You are on page 1of 20

Transportation Research Part A 129 (2019) 52–71

Contents lists available at ScienceDirect

Transportation Research Part A


journal homepage: www.elsevier.com/locate/tra

Bikeway prioritization to increase bicycle network connectivity


T
and bicycle-transit connection: A multi-criteria decision analysis
approach

Ting Zuoa, Heng Weia,b,
a
ART-EngineS Transportation Research Laboratory, Department of Civil & Architectural Engineering & Construction Management, The University of
Cincinnati, Cincinnati, OH 45221-0071, USA
b
Beijing University of Technology, Shandong Jiaotong University, and Beijing Jiaotong University, China

A R T IC LE I N F O ABS TRA CT

Keywords: Due to limited resources such as budget and land space, it remains challenging to identify key
Bikeway prioritization prior corridors on which investments on bikeways (e.g., traditional and protected bike lanes) are
Multi-criteria decision analysis most valuable to increasing low-stress bike network connectivity and bicycle-transit connection.
Bicycle connectivity In built-up urban areas, it’s usually difficult to find the extra space for building bikeways, and
Bicycling access to transit
vehicle delays could be worse off if there is no road widening. Facing these challenges, the paper
Built-up urban area
introduces a systematic planning analytics aiming to increase bicycle connectivity and bicycle-
transit connection with minimized impacts on motor vehicles and cost. The stakeholders con-
cerned include bicyclists, transit users, motor vehicle drivers, and investors. To reduce conflict
potentials due to multiple factors involved in decision making choices between stakeholders, a
Multi-Criteria Decision Analysis (MCDA) technique is applied to compare alternatives based on a
set of performance evaluation criteria. Performance criteria include bicycling connectivity be-
tween origins and destinations, bike share connectivity to destinations, bicycle-transit service
coverage population, bike share connectivity to transit, automobile traffic delay, and capital
costs. Criteria weights are determined by pair-wise comparisons in which the conflicts between
stakeholders and trade-offs among criteria are considered in a quantity-based method. A final
preference score calculated in the sum of the weighted and normalized performance measure-
ments is used to rank alternatives. The approach is applied through a case study in the Uptown
Cincinnati, Ohio to determine the priorities of the proposed new bikeways in the area. The results
indicate that the protected bike lanes exhibit a more significant advantage over bike lanes in
terms of the benefits for bicyclists and transit users. At the same time, however, higher con-
struction cost and countermeasures to reduce the potential of consequent traffic congestion need
to be need carefully considered.

1. Introduction

A connected bike network between many possible origins and destinations (ODs) is a major determinant on bicycle ridership
(Furth et al., 2016; Lowry and Loh, 2017). Meanwhile, providing connected and safe bicycling paths to public transit expands transit
service coverage and encourages more people to bike-and-ride (Griffin and Sener, 2016; Zuo et al., 2018). In addition, as bike-sharing


Corresponding author.
E-mail addresses: zuotg@mail.uc.edu (T. Zuo), heng.wei@uc.edu (H. Wei).

https://doi.org/10.1016/j.tra.2019.08.003
Received 6 September 2018; Received in revised form 28 July 2019; Accepted 5 August 2019
Available online 14 August 2019
0965-8564/ © 2019 Elsevier Ltd. All rights reserved.
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

programs provide the public with affordable and convenient access to bicycles, continuous bike paths connecting bike-sharing sta-
tions with various destinations and transit stops are essential to promote the bike sharing and bicycle-transit integration (El-Assi
et al., 2017; Griffin and Sener, 2016; Martens, 2004). In most U.S. cities where no or only a few bikeways exist to protect bicyclists
from mixed vehicular traffic, the concern of comfort and safety is a critical factor deterring people to bike (Furth, 2012). Statistical
facts have shown that bicycle as a transportation mode has been greatly underutilized in the U.S. The latest Benchmarking Report by
Alliance for Biking & Walking (2018) showed that the average percentage of bike commuters in all states was only 0.6% of all
commuters. The share of bicycling was less than 5% of all transit access trips according to the 2017 U.S. National Household Travel
Survey data (U.S. Department of Transportation and Federal Highway Administration, 2018). The underutilization is mainly asso-
ciated with the concerns of the danger and stress posted by motor traffic (Winters et al., 2011). A well-connected low-stress bike
network is a key part of encouraging more people to switch onto bike and/or bike-and-ride mode.
In the U.S., there are approximately 33 states that have developed statewide bike plans. Due to limited resources, it’s always
appropriate and favorable to determine the key prior bike travel corridors for the implementation purpose (Biton et al., 2014). When
designing bike networks in built-up urban areas, planners always face the barrier that most urban streets are constrained by large
volumes of motor traffic (Federal Highway Administration, 1998). Finding the extra width for implementing bikeways through road
widening is often very difficult. Accordingly, road diets by narrowing down or eliminating automobile lanes become preferable. In
such cases, a traffic analysis should be performed to see how the vehicular traffic would be impacted after the roadway re-
configuration. However, little literature has been reported on the analysis of the space required by bikeways and impacts on auto-
mobile traffic in bike network design problem (such as Duthie and Unnikrishnan, 2014; Mauttone et al., 2017).
Regarding the above concerns, it’s imperative to identify an approach to facilitate bike network design focusing on improved
bicycling connectivity and bicycle-transit integration. The complexity of the problem stems from a multitude of factors influencing
decision choices as well as trade-offs among these criteria. Using the Multi-Criteria Decision Analysis (MCDA) as an effective tool to
evaluate alternatives and trade-offs based on a set of evaluation criteria, the paper presents an approach based on MCDA to aid
decision-makers in assessing bike improvement plans. The process of selecting the most valuable plans for increasing bicycle con-
nectivity and bicycle-transit connection and minimizing negative impacts on vehicular traffic and cost involves multiple stakeholders
and criteria. Stakeholders concerned include bicyclists, transit users, motor vehicle drivers, and investors. Selected performance
measures are OD bicycling connectivity, bike share connectivity to destinations, bicycle-transit service coverage population, bike
share connectivity to transit, automobile traffic delay, and capital costs. Following the MCDA method selection tree by Greene et al.
(2011), the Analytic Hierarchy Process (AHP) is adopted to derive criteria weights though pair-wise comparisons, and a utility
function-based method is used to calculate final preference scores of alternatives. The application of the method is demonstrated
through a case study of proposed bikeways in Uptown Cincinnati, Ohio. The outcome includes identified critical corridors and
priorities of bike plans.
The rest of the paper is organized as follows: Section 2 summarizes the literature relative to the topic; Section 3 presents the
developed methodology of MCDA approach for evaluating and prioritizing bike improvement plans; following that, the case study
area in Uptown Cincinnati, Ohio and relative data to demonstrate the application of the methodology are described in Section 4; case
study results are then presented in Section 5; in the last, conclusions are drawn in Section 6 including the policy implication, the
contributions and limitations of the paper, and the future research.

