International Journal of
Environmental Research
and Public Health
Article
The Association between the Regular Use of ICT
Based Mobility Services and the Bicycle Mode Choice
in Tehran and Cairo
Hamid Mostofi 1,* , Houshmand Masoumi 2,3 and Hans-Liudger Dienel 1
1
Mobility Research Cluster, Department of Work, Technology and Participation, Technische Universität Berlin,
2Center for Technology and Society, Technische Universität Berlin, 10623 Berlin, Germany;
3Department of Transport and Supply Chain Management, College of Business and Economics,
University of Johannesburg, Johannesburg 2006, South Africa
*Correspondence: [email protected]
Received: 8 October 2020; Accepted: 23 November 2020; Published: 25 November 2020
Abstract:
Regarding the sharp growth rate of ICT (information and communication technology)—based
mobility services like ridesourcing, it is essential to investigate the impact of these new mobility services
on the transport mode choices, particularly on active mobility modes like cycling. This impact is more
important in the MENA context (the Middle East and North Africa), where cycling does not constitute
the main mobility mode in the modal split of most MENA cities. This paper studies the relationship
between the regular use of ICT-based mobility services like ridesourcing and the tendency to cycle to
near destinations. This paper contains the analysis of 4431 interviews in two large cities of the MENA
region (Cairo and Tehran). This research uses logistic regression to analyze and compare the odds
of cycling among regular and non-regular users of ridesourcing by considering the socio-economic,
land use, and perception variables. The findings indicate that the odds of cycling among the regular
users of ridesourcing are 2.30 and 1.94 times greater than these odds among non-regular ridesourcing
users in Tehran and Cairo, respectively. Therefore, the regular users of ridesourcing are more likely to
cycle to their near destinations than non-regular ridesourcing users in these cities.
Keywords:
ICT-based mobility services; cycling; the active mobility mode; nonmotorized mode
choices; ridesourcing; ride hailing; MENA region
1. Introduction
Nonmotorized transport modes such as biking are sustainable modes in urban transportation
systems, and are reliable and effective in terms of energy use and healthiness without environmental
pollution [
1
–
7
]. In comparison to motorized modes, such as a private car, a bicycle is a cheaper
door-to-door mobility mode. Moreover, compared to walking, urban biking is 3.6 times faster [
8
,
9
]
and requires less energy (35% of the walking calories) for the same travel [
10
,
11
]. At this time, traffic
congestion and environmental pollution are common in many cities worldwide due to high car
dependency [
12
–
14
]. Therefore, cycling is a practical solution to reduce CO
2
emissions and help cities’
sustainability in economic and social aspects [
15
–
17
]. Therefore, the international advice is to develop
the bicycle’s share in the cities’ modal splits [
18
–
21
]. However, biking is not a substantial mode for
daily travel purposes in some MENA cities, such as Cairo and Tehran, which is entirely different from
the cycling mode share in European cities. For example, the share of biking is less than 1 percent in
the modal split in Iranian cities [
22
]. However, cycling constitutes around 40 percent of daily trips in
bicycle-friendly citiesin Europe, such as CopenhagenandAmsterdam[
23
].Some studiesindicatedifferent
Int. J. Environ. Res. Public Health 2020,17, 8767; doi:10.3390/ijerph17238767 www.mdpi.com/journal/ijerph
Int. J. Environ. Res. Public Health 2020,17, 8767 2 of 19
reasons for the low cycling rate in Tehran and Cairo, such as car-oriented urban forms, topographic
conditions, sociocultural attitudes (women rarely use bikes), and the lack of suitable infrastructure [
24
–
28
].
Moreover, ICT (information and communication technology) has considerably affected theurban mobility
system, as it offers real-time trip information, sourcing, and communication instruments between service
providers and users. Moreover, the ICT-based mobility services have been developed very fast, such as
online ridesharing and sourcing modes, in which services information technologies are the major
component. The ICTs have also changed the concepts of distance, accessibility, and individual lifestyles,
which consequently have a potential influence on mobility behaviors, particularly nonmotorized mode
choice [
29
–
32
]. This influence is gaining more importance in the cities where the share of nonmotorized
modes is low.
