “5Ds” framework (Ewing &
Cervero, 2010), in which density, diversity, design, distance to transit, and destination
accessibility are considered concurrently. In this study, population density, land use
mix, intersection density, the number of bus stops, the number of recreational and
sports facilities are used to represent the “5Ds” variables.
A pairwise correlation analysis was conducted to determine the presence of
multicollinearity between the control variables. The results illustrate the absence of
multicollinearity in the data. Similarly, such an analysis was carried out for the three
urban greenery variables, and the outcomes point out the presence of the
multicollinearity between them. For example, the correlation coefficient between
Street greenery and NDVI is 0.74. Therefore, separate regressions are conducted in
subsequent analyses.
Table 2
Definitions and summary statistics of variables (N=19,703)
Variable
Description
Mean
Standard
Deviation
Data Source
Dependent variable
Travel
propensity
Binary variable; equals 1 for having walked out of
the door at least once (or made at least one trip) on
the reference day, and 0 otherwise
0.83
0.38
TCS
Control variables
Socio-demographic characteristics
Housing type
Binary variable; equals 1 for person living in a
private housing estate, and 0 otherwise
0.46
0.50
TCS
Household size
Number of persons living together
2.90
1.38
TCS
Male
Binary variable; equals 1 for male, and 0 for
female
0.48
0.50
TCS
Age
Age of older adults (years)
70.44
8.58
TCS
Employment
status
Binary variable; equals 1 for person with a job,
and 0 otherwise
0.15
0.36
TCS
Automobile
Binary variable; equals 1 for person with
household automobile availability, and 0
otherwise
0.08
0.28
TCS
Illness
Binary variable; equals 1 for person with illness,
and 0 otherwise
0.01
0.11
TCS

14
Geographic location
Hong Kong
Island
Binary variable; equals 1 for living in Hong Kong
Island, and 0 otherwise
0.21
0.40
TCS
Kowloon
Binary variable; equals 1 for living in Kowloon,
and 0 otherwise
0.34
0.47
TCS
Non-rural New
Territories
(reference
group)
Binary variable; equals 1 for living in the New
Territories non-rural area, and 0 otherwise
0.36
0.48
TCS
Rural New
Territories
Binary variable; equals 1 for living in the New
Territories rural area, and 0 otherwise
0.10
0.29
TCS
Built environment characteristics
Population
density
Population density of the Tertiary Planning Unit
(104/km2)
4.66
3.25
GIS
Land use mix
Mean entropy1for land use categories in the
Tertiary Planning Unit
0.53
0.28
GIS
Intersection
density
Street intersection density in the neighborhood
(1/km2)
46.12
23.48
GIS
Number of bus
stops
Number of bus stops in the neighborhood
166.88
99.75
GIS
Number of R&S
facilities
Number of recreational and sports facilities in the
neighborhood
163.64
79.57
GIS
Explanatory variables
Street greenery
Ratio of green pixels to total pixels
0.15
0.03
GSV
Number of
parks
Number of parks in the neighborhood
5.28
3.98
GIS
NDVI
NDVI value of the neighborhood
0.18
0.07
Multispectral
imagery
4. Methods
This section introduces the standard logistic regression model and its two
advanced versions, namely the multilevel logistic regression model and the
geographically weighted logistic regression model. The two advanced models use
more information embedded in the empirical data. The multilevel logistic regression
model addresses the hierarchical structure in the data, and it is still a global modeling
approach. By contrast, the geographically weighted logistic regression model uses the
location information of all observations and estimates the relationship between the
1In this study, land use mix is measured by entropy index, which is calculated as - i(pi
lnpi)/lnN,
where pidenotes the proportion of land-use category iand N is the number of land-use categories. In
this study, three land use categories were considered: residential, retail, and office.

15
dependent and independent variables that may vary across space. The presence of a
hierarchical structure and the availability of location attributes are the preconditions
of using the two models, respectively.
4.1. Logistic regression model
The (binary) logistic regression model, as a type of discrete choice model, is
extensively used to analyze conditions where the dependent variable has only two
outcomes. It is used in this study to relate the travel propensity to the aforementioned
independent variables. The model takes the following form:
=
1+,
Or equivalently, =()=( 
1−),
where is the probability that older adult imakes at least one trip on the reference
day, 
1−is often called the odds, and is the utility for older adult i, which
captures the factors influencing the travel decisions of older adult i. The relationship
between and the independent variables can be expressed as follows:
=0+
+,
where  is the kth independent variable for older adult i,is the coefficient of
,0is an intercept, and is the random error that reflects the stochastic
behavior of individuals and uncaptured variables and follows the logistic distribution.
0+
is the deterministic component of the utility, while is the random
component.
If the coefficient of a variable is significantly positive, the variable has a positive
effect on the travel propensity. Conversely, if the coefficient of a variable is
significantly negative, the variable adversely influences the travel propensity.
4.2. Multilevel logistic regression model
The standard logistic regression model overlooks the hierarchical structure in the
data. The multilevel (or random-effects) logistic regression model allows parameters
to vary at over one level. In this study, the random-intercept multilevel logistic
regression model is used, which allows random effects for each group (or street block,
as in this study). It can be written as follows:
 =0+
 ++,
where  is the utility for older adult iliving in street block j,  is the kth
independent variable for older adult iliving in street block j,is the random error
component for the deviation of the intercept of street block jfrom the overall intercept
0, is the random error, and other variables are as defined previously.
4.3. Geographically weighted logistic regression model
The two models above assume that the relationship between the dependent
variable and independent variables are fixed over space and thus employ merely one
equation to depict such a relationship. By contrast, geographically weighted models
relax this assumption and thus create a variety of equations to represent the spatially
varying (or non-stationary, heterogeneous) relationships (Xu & Huang, 2015; Yang,
Chau, Szeto, Cui, & Wang, 2020; Yang, Zhang, Zhong, Zhang, & Ling, 2020; Pan,
Yang, Xie, & Wen, 2020), which makes the visualization of the global pattern of
coefficient estimates possible. Geographically weighted models were initially put