2. Literature review

2.1. Bicycling connectivity

Factors measuring the bicycling connectivity of two points over a transportation network have been extensively studied. Among
the factors, travel distance or travel time has been widely used as an important one (Broach et al., 2012; Hood et al., 2011; Zuo et al.,
2018). The distance factor can be conventionally measured willingness distance or time to travel by bike (Mistretta et al., 2009),
which can be determined through an investigation of empirical travel data, such as 85th percentile distance or time (El-Geneidy et al.,
2014). Studies showed that the bicycling can easily cover the distance shorter than 3 miles between ODs (Buehler, 2012; Pucher and
Dijkstra, 2000). Compared with the commonly recommended 0.25-mile walking distance to access transit (Kim et al., 2005; Kittelson
& Associates et al., 2013; Murray and Davis, 2001), riding a bike can extend the access distance to 1–2.62 miles in first-and-last-mile
trips to transit lines (Flamm and Rivasplata, 2014; Hochmair, 2015; Lee et al., 2016; Rietveld, 2000).
Concerns with traffic stress associated with safety and comfort have a direct influence on the bicycling connectivity (Broach et al.,
2012). In the auto-dependent environment where no or few on-street bikeways are available, bicyclists have to share automobile
lanes with motor vehicles. A higher volume of automobile traffic produces higher stress that makes those streets less appealing and
not comfortable for most bicyclists (Heinen et al., 2010; Parkin et al., 2008; Winters et al., 2011), which easily results in fragmented
bicycling paths and poorly connected bike network (Furth et al., 2016). The poor bike network connectivity further discourages
bicycling and bike-n-ride, as well as the bicycling distance.
On-street bikeways include bike lanes and protected bike lanes. A bike lane is an exclusive space for bicyclists, typically, de-
signated through the use of pavement markings and signage without physical barriers. A protected bike lane is referred to the
exclusive space for bicyclists with the physical protection from vehicular traffic. Exclusive bike lanes on major roads may reduce the
risks for bicyclists and encourage participation by offsetting the negative effects of higher traffic volumes (Broach et al., 2012) while
improving the connectivity and travel times (Winters et al., 2011). Using a stated preference survey conducted via the Internet,
Stinson and Bhat (2003) revealed that the presence of a bicycle facility (especially a separate path), the level of automobile traffic,

53
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

and the presence of a bicycle facility on a bridge are a very important determinant. Hood et al. (2011) used the data collected in San
Francisco to reveal that streets with bike lanes were preferred, especially by infrequent cyclists. According to the National Association
of City Transportation Officials (NACTO) (2016): “In cities that are building protected bike lane networks, cycling is increasing, and the risk
of injury or death is decreasing.”

2.2. Methods of measuring bicyclist’s perceived safety and comfort

Several evaluation methods have been developed with roadway, traffic, and environment attributes that affect bicyclists’ per-
ceived safety and comfort level. Commonly used approaches are reviewed in this section and summarized as follows.

(1) Bike Level of Service (BLOS)

The BLOS developed by Landis et al. (1997) measures the perceived comfort and convenience for bicycle mode. Highway Capacity
Manual (HCM)’s BLOS (Transportation Research Board, 2010) evaluates bicyclist perceived safety and comfort with a function of
BLOS = 0.76 − 0.005we2 + 0.507ln(vma/4Nth) + 0.199[1.1199ln(SRa − 20) + 0.8103](1 + 0.1038PHVa) + 7.066/ Pc2 (1)
where We is effective width of outside through lane, vma is adjusted mid-segment demand flow rate, veh/h, Nth is number of through
lane, SRa is adjusted motorized vehicle speed, mph, PHVa is adjusted heavy vehicle flow rate in midsegment demand flow rate, %, Pc is
pavement condition rating.
The BLOS is classified into six levels: A (BLOS ≤ 2.00, extremely high compatibility), B (BLOS 2.01–2.75, very high compat-
ibility), C (BLOS 2.76–3.50, moderate high compatibility), D (BLOS 3.51–4.25, moderate low compatibility), E (BLOS 4.26–5.50, very
low compatibility), F (BLOS > 5.50, extremely low compatibility). Streets with a LOS D or lower (with a BLOS > 3.50) are sug-
gested as deficiencies of the bike network (Ilie et al., 2016).
Lowry et al. (2012) developed a method to assess the quality of bicycle travel throughout a community based on link-specific bike
suitability measured by HCM BLOS method. Callister and Lowry (2013) introduced tools and strategies to assess the perceived
comfort and safety of bicycle travel across a large geographic area based on BLOS. They stated that the data requirements are hard to
meet for most communities. A simplified BLOS was developed (Ali et al., 2012), which only use four variables, i.e., the presence of a
bicycle lane, the posted speed limit, the number of traffic lanes, and the number of unsignalized conflicts per mile. One drawback of
the HCM’s and simplified BLOS methods is that protected bike lanes are not included. To address this issue, a BLOS model for
protected bike lanes was proposed by Foster et al. (2015) as a compliment to the HCM’s BLOS model. This model is expressed as a
function of the type of buffer, direction of travel, speed limit, and annual daily traffic (ADT).

(2) Bicycle Compatibility Index (BCI)

The BCI was developed to evaluate the capability of roadway segments to accommodate both motorists and bicyclists (Harkey
et al., 1998). It reflects the comfort levels of bicyclists based on geometric and operational conditions of a variety of roadways, as
measured by Eq. (2).
BCI = 3.67 − 0.966BL − 0.410BLW − 0.498CLW + 0.002CLV + 0.0004OLV + 0.022SPD + 0.506PKG − 0.264AREA + AF (2)
where BL is the presence of a bicycle lane or paved shoulder ≥0.9 m, BLW is the bicycle lane or paved shoulder width, m, CLW is the
curb lane width, m, CLV is the curb lane volume, vph in one direction, OLV is the volume of other lane(s) at the same direction with
the curb lane, vph, SPD is the 85th percentile speed of traffic, km/h, PKG is the presence of a parking lane with more than 30 percent
occupancy, AREA is the type of roadside development, AF is the adjustment factor.
To remain consistent with the HCM’s BLOS, six BCI levels designations were defined in the model: A (BCI ≤ 1.50), B (BCI
1.51–2.30), C (BCI 2.31–3.40), D (BCI 3.41–4.40), E (BCI 4.41–5.30), and F (BCI > 5.30). Similar to the BLOS, streets with moderate
high or higher compatibility are recommended to be used as bike network (Duthie and Unnikrishnan, 2014; Ilie et al., 2016). The BCI
model has been used in studies of bike network design problems (e.g., Duthie and Unnikrishnan, 2014) and bicycle routing (e.g., Ziari
and Khabiri, 2010) to evaluate the bicycling suitability. Similar to the HCM’s BLOS method, intensive data are required, and no
protected bike lanes are addressed in the BCI model.

