Content uploaded by Jeffrey M. Casello
Author content
All content in this area was uploaded by Jeffrey M. Casello on Apr 10, 2015
Content may be subject to copyright.
An Analysis of Stated Preference and GPS Data for Bicycle Travel Forecasting
Jeffrey M. Casello
Associate Professor
School of Planning and Department of Civil and Environmental Engineering
University of Waterloo
200 University Ave West
Waterloo, ON Canada N2L 3G1
jcasello@fes.uwaterloo.ca
519 888 4567 ext 37538
Akram Nour
Ph.D. Candidate
Department of Civil and Environmental Engineering
University of Waterloo
200 University Ave West
Waterloo, ON Canada N2L 3G1
anour@engmail.uwaterloo.ca
Kyrylo Cyril Rewa
Master’s Candidate
Department of Civil and Environmental Engineering
University of Waterloo
200 University Ave West
Waterloo, ON Canada N2L 3G1
krewa@uwaterloo.ca
John Hill
Principal Planner
Regional Municipality of Waterloo
150 Frederick Street
Kitchener, ON Canada N2G 4J3
jhill@regionofwaterloo.ca
November 15, 2010
Word count:
Abstract: 178
Text: 5112
Figures 5 (x250): 1250
Tables 5 (x250): 1250
Total: 7790
ABSTRACT
In this paper, we present preliminary results from an ongoing study of cyclists and cycling in the
Region of Waterloo, Ontario Canada. The paper describes two data collection efforts. The first
is an on-line survey that provides information on cyclists’ demographics as well as their
household composition. The survey also gathers data on respondents’ motivation for and
obstacles to cycling. The second activity collects data on actual cycling trips using GPS units.
We describe these units and the steps taken to validate the data. We use the GPS data to produce
trip generation and attraction rates for cycling as a function of land use. We also generate a
histogram of observed cycling trip lengths that can be used to calibrate a gravity-type model of
trip distribution. We then explore the methods by which the survey and GPS data may be
combined to develop multi-class and multi-trip purpose generalized cost functions. These
formulations may be applied to prioritizing infrastructure investments, as well as for mode and
path choice models. We conclude with a discussion of ongoing research work.
INTRODUCTION
Cities throughout the world are challenged by the planning, design and operation of their
transportation systems, particularly in light of growing per capita auto use. Many cities are
attempting to expand the role of higher-efficiency modes - walking, cycling and public
transportation - to accommodate growing travel demand with lower energy consumption, fewer
negative environmental impacts and lower infrastructure costs. This paper presents methods to
collect data on bicyclists and bicycling to improve the planning and evaluation of cycling policy
and investment.
To this end, we have engaged 100 cyclists in two parallel data collection efforts. The first
process asks the cyclists to complete a web-based survey about their demographics, household
compositions, and motivations for (and obstacles to) cycling. The second data collection
technique equips the cyclists with low-cost GPS units to record origins and destinations, paths,
and travel times from which observations can be drawn on the cyclists’ activities. This paper
reports preliminary results from these data collection efforts.
Using the data gathered in the survey, we are able to contribute an overview of the cycling
community - in terms of age, incomes, and auto ownership - for a mid-sized Canadian
metropolitan area. We also identify and rank the relative importance of respondents’ impetus for
and obstacles to cycling. These data can improve the way in which bicycling mode choice and
path choice models are formulated.
From the GPS data, we produce four important results that correspond to the standard, four-step
transportation planning process. By correlating the observed origins and destinations of trips to
land uses, we construct quantitative relationships between cycling trip generation (and trip
attraction) and land use density. We also develop a histogram of observed trip lengths which has
value in calibrating gravity-type models for cycling trip distribution.
For a subset of origins and destinations, we compute the user costs (in terms of time and out-of-
pocket expenses) of travel by cycling and competing modes - auto and transit. The results
validate empirically the survey responses that simple cost representations are insufficient in
predicting cycling mode choice. Finally, we present a similar analysis for path choice. We
compare observed cycling paths between an origin and destination with GIS generated shortest
network paths for the same trips. Our empirical data suggests that shortest path assignment for
cyclists is also not sufficient.