16
forward by Brunsdon, Fotheringham, & Charlton (1996) and Fotheringham, Brunsdon,
& Charlton (2002). The geographically weighted logistic regression model is
expressed as follows: =0(,)+ (,)
+,
where (,)denotes the coordinates of point i,(,)is the coefficient of ,
0(,)is the intercept of point i, and other variables are as defined previously.
(,)and 0(,)are parameters to be jointly estimated.
For any given observation, a kernel function is needed to estimate the weights of
geographically close observations. Commonly used kernel functions include the fixed
Gaussian and the adaptive bi-square kernel functions. The former assigns the weight
as a continuous function of distance. By contrast, the latter does not coerce the
bandwidth into a constant but permits the spatial extent (bandwidth) to vary across
space. The fixed Gaussian kernel function is expressed as
 =
(−2/2),
where  is the weight of nearby point jfor estimating the coefficients of regression
point i, is the straight-line (or Euclidean) distance between points iand j, and 
is the fixed bandwidth.
The adaptive bi-square kernel function is expressed as
 =(1−2/()
2)2  ≤()
0   >() ,
where () denotes the adaptive bandwidth calculated by the kth nearest-neighbor
distance, and other variables are as defined previously.
5. Results
5.1. Results of multilevel logistic regression
The statistical software Stata (v. 16) was used to estimate three separate
multilevel logistic regression models with Street greenery, Number of parks, or NDVI
by using the maximum likelihood estimation method. The results of the three models
are shown in Table 3. The model with the variable street greenery (Model 1)
outperforms the other two models (Models 2 and 3). The intraclass correlation
coefficient (ICC) is roughly 0.2, indicating that approximately 20% of the total
variance is explained by between-group variance while the other 80% is explained by
within-group variance.
Table 3
Results of multilevel logistic regression models for older adults (60+ years of age)
(N=19,703)
Variable
Model 1
Model 2
Model 3
Coefficient
z-value
Coefficient
z-value
Coefficient
z-value
Housing type
0.204*
2.55
0.197*
2.45
0.209**
2.61
Household size
-0.176**
-11.36
-0.177**
-11.43
-0.177**
-11.40
Male
0.181**
4.21
0.181**
4.21
0.181**
4.22
Age
-0.064**
-25.08
-0.064**
-25.04
-0.064**
-25.06
Employment status
1.585**
13.65
1.585**
13.64
1.584**
13.64
Automobile
-0.099
-1.15
-0.092
-1.06
-0.095
-1.10

17
Illness
-1.489**
-9.65
-1.491**
-9.67
-1.488**
-9.65
Hong Kong Island
0.373**
3.11
0.254*
2.31
0.260*
2.37
Kowloon
-0.004
-0.03
-0.115
-0.92
-0.102
-0.83
Rural New Territories
0.765**
5.18
0.827**
5.67
0.797**
5.40
Population density
0.028*
2.20
0.024
1.89
0.030*
2.22
Land use mix
0.210
1.39
0.166
1.10
0.201
1.31
Intersection density
0.010*
2.40
0.010*
2.31
0.011**
2.70
Number of bus stops
-0.001
-0.91
-0.001
-1.28
-0.001
-1.38
Number of R&S facilities
0.001
0.72
0.001
0.94
0.001
0.80
Street greenery
3.600*
2.34
Number of parks
-0.004
-0.30
NDVI
0.807
1.26
Constant
5.324**
14.23
6.034**
26.10
5.729**
17.55
Performance statistic
ICC
0.194
0.195
0.195
AIC
15663.36
15668.81
15667.31
BIC
15805.35
15810.80
15809.30
Log-likelihood
-7813.68
-7816.41
-7815.66
Note: ** Significant at the 1% level. *Significant at the 5% level.
The control variables perform consistently across the three models, and their
coefficients have the expected signs. In general, older adults who live in private
housing, are male, and have a job have a higher travel propensity. In comparison,
older adults who are older, ill, and live with more people have a lower travel
propensity. Besides, automobile availability does not significantly affect the mobility
of older adults. This outcome is reasonable because Hong Kong is a transit-dependent
city with a high transit usage (90% for daily commuters), and it is in line with results
from studies conducted in transit-dependent cities (Yang, 2018; Feng et al., 2013,
2015; Feng, 2017). However, the outcome deviates from those in car-dominant cities
in the West (Paez et al., 2007; Roorda et al., 2010; Schwanen et al., 2001), where cars
are a mobility tool for residents.
Compared with those living in the New Territories non-rural areas, older adults
living in Hong Kong Island and the New Territories rural areas have a higher
willingness to travel. The performance of built environment characteristics is
reasonable, as population density and intersection density are positively related to the
willingness to travel of older adults. Other variables, however, are found to
insignificantly affect older adults' mobility. The effects of the control variables are
mostly consistent with existing literature.
The interpretation of the three greenery variables is the primary focus of this
study. Street greenery is significant at the 5% level, while the number of parks and the
NDVI are insignificant. This outcome indicates that street greenery, which reflects
what residents see, plays a role in shaping the travel propensity of older adults, but the
other two greenery variables are too weak to affect the travel propensity. Our outcome
is modestly supported by the results of many other studies conducted in the same city
and elsewhere (e.g., Shenzhen, China) that street greenery is significantly correlated
with several travel outcome measures (e.g., odds of walking/cycling, cycling