(3) Level of Traffic Stress (LTS)

Sorton and Walsh (1994) developed the Bicycle Stress Level (BSL) index to assess the bicycle compatibility of roadway segments.
The BSL outlines a simplistic calculation of ordinal rating score (1 to 5). Based on the BSL approach, Mekuria et al. (2012) proposed
LTS to measure the stress felt by cyclists and bicycle compatibility. In the approach, four LTS levels are designated as follows

• LTS 1: tolerated by most children,


• LTS 2: tolerated by the mainstream adult population,
• LTS 3: tolerated by American cyclists who are “enthused and confident” but still prefer having their own dedicated space for
riding, and
• LTS 4: tolerated only by those characterized as “strong and fearless”.
54
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Table 1
Mekuria et al.’s (2012) LTS criteria in mixed traffic.
Speed Limit Street width

2–3 lanes 4–5 lanes 6+ lanes

Up to 25 mph LTS 1* or 2* LTS 3 LTS 4


30 mph LTS 2* or 3* LTS 4 LTS 4
35+ mph LTS 4 LTS 4 LTS 4

* Use lower value for streets without marked centerlines or classified as residential and with fewer than 3 lanes; use higher
value otherwise.

For the protected bike lanes, LTS = 1; for bike lanes and mixed traffic, LTS can vary from 1 to 4. Details of LTS criteria are enlisted
in Tables 1 and 2. The bicycle LTS was designed to guide bicycle infrastructure development (Mekuria et al., 2012) and requires less
intensive data (Wang et al., 2016). The LTS method has been applied in some recent bicycle connectivity studies. For example, Lowry
et al. (2016) introduced a new method to prioritize bicycle improvement projects based on low-stress accessibility to important
destinations.

(4) Other alternatives

Davis (1987) developed the Bicycle Safety Index Rating (BSIR) as one of the pioneer endeavors to evaluate the comfort and safety
levels of bicyclists on roadway segments using mathematical models. Factors considered in the BSIR include number of travel lanes,
width of outside lane, speed limit, pavement condition, and average motor vehicle traffic. Based on the BSIR, Davis (1995) proposed
the Bicycle Suitability Rating. Besides all the variables in the BSIR, other attributes, i.e. on-street parking, grade, sight distance,
adjacent land use, drainage grate presence, and rough railroad crossing presence, are included in the model. Bicycle Suitability Score
was developed by Turner et al. (1997) as a bicycle suitability criterion for evaluating state roadways in Texas. It evaluates the
roadway bike suitability with a rating method (−8 to 8) based on attributes of adjacent motor vehicle traffic volume, road shoulder
width, posted speed limit, and pavement condition. Another effort dedicated by Noël et al. (2003) to modeling bicycle compatibility
and safety index produced the Compatibility of Roads for Cyclists (CRC) for routes in rural and urban fringe areas. The CRC method is
computed with a rating system (0–100 scale) comprising car speed, car traffic flow, heavy truck flow, roadsides, road profile, width of
the residual space, shoulder pavement, and so forth. Guided by the bikeability index presented by Winters et al. (2013), a Bike Score
was developed to evaluate neighborhood bikeability on a scale from 0 to 100 (Winters et al., 2016). Bike Score is based on en-
vironmental characteristics consistently associated with cycling: density and quality of cycling infrastructure, topography, desirable
amenities, and road connectivity. After an investigation of 24 U.S. and Canadian cities, Winters et al.’s (2016) found higher bike mode
shares in the census tracts with greater bike scores. Those bicycle suitability measures are developed with data collected at city,
region, or state level, and take various factors into account. However, they do not distinguish different types of bikeways, e.g.
protected bike lanes vs. bike lanes.

2.3. Multi-criteria decision analysis (MCDA)

2.3.1. General process of MCDA


The MCDA uses computational and mathematical techniques to facilitate a selection from a host of alternatives involving multiple
objectives and criteria (Marttunen et al., 2017; Triantaphyllou, 2000; Zavadskas et al., 2014). This is achieved on the basis of the
impacts of the alternatives on weighted criteria. The MCDA has been widely applied to assist decision-making in research and

Table 2
Mekuria et al.’s (2012) LTS criteria for bike lanes.
Items LTS ≥ 1 LTS ≥ 2 LTS ≥ 3 LTS ≥ 4

Bike lanes alongside a parking lane


Street width (through lanes per direction) 1 (no effect) 2 or more (no effect)
Sum of bike lane and parking lane width (includes 15 feet or more 14 or 14.5 feet* 13.5 feet or less (no effect)
marked buffer and paved gutter)
Speed limit or prevailing speed 25 mph or less 30 mph 35 mph 40+ mph
Bike lane blockage rare (no effect) frequent (no effect)

Bike lanes not alongside a parking lane


Street width (through lanes per direction) 1 2, if directions are separated by a more than 2, or 2 without a (no effect)
raised median separating median
Sum of bike lane and parking lane 6 feet or more 5.5 or less (no effect) (no effect)
Speed limit or prevailing speed 30 mph or less (no effect) 35 mph 40+ mph
Bike lane blockage rare (no effect) frequent (no effect)

* If speed limit < 25 mph or roadside land use is residential, then any width is acceptable for LTS 2.

55
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Fig. 1. Decision matrix.

practice such as energy, environment and sustainability assessment (Cinelli et al., 2014; Huang et al., 2011), infrastructure man-
agement (Kabir et al., 2014), supply chain management (Govindan et al., 2015; Sanayei et al., 2008), and transportation project
evaluation (Macharis and Bernardini, 2015; Sirikijpanichkul et al., 2017). The general process of MCDA includes (Triantaphyllou,
2000):

(1) Determining the relevant criteria and alternatives.


(2) Attaching numerical measures to the relative importance of the criteria and to the impacts of the alternatives on these criteria.
(3) Calculating the numerical values to determine the ranking of each alternative.