The remainder of the paper is organized as follows. In the next section we review previously
published literature with an emphasis on methods to collect cycling data and the application of
these data in bicycle planning. We then introduce briefly the study location - the Regional
Municipality of Waterloo in Ontario, Canada. Next we explain our data collection efforts in
more detail. We present the results of our study in two sections, beginning first with a statistical
representation of the participants in the study. The subsequent section presents the results of our
GPS data collection and the means by which these data can be used in transportation planning.
The paper ends with conclusions on our work and suggestions for additional research.
LITERATURE REVIEW
The need to understand cyclists and cycling behavior has been recognized by many previous
studies. Replogle (1) performed a “state of the practice” review in 1995 that summarized the
methods employed in practice to estimate cycling and its impacts (in terms of environment and
costs). He also proposed a number of research directions to improve the process of planning for
cyclists. A similar study was conducted by Porter et al. (2) four years later. In that report, the
authors noted a significant disconnect between a growing understanding of cycling motivation
and the inclusion of this knowledge base in multimodal travel forecasting models.
Since these reviews, several authors have tried to generate meaningful predictions of cycling
mode choice in metropolitan areas. Hunt and Abraham (3) used a stated preference survey
(while noting its shortcomings) to develop utility functions for cycling in Edmonton, Alberta.
They extend their findings to a multinomial logit model to predict cycling use. Eash (4) used the
1990 Chicago Area Transportation Study data to calibrate destination and mode choice models
for different trip types - home- and non-home-based - considering also the qualities of potential
destinations. Dill and Carr (5) created a regression model which predicted cycling use as a
function of several variables, including employment, quality of competing modes, land uses,
socio-economics, weather and social perceptions of cycling. Sener et al. (6) have studied the
characteristics of cyclists using a web-based survey originating in Texas. This work correlated
the factors that influenced cycling - individuals, household, neighborhoods as well as various
safety measures - to the frequency of cycling activity. This model is interesting in that it
contains a “latent propensity to cycle” which may be considered as a mode specific constant for
cycling in typical logit models.
In a second paper (7), these authors used stated preference data to construct a utility function for
route choice. Various characteristics of the cyclists and the cycling network are combined in an
econometric framework to predict the likelihood of choosing a given route. This work contains
an extensive literature review of research related to cycling mode and path choice. Other authors
have used stated preference techniques to evaluate the path choice problem. Stinson and Bhat
(8) use a stated preference method to calibrate a route choice model based on route and link (and
their interactions) factors. Their results suggest that travel time is most important in route choice
followed by roadway type and bicycle facility design.
Howard and Burns (9) studied the city of Phoenix. They asked 150 cyclists about their
commuting route choice including origin and destination. They also gathered information about
the cyclists. From these data, they were able to calculate how chosen paths relate to both the
shortest-distance paths and to the safest paths. Their approach motivates the work we present in
comparing our cyclists paths to shortest paths. Their results suggest that cyclists in their study
have disparate perceptions of “optimal” route choice.
STUDY LOCATION
In this section, we briefly introduce the Region of Waterloo and its cycling facilities. The
Region offers an interesting case study as it has several properties that support cycling - the
presence of several universities, for example - and some characteristics which deter cycling,
most notably the climate of southwestern Ontario.
The Region of Waterloo
Waterloo Region is located approximately 100 km west of the City of Toronto in Ontario
Canada. The Region is comprised of three cities - Waterloo, Kitchener and Cambridge (tri-
cities) - and four rural townships. The Regional population is approximately 500,000 but is
expected to grow to 730,000 by 2031. Similar to population, the Region’s economy is expected
to add 50% more jobs in the same time period. A major challenge for the Region in light of the
projected growth is to accommodate increased housing and employment lands without
diminishing the value of local agricultural activities. The intention of Regional planners is to
develop a balanced, multi-modal transportation system that will both facilitate future travel
demands and positively influence land uses (achieve intensification). As such, strong efforts are
underway to increase the viability of higher-efficiency modes such as transit, cycling and
walking.