This study investigates the association between the regular usage of ICT-based mobility services
such as ridesourcing and the cycling mode choice in Tehran and Cairo. The primary assumption of
this research is related to the principle that the frequent usage of one mobility mode affects other
mode choices [
33
]. We conducted 4431 face-to-face interviews in Cairo and Tehran in 2017. Among
ICT-based mobility services, such as online car-sharing and bike-sharing, this study focuses only on
the online ridesourcing platforms because there was no considerable online bike and car-sharing in
these two cities in 2017. Ridesourcing is a door-to-door mobility service in which commuters and
drivers interact through ICT and GPS platforms, such as Uber and Lyft in many western countries,
“Careem” in Cairo, and “Snapp” in Tehran [
34
]. As commuters are able to “source” a ride from a pool
of drivers by ICT-based platforms, this mobility service is named ridesourcing [
35
]. Passengers use
their smartphone apps to book, pay, and rate the quality of the services. In the MENA countries,
ridesourcing has seen a sharp growth rate among other mobility modes.
Regarding the Uber report, Egypt is the biggest market of this company in the MENA region,
with 157,000 drivers and 4 million users in 2017 [
36
]. The first ridesourcing company in Iran was Snapp,
established in 2014 with a growth rate of 70% per month, with a big network of 120,000 active drivers
to give services to 0.5 million users in 2016 [
37
,
38
]. These figures show the rapid growth of these new
emerging travel modes in Tehran and Cairo, and indicate a potential impact on these cities’ mode
choice behaviors.
Ridesourcing Adaptation and Biking Mode Choice
It is necessary to study whether ICT-based mobility services support or compete with sustainable
modes like cycling, in order to evaluate their role in the sustainability of urban transport systems [
39
–
41
].
There is a debate around ridesourcing that it encourages commuters to shift from sustainable mobility
modes, like nonmotorized modes, to car travels. Regarding the findings in the global north, the adoption
rate of ridesourcing is remarkably higher among young people with higher incomes and educational
degrees [
42
,
43
]. Moreover, Alemi et al. (2017) mentioned a positive correlation between ridesourcing
usage and the regular use of smartphones for daily activities such as shopping, entertainment,
and travel [
42
]. Feigon and Murphy (2018) reported that ridesourcing was used for an average travel
distance of between three and six kilometers in five American cities [
44
], indicating its possible impact on
the cycling mode choice. Alemi et al. (2018) showed that due to the ridesourcing adaptation, the younger
generation decreased their nonmotorized mobility choices, such as walking and biking, more so than
the older people [
42
]. Gehrke et al. (2019) indicated the high probability of a modal shift from walking
and cycling to near destinations or under poor weather conditions to ridesourcing in Boston [
45
].
Becker et al. (2017) mentioned that although ridesourcing can fill gaps in the public transport network,
in many situations, it decreases public transport usage and nonmotorized modes, which indicates
a substitution impact in favor of car dependency [
46
]. Circella et al. (2018) mentioned that around 40%
of ridesourcing users have decreased their walking and biking, while 10% of users have increased these
nonmotorized modes in California [
47
]. On the other hand, ridesourcing services in some countries
provide more motorized mobility options for disabled people with health issues, by employing trained
drivers to help passengers with walkers, wheelchairs, and other equipment [39,48].
Int. J. Environ. Res. Public Health 2020,17, 8767 3 of 19
2. Materials and Methods
This research includes 3 main research questions: (1) Are the socio-economic variables of the
respondents who use bicycles significantly different from those who do not use this mode? (2) Is there
a significant association between regular ridesourcing use and the odds of cycling? (3) what are the main
subjective barriers and reasons for not cycling among regular ridesourcing users?
We conducted a large sample size of face-to-face interviews in Cairo and Tehran in different
neighborhoods to answer these three questions. Regarding the literature review about the urban forms
in Cairo and Tehran, the compactness, population density, urban forms, and road network forms
in these two cities have correlations with the periods of urban construction and development [
49
].
The newly developed neighborhoods are centerless and have less population density and compactness
than older neighborhoods in Tehran and Cairo. As such, we chose six neighborhoods in each city
in three different urban forms (two neighborhoods in each urban form) to gather adequate samples.
These three different urban forms are traditional parts (historical), transitional parts (in between),
and newly developed parts. The traditional (historical) neighborhoods in both cities are discernible
center with high density and compact urban forms. Transitional neighborhoods were constructed
from 1930 till 1980 with lower density and compactness than the old parts. The new parts, which
were developed after 1980, are centerless neighborhoods and are located in the peripheral parts of
Tehran and Cairo. The interviewers conducted face-to-face interviews with the residents of these
neighborhoods who were selected randomly in 2017. The total observations are 4431 interviews for
these two cities, including 2369 interviews in Tehran and 2062 interviews in Cairo.