It's assumed that Ai , i = 1, 2, …, m is the set of alternatives, Cj , j = 1, 2, …, n is the set of criteria, Wj is the weight of the j -th
criterion, and x ij is the performance (or measure of importance) of i -th alternative against the j -th criterion. The Pi , i = 1, 2, …, m
denotes the final preference score of alternative Ai . In most MCDA studies, all weighting coefficients satisfy that 0 ≤ wj ≤ 1 ( j = 1, 2,
n
…, n ) and ∑ j = 1 wj = 1. The weight assessment involves the participation of stakeholders. The performances of an alternative against
all criteria ( x ij , j = 1, 2, …, n ) are aggregated into a final preference score. The most used form of the final score is the sum of all the
n
weighted value of individual criterion performance, that is Pi = ∑ j = 1 wj x ij (i = 1, 2, …, m ) (Josselin and Le Maux, 2017). In that
MCDA process, a decision matrix (see Fig. 1) is obtained to rank all alternatives and help the decision making.

2.3.2. Decision making techniques


MCDA methods cover a wide range of distinct techniques. The commonly used MCDA techniques and their strengths and
weaknesses are reviewed and summarized below.
Multi Attribute Utility Theory (MAUT) is mathematically expressed as multiple attributes utility functions where the criteria are
completely aggregated in a single utility or preference score. The simplest MCDA form is the weighted sum model, developed by
Fishburn (1967) based on the additive utility theory. The MAUT is capable of dealing with both deterministic and stochastic decisions
(Zionts, 1992). AHP was developed by Saaty (1990) based on the pairwise comparison to determine the criteria weights (Saaty,
1990). It belongs to the MAUT variants and is the mostly used MCDA method applied in transportation (Macharis and Bernardini,
2015). Interdependence between criteria and alternatives can lead to inconsistencies between judgment and ranking criteria. To
ensure the consistency between judgment matrix and criteria priority, consistency indexes and adjustment method have been de-
veloped to measure the consistency of the decision makers (Aguarón and Moreno-Jiménez, 2003).
Fuzzy Set Theory is an extension of classical set theory that can deal the imprecise inputs and uncertain data and allows for a few
rules to encompass problems with great complexity (Balmat et al., 2011). But it is difficult in cases of no sufficient data available to
develop fuzzy systems, and numerous simulations are required before use (Velasquez and Hester, 2013).
Outranking methods provide an outranking relation on the set of alternatives instead of a cardinal measure of the preference
relations or aggregated utility (Belton and Stewart, 2002; Kangas et al., 2001). Pairwise comparisons of the alternatives concerning a
criterion can be included directly in the evaluation matrix (Macharis et al., 2012). The Preference Ranking Organization Method for
Enrichment Evaluations (PROMETHEE) (Brans and De Smet, 2016) and Elimination et choice translating reality (ELECTRE) (Figueira
et al., 2016) are two mostly used outranking approaches. The outranking methods allow incomparability between alternatives (Wang
et al., 2009). As pointed out by Velasquez and Hester (2013), the PROMETHEE does not provide a clear method of assigning criteria
weights, and the ELECTRE does not identify the advantages and disadvantages of the alternatives directly.
Technique for Order Preference by Similarity to Ideal Solutions (TOPSIS) method proposed by Hwang and Yoon (1981) looks for
alternatives with the shortest distance from the most ideal option and farthest distance from the least ideal solution. Preferences of
alternatives are calculated based on Euclidean distance but without the relative importance of these distances considered (Opricovic
and Tzeng, 2004). Vise Kriterijumska Optimizacija I Kompromisno Resenje (VIKOR) computes the multicriteria ranking index based
on the particular measure of proximity to the ideal solution. The VIKOR method is an effective tool in situations where the decision
maker is not able to express his/her preference at the beginning of system design (Opricovic and Tzeng, 2007). The VIKOR result
depends on the ideal solution determined by a given set of alternatives. Inclusion or exclusion of an alternative could affect the
VIKOR ranking of the new set of alternatives (Opricovic and Tzeng, 2007).
Multi-Actor Multi-Criteria Analysis (MAMCA), developed by Haezendonck (2008) as an extension of MCDA, explicitly in-
corporates the objectives of different stakeholders in a decision making process. In the MAMCA methodology, the criteria for the

56
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

evaluation are the goals and objectives of the stakeholders rather than impacts of alternatives. These impacts can be reflected in the
objectives if all relevant stakeholders are included (Macharis et al., 2012). However, this more participatory process also creates the
risk of bias (Macharis and Nijkamp, 2011). Additionally, the asymmetry in the number of criteria over stakeholders might generate
some extreme and biased results (Macharis and Nijkamp, 2013). For instance, if a stakeholder has only one criterion, the point of
view from this stakeholder weighs more in the final decision.
A Multi-Objective Optimization (MOO) is a pragmatic programming method that attempts to find the best solution from an
infinite number of alternatives to optimize multiple objectives given constraints (Von Neumann and Morgenstern, 2007). In MOO,
typically, there isn’t a feasible solution that maximizes and/or minimizes all objective functions simultaneously. Therefore, attentions
are paid to the Pareto front consisting of a set of optimal solutions that cannot be improved in any of the objectives without degrading
at least one of the other objectives (Koopmans, 1951). Goal Programming (GP), first developed by Charnes and Cooper (1959), is a
branch of MOO and provides a compromise method for solving MOO. In solving GP problems, objectives are converted into a single
objective with a weighted sum method (Deb, 1999). GP is capable of handling large-scale problems with infinite alternatives
(Velasquez and Hester, 2013). However, it’s incapable of weighting criteria and has to be combined with other MCDA methods such
as AHP (Schniederjans and Wilson, 1991).
None of those MCDA techniques is absolutely superior than others. In applications, selection of a suitable method depends on the
context. Given the diversity of MCDA methods, Greene et al. (2011) developed a decision tree of MCDA methods accounting for a list
of model selection factors, including number of decision makers, decision phase, number of objectives, number of alternatives,
existence of constraints, risk tolerance, uncertainty, measurement scales and units, experience, computational resource capacity, and
direction of problem solving. It’s necessary to compare merits and demerits of MCDA methods to determine which one is most
suitable and applicable.

