Content uploaded by Khandker Nurul Habib
Author content
All content in this area was uploaded by Khandker Nurul Habib on Nov 23, 2015
Content may be subject to copyright.
How Active Modes Compete with Motorized Modes in High-Density Areas: A Case Study 1"
of Downtown Toronto 2"
3"
4"
5"
6"
7"
Mohamed Salah Mahmoud*, M.Sc. 8"
Ph.D. Candidate 9"
Department of Civil Engineering 10"
University of Toronto 11"
35 St. George Street, Toronto, Ontario, M5S 1A4, Canada 12"
Tel: 416-946-8299; Email: mohamed.mahmoud@utoronto.ca"13"
14"
15"
Wafic El-Assi, 16"
B.A.Sc Candidate 17"
Department of Civil Engineering 18"
University of Toronto 19"
35 St. George Street, Toronto, Ontario, M5S 1A4, Canada 20"
Tel: 416-946-8299; Email: wafic.el.assi@mail.utoronto.ca 21"
22"
23"
Khandker Nurul Habib, Ph.D., P.Eng. 24"
Assistant Professor 25"
Department of Civil Engineering 26"
University of Toronto 27"
35 St. George Street, Toronto, Ontario, M5S 1A4, Canada 28"
Tel: 416-946-8027; Email: Khandker.nurulhabib@utoronto.ca 29"
30"
31"
Amer Shalaby, Ph.D., P.Eng. 32"
Professor 33"
Department of Civil Engineering 34"
University of Toronto 35"
35 St. George Street, Toronto, Ontario, M5S 1A4, Canada 36"
Tel: 416-978-5907; Email: amer@ecf.utoronto.ca 37"
38"
39"
40"
41"
42"
Submission Date: February 28, 2014 43"
* Corresponding Author44"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 1"
"
ABSTRACT 1"
This paper provides an investigation on the use of active transportation modes as a 2"
substitute for motorized modes in high-density areas. The downtown core area of the City of 3"
Toronto is selected as a case study. A nested logit mode choice model is developed using data 4"
from the 2011-2012 Transportation Tomorrow Survey (TTS) along with supplementary weather 5"
data and travel modes’ level of service attributes. Variables that potentially affect individuals’ 6"
active mode choice such as cycling infrastructure, pedestrian network, and weather conditions 7"
are carefully considered in the analysis. The empirical model provides meaningful insights 8"
towards understanding short distance commuters’ mode choice behaviour in high-density areas. 9"
The results show that individuals with higher percentages of cycling infrastructure lengths with 10"
respect to their total trip distances are more likely to choose “bike” as their commuting mode. 11"
The developed model can be used as a policy analysis tool to quantify the effects of the built 12"
environment on individuals’ mode choices in high-density areas. 13"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 2"
"
INTRODUCTION 1"
Active transportation modes, namely: walking and biking, are the most sustainable travel modes 2"
in the context of urban transportation systems. As such, the use of active modes for different 3"
activity purposes has been under investigation by travel demand researchers (1-5). This is due, in 4"
part, to the potentially significant positive impacts of active modes on trip makers as well as 5"
urban transportation systems. Previous research efforts showed the various benefits of active 6"
modes with respect to the environment, public health, social equity, and traffic congestion (6-8). 7"
In terms of environmental benefits, walking and cycling produce nearly zero air and noise 8"
pollution compared to motorized modes. This has a direct effect on public health in addition to 9"
other benefits that an individual can achieve by using active modes such as fighting obesity and 10"
chronic illnesses. Active modes offer independent mobility for individuals who cannot drive or 11"
take transit due to inaccessibility issues (e.g. children and low-income families). 12"
Depending on the quality of the pedestrian and cycling networks, active modes can provide a 13"
viable substitute for motorized modes especially for short distance trips. High density, mixed-14"
use, and well-connected neighbourhoods encourage individuals to walk and/or bike instead of 15"
relying on motorized modes (9). Empirical evidence from a study conducted on 90 of the 100 16"
largest cities in the United States shows that areas with improved cycling infrastructure have 17"
higher commuting bike mode shares which contributes partially to the reduction of traffic 18 "
congestion (10). Similarly, another study showed that a modest shift in short trips (less than 5 19"
miles) to active transportation modes could significantly reduce the annual vehicle-miles driven 20"
(11). This explains the drive behind promoting for the development of active communities in 21 "
large cities throughout the last few decades and the corresponding action plans of expanding 22"
pedestrian and cycling infrastructures. 23"
Weather conditions and the built environment along with other several factors may affect 24 "
individuals’ decisions of choosing active modes as their commuting travel modes. For instance, 25"
Canada has a relatively cold climate in comparison with the United States and some European 26 "
countries. Nonetheless, Canadians tend to cycle more than their American counterparts. This is 27"
partially due to land-use policies that target: higher percentages of short distance trips by 28 "
developing mixed-use neighbourhoods, safer cycling conditions by improving cycling 29"
infrastructure, and lower driving mode shares by introducing higher costs of owning, driving and 30"
parking a car (12). Overall, active modes of transport are on the rise in North American cities; 31"