To this end, the Region has launched “Travel Wise” - the overarching Transportation Demand
Management program for the municipality. Under the Travel Wise program, the Region has
developed a cycling master plan (10) that sets modal targets for cycling - 2% of total trips by
2016 - and has set the budget ($33M) to construct a full Regional network of cycling facilities
(732 km). The Region also conducts and funds cycling education programs, and develops and
implements programs to strengthen the ties between cycling, walking and transit. Table 1
summarizes the Regional characteristics and the cycling network.
Table 1 Characteristics of the Region of Waterloo
Regional Municipality of Waterloo
Population (2009) 534,900
Employment (2006) 257,655
Area (tri-cities) 318 km2
Population density 1682 prs / km2 (16.8 prs / ha)
Length of road network 3,342 km
Cycling Data for the Region of Waterloo
On-road network length 73 km
Off-road network length 457 km
Cycling mode share (commuting - 2006) 6.7%
SURVEY DATA COLLECTION METHODS AND RESULTS
As noted earlier, we conducted two concurrent data collection efforts with 100 volunteer cyclists.
This section presents a summary of the survey used and the data collected.
Demographic Data Collection
Typical variables which are known to influence traveler behavior include age, income, household
composition and auto ownership. We were also interested in understanding specifically what
motivates travelers to choose cycling and what challenges exist to cycling. To collect these data,
we asked participants to complete an extensive survey containing more than 50 questions in
seven categories. The categories of questions and summaries of data collected are shown in
Table 2. We used a commercially available web service to host the survey; we also provided
participants with a paper copy to complete the questions if they preferred.
Table 2 Demographic data collected
Category Data collected
A. Demographics, auto ownership
and vehicle ownership
Respondent’s age, income, frequency of cycling by
season, frequency of other modes; household # autos
owned, # licensed drivers
B. Characteristics of regular
cycling route
Satisfaction with current route, assessment of skill level,
frequency of bike-transit trips (racks on buses), helmet use
C. Cycling behavior Factors that motivate to cycle, factors in current route
choice, obstacles to increased cycling
D. Specific hazards Relative importance of various interactions with cars,
interaction with other cyclists, road (facility) maintenance,
weather
E. Cycling economics $ spent on annual maintenance, level of investment for new
bicycle, and willingness to pay to join a bike-sharing
program
F. Necessary cycling
infrastructure
Prioritized list amongst: paths (on-road, off-road,
boulevard), parking facilities, lighting, shower facilities,
bicycle signage and path maintenance.
G. Miscellaneous Use of GPS / cell phone when cycling; evidence of bicycle
theft; bicycle collisions; level of training; opinions on
motorized bicycles.
Cyclists and Cycling in the Region of Waterloo
The 100 cyclists who participated in the study were volunteers. They were self-proclaimed
winter cyclists who mostly participated (60%) as a response to local newspaper article which
asked winter cyclists to participant in a winter GPS cycling initiative. Word of mouth also
generated much interest in the survey. The geographic location of the participants’ assumed
origins (based on GPS data) are shown in Figure 1.
Figure 1 Location of participant's origins within the Region of Waterloo
The age distribution of respondents was nearly uniform between ages 19 and 60 with 47% being
younger than 40 years old. A comparison of the distribution of study participants with the
overall population of the Region of Waterloo (as reported in the 2006 Census, 11) suggests that
the distribution of ages in our survey under represents seniors and children. This is perhaps
logical since these age groups may be considered less likely (relative to the population as a
whole) to cycle regularly. The mean income in 2005 for the Region of Waterloo was $29,449
(before taxes) according to Statistics Canada. The most common (23%) income range in our
survey was between $50,000 and $75,000; 34% of respondents indicated incomes greater than
$75,000. Generally, the participants in our study can be considered higher-than-average earners.
We interpret this result to suggest that our participants choose to cycle rather than being limited
to cycling based on their financial means.
We examine the relationship between auto ownership and the driving demand from households
in which our participants reside. In our study, 96% of the participants have a valid driver’s
license. Table 3 shows the relationship of auto ownership to number of licensed drivers in the
study participants’ households. In our study, 22% of households own the same number of cars as
licensed drivers. Seventy-eight percent of our households own fewer cars than there are licensed
drivers; of particular note is the fact that 28% of households function with only one car for three
licensed drivers. Note that in no household does auto ownership exceed the number of licensed
drivers. We interpret the data in table 3 to support the conclusion that regular cycling can reduce
a household’s auto ownership needs and, by extension, reduce the household’s expenditures on
transportation.