2.3.3. Applications of MCDA in bicycle-related planning


The MCDA techniques have been applied in many studies of bike network planning, bike sharing system design, and bike routing.
Using GIS and MCDA, Rybarczyk and Wu (2010) integrated supply and demand criteria for identifying mismatch between supply and
demand, and planning bike facilities infrastructures. Guerreiro et al. (2018) developed a method to design and compare bike network
based on data-mining of disaggregated trip OD data, GIS, and MCDA to find the routes well covering bike trip generators. Saplıoğlu
and Aydın (2018) implemented an AHP method to decide effective parameters on safe and serviceable bicycle routes with survey
data. With the AHP results, the suitability of bicycle routes is evaluated to establish safe and comfortable bicycle-bus integration
routes. Kabak et al. (2018) proposed GIS-multi-criteria decision-making model for selecting suitable locations for bike-share stations.
In this model, the AHP is implemented to determining the weight coefficients of the evaluation criteria, and a MOO based on ratio
analysis is used to rank the alternative bike-share stations. Terh and Cao (2018) proposed a GIS-MCDA based route planning support
framework to calculate suitable bike routes based on preferences from the public, transportation experts, and planners. The MOO
techniques have been popular in recent years’ bike network design. For instance, Lin and Yu (2013) developed a bike network design
model for bicycling in built-up urban areas to minimize cyclist risk, maximize cyclist comfort, maximize service coverage for re-
sidents, and minimize the impact of the bikeway on existing traffic. Mesbah et al. (2012) proposed a bi-level optimization bike
network design to encompass the benefits of cyclists and car users at the upper level and to assign car and bike demand at the lower
level. The objective function was defined by a weighted sum of the total travel time by car versus total travel distance on bike lanes.
Duthie and Unnikrishnan (2014) presented a formulation for the bike network design as it relates to retrofitting existing roadway
infrastructure for bicycles. The goal of the problem is to connect all ODs with bicycling paths at a minimum cost.

2.4. Research gap and contribution of the paper

To fully take advantage of the mobility benefits of bicycling in multimodal transportation systems, it is a critical need to in-
corporate the bicycling connectivity of ODs and bicycling access to public transit into the planning and prioritization of new bike-
ways. Meanwhile, with the increasing popularity of bike-sharing programs, it’s necessary to include the bicycling connectivity for
bike sharing stations in that network planning process. On the other hand, as existing transportation systems are historically de-
signated for automobiles in most U.S. cities, no spare space was preserved for adding new bikeways without widening the roadway.
Under such circumstances, road diet is a more feasible solution, and vehicle delays should be examined after reallocating the right-of-
way. Those lack comprehensive consideration in previous bike network planning.
In light of the review on preceding work and the discussions above, this paper contributes to developing a bikeways prioritization
method based on the MCDA to improve low-stress bike network connectivity and low-stress bicycle-transit connection in built-up
environments, so as to improve the bike sharing condition. When evaluating the impacts, the automobile users are involved in the
decision process to resolve the conflicts between automobile users and bicyclists. In the approach, the difference in bike suitability
and bicycling connectivity between bike lanes and protected bike lanes will be examined with the LTS approach.
The decision-making process of bikeway prioritization is concerned with one objective that is rating a finite number of pre-
determined bikeway projects. Relevant criteria or factors should be quantitatively measurable. To calculate final preferences, the
criteria and weighting should be quantified with certainty, and aggregated weighting method reflecting criteria trade-offs is required
to derive alternatives’ ratings. With MCDA method selection process developed by Greene et al. (2011), the AHP is applied in the
paper to generate criteria weights though pair-wise comparisons, and utility function based on MAUT is used to calculate final
preferences of alternatives.

57
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Table 3
Selection of criteria.
Stakeholders Dimension Criteria

Bicyclists Low-stress bicycling connectivity (D1) C1: Fraction of trips between ODs connected by bicycling
C2: Fraction of bike-sharing stations connected to all possible destinations by bicycling
Transit users Low-stress bicycling access to public transit (D2) C3: Fraction of population with access to transit by bicycling
C4: Fraction of bike-sharing stations connected to transit by bicycling
Car drivers Automobile traffic mobility (D3) C5: Automobile traffic delay
Investors Cost (D4) C6: Capital cost of implementing bikeways

3. Method

3.1. Defining criteria

Before planning and implementing new bikeways, it’s important to involve stakeholders who could be affected to evaluate
impacts of the planned bike facilities from different perspectives. Stakeholders that may be involved in the decision-making process
include bicyclists, transit users, motor vehicle drivers, and investors. For bicyclists, they desire better connectivity of low-stress bike
paths between OD pairs; for transit users, they prefer improved optional bicycle access to transit service; for motor vehicle drivers,
they wish less congestion and delay; for investors, they are concerned with the benefits over costs of deploying new bike facilities.
Regarding these considerations, four dimensions of the bikeway performance evaluation are identified in the paper. They are the low-
stress bicycling connectivity, low-stress bicycling access to public transit, automobile traffic mobility, and cost.
When measuring the bike network connectivity, number of trips or nodes connected by bicycling is commonly applied (Mesbah
et al., 2012). In the calculation of the access to public transit, the population that can reach transit service by bicycling is estimated
(El-Geneidy et al., 2014; Lee et al., 2016; Zuo et al., 2018). The low-stress bicycling connectivity and low-stress bicycling access to
public transit for bike-sharing stations are also considered. Number of transit stops with bike-sharing access is used as the perfor-
mance measure indicating the connectivity between transit and bike-sharing stations (Griffin and Sener, 2016). The automobile
traffic delay is one of the most commonly used indexes (Schrank et al., 2015), particular in examining roadway capacity on traffic
mobility at the roadway segment-level (Chin et al., 2004; Karim and Adeli, 2003). It reflects the total traffic and extra travel time
spent on a link. The capital cost including all costs (e.g. construction, labor, and purchase of equipment and materials) associated
with implementing new bikeways is used.
The evaluation dimensions and bikeway prioritization criteria versus stakeholders are summarized in Table 3. It is suggested that
the decision makers consider trade-offs between criteria when rating preferences of bikeway alternatives.

3.2. Developing numerical performance measures of criteria

3.2.1. Low-stress bicycling connectivity


In measuring the low-stress bicycling connectivity, the stress level of any route is determined by the most stressful link of the
entire route (Furth et al., 2016). To check if an OD pair (o and d ) is connected at LTS k , two criteria are used: (1) there is a route
connected from o to d with links whose LTS ≤ k and (2) the total length of the route is no greater than the bicycling distance
threshold, as expressed in Eq. (3). The 85th percentile bicycling distance is used as the distance threshold.

Lod, k ≤ Lmax (3)

where Lod, k is the path between od with links at LTS ≤ k , and Lmax is the bicycling distance threshold.
Given the daily volume of trips between ODs, a summary measure of bike network to provide connectivity to all trips is the
fraction of trips in the trip table whose ODs are connected at a given LTS value, which is mathematically expressed as

1, if Lod, k ≤ Lmax
x i1 = 100% × ∑o,d fod δod,k / ∑o,d fod , and δod, k = ⎧
⎩ 0, if Lod, k > Lmax
⎨ (4)

where x i1 is the fraction of trips connected by bicycling at LTS ≤ k , δod, k is the indicator whether origin o is connected to destination d
at LTS ≤ k , and fod is the number of trips between od .
For the bike-sharing stations, bicycling connectivity is measured by the fraction of connected bike-sharing stations to destinations
by bicycling and represented by

1, if Lbd, k ≤ Lmax
x i2 = 100% × ∑b,d δbd,k /B/D, and δbd, k = ⎧
⎩ 0, if Lbd, k > Lmax
⎨ (5)

where x i2 is the fraction of bike-sharing stations connected to destinations at LTS ≤ k , δbd, k is the indicator whether bike-sharing
station b is connected to destination d at LTS ≤ k , B is the total number of bike-sharing stations in the area, D is the total number of
destinations, and Lbd, k is the distance between bike-sharing station b and destination d .