Toronto is no exception (13). The City of Toronto has embarked on an ambitious plan to expand 32"
its bike network along most of the major routes in the downtown core (14). In order to test the 33"
effectiveness of such policies on maintaining as well as encouraging the use of active modes, an 34"
in-depth understanding of individuals’ travel behaviour and how active modes compete with 35"
motorized modes in high-density areas is essential. That is, transportation planners can develop 36"
efficient policies and/or programs that promote for the increase of active modes usage. 37"
This paper is organized as follows. A review of the literature on the current investigation of short 38"
distance trips in dense neighbourhoods is presented. The following sections present a description 39"
of the dataset used in the analysis, the econometric modelling framework, and the development 40"
of the empirical model. Finally, key findings and possible implications for developing effective 41"
policies for active modes are identified. 42"
43"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 3"
"
LITERATURE REVIEW 1"
Studying commuters’ mode choice behaviour has been under continued investigation by 2"
transportation planners. Nonetheless, few studies focused on studying commuting mode choice 3"
behaviour for short distance trips in cases where active modes are competitive options (9), 4"
especially within the North American Context. In this section, relevant research efforts to the 5"
current investigation’s context are presented. 6"
A recent study in the Greater Copenhagen Area analyzed mode choice behaviour for trips shorter 7"
than 22 km (15). A dataset including travel diaries and socio-economic variables of a 8"
representative sample of the population was used for the analysis. A mixed-logit mode choice 9"
model was developed to investigate the effect of travel modes’ level of service attributes and trip 10 "
makers’ characteristics on their mode choices. The study concluded that the use of active modes 11"
of transportation is positively correlated with temperature and short trip distances. However, 12"
while there is no distance threshold defined for active modes, considering walking or biking for 13"
commuting trips that are more than 10 km is not realistic in most cases (6). Further, the study did 14"
not investigate the effect of pedestrian and cycling infrastructures on selecting active modes as a 15"
mode of transportation. Another study on short distance trips was conducted in the Netherlands, 16 "
examining the effect of personal and neighbourhood characteristics on active modes choice in 17"
comparison with motorized modes (9). A multilevel logistic model is developed using a dataset 18"
that included travel records with various trip purposes with a maximum trip length of 7.5 KM. 19"
The results indicate that educated middle age urban residents are more likely to use active modes 20"
of transport. Nevertheless, similar to the Copenhagen study, the built environment characteristics 21"
were not considered. 22"
Using data of individuals’ travel diaries, a similar study assessed the competitiveness of biking 23"
and driving alongside other modes in the City of Ghent (16). The study concluded that cycling 24 "
might only be competitive within a range of 5 KM, while walking was only competitive within a 25"
1 KM range. In addition, Kim and Ulfarsson conducted a study on trips that are less than 2.25 26"
KM in Washington D.C. (17). The results indicated that individuals are more likely to drive if 27"
they can or are accustomed to. Moreover, the authors indicated that individuals are less likely to 28"
walk or bike as they age. Nonetheless, the study did not consider the effect of bike infrastructure 29"
and street walkability on individuals’ mode choices for such short distance trips. 30"
Over the past decade, a significant amount of research on studying individuals’ behaviour of 31 "
choosing biking as their commuting travel mode has been conducted. Heinen, van Wee and Maat 32"
(18) discussed the role of personal attitudes and built-environment on commuters’ mode choice 33"
decision. The study showed that “safety” and “awareness” are strong determinants for the choice 34"
of biking as a travel mode. Xing, Handy and Mokhtarian (19) provided an extensive 35"
investigation of factors associated with commuters’ choice of biking as a mode of travel in six 36"
cities in the United States. The study revealed that short distances to destinations and supporting 37"
biking infrastructure are key factors that explain the higher bike mode share among commuters. 38"
Similarly, Pucher, Dill and Handy (20) studied the effects of bike infrastructure and bike 39 "
programs on bike usage as a travel mode. Habib et al. (5) investigated biking behaviour in terms 40"
of choice of biking for utilitarian and/or recreational purposes as well as bike ownership level for 41"
the City of Toronto. Howard and Burns (21) examined the effect of bike infrastructure, distance 42"
travelled and safest routes on biking route selection. The results indicate that cyclists tend to alter 43"
their routes to maximize on the utility of the aforementioned variables. 44"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 4"
"
Similarly, research efforts on studying individuals’ behaviour of engaging in physical activities 1"
such as commuting by bike or on foot with respect to the built environment have been 2"
investigated (7, 9). Empirical evidence from 36 environmentally diverse but equivalent-sized 3"