Table 3 Licensed drivers and auto ownership for respondents' households
Number of autos owned by household members
Number of licensed
drivers in household
0 1 2 3+
0 2% 0% 0% 0%
1 8% 8% 0% 0%
2 3% 28% 10% 0%
3+ 0% 28% 10% 2%
When asked what mode they would use if they were not able to ride their bicycle for an extended
period of time, nearly 40% of participants responded that they would use a currently-owned
personal vehicle. We interpret the results to mean that for many respondents the car is a viable
option for many trips. This observation advances the notion that cyclists are not “captive”
cyclists but derive more personal utility (or experience lower cost) from cycling than from auto
for many trips. Other respondents indicated that they would use transit (28%), walk (22%), or
buy a new car (8%).
GPS DATA COLLECTION METHODS AND RESULTS
Global positioning systems (GPS) are becoming increasingly common and significantly less
expensive. We purchased and distributed 50 hand held GPS devices to two study groups (100
volunteers total) over a five week period. For our study, we collected location (x,y,z) data every
three seconds. The data we collected represent 1,232 trips covering 7,248kilometers; the
average speed observed in our data is approximately 15 km per hour.
The data are stored in a simple column form; all data collected from the time the unit is turned on
until it is turned off is contained as a single trip or, in GPS terms, a “trace.” The GPS loggers
also come with their own software that allows the user to overlay commercial mapping on the
traces that are created; the software also produces speed profiles.
The accuracy of the GPS data varies in some cases by 10-15 meters from the roadway. The
quality of the data is influenced by external conditions. In open spaces, with clear skies the units
gathered the most accurate data. On overcast days or adjacent to large structures, the data were
often less reliable. The degree of exposure for the unit itself - whether fully exposed or enclosed
in a pocket or bag - influences data quality. In subsequent efforts, we have provided participants
with a transparent pouch which maximizes the unit’s exposure.
Data Validation
The data collected often contained errors in both the x, y and z dimensions which resulted in
erroneous distances and speeds. To validate the data we developed an automated technique to
identify and eliminate erroneous points while still maintaining continuous traces. The first data
correction method was to eliminate points that exceeded elevation limits observed in the Region
- in our case between 0 and 600 meters. Next, we deleted points between which unrealistic
elevation changes occurred; if two subsequent points differed in elevation by 35 meters, these
points were deleted.
Similarly, a maximum speed threshold was established as 75 km/hr; any points generating speeds
higher than 75 km/hr were deleted. We also tested each point for consistency in speed with
previous and subsequent points. The rules associated with deleting points based on inconsistent
speed are shown in Figure 1. If we consider four points, 0 through 3, and the three segments that
connect them (01, 12, 23), we can compute observed speeds on each segment. We also compute
the differences in speeds amongst consecutive segments. If we observe a high difference in
speeds amongst three segments, or if we observe a very high speed difference between two
consecutive segments, we eliminate the common point in the sequence.
Figure 2 Logic used in assessing consecutive speeds
It was our intention to maintain as many traces as possible. As such, we eliminate only
individual suspect points from the traces and recreate continuous traces connecting points
immediately prior and immediately following the eliminated data. We reapply the validation
rules to ensure the suitability of this connection. For some traces, the ratio of non-suspect to
suspect points was too low and, as a result, these traces were not kept in the final data set.
We were also interested in understanding trip purpose and limiting our analyses to non-
recreational cycling because trips for recreation will have sufficiently different characteristics
than cycling for more utilitarian purpose. We developed a methodology which analyzed each
trip to identify trips with: 1) the same origin and destination – i.e. a full cycle; and 2) no
prolonged stops – i.e. times of one hour or longer with very little or no motion. Of the more than
1100 traces we analyzed, approximately 9% were identified as recreational trips. These trips
were eliminated from future analysis.