Many studies reported that the cycling mode choice has an association with age [
49
–
51
],
socio-economic parameters, road network and land use parameters [
52
,
53
], the safety of cycling
paths with physical separation from the motorized network [
54
], efficient facilities [
55
,
56
], and the gently
graded topography of cities [
57
]. Therefore, we designed three main sections in the questionnaire, which
are (1) socio-economic variables, (2) mobility behavior variables, and (3) the land use parameters of
the neighborhood. The socio-economic factors included gender, age, monthly household expenditure
and income, car ownership, and having a driving license. The variable of car ownership is a binary
variable, which indicates whether the household has at least one car or not. The variable of possession of
a driving license is also a binary variable, indicating whether the respondent has a driving license or not.
The economic factors of income and monthly expenses were assessed in Egyptian pound (the Egyptian
currency) and Iranian rial (Iranian currency).
2.1. The Mobility Behavior Variables
The mobility behavior section includes questions in two main subsections, which are the main
modes for common daily travel purposes and the tendency towards cycling to near destinations.
The regular trip purposes include study or work trips and non-work/study trips within and out
of the respondents’ neighborhoods. We defined a binary variable, “ridesourcing use”, to study
the impact of ridesourcing adaptation as the main mode on the tendency to cycle. This variable
categorizes the interviewees into 2 groups: (1) regular users; (2) non-regular users. The regular users
of ridesourcing are defined as the commuters who frequently use ridesourcing as the major mobility
mode for a minimum of one of their daily trip purposes. The non-regular users are the commuters
who do not use and adopt ridesourcing as their major mobility mode for their everyday trip purposes.
Therefore, “ridesourcing use” is a dichotomized variable that is collectively exhaustive and mutually
exclusive. Therefore, this variable classified all observations of the Tehran and Cairo samples into
2 classes, and these two classes do not have the same observation.
In the second subsection, the respondents answered whether they cycle to a near destination inside
their neighborhood. This question aims to capture the tendency of respondents to cycle inside their
neighborhoods to a destination that is near based on their subjective perception of distance. Therefore,
in this question, we did not mention the distance in meters or kilometers. We mentioned the term
of “near destination” to understand whether the respondent tends to cycle even to a destination that
Int. J. Environ. Res. Public Health 2020,17, 8767 4 of 19
he/she perceives as a near destination. The dichotomized variable is defined as “bicycle use for a near
destination,” coded 1 and 0 for the yes and no responses, respectively. Additionally, the respondents
were asked about their main reason for not cycling to near destinations. The respondents could
only choose one major reason for the multiple-choice question. These options were designed in
the present form to collect subjective reasons for not cycling. This question asks for the interviewees’
perceptions about cycling to a near destination. We designed the answers to this multiple-choice
question following the review of research and reports about cycling barriers in the MENA cities [
58
–
63
].
The options include four factors, which are (1) cultural and social problems, (2) lack of biking facilities,
(3) disabled/too old, (4) takes too much time/it is slow.
2.2. The Land Use Parameters of the Neighborhood
Regardingtheroleofthelanduseandroad networkfactorsinthetendencytocycle, twoparameters
were measured, indicating the neighborhoods’ connectivity. These parameters are “link–node ratio
(%)” and “intersection density”. Link–node ratio (%) is the number of links (street segments) divided
by nodes (street intersections) within the 600 m catchment of every interviewee’s house. A bigger
link–node ratio indicates the greater connectivity of the road network in the interviewee’s neighborhood.
This parameter illustrates how many possible routes there are in the neighborhood per each node
for cycling. However, the link–node ratio is unrelated to the size of the intersections or blocks [
64
].
Intersection density (nodes/ha is the sum of intersections per unit area in a 600 m catchment area of
the interviewee’s house) is related to the size of blocks in a neighborhood [
65
]. The bigger intersection
density suggests more connectivity in a neighborhood because of the smaller blocks and the shorter
cycling distances inside this neighborhood. The full details of the neighborhood characteristics in this
survey were published [66].
2.3. Analysis Methods
2.3.1. Comparison of the Demographic Variables
We compared the socio-economic parameters of two groups of respondents who use bicycles
and who do not use this mode for a near destination in the samples of Cairo and Tehran. First,
the Kolmogorov–Smirnov test was applied to check whether the distribution of the continuous
socio-economic variables is normal. The result indicates a p-value less than 0.001 for the variables of
age, monthly household income, and monthly household living costs, indicating their distributions
are not normal. Therefore, we applied nonparametric tests, such as the Mann–Whitney U test and
the median test, to evaluate whether the differences in the distribution and median of the mentioned
continuous variables are significant across the binary variable (bicycle use) at the confidence level
of 95%. The H
0
of the Mann–Whitney U test indicates that the distribution of each socio-economic
variable is the same between two values of the binary variable “bike use”. The H
0
of the median
test assumes that the continuous variable has a similar median between two groups of respondents
who ride a bicycle and do not ride it for a near destination. Regarding gender and household car
ownership variables, the hypotheses are defined for each variable about the significant correlation
between bicycle use and gender variables. We applied the Chi-square test to test this hypothesis at
a confidence level of 95%.