neighbourhoods showed that individuals walk more for activities in high-density areas (22). 4"
Another group of studies targeted the effect of land-use, the built environment and 5"
neighbourhood characteristics on the use of active modes of transportation. Cervero and Radisch 6"
compared the effect of suburban and neo-traditional neighbourhoods on the use of non-motorized 7"
mode for different activity purposes (23). The results showed that neo-traditional 8"
neighbourhoods have a stronger effect on selecting active transportation modes specifically for 9"
shopping purposes. Guo, Bhat and Copperman investigated the effect of the built environment on 10"
motorized and non-motorized trip making behaviour (24). A bivariate ordered probit model was 11 "
developed. The model accounts for complementary and synergistic relationships between 12"
motorized and non-motorized modes. The study concluded that business density, street 13"
connectivity, and bike lane density are positively correlated with the use of active modes of 14 "
transportation. 15"
Other studies focused on studying the effect of weather conditions on transportation. Saneinejad, 16"
Roorda and Kennedy (25) developed a series of models to explore the effect of weather on 17"
home-based work trips in the City of Toronto. Results from their study showed that weather 18"
conditions including temperature and precipitation have a significant impact on all travel modes 19"
and more specifically on active modes of transportation. Another study on the effect of climate 20"
and weather on bike use in Melbourne City confirmed the negative correlation between cycling 21"
and precipitation (26). 22"
Recently, bike sharing as a competitive active mode of transportation in urban centres have been 23"
under investigation. Habib et al. (27) investigated the determinants of bike share demand in 24"
Toronto. The study considered weather effects, socio-demographic variables, and built 25"
environment attributes on bike share trip activity. Results indicated that bike share ridership is 26 "
positively correlated with temperature while being negatively correlated with snow on ground, 27"
humidity and precipitation. In addition, the study concluded that bike share users preferred to use 28"
routes that exhibit a high density of bike lanes. Finally, Mahmoud et al (28) examined the “mode 29"
culture” in the City of Toronto with a particular focus on the factors that influence the cycling 30"
culture. The results showed that individuals who lived and worked in the downtown core of 31"
Toronto were more likely to be rely on biking as their mode of travel. 32"
This paper focuses on studying short distance commuting trips in high-density neighbourhoods, 33"
in particular, investigating on how active modes of transportation compete with motorized 34 "
modes. Lessons learned from the literature are included and investigated in this study. Data on 35"
trip makers and their household attributes, trip characteristics, land-use and built environment 36 "
attributes, and weather conditions are used to develop a nested logit mode choice model for 37 "
commuting trips in the downtown area of the City of Toronto. The developed model reveals 38"
meaningful insights that answer the research question of this study. 39"
STUDY AREA AND DATA DESCRIPTION 40"
The City of Toronto is located in the heart of the Greater Toronto and Hamilton Area (GTHA) 41"
which forms Canada’s largest urban region (29). Figure 1 shows a map of the City of Toronto 42"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 5"
"
with a detailed section of the city’s Planning District 1, featuring the downtown core area. In 1"
addition, the map shows the distribution of trip origins categorized by distance travelled. 2"
3"
4"
5"
6"
7"
8"
9"
10"
11"
12"
FIGURE 1 Study Area and Distribution of Trip Origins by Distance Travelled 13"
Trip characteristics and socio-demographic attributes of the study were extracted from the 2011-14"
2012 Transportation Tomorrow Survey (TTS) dataset. The TTS is a trip-based household survey 15"
that is conducted every five years in the Greater Toronto and Hamilton Area (GTHA) among 5% 16"
of its population. This study focuses on short distance commuting trips in which active modes 17"
are truly competing with motorized modes. In order to account for short distance commuting 18"
trips, only home-work trips with both trip ends in the downtown Toronto area are considered in 19"
City of Toronto
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 6"
"
this analysis. This subset of trips represents around 10% of the total commuting trips in the City 1"
of Toronto. The dataset provides detailed disaggregate individual trip records with geo-coded 2"
locations of individuals’ households (to the nearest midblock), places of employment and their 3"
observed travel mode. The total number of complete trip records used in this study is 1,956. "4"
Four travel modes are considered in this analysis, namely auto driver, transit, bike and walk. The 5"
four modes are assumed to be available to all individuals in the dataset. Travel distances and 6"
times for the auto driver mode were obtained from the 2012 EMME traffic assignment network 7"
model of the GTHA. However, for the transit mode, travel distances and times were obtained 8"
from Google Maps® Application Programming Interface (API) based on the General Transit 9"
Feeds Specifications (GTFS) data. Data obtained from the EMME assignment model and Google 10"