GPS Study Results
The intentions of the data collection are two-fold: to gain a better understanding of cyclists and
cycling in the Region of Waterloo and to use the data collected to begin to improve bicycle
planning. The well-known four-step transportation planning procedure contains sub-models of:
Trip production and attraction;
Trip distribution;
Mode choice, and;
Trip assignment which allocates individual trips to paths or routes based on an assumed
user objective, typically minimum cost.
Each of the following sections addresses these steps using empirical data.
Trip Production and Attraction
From our data, we are able to estimate the number of cycling trips produced from and attracted to
areas of differing land use. To do so, we first link a cycle trip’s origin and destination to the
appropriate Regional Traffic Analysis Zone (TAZ). We then sort the TAZs by the number of
observed trips which begin in that zone and by the number of observed trips that end in that
zone. We classified the Region’s TAZs into five categories of origins and five categories of
destinations. Table 4 shows the limits on trip activity for each category and shows the number of
TAZs which fall into each category. To interpret Table 4, consider the following example. If for
a given TAZ we observed eight trip origins and 12 trip destinations, then that TAZ would be
considered a category 4 origin and a category 5 destination.
Table 4 Category definitions for land use analysis
Categories
1 2 3 4 5
Number of observed bicycle trips originating
from or destined to zones in this category 0 1 2 to 4 5 to 9 10+
Number of origin TAZs in category 385 84 69 31 7
Number of destination TAZs in category 435 74 54 9 4
Average land use density of origins 1,391.2 3,321.4 3,829.0 6,883.0 12,488
Average land use density of destinations 1,535.5 4,155.5 4,851.0 7,204.8 15,155
We next calculated the land use density for each TAZ in the Region of Waterloo. The Province
of Ontario (12) has established the following definition of land use density, D:
(1)
where area is calculated in km2. We calculated the average land use densities for zones belonging
to each category for both origins and destinations. These are also shown in Table 4. As is
expected, the land use density increases with the number of observed trips. We then calculated
trip generation and trip attraction rates as a function of population and employment for each
category. Quantitatively, we computed the trip generation from zone category i, TGi and trip
attraction rates to zones in category j, TAj as:
(2)
(3)
In Figure 2, we plot these rates as a function of the land use density in each category. (For
clarity, we present the rates per 10,000 population and employment). Also in Figure 2, we plot a
logarithmic trend line that best fits the data to provide a generalized estimate of cycling trip
generation and attraction as a function of land use.
Figure 3 Cycling trip attraction and generation rates as a function of land use density
TG = 3.84ln(land use density) - 27.45
R² = 0.935
TA = 3.342ln(land use density) - 23.97
R² = 0.892
0
1
2
3
4
5
6
7
8
9
10
0 2000 4000 6000 8000 10000 12000 14000
Observed Trip Rate per 10,000 Population +Employment
Land Use Density (population+employment)
Trip Generation Trip Attraction Log. (Trip Generation) Log. (Trip Attraction)
The expressions for trip generation and attraction shown in Figure 2 require additional
explanation. The absolute values only reflect the rates generated by our (very small) subset of
Regional travelers. In order to accurately predict total trip generation rates for the Region as a
whole, these equations need to be “scaled up” and adjusted to account for the higher riding habit
of our study participants.
What can be discerned from the expressions presented in Figure 2 are the relative trip generation
rates for various land use densities. In other words, we interpret Figure 2 to mean that increasing
land use densities (as defined above) from 2000 to 4000 is likely to double the trip generation
and attraction rates for those zones.
Trip Distribution
One input necessary to calibrate traditional travel forecasting models is the distribution of trip
lengths observed in the study area and in the time period being modeled. Figure 3 shows the
distribution of trips observed in our study. The mean distance traveled is 6.9 km and the
standard deviation of trip lengths is 5.5 km. Approximately 77% of the observed trips are shorter
than 12km in length.
Figure 4 Distribution of trip distances
One definition of transportation accessibility is the percentage of local destinations accessible
within a given travel distance. Based on this definition and the range of trip lengths shown in
Figure 3, we are able to demonstrate the extent to which cycling provides access to Regional
destinations. We overlaid the Region of Waterloo with three 12 km diameter circles centered at
the points from which survey units were distributed. These circles represent the areas which may
be accessed in a trip of less than 6 km if a cyclist began a trip at one of the distribution locations.