58
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

3.2.2. Low-stress bicycling access to public transit


The bicycling access to public transit can be measured as percentage of population in all origins connected to transit lines by low-
stress bicycling paths at a given LTS value. This spatial boundary of transit service coverage is determined by the distance threshold
and network distance between an origin and a transit stop. Mathematically, it’s expressed as Eq. (6).

1, if Losl, k ≤ Lacc, max


x i3 = 100% × ∑ ∑ Po δosl,k /L/ ∑ Po, and δosl, k = ⎧
l, s o o ⎩ 0, if Losl, k > Lacc, max
⎨ (6)

where x i3 is the fraction of population with access to transit by bicycling at LTS ≤ k , Po is the number of population in the origin o ,
δosl, k is the indicator of whether o is connected to nearest stop s of transit line l at LTS ≤ k , L is the total number of transit lines, Losl, k is
the network distance computed using the street network from o to reach the nearest stop s of transit line l at LTS ≤ k , and Lacc, max is
the distance threshold to access transit by bicycling. If Losl, k is no greater than the distance threshold, then δosn, k = 1 and the po-
pulation Po in o have access to transit line l .
Bike share provides a feeder service to make the first-and-last mile connection to public transit in an much convincement manner
with no need to own a bike. Convenient and connected bike paths between bike sharing stations and transit stops make it possible for
more people to use shared bikes to reach transit services. The bicycling access to transit for bike share is measured as

δbsl, k = 1, Lbsn, k ≤ Lacc, max


x i 4 = 100% × ∑ ∑ δbsl,k /B/L, and ⎧
l, s b ⎩ δbsl, k = 0, Lbsn, k > Lacc, max
⎨ (7)

where x i4 is the fraction of bike-sharing stations connected to transit by bicycling at LTS ≤ k , δbsl, k indicates whether bike-sharing
station b is connected to nearest transit stop s of transit line l at LTS ≤ k , and Lbsl, k is the distance between b to nearest stop s of transit
line l at LTS ≤ k .

3.2.3. Automobile traffic delays


In a built-up urban environment, designing on-street bikeways through road diets is more feasible than roadway widening. Road
dieting can be accomplished by either narrowing automobile lanes or eliminating an automobile travel lane. To minimize the ne-
gative impact on automobile traffic, narrowing down lanes is more preferred. The condition for that the existing roadway can provide
enough space without leaving a motor lane for a new bike lane is set forth as:
na Wa ≥ na Wmin + Wb (8)

where na is the number of automobile lanes, Wmin is the minimum automobile lane width, Wa is the average lane width of link a , and
Wb is the bikeway width. If there isn’t enough space to add a bike lane and keep the same number of automobile lanes, an automobile
travel lane will be eliminated. This condition is represented by Eq. (9).
na Wa ≥ (na − 1) Wmin + Wb (9)

On-street bikeways discussed in the paper include traditional bike lanes (i.e., bike lanes here) and protected bike lanes. The
spacing requirements of these two types of on-street bikeways are summarized in Table 4 (NACTO, 2011). New bikeways should be
planned with careful consideration of automobile traffic capacity concerns. The American Association of State Highway and
Transportation Officials (AASHTO)’s Green Book (AASHTO, 2011) suggests that 12-foot lanes should be used where practical on
higher speed, free flowing, and principal arterials; under interrupted-flow at speed of 45 mph or less, narrower lane widths are
normally quite adequate. Due to the narrowed lane width, the roadway capacity might be reduced. According to HCM
(Transportation Research Board, 2010), there is no observed capacity reduction in roadways until the lane width falls below 10 feet
and 10% drop in capacity of streets with 9 feet.
For automobile traffic, the travel time on each link is estimated by the Bureau of Public Roads (BPR) function, as shown by Eq.
(10). The traffic delay (del ) is calculated as the additional travel time for all vehicles in Eq. (11). If the capacity is reduced, traffic
delay could be increased.

tac = tac,0 (1 + α (uac / Ca) β) (10)

Table 4
LTS and spacing for traditional and protected bike lanes.
Bikeway type LTS Designate spacing requirements

Traditional bike lane 1–4 • When adjacent to a parking lane, the desirable reach from the curb face to the edge of the bike lane is 14.5 feet; the absolute
minimum is 12 feet.
• minimum
A bike lane next to a parking lane shall be at least 5 feet wide, unless there is a marked buffer between them. The desired
is 6 feet.
• The desirable ridable surface adjacent to a street edge is 4 feet, with a minimum width of 3 feet.
Protected bike lane 1 • The desired width for a protected bike lane should be 5 feet and 7 feet for areas with high bike volumes or uphill.
• Three feet is the desired width for a parking buffer.
• Where adjacent to a parking lane, the desired combined width of parking lane and buffer is 11 feet.
59
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

x i5 = ∑a uac (tac − tac,0) (11)


where tac
is the automobile travel time on link a, tac,0
is the free-flow travel time on link a, α and β are coefficients, uac is the automobile
traffic volume on link a, and Ca is the roadway capacity on link a.

3.2.4. Capital cost


The capital cost of implementing bikeways on streets is calculated as

x i6 = ∑a δa la c (12)
where δa is the indicator if a bikeway will be implemented on link a (0 = no, 1 = yes), la is the length of link a , and c is the average
unit cost of implementing bikeways per mile. The capital cost depends on length, type of bikeway, site conditions, and so forth.