Maps API consider the interactions between transit and traffic to generate realistic mode-specific 11"
travel times that take into account traffic congestion. Similarly, for the bike and walk travel 12"
modes, mode-specific paths were generated using Google Maps® API. Suggested bicycle paths 13 "
were generated such that the travelled distance was minimized and percent of bicycle 14 "
infrastructure along the route was maximized. Accordingly, walk and bike travel distances and 15"
times were obtained for each individual using the locations of individuals’ trip origins and 16"
destinations. 17"
Travel costs for the auto driver mode were obtained based on the distance travelled and average 18"
parking costs at the traffic analysis zone of the trip destination. The average parking cost in the 19"
downtown area of the City of Toronto is $23 per day and $8 per hour (30). As such, the 20"
relatively low-cost car trips (due to the short distance travelled) are offset with high parking cost. 21"
On the other hand, transit cost is defined based on the fares set by the Toronto Transit 22"
Commission (TTC), the public transport agency that operates transit services in the City of 23"
Toronto, of $3 for adults or $2 for seniors (+65 years old) and students (13 to 19 years old) as a 24 "
flat fare per trip
1
. In addition, hourly weather data collected at the Billy Bishop Toronto city 25 "
airport (also known locally as the ‘Toronto Island Airport’) weather station was provided by 26"
Environment Canada. The data included weather temperature, wind speed, precipitation, and 27 "
snow on ground. The average weather temperature for the trip records in the dataset during the 28"
fall season is 7.4
o
C. Further, using an updated version of a street network with details on bicycle 29"
infrastructure types, the suggested bicycle paths were used to generate the percent of bike facility 30"
length compared to the total trip distance. In addition, the total number of intersections with 31 "
major roads was determined. The spatial analysis was conducted in ArcMap® 10.2. 32"
The TTC provides extensive transit coverage in the downtown core area of the City of Toronto. 33"
Within the study area, the average airline distance from any household location (in the dataset) to 34"
any transit stop/station is 100 m signifying the ease of transit access. In addition, the study area is 35"
considered to be a walkable/bikeable neighbourhood that has well-developed pedestrian paths 36"
and cycling infrastructure including bike racks, bike lanes (separated bike lanes), sharrows 37"
(marked bike lanes), park roads, signed routes and multi-use pathways. One of the unique 38"
features of Toronto’s downtown pedestrian network is the “PATH” - an underground walkway 39"
that connects more than 50 buildings/office towers, 20 parking garages, six subway stations, 40"
Toronto Coach Terminal, and Union Station. As the largest underground shopping complex in 41 "
the world with a network length of 30 km (31), the PATH accommodates more than 200,000 42"
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
1
"http://www.ttc.ca/Fares_and_passes/Prices/"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 7"
"
business-day commuters and thousands of tourists and residents in weatherproof comfort. Figure 1"
2 shows a heat map of the Walk Score® for the City of Toronto as an index of public access 2"
walkability. The average Walk Score® in the City of Toronto is 71 and it goes up to 100 in the 3"
downtown area (32). Similarly, Figure 3 shows the density of bike infrastructure in the 4"
neighbourhoods of the City of Toronto. Clearly, the downtown area is more walkable and 5"
bikeable than the surrounding areas of the city. That is, the study area with such compact, mixed-6"
used, and well-connected pedestrian and cycling network provides a perfect case to study short 7"
distance commuting trips in which active modes are truly competing with motorized modes. 8"
Figure 4 shows the travel mode shares within the study area. One of the major factors that affect 9"
individuals’ mode choice is travel cost. The average cost per unit distance for transit users in the 10"
downtown area is $1.15 per KM. In addition, as explained earlier, parking costs are relatively 11"
higher in the downtown area. Further, the average trip length of trips that originate from and 12"
destined to the downtown area is 2.25 km. That is, shorter trip distances, higher parking costs, 13"
and the flat transit fares are important factors that explain the dominance of the active travel 14 "
mode with more than a 50% modal share. Figure 5 shows the average observed travel distances 15"
by each mode. The average biking distance is shorter by only 0.1 and 0.4 KM compared to the 16"
average driving and transit distances, respectively. Figure 6 shows a density chart of observed 17"
walking and biking trips distributed by travel distance. This distribution suggests that walking 18"
commuting trips are often shorter than 5KM, while biking commuting trips can be longer. 19"
Similarly, Figure 7 shows the average generated travel time by mode for all trips (i.e., travel 20"
times by each mode for the same origin-destination (O/D) pair) in the sample data. Figure 7 21"
shows that on average for the same O/D pairs, the bike mode is 5 minutes shorter than the transit 22"
mode. In terms of the effect of the built-environment on bike trips, Figure 8 shows a density 23"