18
frequency, or physical activity) (e.g., Lu, Sarkar, & Xiao, 2018; Lu et al., 2019; Lu,
2019; Zang et al., 2019; Wang et al., 2020).
5.2. Robustness checks
To increase confidence in and uplift the plausibility of our findings, we decide to
conduct two robustness checks by narrowing the scope of the object of study.
5.2.1. Redefining older adults by shifting the age threshold from 60 to 65 years
Following the guidelines of TCS 2011, this study chooses to define older adults
as people aged 60 years or over. However, 65-year is a commonly used age threshold
for the definition of older adults (Moniruzzaman et al., 2015). Therefore, we decide to
conduct a robustness check by shifting the cut-off age from 60 to 65 years (redefining
older adults) and re-estimate the multilevel logistic regression models. The results of
these models are presented in Table 4. These results indicate that the model using
street greenery (Model 4) still outperforms the other two models (Models 5 and 6)
with the more stringent definition of older adults. Street greenery is significant at the
10% level, while the models using the number of parks and the NDVI are still not
significant at the 10% level. This further supports the robustness of our findings.
Table 4
Results of multilevel logistic regression models for older adults (65+ years of age)
(N=13,468)
Variable
Model 4
Model 5
Model 6
Coefficient
z-value
Coefficient
z-value
Coefficient
z-value
Housing type
0.148
1.75
0.143
1.68
0.152
1.79
Household size
-0.191**
-11.02
-0.192**
-11.09
-0.192**
-11.07
Male
0.246 **
5.12
0.246**
5.11
0.246**
5.11
Age
-0.070**
-20.90
-0.070**
-20.85
-0.070**
-20.87
Employment status
1.690**
8.26
1.690**
8.26
1.689**
8.26
Automobile
-0.029
-0.28
-0.021
-0.21
-0.024
-0.24
Illness
-1.486**
-8.39
-1.490**
-8.41
-1.487**
-8.39
Hong Kong Island
0.370**
2.95
0.271*
2.36
0.274*
2.39
Kowloon
0.007
0.05
-0.094
-0.72
-0.078
-0.61
Rural New Territories
0.765**
4.94
0.818**
5.36
0.794**
5.13
Population density
0.025*
1.90
0.022*
1.65
0.026*
1.88
Land use mix
0.234
1.48
0.197
1.25
0.224
1.40
Intersection density
0.009*
2.07
0.009*
2.03
0.010*
2.29
Number of bus stops
-0.0005
-0.46
-0.001
-0.81
-0.001
-0.84
Number of R&S facilities
0.0004
0.42
0.001
0.56
0.001
0.49
Street greenery
3.066*
1.90
Number of parks
-0.001
-0.07
NDVI
0.796
0.94
Constant
5.919**
13.95
6.513**
22.20
6.282**
16.60
Performance statistic
ICC
0.186
0.187
0.187

19
AIC
12235.62
12239.25
12238.37
BIC
12370.77
12374.39
12373.51
Log-likelihood
-6099.81
-6101.62
-6101.18
Note: ** Significant at the 1% level. *Significant at the 10% level.
5.2.2. Excluding older adults with a job
This study defines travel propensity as the willingness/propensity to “travel by
any mode of transport for any purpose” on the reference day, which follows Beaver
(2012). As such, any “trips” on the reference day may be made for work. Including or
excluding the segment of older adults who have to commute would produce different
modeling results, though this group of older adults is relatively limited. This may be
questioned and challenged by our peers. To gain confidence in our core findings, we
further exclude older adults with a full- or part-time job and only retain those without
a job for analysis. The multilevel logistic regression models are re-estimated, and their
results are shown in Table 5. We can reach two conclusions that are fairly similar to
former analyses: (1) the explanatory power of the model with the street greenery
variable (Model 7) is higher than that of the other two models (Models 8 and 9); and
(2) older adults' mobility is significantly affected by eye-level street greenery, but not
by the number of parks and the NDVI.
Table 5
Results of multilevel logistic regression models for older adults without a job
(N=16,752)
Variable
Model 7
Model 8
Model 9
Coefficient
z-value
Coefficient
z-value
Coefficient
z-value
Housing type
0.210*
2.50
0.206*
2.42
0.215*
2.54
Household size
-0.186**
-10.91
-0.187**
-10.99
-0.187**
-10.96
Male
0.184**
3.94
0.184**
3.94
0.184**
3.94
Age
-0.064**
-22.82
-0.063**
-22.78
-0.063**
-22.79
Employment status
1.612**
12.82
1.613**
12.82
1.612**
12.82
Automobile
-0.080
-0.85
-0.070
-0.74
-0.075
-0.79
Illness
-1.488**
-8.82
-1.491**
-8.84
-1.488**
-8.82
Hong Kong Island
0.455**
3.56
0.329**
2.80
0.331**
2.83
Kowloon
0.041
0.30
-0.098
-0.74
-0.069
-0.53
Rural New Territories
0.795**
5.04
0.857**
5.49
0.825**
5.22
Population density
0.032*
2.34
0.027*
2.03
0.034*
2.37
Land use mix
0.295*
1.83
0.247
1.54
0.286*
1.75
Intersection density
0.009*
2.09
0.010*
2.10
0.011*
2.42
Number of bus stops
-0.001
-1.02
-0.002
-1.55
-0.002
-1.53
Number of R&S facilities
0.001
1.09
0.001
1.23
0.001
1.19
Street greenery
4.009*
2.45
Number of parks
0.002
0.12
NDVI
1.138
1.32
Constant
5.131**
12.85
5.904**
23.71
5.577**
15.88