The area contained in the circles is approximately 296.3 km2 or 93% of the tri-cities’ area. The
area that is “inaccessible” by cycling - the area between Kitchener and Cambridge - includes a
generally low-density, former industrial area.
17%
42%
18%
12%
4% 3% 1% 1% 0% 1%
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
<6 6-9 9-12 12-15 15-18 18-21 21-24 24-27 27-30 >30
Frequency (%)
Distance Traveled (km)
Mode and Path Choice
Mode and path choice models typically assume that travelers choose the mode and then route
that maximize their utility (or, alternatively, minimizes their disutility) for a trip between an
origin and destination. Utility functions typically include a number of variables of different
magnitudes and units; the approach most often taken is to convert all of the input variables to a
single unit, usually dollars, and to compute and compare generalized costs (GC) for each mode
or path. Much research has been conducted through stated preference surveys to develop
appropriate generalized cost functions - in terms of the variables considered and their relative
weightings - for various modes. Our study adds to this literature.
In completing the survey, study participants indicated on a scale from 1 to 5 the relative
importance of many variables related to both mode and path choice. To determine a priority
ranking of motivations, we calculated the sum of the product of the rank and the percentage of
respondents choosing that value. For example, responses and calculations for convenience are:
Convenience rank = 1.3% ·1 + 3.8% · 2 + 11.3% · 3 + 36.3%·4 + 47.5%· 5 = 4.26
Our results, shown in Table 5, suggest that the primary reason for cycling is its convenience
relative to other modes. No formal definition of convenience was presented in the survey. We
presume that convenience may include the ability to choose exactly your departure time (as
opposed to waiting for a scheduled transit departure time); shorter access and egress distances (as
opposed to longer walking distances to / from parking areas); or shorter travel time (when
compared to walking.)
Table 5 Relative ranking of influences on cycling behavior
Motivations for cycling
Variable Importance
Convenience compared to other modes 4.26
Contribution to environment 4.19
Lower cost compared to other modes 3.80
Allows for recreation 3.42
Improves health 3.40
Obstacles to cycling
Feels unsafe 2.70
Poor motorist behaviour 2.66
High traffic volumes 2.65
Poor road conditions 2.32
Travel time is long 1.88
Poor weather 1.63
Many stops 1.62
Distance travelled is long 1.54
Lack of bike parking 1.44
Route not scenic 1.23
Factors influencing route choice
Feels safe 2.91
Shortest by time 2.90
Low amount of traffic 2.83
Best road conditions 2.64
Shortest by distance 2.55
Fewest stops 2.10
Route is scenic 1.99
There are strong correlations between the deterrents to cycling and the variables which influence
path selection. Safety is the most important consideration in electing whether and where to
cycle. The study also reveals that motorist behavior and vehicular volume are more important
than road conditions. We do note one difference in the relative importance of these variables.
Travel time has a much weaker influence on mode choice than on path choice.
Applying survey results in estimating generalized costs
It is our intention to use these results to formally develop and calibrate generalized cost functions
for cycling to be applied in the mode choice function of the Region of Waterloo’s multimodal
travel forecasting model. Several issues have arisen in these efforts. Mode choice models
estimate mode shares between traffic analysis zones; therefore, we are required to develop
interzonal generalized cost representations for cycling. To do so necessitates a composite, zonal-
level representation of the cost representations. In other words, to generate meaningful mode
share estimates, we have to develop a method to quantify the average generalized cost of travel
from zone i to zone j considering traffic volumes, speeds, motorists’ behavior, roadway
conditions, etc. on all (many) possible paths between i and j. That research is ongoing.
To demonstrate the importance of generalized costs in bicycle planning, we present the following
application. We chose 20 different origins and destinations and calculated bicycling generalized
costs for the 380 origin destination pairs (eliminating the case where the origin and destination
are the same). We compute the generalized cost of cycling, GCc as the product of travel time
and value of time; we use the Region’s estimate of $10.30 per hour for value of time in our
calculations. We then used the output from the Region’s travel forecasting model to develop a
matrix of generalized costs between the same OD pairs by auto, GCa and transit, GCt. For all the
OD pairs, we calculate the ratio of generalized costs. A ratio of less than one suggests that
bicycling is competitive - i.e. will generate a significant mode share. Figure 4 shows the range of
ratios observed.