chart of biking trips distributed by percent of bike facility length (e.g., bike lanes) to the total 24 "
travel distance. Clearly, higher percent of bike facility length compared to the total travel 25 "
distance is a significant factor that motivates individuals to bike. 26"
27"
FIGURE 2 Walk Score Map of the City of Toronto (32) 28"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 8"
"
1"
FIGURE 3 Bike Infrastructure Density in the City of Toronto 2"
3"
FIGURE 4 Travel Mode Shares 4"
5"
13%$
31%$
6%$
50%$
Car"
Transit"
Bike"
Walk"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 9"
"
1"
FIGURE 5 Average (Observed) Travel Distance by Mode 2"
3"
FIGURE 6 Density of Walking and Biking Trips by Total Travel Distance 4"
5"
Mahmoud,"M.S.,"El?Assi,"W.,"Habib,"K.M.N.,"Shalaby,"A." " " " 10"
"
1"
FIGURE 7 Average (Generated) Travel Time by Mode 2"
3"
4"
FIGURE 8 Density of Biking Trips by Percent of Bike Facility Length 5"
to Total Travel Distance6"
0"
5"
10"
15"
20"
25"
30"
Auto" Transit" Bike" Walk"
Average"Travel"Time"(Minutes)"
Travel"Mode"
Mahmoud,"M.S.,"Habib,"K.M.N.,"Shalaby,"A." " " "11"
"
ECONOMETRIC MODEL 1"
This section presents the econometric model formulation of the Nested Logit (NL) model. The 2"
NL model formulation allows for relaxing the assumption of independence of irrelevant 3"
alternatives (IIA) and capturing preference heterogeneity among respondents. The model 4"
formulation categorizes common alternatives in “nests”. The NL model formulation considers 5"
partially common error term component for within-nest alternatives. Individuals are assumed to 6"
gain a certain level of utility by choosing one travel mode over the other three available modes. 7"
The utility function (U) for each mode consists of systematic and random components. The error 8"
term component of nested alternatives can be divided into two portions" to adopt nesting 9"
structures; within-nest and alternative-specific error terms. Based on the the fundamental 10 "
Random Utility Maximization (RUM) Theory, trip makers are assumed to be rational in selecting 11 "
their travel modes by choosing the alternative with the highest utility value (33): 12"
U
m|n
= V
m
+
ε
m
+
ε
n
= (
β
⋅x)
m
+
ε
m
+
ε
n
[1] 13"
Where the subscript “m” indicates one of the travel modes in a choice set of “M” modes, “X” is 14"
the observed variables and their corresponding coefficients “β”, “Ԑ
m
” is the alternative-specific 15"
random error term, and “Ԑ
n
” is the within-nest random error term. As such, the conditional 16"
probability that a person “i” selects a mode alternative “m” from a nest “n” follows the logit 17"
formula of: 18"
Pr(m | n) =
exp(V
m
/
λ
n
)
exp(V
m
'
/
λ
n
)
m
/
=1
M
∑
[2] 19"
Where “Pr(m|n)” is the conditional probability of choosing mode “m” from nest “n” and “
λ
n
” is a 20"
measure of the degree of independence in the unobserved utility among the alternatives in nest “n”. 21"
The marginal choice probabilities are calculated as: 22"
Pr(n) =
exp(V
n
+
λ
n
⋅ I
n
)
exp(V
n
+
λ
n
⋅ I
n
)
n
/
=1
N
∑
[3] 23"
Where “Pr(n)” is the probability of choosing alternative from nest “n”, “V
n
” is a function of 24 "
common attributes within nest “n” (if any), “N” indicates the total number of nests “n”, and “I
n
” 25"
is the “logsum” variable of nest “n”. The logsum represents the expected utility of the within-26"
nest alternatives as: 27"
I
n
= log exp(V
m
'
/
λ
n
)
m
/
=1
M
∑
"
#
$
%
&
'
[4] 28"
Therefore, the unconditional probabilities of travel modes can be obtained as: 29"
Pr(m) = Pr(m | n)⋅ Pr(n)
[5] 30"
Mahmoud,"M.S.,"Habib,"K.M .N.,"Shalaby,"A." " " "12"
"
In this paper, the empirical models were estimated using the “mlogit” package in the statistical 1"
software “R” and using the “MAXLIK” component for maximum likelihood estimation (34, 35). 2"
EMPIRICAL MODEL 3"
Previous studies showed that trip distance/time and travel cost along with personal attributes are 4"
important variables to consider while developing mode choice models (25-27). In addition to the 5"
typically used variables, the effect of weather conditions and built environment variables on 6"
active modes is investigated in this study. Table 1 presents definitions of variables that are used 7"
in this analysis. 8"
TABLE 1 Definitions of Variables 9"
Variable Name
Description
Distance
Mode-specific network distance, in KM, from individuals’
household location to their work location
Cost
Mode-specific travel cost, in Canadian Dollars (CAD), from
individuals’ household location to their work location
Time
Mode-specific travel time, in minutes, from household
individuals’ location to their work location
Male
=1 if individual is male; =0 Otherwise
Free Parking
=1 if individual has access to free parking at work location;
=0 Otherwise
Age<35
=1 if individual age is equal to or less than 35 years old;
=0 Otherwise
Transit Pass
=1 if individual holds a TTC Metro Pass; =0 Otherwise
Number of Vehicles
Number of vehicles per household
Number of Persons
Number of persons per household
Temperature
Weather Temperature, in
o
C, by time of day of the individuals’
trips
Number of Intersections
Number of intersections with major roads along individuals’ bike
path from their household location to their work location
Length of Bike Facilities
The total length of bike lanes along individuals’ bike path from
their household location to their work location
Walk Score at
Employment Zone
The walk score of individuals’ work location traffic analysis zone
(measured at the neighbourhood level)