20
Performance statistic
ICC
0.195
0.196
0.196
AIC
13316.31
13322.34
13320.62
BIC
13455.38
13461.42
13459.69
Log-likelihood
-6640.15
-6643.17
-6642.31
Note: ** Significant at the 1% level. *Significant at the 10% level.
5.3. Results of geographically weighted logistic regression
In the full dataset (19,703 observations in 1,815 street blocks), a street block may
have many observations. To ease the model estimation process, we randomly chose
one observation from each street block. 1,815 observations were chosen.
GWR 4.0 was employed to estimate the geographically weighted logistic
regression model that uses the fixed Gaussian kernel function and the golden
bandwidth selection method that aims to automatically determine the best bandwidth
size. (Other kernel functions were tested, and the results are similar.) The local
regression model results are presented in Table 6. The global regression model result
for the 1,815 observations is provided in Appendix, which illustrates that street
greenery is significant at the 5% level. This observation also supports the robustness
of our core findings.
Table 6
Results of the geographically weighted logistic regression model for the sub-sample
(N=1,815)
Variable
Min
Lower quartile
Mean
Upper Quartile
Max
Housing type
0.463
0.482
0.498
0.507
0.551
Household size
-0.219
-0.213
-0.212
-0.211
-0.207
Male
0.393
0.398
0.401
0.403
0.420
Age
-0.069
-0.067
-0.067
-0.066
-0.065
Employment status
1.860
1.892
1.905
1.912
1.983
Automobile
0.041
0.145
0.150
0.166
0.201
Illness
-1.134
-1.098
-1.082
-1.076
-0.826
Hong Kong Island
0.504
0.537
0.560
0.575
0.640
Kowloon
0.175
0.216
0.245
0.265
0.354
Rural New Territories
0.573
0.678
0.691
0.705
0.880
Population density
0.029
0.035
0.036
0.038
0.040
Land use mix
0.105
0.129
0.141
0.147
0.221
Intersection density
0.003
0.005
0.005
0.005
0.006
Number of bus stops
0.000
0.000
0.000
0.000
0.000
Number of R&S facilities
0.002
0.002
0.002
0.002
0.002
Street greenery
5.175
5.706
6.049
6.292
7.223
Constant
4.075
4.290
4.317
4.349
4.638
The coefficient of street greenery is positive throughout the study area, ranging
from 5.175 to 7.223. This implies that the effect of street greenery on the travel
propensity is not fixed but is spatially non-stationary, indicating that street greenery

21
exerts higher effects on older adults' mobility in specific locations, but lower effects
in others.
To adequately reflect the distribution of the effect of street greenery on the travel
propensity, local estimates of the coefficient are mapped in Fig. 3 by using the inverse
distance weighted (IDW) interpolation in ArcGIS 10. Observe that street greenery has
a larger effect in suburban areas (northern regions) than in urban areas (southern
regions). Existing literature indicates that the low-income population favors public
greenery more than the high-income (Xiao et al., 2017); and that people from
low-socioeconomic status households benefit more from exposure to greenery than
others (James, Banay, Hart, & Laden, 2015). As such, we suspect but cannot firmly
conclude that this differential impact can be attributed to socioeconomic status; and
that street greenery has a larger effect on those with lower socioeconomic status.
Fig. 3. Map of local estimates of the coefficient of Street greenery
6. Discussion
6.1. Strengths
This study is one of the first to apply human-scale, big-data-assessed street
greenery in the arena of older adults' mobility research. It offers some scientific
evidence supporting the positive mobility effect of street greenery.
In a voluminous body of older adults' mobility research, the “average”
relationship between mobility measures and built environment characteristics has
been extensively scrutinized. However, it is, of course, insufficient. Examining the
spatially varying relationship benefits future policy and planning. It is perhaps the
first study to analyze the spatially varying relationship between big-data-assessed
street greenery and older adults' mobility or travel behavior. The results illustrate
spatial variations in regression parameters and indicate that street greenery has a

22
greater impact on the travel propensity of older adults in suburban areas than in urban
areas.
Similar to many recently emerged studies, this study uses GSV images and deep
learning techniques to assess a built environment variable (eye-level street greenery)
that is traditionally evaluated by questionnaire surveys and field audits. Therefore,
this study reveals the potential of using big data in urban and regional studies (e.g.,
Helbich et al., 2019; Lu et al., 2019; Wang et al., 2020).
6.2. Limitations
There are several shortcomings in this study. First, according to the
geographically weighted logistic regression model, street greenery has a more
significant effect on the travel propensity of older adults in suburban areas than in
urban areas. However, geographically weighted models (not limited to geographically
weighted logistic regression models) are useful to reflect a pattern of marginal effects
of independent variables. Still, they cannot offer definite reasons for such a pattern,
which is an inherent drawback of geographically weighted models. The interpretation
of geographically weighted modeling results should be carried out with due caution.
Therefore, this study only provides a possible explanation to the interesting and
serendipitous observation (socioeconomic status) and does not rule out alternative
explanations. More sophisticated approaches, such as incorporating the interaction
term between street greenery and socioeconomic status or separate regressions, are
necessary to support or dismiss this claim.
Second, this study is cross-sectional, so it can detect statistical association and
identify “correlates” of older adults' mobility. However, such a study cannot infer
causality, rule out reverse causality, and reveal “determinants.” It is possible that old
adults who have a preference for travel choose to live in greenery-abundant
communities. Technically speaking, residential self-selection may exist, although it is
too much the case for older adults because they often have a limited capacity to
choose residence freely and little desire for relocation. As such, longitudinal
observation studies tracking the same individuals over more extended periods or
experimental (or quasi-experimental) studies are needed to determine a causal
relationship and provide a more persuasive conclusion.
Third, we used the “5Ds” framework to represent the built environment.
However, because built environment variables are inexhaustible, most, if not all,
travel-built environment studies suffer from the missing variables bias. More
variables, such as high integration (from a space syntax perspective) and direct access
to relevant destinations, need to be included in future research since their association
with travel as well as eye-level street greenery is worth exploring.
Fourth, the TCS 2011 survey was acquired between September 2011 and January
2012 and did not cover the three hottest months (June, July, and August) of Hong
Kong, during which the mean daytime temperature often exceeds 32 ℃(90 ℉).
Therefore, seasonal characteristics in travel patterns and the propensity of older adults
cannot be captured by the TCS 2011 data. From our perspective, this should be
remedied in future research with first-hand empirical data to gain a broader picture of
older adults' mobility and its correlates.
Last, many recent studies (Lu, 2019; Wang et al., 2019; Yang et al., 2020),
including this study, measure the average eye-level street greenness of the
neighborhood around a dwelling location. Shifting the research focus from “dwelling
location” to “origin-destination route” or accessing the street greenness of each
origin-destination route may produce more meaningful findings.