Figure 5 Ratio of generalized costs of cycling, bus and auto
Using this formulation of GCc, cycling is superior to auto and transit in nearly every case which
would result in cycling mode shares much higher than the Region’s 6.7%. This suggests that
only considering travel time in estimating GCc greatly underestimates the true cost of cycling
perceived by travelers and further motivates our research into a composite GCc estimate.
Despite the underestimates of generalized cost for cycling, Figure 4 does hold value in
prioritizing bicycling investments. The relative values of the GC ratios give insight into for
which OD pairs bicycling currently competes well (as defined by (13)), and for which pairs
infrastructure investments may produce the biggest improvements in attracting cyclists.
The path choice problem is more tractable. For travel between any origin and destination
address, there exists a finite set of paths - both on-road and off-road - for which the attributes
(length, vehicle speeds and volumes, etc.) can be known or estimated directly. Because our GPS
data provide observations on origin, destination and path, we can compute GCc - under various
formulations - for every trace. Using GIS, we can calculate the path between origin and
destination hat results in the lowest GCc (again under various formulations). The GCc
formulation that produces the lowest disparity between observed and predicted is the best
representation of the study participants’ perception of costs.
As an initial attempt to validate this approach, we analyzed 50 observed commuting trips which
we identify by departure time (am peak period) and the pattern of destination. In this case, we
expect that travel time and distance may be more important than the other variables presented in
Table 5. As such, for each trip we computed a distance-only generalized cost (shortest path)
between origins and destinations. We then calculated the ratio of the observed path and shortest
path distance. A value of 1 for this ratio indicates the cyclist chose the shortest path. Of the 50
trips, only 22 had ratios of observed travel distances to shortest travel distances of less than 1.25;
0
220
87
24
4704
20
164
138
39
19 866
0
50
100
150
200
250
<0.25 0.25-0.5 0.5-0.75 0.75-1 1-1.25 1.25-1.5 1.5-1.75 >1.75
Number of Trips
Generalized Cost Ratio
Biking Over Transit Biking over Auto
12 trips had ratios greater than 2.0. These results imply that even for commuting trips, where
conceivably directness of travel should be most important, many cyclists in our study choose
paths based on different criteria.
ONGOING RESEARCH
This paper describes the data collected with 100 participants. We intend to continue these data
collection processes for the next six months. We have also purchased 50 additional GPS units so
that in each two week period we are able to collect data from 100 participants. In total, we hope
to collect data on approximately 1000 cyclists in the Region of Waterloo.
With this level of participation, we expect to be able to produce disaggregate models of cycling
behavior based on trip purpose - commuting, shopping or recreation - and traveler characteristic -
both income level and household composition. We have also introduced a travel diary to help in
the validation of trip purpose which is now completed by an analysis of the data.
CONCLUSIONS
This paper describes two methods of collecting data on cycling and cyclists in the Region of
Waterloo, Ontario. The first method is an on-line survey of cyclists. Our results indicate that
cycling supports the travel demands of participants’ in all age groups, with more than half of our
respondents being older than 40. Our data also suggest that the respondents’ incomes are higher
than the Regional average; despite these earnings, the study participants’ households own fewer
cars than there are licensed drivers. We interpret these results to mean that cycling facilitates
lower auto ownership and thereby financial savings. We also note that 40% of the respondents
indicate that when not cycling, their primary mode is private auto. This suggests that a high
proportion of participants have a car available to them, but elect to cycle frequently.
The second data collection method described in the paper employs low cost GPS units to collect
origins, destinations, paths and times for cycle trips in the Region. Our data cleaning involves
simple, automated rules that eliminate points based on unreasonable speeds and elevations;
individual trips are recreated from the remaining points. From the cleaned data, we are able to
produce relationships between land use density (at the origin and destination) and the number of
cycling trips generated and attracted. We are also able to construct a trip distribution model
based on our observations.