A traditional multinomial logit (MNL) model as well as a nested (NL) logit model were 10"
developed. The empirical results showed that the NL model outperformed the MNL model and 11"
therefore results of the NL model are only presented herein. The NL model is developed with 12"
three nests, namely: auto driver, transit, and active modes. Different model structures and 13 "
specifications were tested and the final model specifications are reported in Table 2. A total of 16 14"
parameters were estimated using a sample of 1,956 trips that originate from and destined to the 15"
downtown Toronto area. The reported adjusted Rho-Squared value, as a measure of goodness-of-16 "
fit (36), is 0.36. All the reported parameters are estimated with the expected signs and found to 17 "
be statistically significant (with t-statistics higher than 1.96) at the 95% confidence interval, 18 "
except for the “walk score” variable for the walk mode and the “Male” variable for the transit 19"
mode which are statistically significant at the 90% confidence interval. As expected, travel 20"
Mahmoud,"M.S.,"Habib,"K.M .N.,"Shalaby,"A." " " "13"
"
modes with higher unit travel costs are less likely to be chosen over modes with lower unit travel 1"
cost or modes of no travel cost. Similarly, modes with shorter travel times are preferred more. 2"
TABLE 2 Parameter Estimation Results – Mode Choice NL Model
Number of Observations
1956
Log-Likelihood (Full Model)
-1399.8
Log-Likelihood (Null Model)
-217.8
Rho-Squared Value
0.36
Systematic Utility Function:
Variables
Parameter
t-Statistics
Alternative Specific Constant
Auto Drive
?4.85800"
?12.069*
Transit
?0.60736"
?1.975*"
Bike
?2.91619"
?6.519*
Travel Cost/Distance
?0.52030
?20.694*
Travel Time
?0.10242
?15.564*
Auto Drive
Free Parking
1.91308"
9.479*"
Number of Vehicles/Number of Persons
2.23915
9.921*
Transit
Male
?0.25804
?1.922’
Transit Pass
2.75873
15.447*
Bike
Male
0.43767
2.416*
Temperature
0.02540
1.977*
Number of Intersections/Distance
?0.11461
?3.334*
Length of Bike Facilities/Distance
0.49890
2.094*
Age<35
0.32795
2.048*
Walk
Walk Score at Employment Zone
0.39377
1.695’
Log-Sum of Active Modes Nest
0.81572"
1.978*
* Significant at the 95% level of confidence, ’ Significant at the 90% level of confidence 3"
In terms of individual-specific variables, the increase of the car ownership level as compared to 4"
the number of persons per household has a positive effect on the probability of choosing the auto 5"
driver mode. In addition, individuals who have access to free parking at their work locations are 6"
more likely to drive to work. Similarly, holding a transit pass is a significant variable in 7"
increasing the probability of choosing transit as a travel mode. Consistent with previous research 8"
findings, the model shows that males are more likely to bike than females (37). In addition, 9"
individuals who are 35 years old or less are more likely to bike more than older individuals. 10"
While different weather variables were tested, only the atmospheric temperature showed 11"
significant correlation with bike usage. Data from the household survey was collected during the 12 "
fall season (September-December) in which severe weather events were not recorded. Variables 13 "
such as precipitation, snow on ground and wind speed were tested but were not found 14 "
statistically significant; therefore, only the weather temperature was included in the model. 15"
During the fall season, higher temperatures are desirable for individuals who are using active 16"
Mahmoud,"M.S.,"Habib,"K.M .N.,"Shalaby,"A." " " "14"
"
modes. This explains the positive effect of the increase in temperature on the probability of 1"
choosing bike as a travel mode. The reported temperature during the analysis period over the 2"
sampled trip records ranged from -12
o
C to 25
o
C with an average of 7.5
o
C. 3"
The effect of the built-environment on active modes was carefully considered while developing 4"
this model. The average slope of bike routes was calculated using the digital elevation model 5"
(DEM) of Toronto. However, it did not show statistical significance when it was included in the 6"
model. This result is perhaps due to the relatively flat terrain of the downtown core area. As 7"
explained above, a detailed bike network was used to generate the percent of bike facilities (i.e., 8"
bike lanes) compared to individuals total travel distance. In addition, the total number of 9"
intersections with major roads was determined. These metrics were obtained according to the 10"
suggested bike path using Google Maps® API which considers the safest (by utilizing the 11"
surrounding bike infrastructure) and shortest bike route. Model results show that higher 12"
percentages of bike lanes compared to the total travel distance has a significant positive effect on 13"
the probability of choosing biking as a travel mode. On the other hand, higher number of 14 "
intersections with major roads has a negative effect on bike usage. Similarly, data on 15"
neighbourhood walk scores were obtained at the employment zone. Higher walk score values 16"
indicates more walkable neighbourhoods. The model results suggest that individuals are more 17"
likely to walk to work if the walk score at the employment zone is high. 18"
Model results show that females, transit pass holders, individuals who live in households with 19"
high car ownership levels, and individuals who have access to free parking at work locations are 20"