23
7. Conclusions
Many countries and cities struggle with the aging of their populations. Due to the
high importance of mobility to older adults, identifying the correlates of older adults'
mobility is essential to serve the needs of this growing population. Existing literature
emphasizes the role of several built environment attributes (e.g., density, accessibility
to transit, and destination accessibility) in shaping older adults' mobility. Still, it
largely overlooks the effect of urban greenery that may have the potential to facilitate
travel.
Urban greenery has recently received increasing scholarly attention compared
with other physical environment factors (e.g., population density and land use
diversity). This study identifies the associations between urban greenery as
represented by three variables (street greenery, the number of parks, and NVDI) and
the travel propensity of older adults, an established measure of mobility, in Hong
Kong. Therefore, we developed multilevel and geographically weighted logistic
regression models to estimate the global and spatially varying relationships between
urban greenery and travel propensity. The results from the Hong Kong study show
that: (1) street greenery is positively associated with the travel propensity of older
adults, while the number of parks and the NDVI of the neighborhood do not
significantly affect the travel propensity. (2) The association between street greenery
and the travel propensity of older adults is spatially varied. (3) Street greenery has a
larger impact on older adults’ travel propensity in suburban areas than in urban areas.
However, caution is needed to extend the empirical findings that may be
location-specific to other study areas. More studies conducted elsewhere are, therefore,
indispensable in reaching a stronger conclusion (Cervero, Sarmiento, Jacoby, Gomez,
& Neiman, 2009).
Exposure to street greenery is a feasible and viable way to promote the travel of
older adults (or attract older adults outdoors) and improve the mobility. The
improvement of older adults' mobility likely benefits physical, mental, and social
well-being, partially because they are exposed to more opportunities for physical
activities and social interactions (Delbosc & Currie, 2011; Wentowski, 1981).
Therefore, street greenery is suggested to be offered in suburban areas and clustering
areas of older adults.
Appendix
Table A1. Results of the (global) logistic regression model for the sub-sample
(N=1,815)
Variable
Coefficient
z-value
Housing type
0.519**
3.00
Household size
-0.210**
-4.65
Male
0.411**
2.93
Age
-0.067**
-8.53
Employment status
1.948**
4.19
Automobile
0.129
0.56
Illness
-0.978
-1.57
Hong Kong Island
0.537*
2.33
Kowloon
0.253
0.99

24
Rural New Territories
0.722**
2.65
Population density
0.035
1.47
Land use mix
0.150
0.54
Intersection density
0.005
0.64
Number of bus stops
0.000
-0.05
Number of R&S facilities
0.002
1.06
Street greenery
5.893*
2.08
Constant
4.351**
5.37
Performance statistics
McFadden R-squared
0.14
Log-likelihood
-711.03
Note: ** Significant at the 1% level. *Significant at the 5% level.
Acknowledgments
This research was jointly supported by grants from the National Natural Science
Foundation of China (No. 51778530 and No. 4130163), Sichuan Provincial Young
Talent Program (No. 2019JDJQ0020), Chengdu Softscience Fund (No.
2020-RK00-00246-ZF), and Sichuan Province Circular Economy Research Center
Fund (No. XHJJ-2002). The authors are grateful to the editor and the three reviewers
for their constructive comments.