3.3. Determining weights of criteria

Pair-wise comparison is applied to evaluate the relative importance of criteria. In the pair-wise comparisons, decision makers
need to compare “how important is criterion j relative to criterion j' ?” There are several scale techniques for the comparison, such as
Saaty’s (1980) linear scale, Lootsma’s (1989) geometrical scale, Salo and Hämäläinen’s (1997) balanced scale, Ma and Zheng’s (1991)
inverse linear scale, and Ishizaka et al.’s (2011) logarithmic scale. Based on the performance of commonly used scale methods
regarding the consistency of judgment and criteria priority and variance of allocation of priorities’ values discussed by Franek and
Kresta (2014), Saaty’s scale is used. In Saaty’s (1980) linear scale, the preference is expressed on an ordinal scale: 1 = equally
important, 3 = moderately more important, 5 = strongly more important, 7 = very strongly more important, 9 = absolute more
important, and 2, 4, 6, 8 = intermediate values. Thereby, a judgment matrix Y is created with the comparison result yjj' between
criteria j and j' ( j , j' = 1, 2, …, n , and yjj' = 1/ y j' j ), illustrated as (see Fig. 2)
From the judgment matrix, a criteria priority vector W = (w1, w2, ⋯, wn ) indicating the priority or weight of each criterion is
calculated. Commonly, the eigenvalue method (EVM) (Saaty, 1980) and the row geometric mean method (RGMM) (Williams and
Crawford, 1980) are used to scale the judgment matrix. It’s suggested that those two methods are numerically quite close (Budescu
et al., 1986; Crawford, 1987; Dong et al., 2008); however, the latter is less computational complex. The RGMM is applied to scale
n
weights of criteria in the study. Denoting πj as the geometric mean of {yjj' , j' = 1, 2, ⋯n} , that is πj =( ∏ j' = 1 yjj' )1/ n . The weight of
criterion j is calculated as
πj
wj = n
∑ j = 1 πj (13)
The consistency of a pairwise comparison judgment matrixY is defined as the cardinal transitivity between the judgments, that is,
yjj' = yjk ykj' ( j , j' , k = 1, 2, …, n ) (Saaty, 1980). To provide a consistency measure related to EVM, Saaty (1980) proposed the use of the
Consistency Ratio (CR). The lower the value of CR, the better the consistency of the judgments. A 10% threshold has been suggested
for the CR for n > 4 (5% and 8% for the n = 3 and n = 4, respectively) to accept the estimation of w (Saaty, 1994). For RGMM,
Crawford and Williams (1985) proposed a Geometric Consistency Index (GCI) as a measure of the consistency, represented by

GCI = 2(n − 1)(n − 2) ∑j < j' log2 (yjj' w j'/wj) (14)


GCIs corresponding to the values of the CR are: GCI = 0.31 for n = 3, GCI = 0.35 for n = 4, and GCI = 0.37 for n > 4 (Aguarón
and Moreno-Jiménez, 2003). When the value of the GCI exceeds its threshold, the most inconsistent judgment that has the greatest
difference between yjj' and wj / w j' needs to be modified, and a new priority vector will then be calculated.

3.4. Calculating preference scores

The MCDA usually requires determined values of all x ij in a comparable context. If criteria are not expressed in similar terms,
transformation of measurements so that each individual performance value is comparable is required. A partial value function by
unity-based normalization based on MAUT is applied to rescale performance values, expressed as Eq. (15) (Josselin and Le Maux,
2017). Performance measurements of alternatives with regard to each criterion are normalized between 0 and 1.

Fig. 2. Judgment matrix.

60
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Fig. 3. Study area.

x −x
ij min, j
⎧ xmax , j − xmin, j If a higher value indicates better performance
v (x ij ) =
⎨ xmax , j − xij If a higher value indicates worse performance
⎩ xmax , j − xmin, j (15)
where x max , j and x min, j are the maximum and minimum values of {x ij , i = 1, 2, ⋯, m} regarding criterion j , respectively. The minimum
and maximum reference points are determined with a local scaling method based on the estimated performance scores of alternatives
(Josselin and Le Maux, 2017). The final preference score Pi of alternative i is then aggregated by
n n
Pi = ∑ wj v (xij) and ∑ wj = 1
j=1 j=1 (16)

4. Study context and data

4.1. Study area

In the case study presented, the developed approach is applied in the Uptown Cincinnati, Ohio, U.S. with an area of 6.56 mi2
(Fig. 3). According to the 2012–2016 American Community Survey by U.S. Census Bureau, the study area has a population of 35,959
people. There are 25 bus lines and 454 bus stops serving this area. In the case study, census block is chosen as the geographic analysis
unit, and the study area is divided into 445 census blocks. No on-street bikeway is available. In the area, there are 10 Cincinnati Red
Bike kiosk stations providing shared bikes.

4.2. Empirical travel distance of bicycling trips and transit access trips by bicycling

In the paper, the 2009–2010 Cincinnati GPS-based Household Travel Survey (HTS) is adopted to examine the empirical travel
behaviors. This survey was conducted by the Ohio Department of Transportation Research Division in cooperation with the Ohio-
Kentucky-Indiana (OKI) Council of Governments. In the survey, three-day’s personal trip diaries from recruited households on a
second-by-second basis were collected, and 77,209 trips were recorded (Wargelin et al., 2012). In those trip records, 1330 trips are
transit access trips, of which only 16% are finished by bicycling, and 553 trips are bicycling trips connecting ODs directly. With the
travel distance data from the GPS-based HTS, the cumulative curves of travel distance of bicycling trips connecting ODs directly and
transit access trips by bicycling are plotted in Fig. 4. In the figure, 85th percentile travel distances are identified. The 85th percentile
distance for bicycling trips connecting ODs directly (Lmax ) is 2.39 miles, and the 85th percentile bicycling distance to access transit
(Lacc, max ) is 1.59 miles. The values of Lmax and Lacc, max are used as distance thresholds to identify if two points are connected in Eqs.
(3)–(7).

4.3. Bike network planning strategies and relative importation of criteria

Criteria weights should be determined in a manner of reflecting the relative importance of the criteria and abiding with the

61
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Fig. 4. Travel distance of bicycling trips and transit access trips by bicycling.

decision makers’ preferences under planning goals. In the case study, four development strategies are formulated to direct the
evaluation of bike improvement plans considering trade-offs between criteria. The development strategies are described below.

Strategy I: Balancing benefits and costs between all stakeholders with equal importance of all criteria. It’s assumed that all criteria
are of equal importance.
Strategy II: Improving bicycling connectivity and bicycle-transit integration. It’s focused on low-stress bicycling connectivity (D1)
and low-stress bicycling access to transit (D2 ) simultaneously. Less attention is paid on automobile traffic delays (D3 ) and capital
costs (D4 ). In this case, criteria under the D1 and D2 are given equal importance; criteria under the D3 and D4 are equally
important; criteria under the D1 and D2 are absolute more important than criteria under the D3 and D4 .
Strategy III: Primary goal is to improve bicycling paths connecting more ODs, and secondary goal is to increase bicycle-transit
connection. In this scenario, criteria under the D1 are more important than criteria under D2 and absolute more important than
criteria under the D3 and D4 ; D2 is weakly more important than criteria under the D3 and D4 ; criteria under the same dimension are
equally important.
Strategy IV: Primary goal is to increase bicycle-transit connection, and the secondary goal is to improve bicycling connectivity
between ODs. In this scenario, it’s assumed that criteria under the D2 are more important than criteria from D1 and absolute more
important than criteria under the D3 and D4 ; D1 is weakly more important than criteria under the D3 and D4 ; criteria under the
same dimension are equally important.