The survey results also provide guidance on how our respondents perceive the generalized costs
of cycling; lack of safety, particularly interactions with private automobiles is a major factor that
both dissuades cycling and influences the path taken when cycling. We demonstrate empirically
using our observed data that generalized costs consisting of only travel time or distance are
inadequate in predicting cyclists mode or path choice. Our ongoing research attempts to
calibrate improved generalized cost formulations using the GPS data; we further intend to
combine the survey data with the GPS data to produce multi-class, multi-trip purpose models of
mode and path choice.
ACKNOWLEDGEMENTS
This research is supported by the Regional Municipality of Waterloo. We gratefully
acknowledge their contribution.
REFERENCES
1. Replogle, M. Integrating Pedestrian and Bicycling Factors into Regional Transportation
Planning Models: Summary of the State-of-the-Art and Suggested Steps Forward.
Environmental Defense Fund, Washington, D.C., 1995, pp. 75 – 96.
2. Porter, C., J. Suhrbrier, and W. Schwartz. Forecasting Bicycle and Pedestrian Travel: State of
the Practice and Research Needs. In Transportation Research Record, Journal of the
Transportation Research Board, No. 1674, Transportation Research Board of the National
Academies, Washington, D.C., 1999, pp. 94-101.
3. Hunt, J. and J. Abraham. Influences on Bicycle Use. Transportation, Vol. 34, No. 4, 2007,
pp.453 – 470.
4. Eash, R. Destination and Mode Choice Models for Nonmotorized Travel. In Transportation
Research Record, Journal of the Transportation Research Board, No. 1674, Transportation
Research Board of the National Academies, Washington, D.C., 1999, pp. 1-8.
5. Dill, J., and T. Carr. Bicycle Commuting and Facilities in Major U.S. Cities: If You Build
Them, Commuters Will Use Them – Another Look. In Transportation Research Record,
Journal of the Transportation Research Board, No. 1828, Portland State University,
Portland, OR., 2003, pp. 116 – 123.
6. Sener, I., N. Eluru, and C. Bhat. Who are Bicyclists? Why and How Much Are They
Bicycling? In Tranportation Research Record, Journal of the Transportation Research
Board, No. 2134, Transportation Research Board of the National Academies, Washington,
D.C., 2009, pp. 63 – 72.
7. Sener, I., N. Eluru, and C. Bhat. An Analysis of Bicycle Route Choice Preferences in Texas,
US. Transportation, Vol. 36, No. 5, 2009, pp. 511-539.
8. Stinson, M., and C. Bhat. Commuter Bicycling Route Choice: Analysis Using a State
Preference Survey. In Transportation Research Record, Journal of the Transportation
Research Board, No. 1828, Transportation Research Board of the National Academies,
Washington, D.C., 2003, pp. 107 -115.
9. Howard, C., and E. Burns. Cycling to Work in Phoenix: Route Choice, Travel Behaviour,
and Commuter Characteristics. In Transportation Research Record, Journal of the
Transportation Research Board, No. 1773, Maricopa County Department of Transportation,
Phoenix, AZ., 2001, pp. 39 – 46.
10. Region of Waterloo. Cycling Master Plan, 2004.
http://www.region.waterloo.on.ca/web/region.nsf/97dfc347666efede85256e590071a3d4/BB
A9F35E6E019ADA8525760F0055AC05/$file/CYCLING_MASTER_PLAN_2004.pdf?Ope
nElement. Accessed June 2010.
11. Statistics Canada. Waterloo, Ontario (table). Community Profiles. 2006 Census. Statistics
Canada no. 92-591-XWE. Ottawa. Http://www12.statcan.ca/census-recensement/2006/dp-
pd/prof/92-591/index.cfm?Lang=E. Accessed June 2010.
12. Places to Grow Act. Growth Plan for the Greater Gold Horseshoe, 2006.
https://www.placestogrow.ca/images/pdfs/FPLAN-ENG-WEB-ALL.pdf. Accessed June
2010
13. Casello, J. Transit Competitiveness in a Polycentric Metropolitan Region, Transportation
Research A, Policy and Practice Vol. 41, No.1, 2007, pp.19-40.