25
References
Ao, Y., Yang, D., Chen, C., & Wang, Y. (2019a). Exploring the effects of the rural
built environment on household car ownership after controlling for preference and
attitude: Evidence from Sichuan, China. Journal of Transport Geography,74, 24-36.
Ao, Y., Yang, D., Chen, C., & Wang, Y. (2019b). Effects of rural built environment
on travel-related CO2 emissions considering travel attitudes. Transportation Research
Part D: Transport and Environment,73, 187-204.
Bao, Z., Lu, W., Chi, B., Yuan, H., & Hao, J. (2019). Procurement innovation for a
circular economy of construction and demolition waste: Lessons learnt from Suzhou,
China. Waste Management,99, 12-21.
Bao, Z., & Lu, W. (2020). Developing efficient circularity for construction and
demolition waste management in fast emerging economies: Lessons learned from
Shenzhen, China. Science of The Total Environment,724, 138264.
Beaver, A. (2012). Dictionary of Travel and Tourism. Oxford University Press.
Böcker, L., van Amen, P., & Helbich, M. (2017). Elderly travel frequencies and
transport mode choices in Greater Rotterdam, the Netherlands. Transportation,44(4),
831-852.
Banister, D., & Bowling, A. (2004). Quality of life for the elderly: The transport
dimension. Transport Policy,11(2), 105-115.
Brunsdon, C., Fotheringham, A. S., & Charlton, M. E. (1996). Geographically
weighted regression: A method for exploring spatial nonstationarity. Geographical
Analysis,28(4), 281-298.
Buehler, R., & Nobis, C. (2010). Travel behavior in aging societies: Comparison of
Germany and the United States. Transportation Research Record: Journal of the
Transportation Research Board,2182(1), 62-70.
Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: Density, diversity,
and design. Transportation Research Part D: Transport and Environment,2(3),
199-219.
Cervero, R., Sarmiento, O. L., Jacoby, E., Gomez, L. F., & Neiman, A. (2009).
Influences of built environments on walking and cycling: Lessons from
Bogotá. International Journal of Sustainable Transportation,3(4), 203-226.
Cheng, L., Caset, F., De Vos, J., Derudder, B., & Witlox, F. (2019). Investigating
walking accessibility to recreational amenities for elderly people in Nanjing, China.
Transportation Research Part D: Transport and Environment,76, 85-99.
Cheng, L., Chen, X., Yang, S., Cao, Z., De Vos, J., & Witlox, F. (2019). Active travel
for active ageing in China: The role of built environment. Journal of Transport
Geography,76, 142-152.
Cheng, L., De Vos, J., Shi, K., Yang, M., Chen, X., & Witlox, F. (2019). Do
residential location effects on travel behavior differ between the elderly and younger
adults? Transportation Research Part D: Transport and Environment,73, 367-380.
Chi, B., Lu, W., Ye, M., Bao, Z., & Zhang, X. (2020). Construction waste
minimization in green building: A comparative analysis of LEED-NC 2009 certified
projects in the US and China. Journal of Cleaner Production,256, 120749.
Choi, J., Do Lee, W., Park, W. H., Kim, C., Choi, K., & Joh, C.-H. (2014). Analyzing
changes in travel behavior in time and space using household travel surveys in Seoul
Metropolitan Area over eight years. Travel Behaviour and Society,1(1), 3-14.
Cirella, G., Bąk, M., Kozlak, A., Pawłowska, B., & Borkowski, P. (2019). Transport
innovations for elderly people. Research in Transportation Business & Management,
30, 100381.

26
Collia, D. V., Sharp, J., & Giesbrecht, L. (2003). The 2001 national household travel
survey: A look into the travel patterns of older Americans. Journal of Safety Research,
34(4), 461-470.
de Keijzer, C., Bauwelinck, M., & Dadvand, P. (2020). Long-term exposure to
residential greenspace and healthy ageing: A systematic review. Current
Environmental Health Reports,7(1), 65-88.
Delbosc, A., & Currie, G. (2011). Exploring the relative influences of transport
disadvantage and social exclusion on well-being. Transport Policy,18(4), 555-562.
Du, M., Cheng, L., Li, X., & Yang, J. (2020). Factors affecting the travel mode choice
of the urban elderly in healthcare activity: Comparison between core area and
suburban area. Sustainable Cities and Society,52, 101868.
Evans, E. L. (2001). Influences on mobility among non-driving older
Americans. Transportation Research Circular E-C026, 151-168.
Ewing, R., & Cervero, R. (2010). Travel and the built environment: A meta-analysis.
Journal of the American Planning Association,76(3), 265-294.
Feng, J. (2017). The influence of built environment on travel behavior of the elderly
in urban China. Transportation Research Part D: Transport and Environment,52,
619-633.
Feng, J., Dijst, M., Wissink, B., & Prillwitz, J. (2013). The impacts of household
structure on the travel behaviour of seniors and young parents in China. Journal of
Transport Geography,30, 117-126.
Feng, J., Dijst, M., Wissink, B., & Prillwitz, J. (2015). Elderly co-residence and the
household responsibilities hypothesis: Evidence from Nanjing, China. Urban
Geography,36(5), 757-776.
Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2002). Geographically Weighted
Regression: The Analysis of Spatially Varying Relationships. Chichester, West Sussex,
England: John Wiley & Sons.
Giuliano, G., Hu, H.-H., & Lee, K. (2003). Travel Patterns of the Elderly: The Role of
Land Use, METRANS Transportation Center.
Handy, S., Cao, X., & Mokhtarian, P. L. (2006). Self-selection in the relationship
between the built environment and walking: Empirical evidence from Northern
California. Journal of the American Planning Association,72(1), 55-74.
Hartig, T., Mitchell, R., De Vries, S., & Frumkin, H. (2014). Nature and health.
Annual Review of Public Health,35, 207-228.
He, S. Y., Cheung, Y. H., & Tao, S. (2018). Travel mobility and social participation
among older people in a transit metropolis: A socio-spatial-temporal perspective.
Transportation Research Part A: Policy and Practice,118, 608-626.
Helbich, M., Yao, Y., Liu, Y., Zhang, J., Liu, P., & Wang, R. (2019). Using deep
learning to examine street view green and blue spaces and their associations with
geriatric depression in Beijing, China. Environment international,126, 107-117.
Hess, D. B. (2009). Access to public transit and its influence on ridership for older
adults in two US cities. Journal of Transport and Land Use,2(1), 3-27.
Hou, Y. (2019). Polycentric urban form and non-work travel in Singapore: A focus on
seniors. Transportation Research Part D: Transport and Environment,73, 245-275.
Ipingbemi, O. (2010). Travel characteristics and mobility constraints of the elderly in
Ibadan, Nigeria. Journal of Transport Geography,18(2), 285-291.
James, P., Banay, R. F., Hart, J. E., & Laden, F. (2015). A review of the health
benefits of greenness. Current Epidemiology Reports,2(2), 131-142.