the least prone to using active modes. From a policy-making perspective, customized 21"
strategies that are specifically targeted to such markets are expected to be more efficient. For 22"
instance, policies such as restricted parking allowance per household in the downtown residential 23"
buildings, and limitation of offered free parking space within the downtown area may contribute 24"
to an overall lower driving modal share. On the other hand, seasonal marketing programs 25 "
targeting the middle aged and females can be utilized to promote for sustainable commuting 26"
alternatives. The targeted market shares are potentially expected to change their culture and 27"
perception towards active modes. Moreover, the model can be used for policy analysis to 28 "
investigate the effect of improving the built-environment on active modes usage. Introducing 29"
new bike lanes, enhancing intersection crossings safety, and providing protected and direct 30"
pedestrian paths are few initiatives that can contribute to a more sustainable community. The 31 "
developed model can help in evaluating the effectiveness of some of the above-mentioned 32"
policies as a tool to support the decision making process. 33"
CONCLUSIONS AND FUTURE WORK 34"
This paper studies commuters’ mode choice behaviour in high-density areas. The study focuses 35"
on short distance commuting trips where active modes (i.e., bike and walk) truly compete with 36"
motorized modes. The downtown area of the City of Toronto as one of the most vibrant active 37"
neighbourhoods in North America is selected as a case study. Data from the 2011-2012 38"
Transportation Tomorrow Survey (TTS) is used for the empirical analysis. A Nested Logit (NL) 39"
mode choice model with three nests, namely: auto driver, transit, and active modes is developed. 40"
In addition to personal and household attributes, the model includes variables that explain 41 "
individuals’ active mode choice behaviour such as mode-specific travel distances and times, the 42"
surrounding built environment, and weather conditions. The empirical investigation reveals 43 "
useful insights that are helpful in understanding how active modes compete with motorized 44"
Mahmoud,"M.S.,"Habib,"K.M .N.,"Shalaby,"A." " " "15"
"
modes in high-density areas. The built-environment and weather conditions have a strong effect 1"
on active mode shares. In addition, shorter distances to destinations and lower travel cost per unit 2"
distance contribute significantly to the increase of bike and walk mode shares. 3"
Next steps of this research include the application of the developed model for policy analysis. As 4 "
such, the effect of different variables on short distance trips modal shares can quantified. In 5"
addition, future research may consider comparing short distance commuting trips in different 6"
planning districts of the GTHA. This will provide more insights on how the built environment 7"
affects commuters’ decisions in different parts of the region. 8"
ACKNOWLEDGEMENTS 9"
The authors acknowledge the Data Management Group (DMG) of the Department of Civil 10"
Engineering at the University of Toronto for sharing the 2011-2012 TTS household travel survey 11 "
data for research purposes. 12"
Mahmoud,"M.S.,"Habib,"K.M .N.,"Shalaby,"A." " " "16"
"
REFERENCES 1"
[1] Moudon, A. V., C. Lee, A. D. Cheadle, C. W. Collier, D. Johnson, T. L. Schmid, and R. D. 2"
Weather. Cycling and the built environment, a US perspective. Transportation Research Part D: 3"
Transport and Environment, Vol. 10, No. 3, 2005, pp. 245-261. 4"
[2] Bourgondiën, M. Perceived physical environmental factors related to walking and cycling for 5"
transport in rural and urban areas in the Netherlands. 2011. 6"
[3] Iftekhar, S., and S. Tapsuwan. Review of transportation choice research in Australia: 7"
Implications for sustainable urban transport design.In Natural Resources Forum, No. 34, Wiley 8"
Online Library, 2010. pp. 255-265. 9"
[4] Wigan, M. Bicycle ownership use and exposure: Participation and activity patterns in 10 "
Melbourne, Australia. Transportation Research Part A: General, Vol. 18, No. 5, 1984, pp. 379-11"
398. 12"
[5] Habib, K. M. N., J. Mann, M. S. Mahmoud, and A. Weiss. Synopsis of Bicycle Demand in 13 "
the City of Toronto: Application of an Integrated Econometric Model with Latent Variables.In 14"
Transportation Research Board 93rd Annual Meeting, Washington D.C., 2014. 15"
[6] Baxter, J. T. Essential Factors of Active Transportation: Analysis and Recommendations for 16"
Downtown Atlanta. 2011. 17"
[7] Sallis, J. F., L. D. Frank, B. E. Saelens, and M. K. Kraft. Active transportation and physical 18"
activity: opportunities for collaboration on transportation and public health research. 19"
Transportation Research Part A: Policy and Practice, Vol. 38, No. 4, 2004, pp. 249-268. 20"
[8] Gordon-Larsen, P., M. C. Nelson, P. Page, and B. M. Popkin. Inequality in the built 21"
environment underlies key health disparities in physical activity and obesity. Pediatrics, Vol. 22 "
117, No. 2, 2006, pp. 417-424. 23"
[9] Scheepers, E., W. Wendel-Vos, E. van Kempen, L. I. Panis, J. Maas, H. Stipdonk, M. 24"
Moerman, F. den Hertog, B. Staatsen, and P. van Wesemael. Personal and Environmental 25 "
Characteristics Associated with Choice of Active Transport Modes versus Car Use for Different 26 "
Trip Purposes of Trips up to 7.5 Kilometers in The Netherlands. PloS one, Vol. 8, No. 9, 2013, p. 27"
e73105. 28"
[10] Buehler, R., and J. Pucher. Cycling to work in 90 large American cities: new evidence on 29 "