2.3.2. Association between Frequent Ridesourcing Use and Odds of Cycling
We used binary logit models (logistic regression) to compare the probability of cycling between
regular and non-regular users of ridesourcing in each sample of these MENA cities at significance levels
of 0.05. The odds of bicycle use for a near destination are the probability of cycling over the probability
of not cycling, which are response variables in the logit models. The transformation from probability
to odds is monotonous, indicating the odds increase (decrease) as the probability increases (decrease).
The binary logit model is structured by the equations below, where P is the probability of cycling to
Int. J. Environ. Res. Public Health 2020,17, 8767 5 of 19
a near destination, 1
−
Pis the probability of not cycling,
β0
is the constant, and
βi
is the coefficient
related to each explanatory variable.
lnP
1−P=β0+β1x1+β2x2+β3x3+. . . +βn(1)
P=e(β0+β1x1+... +βnxn)
1+e(β0+β1x1+... +βnxn)(2)
The logistic model reveals the effects of the independent variables on the odds of bike use (probability
of bike use/probability of no bike use). The exponentiated coefficient changes the cycling odds to a unit
increase in the independent variable by holding other regressors constant. Suppose the odds ratio of
an independent variable is greater than 1. In that case, it is suggested that by keeping other regressors
constant, the odds of bike use increase (decrease) by increasing (decreasing) this independent variable.
If the odds ratio is 1 for an estimator, it suggests that this estimator’s change does not change the cycling
odds. When the odds ratio of one estimator is less than 1, it indicates a decrease in biking odds caused by
an increase in this estimator if other independent variables are constant. Some transport studies estimate
the odds ratio to show the impacts of the explanatory variables on the odds of mode choices. However,
the estimation of average marginal effects is an intuitive technique and a useful way to directly explain
the probability changes and interpret the results of the estimations more understandably. The logistic
regression is nonlinear; therefore, the effect of one unit change in an explanatory variable is averaged over
all observations to estimate the average marginal effects. The average marginal effects of an independent
variable are the average change in probability of the dependent variable when the given independent
variable changes by one unit and the other is constant. We used the add-on package “margins” in R to
calculate the average marginal effects. Therefore, in addition to the odds ratio, we also report the average
marginal effects of each explanatory variable.
We checked the risk of multicollinearity among the independent variables of one pair or
more of explanatory variables being highly correlated together and causing unreliable estimations.
The regressors were selected to avoid high multicollinearity among variables and to control confounding
effects. Therefore, the regressors (independent variables) in the logistic regression are regular ridesourcing,
household car ownership, gender, age, monthly household income, intersection density, and link–node
ratio. Checking the correlation matrix might be useful in order to detect multicollinearity, but it is
not enough. The sufficient diagnostics are performed by linear regression between the variables to
check the variance inflation factor (VIF), tolerance, and the condition index. If the correlation coefficient
among two independent variables is greater than 0.90, it indicates a high multicollinearity risk in
the logit model [
67
,
68
]. We checked the correlation matrix, and we did not find a correlation greater
than 0.8 for both samples of Cairo and Tehran. Furthermore, we performed the linear regression among
the different combinations of explanatory variables and then checked the VIF, tolerance, and condition
index. The VIF for an estimator in the regression model is the ratio of the variance of the overall model to
the variance of a model that includes only the given estimator. Hair et al. (2010) indicated that if the VIF is
greater than 5, there is a concern for multicollinearity among independent variables [
68
]. As we checked
the VIFs for different combinations of the estimators, the VIFs of all estimators in the combination of
regular ridesourcing, household car ownership, gender, age, income, intersection density and link–node
ratio are less than 4.0 in both the Cairo and Tehran models. Then, we checked the tolerances of
the mentioned variables, which is the amount of variability in one estimator that is not explained
by the other estimators. Tolerance values less than 0.2 indicate the risk of multicollinearity among
estimators. Having checked the tolerance values for the mentioned independent variables, all of them
were bigger than 0.3, indicating the low risk of multicollinearity. Moreover, we checked the condition
indices. Condition indices above 15 indicate a risk of multicollinearity. The estimated condition indices
were below 15 in the linear regressions among the aforementioned estimators. Moreover, we checked
the interactions between the variables, such as monthly household income–age, monthly household
income–gender and gender–age, in the two logit models of Cairo and Teheran, and estimated their
Loading more pages...