27
Jorgensen, A., & Anthopoulou, A. (2007). Enjoyment and fear in urban
woodlands–Does age make a difference? Urban Forestry & Urban Greening,6,
267–278.
Kaczynski, A. T., & Henderson, K. A. (2008). Parks and recreation settings and active
living: A review of associations with physical activity function and intensity. Journal
of Physical Activity and Health,5(4), 619-632.
Kellert, S. R., & Wilson, E. O. (Eds.). (1993). The biophilia hypothesis. Washington,
DC: Island Press.
Kemperman, A., & Timmermans, H. (2006). Heterogeneity in urban park use of aging
visitors: A latent class analysis. Leisure Sciences,28, 57-71.
Kim, S. (2003). Analysis of elderly mobility by structural equation
modeling. Transportation Research Record: Journal of the Transportation Research
Board,1854(1), 81-89.
Koh, P. P., Leow, B. W., & Wong, Y. D. (2015). Mobility of the elderly in densely
populated neighbourhoods in Singapore. Sustainable Cities and Society,14, 126-132.
La Rosa, D., Takatori, C., Shimizu, H., & Privitera, R. (2018). A planning framework
to evaluate demands and preferences by different social groups for accessibility to
urban greenspaces. Sustainable Cities and Society,36, 346-362.
Li, S. (2020). Living environment, mobility, and wellbeing among seniors in the
United States: A new interdisciplinary dialogue. Journal of Planning Literature,
0885412220914993. doi: 10.1177/0885412220914993
Liu, Z., Kemperman, A., & Timmermans, H. (2020). Correlates of older adults’
walking trip duration. Journal of Transport & Health,18, 100889.
Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for
semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision
and Pattern Recognition (pp. 3431-3440).
Lu, Y. (2019). Using Google Street View to investigate the association between street
greenery and physical activity. Landscape and Urban Planning,191, 103435.
Lu, Y., Chen, L., Yang, Y., & Gou, Z. (2018). The association of built environment
and physical activity in older adults: Using a citywide public housing scheme to
reduce residential self-selection bias. International Journal of Environmental
Research and Public Health,15(9), 1973.
Lu, Y., Sarkar, C., & Xiao, Y. (2018). The effect of street-level greenery on walking
behavior: Evidence from Hong Kong. Social Science and Medicine,208, 41-49.
Lu, Y., Yang, Y., Sun, G., & Gou, Z. (2019). Associations between overhead-view
and eye-level urban greenness and cycling behaviors. Cities,88, 10-18.
Mercado, R., & Páez, A. (2009). Determinants of distance traveled with a focus on the
elderly: A multilevel analysis in the Hamilton CMA, Canada. Journal of Transport
Geography,17(1), 65-76.
Metz, D. H. (2000). Mobility of older people and their quality of life. Transport
Policy,7(2), 149-152.
Moniruzzaman, M., Páez, A., Scott, D., & Morency, C. (2015). Trip generation of
seniors and the geography of walking in Montreal. Environment and Planning A:
Economy and Space,47(4), 957-976.
Newbold, K. B., Scott, D. M., Spinney, J. E., Kanaroglou, P., & Páez, A. (2005).
Travel behavior within Canada’s older population: A cohort analysis. Journal of
Transport Geography,13(4), 340-351.
Nordbakke, S., & Schwanen, T. (2015). Transport, unmet activity needs and
wellbeing in later life: Exploring the links. Transportation,42(6), 1129-1151.

28
Olawole, M. O., & Aloba, O. (2014). Mobility characteristics of the elderly and their
associated level of satisfaction with transport services in Osogbo, Southwestern
Nigeria. Transport Policy,35, 105-116.
Pa´ez, A., Scott, D., Potoglou, D., Kanaroglou, P., & Newbold, K. B. (2007). Elderly
mobility: Demographic and spatial analysis of trip making in the Hamilton CMA,
Canada. Urban Studies,44(1), 123-146.
Pan, R., Yang, H., Xie, K., & Wen, Y. (2020). Exploring the Equity of Traditional
and Ride-Hailing Taxi Services during Peak Hours. Transportation Research Record:
Journal of the Transportation Research Board, 0361198120928338.
Pettersson, P., & Schmöcker, J.-D. (2010). Active ageing in developing
countries?–trip generation and tour complexity of older people in Metro Manila.
Journal of Transport Geography,18(5), 613-623.
Planning Department (2015). Hong Kong Planning Standards and Guidelines. Hong
Kong SAR Government.
Population Division (2017). World Population Ageing 2017. New York: Department
of Economic and Social Affairs, United Nations.
Rantakokko, M., Mänty, M., & Rantanen, T. (2013). Mobility decline in old age.
Exercise and Sport Sciences Reviews,41(1), 19-25.
Roorda, M. J., Páez, A., Morency, C., Mercado, R., & Farber, S. (2010). Trip
generation of vulnerable populations in three Canadian cities: A spatial ordered probit
approach. Transportation,37(3), 525-548.
Rosenbloom, S. (2001). Sustainability and automobility among the elderly: An
international assessment. Transportation,28(4), 375-408.
Rosenbloom, S., & Morris, J. (1998). Travel patterns of older Australians in an
international context: Policy implications and options. Transportation Research
Record: Journal of the Transportation Research Board,1617(1), 189-193.
Schmöcker, J.-D., Quddus, M. A., Noland, R. B., & Bell, M. G. (2005). Estimating
trip generation of elderly and disabled people: Analysis of London data.
Transportation Research Record: Journal of the Transportation Research Board,
1924(1), 9-18.
Schwanen, T., Dijst, M., & Dieleman, F. M. (2001). Leisure trips of senior citizens:
Determinants of modal choice. Tijdschrift voor economische en sociale geografie,
92(3), 347-360.
Su, F., & Bell, M. G. (2009). Transport for older people: Characteristics and solutions.
Research in Transportation Economics,25(1), 46-55.
Su, F., Schmöcker, J. D., & Bell, M. G. (2009). Mode choice of older people before
and after shopping: A study with London data. Journal of Transport and Land Use,
2(1), 29–46.
Su, F., & Bell, M. G. (2012). Travel differences by gender for older people in London.
Research in Transportation Economics,34(1), 35-38.
Szeto, W. Y., Yang, L., Wong, R. C. P., Li, Y. C., & Wong, S. C. (2017).
Spatio-temporal travel characteristics of the elderly in an ageing society. Travel
Behaviour and Society,9, 10-20.
Tacken, M. (1998). Mobility of the elderly in time and space in the Netherlands: An
analysis of the Dutch National Travel Survey. Transportation,25(4), 379-393.
Truong, L. T., & Somenahalli, S. V. (2015). Exploring frequency of public transport
use among older adults: A study in Adelaide, Australia. Travel Behaviour and
Society,2(3), 148-155.