the role of bike paths and lanes. Transportation, Vol. 39, No. 2, 2012, pp. 409-432. 30"
[11] Gotschi, T., and K. Mills. Active transportation for America: The case for increased federal 31"
investment in bicycling and walking. 2008. 32"
[12] Pucher, J., and R. Buehler. Why Canadians cycle more than Americans: a comparative 33"
analysis of bicycling trends and policies. Transport Policy, Vol. 13, No. 3, 2006, pp. 265-279. 34"
[13] Pucher, J., R. Buehler, and M. Seinen. Bicycling renaissance in North America? An update 35"
and re-appraisal of cycling trends and policies. Transportation Research Part A: Policy and 36"
Practice, Vol. 45, No. 6, 2011, pp. 451-475. 37"
Mahmoud,"M.S.,"Habib,"K.M .N.,"Shalaby,"A." " " "17"
"
[14] Metrolinx. The Regional Transportation Plan for the Greater Toronto and Hamilton Area 1"
(GTHA): “The Big Move”. <http://www.metrolinx.com/thebigmove/en/default.aspx>.In, 2008. 2"
[15] Halldórsdóttir, K., L. Christensen, T. C. Jensen, and C. G. Prato. Modelling mode choice in 3"
short trips-shifting from car to bicycle. ETC 2011, 2011. 4"
[16] Witlox, F. Evaluating bicycle-car transport mode competitiveness in an urban environment. 5"
An activity-based approach Frank Witlox & Hans Tindemans. Editorial board, Vol. 10, No. 4, 6"
2004, pp. 32-42. 7"
[17] Kim, S., and G. F. Ulfarsson. Curbing automobile use for sustainable transportation: 8"
analysis of mode choice on short home-based trips. Transportation, Vol. 35, No. 6, 2008, pp. 9"
723-737. 10"
[18] Heinen, E., B. van Wee, and K. Maat. Commuting by bicycle: an overview of the literature. 11"
Transport reviews, Vol. 30, No. 1, 2010, pp. 59-96. 12"
[19] Xing, Y., S. L. Handy, and P. L. Mokhtarian. Factors associated with proportions and miles 13"
of bicycling for transportation and recreation in six small US cities. Transportation Research 14"
Part D: Transport and Environment, Vol. 15, No. 2, 2010, pp. 73-81. 15"
[20] Pucher, J., J. Dill, and S. Handy. Infrastructure, programs, and policies to increase 16"
bicycling: an international review. Preventive medicine, Vol. 50, 2010, pp. S106-S125. 17"
[21] Howard, C., and E. K. Burns. Cycling to work in Phoenix: route choice, travel behavior, and 18"
commuter characteristics. Transportation Research Record: Journal of the Transportation 19"
Research Board, Vol. 1773, No. 1, 2001, pp. 39-46. 20"
[22] Forsyth, A., J. Michael Oakes, B. Lee, and K. H. Schmitz. The built environment, walking, 21"
and physical activity: is the environment more important to some people than others? 22"
Transportation Research Part D: Transport and Environment, Vol. 14, No. 1, 2009, pp. 42-49. 23"
[23] Cervero, R., and C. Radisch. Travel choices in pedestrian versus automobile oriented 24 "
neighborhoods. Transport Policy, Vol. 3, No. 3, 1996, pp. 127-141. 25"
[24] Guo, J. Y., C. R. Bhat, and R. B. Copperman. Effect of the built environment on motorized 26"
and nonmotorized trip making: substitutive, complementary, or synergistic? Transportation 27"
Research Record: Journal of the Transportation Research Board, Vol. 2010, No. 1, 2007, pp. 1-28"
11. 29"
[25] Saneinejad, S., M. J. Roorda, and C. Kennedy. Modelling the impact of weather conditions 30"
on active transportation travel behaviour. Transportation Research Part D: Transport and 31"
Environment, Vol. 17, No. 2, 2012, pp. 129-137. 32"
[26] Nankervis, M. The effect of weather and climate on bicycle commuting. Transportation 33 "
Research Part A: Policy and Practice, Vol. 33, No. 6, 1999, pp. 417-431. 34"
[27] Mahmoud, M. S., El-Assi, W.; Habib, K.M.N. Effects of Built Environment and Weather on 35"
Bike Sharing Demand: Station Level Analysis of Commercial Bike Sharing in Toronto. 36 "
Presented at the Transportation Research Board 94th Annual Meeting, Washington, D.C., 2015. 37"
Mahmoud,"M.S.,"Habib,"K.M .N.,"Shalaby,"A." " " "18"
"
[28] Mahmoud, M. S. Weiss, A.; Habib, K.M.N. Latent Captivation or Mode Culture? An 1"
Investigation into Mode Choice Preference Structures in Competitive Modal Arrangements 2"
(Multimodal Urban Areas). Presented at the Transportation Research Board 94th Annual 3"
Meeting, Washington, D.C., 2015. 4"
[29] StatisticsCanada. Census Profile, http://www.statcan.gc.ca/. Accessed February, 24th, 2014. 5"
[30] ColliersInternational. North American CBD Parking Rate Survey, http://www.colliers.com/. 6"
Accessed February, 28th, 2014. 7"
[31] CityofToronto. The PATH, http://www.toronto.ca/. Accessed February, 25th, 2014. 8"
[32] WalkScore. http://www.walkscore.com/. Accessed February, 25th, 2014. 9"
[33] Ben-Akiva, M. E., and S. R. Lerman. Discrete Choice Analysis: Theory and Application to 10"
Travel Demand. The MIT Press, 1985. 11"
[34] Croissant, Y. Estimation of multinomial logit models in R: The mlogit Packages.In, R 12"
package version 0.2-3, URL http://cran.r-project.org/web/packages/mlogit/index.html, 2012. 13"
[35] Train, K., and Y. Croissant. Kenneth Train's exercises using the mlogit package for R.In, R 14 "
package version 0.2-3, URL http://cran.r-project.org/web/packages/mlogit/index.html, 2012. 15"
[36] Train, K. Discrete choice methods with simulation. Cambridge university press, 2009. 16"
[37] Bachand-Marleau, J., B. H. Lee, and A. M. El-Geneidy. Better Understanding of Factors 17"
Influencing Likelihood of Using Shared Bicycle Systems and Frequency of Use. Transportation 18"
Research Record: Journal of the Transportation Research Board, Vol. 2314, No. 1, 2012, pp. 19"
66-71. 20"
21"
22"