29
Vickerman, R. W. (1974). Accessibility, attraction, and potential: A review of some
concepts and their use in determining mobility. Environment and Planning A,6(6),
675-691.
Wang, R., Helbich, M., Yao, Y., Zhang, J., Liu, P., Yuan, Y., & Liu, Y. (2019). Urban
greenery and mental wellbeing in adults: Cross-sectional mediation analyses on
multiple pathways across different greenery measures. Environmental Research,176,
108535.
Wang, R., Lu, Y., Wu, X., Liu, Y., & Yao, Y. (2020). Relationship between eye-level
greenness and cycling frequency around metro stations in Shenzhen, China: A big
data approach. Sustainable Cities and Society,59, 102201.
Wen, C., Albert, C., & Von Haaren, C. (2018). The elderly in green spaces: Exploring
requirements and preferences concerning nature-based recreation. Sustainable Cities
and Society,38, 582-593.
Wentowski, G. J. (1981). Reciprocity and the coping strategies of older people:
Cultural dimensions of network building. The Gerontologist,21(6), 600-609.
Włodarczyk-Marciniak, R., Sikorska, D., & Krauze, K. (2020). Residents’ awareness
of the role of informal green spaces in a post-industrial city, with a focus on
regulating services and urban adaptation potential. Sustainable Cities and Society,59,
102236.
Wong, R. C. P., Szeto, W. Y., Yang, L., Li, Y. C., & Wong, S. C. (2018). Public
transport policy measures for improving elderly mobility. Transport Policy,63,
73-79.
Wong, R. C. P., Szeto, W. Y., Yang, L., Li, Y. C., & Wong, S. C. (2017). Elderly
users’ level of satisfaction with public transport services in a high-density and
transit-oriented city. Journal of Transport & Health,7, 209-217.
Xiao, Y., Lu, Y., Guo, Y., & Yuan, Y. (2017). Estimating the willingness to pay for
green space services in Shanghai: Implications for social equity in urban China.
Urban Forestry & Urban Greening,26, 95-103.
Xu, P., & Huang, H. (2015). Modeling crash spatial heterogeneity: Random parameter
versus geographically weighting. Accident Analysis & Prevention,75, 16-25.
Yang, H., Zhang, Y., Zhong, L., Zhang, X., & Ling, Z. (2020). Exploring spatial
variation of bike sharing trip production and attraction: A study based on Chicago’s
Divvy system. Applied Geography,115, 102130.
Yang, L. (2018). Modeling the mobility choices of older people in a transit-oriented
city: Policy insights. Habitat International,76, 10-18.
Yang, L., Chau, K. W., Szeto, W. Y., Cui, X., & Wang, X. (2020). Accessibility to
transit, by transit, and property prices: Spatially varying relationships. Transportation
Research Part D: Transport and Environment,85, 102387.
Yang, L., & Cui, X. (2020). Determinants of elderly mobility in Hong Kong:
Implications for elderly-friendly transport. China City Planning Review,29(1), 74-83.
Yang, Y., He, D., Gou, Z., Wang, R., Liu, Y., & Lu, Y. (2019). Association between
street greenery and walking behavior in older adults in Hong Kong. Sustainable Cities
and Society,51, 101747.
Ye, Y., Richards, D., Lu, Y., Song, X., Zhuang, Y., Zeng, W., & Zhong, T. (2019).
Measuring daily accessed street greenery: A human-scale approach for informing
better urban planning practices. Landscape and Urban Planning,191, 103434.
Ye, Y., Zeng, W., Shen, Q., Zhang, X., & Lu, Y. (2019). The visual quality of streets:
A human-centred continuous measurement based on machine learning algorithms and
street view images. Environment and Planning B: Urban Analytics and City Science,
46(8), 1439-1457.

30
Zang, P., Lu, Y., Ma, J., Xie, B., Wang, R., & Liu, Y. (2019). Disentangling
residential self-selection from impacts of built environment characteristics on travel
behaviors for older adults. Social Science & Medicine,238, 112515.
Zhang, A., Xia, C., Chu, J., Lin, J., Li, W., & Wu, J. (2019). Portraying urban
landscape: A quantitative analysis system applied in fifteen metropolises in China.
Sustainable Cities and Society,46, 101396.
Zhang, S., & Ramírez, F. M. (2019). Assessing and mapping ecosystem services to
support urban green infrastructure: The case of Barcelona, Spain. Cities,92, 59-70.
Zhou, H., He, S., Cai, Y., Wang, M., & Su, S. (2019). Social inequalities in
neighborhood visual walkability: Using street view imagery and deep learning
technologies to facilitate healthy city planning. Sustainable Cities and Society,50,
101605.