Directed by each assumed development strategy, the corresponding criteria importance judgment matrix that satisfies the con-
sistency criteria is determined following procedures introduced in Section 3.3. The cell value in a judgment matrix follows the
principle that yjj' = 1/ y j' j . With the calculated judgment matrix Y , the criteria priority vector {wj, j = 1, 2, ⋯, n} is calculated with Eq.
(13). The associated consistency index GCI is measured with Eq. (14), and its value should not exceed 0.37 (Aguarón and Moreno-
Jiménez, 2003). The resulting judgment matrices, criteria priority vectors, and consistency indexes are enlisted in Table 5.

4.4. Bike network improvement plans

On the basis of the local bike improvement plans proposed by the City of Cincinnati (Toole Design Group, 2010), locations of new
bikeways in the study area are split into 10 project plans and illustrated in the Fig. 5. Details of the length, bikeway type, and
locations of bike plans are given as follows:

• Bike plan 1 (BL_1): 3.21-mile bike lanes on Clifton AV;


• Bike plan 2 (BL_2): 2.70-mile bike lanes on Burnet AV;
• Bike plan 3 (BL_3): 3.48-mile bike lanes on Reading RD;
• Bike plan 4 (BL_4): 5.04-mile bike lanes on Gilbert AV;
• Bike plan 5 (BL_5): 3.77-mile bike lanes on Victory PKWY;
• Bike plan 6 (BL_6): 3.12-mile bike lanes on Woodburn AV;
• Bike plan 7 (BL_7): 5.61-mile bike lanes on Woolper AV-Forest AV-Rockdale AV;
• Bike plan 8 (BL_8): 11.45-mile bike lanes on Madison RD-Martin Luther King JR;
• Bike plan 9 (BL_9): 6.12-mile bike lanes on Calhoun ST-William H Taft; and
• Bike plan 10 (BL_10): 5.87-mile bike lanes on Ludlow AV-Nixon ST-Vine ST-Jefferson AV.
Other information including street width, parking lane availability and width, speed limit, bike lane blockage on proposed
locations are listed in Table 6.
In those bike improvement plans, only bike lanes are proposed. According to the LTS definition, bike lanes can’t guarantee a
desired low LTS (i.e., LTS ≤ 2) on streets with large traffic volume and high traffic speed. Compared with bike lanes, protected bike

62
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Table 5
Judgment matrix, criteria priority vector, and consistency index by bike network development strategy.
Strategy Judgment Matrix (Y ) Criteria Priority Vector (W ) Consistency Index

I [0.167, 0.167, 0.167, 0.167, 0.167, 0.167] GCI = 0

II [0.237, 0.237, 0.237, 0.237, 0.026, 0.026] GCI = 0

III [0.380, 0.380, 0.079, 0.079, 0.041, 0.041] GCI = 0.11

IV [0.079, 0.079, 0.380, 0.380, 0.041, 0.041] GCI = 0.11

Fig. 5. Locations of proposed new on-street bike facilities in the study area.

lanes are more effective on reducing LTS as they provide separated and exclusive space from automobile traffic. However, one
limitation of protected bike lanes is that they require more space than bike lanes (see Table 4 for details), which may cause a capacity
drop of automobile traffic and cause more traffic delay. Besides that, protect bike lanes usually produce higher costs than bike lanes.
To investigate the benefits, costs, and the negative effect on automobile traffic of protected bike lanes, the paper also evaluates the
performance if using protected bike lanes instead of bike lanes on the proposed streets. The bike improvement planes with protected
bike lanes on those locations are denoted as PBL_1, PBL_2, …, PBL_10, respectively.

5. Case study results and discussion

5.1. Bicycling stress level before and after implementing new bike lanes

With the LTS criteria, the stress level of the existing bicycling network is assessed (see Fig. 6). In the study area, there are 22.95%

63
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Table 6
Street width, parking availability, speed limit, and bike lane blockage on streets with bike lanes planned.
Bike plan Street Width (feet) Speed Limit (mph) Parking Availability Bike Lane Blockage

BL_1 30–48 30–35 Yes No


BL_2 30–50 30 Yes No
BL_3 30–72 25–35 Yes No
BL_4 28–50 25–35 Yes Yes
BL_5 30–50 25–35 No No
BL_6 28–50 25–35 Yes Yes
BL_7 28–50 25–35 Yes No
BL_8 28–50 25–45 No No
BL_9 28–50 30–35 Yes Yes, on Calhoun ST
BL_10 28–48 25–35 Yes Yes, on Ludlow AV

Fig. 6. Existing bicycling LTS in the study area.

of streets with an LTS = 1 and 39.58% with an LTS = 2. Those are low-speed and fewer-lane local streets and are mostly located in
residential areas. There are 12.02% and 19.63% of streets with an LTS = 3 and LTS = 4, respectively. Those streets have a speed no
less than 30 mph, and many of them are in non-residential areas such as commercial and office areas. For the rest (5.83%) are
freeways and not accessible by bicyclists. According to the LTS definition, it’s suitable for most travelers. Hence, streets with LTS ≥ 3
that are non-suitable for bicyclists will be removed from the bicycling network. The removed links result in fragmented and dis-
connected bicycling paths.
With all the planned new bike lanes, percentages of streets with an LTS = 1 and an LTS = 2 are expected to increase to 24.67%
and 44.69%, respectively. As streets at an LTS ≤ 2 are suitable for bicycling, 69.36% of the streets can be used for the bike network,
and there is an increase of 6.83% (1.72% for LTS = 1, and 5.11% for LTS = 2) compared with the figure of the existing condition. A
detailed comparison of LTS on streets where new bikeways are planned before and after are presented in Fig. 7. Most of those existing
streets have a high bicycling stress level (LTS > 2) and not suitable for most population to bike on them. After implementing new
bike lanes, the LTS is reduced for most streets. In the bike network improvement plans, streets such as in the BL_2 and BL_4 receive
significant improvements in LTS after installing new bike lanes. However, as bike lanes are designated by striping, signage, and
pavement markings and provide no separation from pass automobile traffic for bicyclists, riding a bike adjacent to high-speed and/or
large-volume motor vehicle traffic still causes a high stress level to bicyclists, particularly on streets with the motor traffic speed limit
≥35 mph such as in BL_5. In such cases, those bike lanes are still not considered as suitable for bicycling and contribute little to
improving low-stress bike network connectivity and bicycle-transit connection.

5.2. Final preference scores of bike improvement plans

The bicycling connectivity and bicycling access to public transit are calculated for before (the existing condition) and after the
installation of new bike lanes with Eqs. (3)–(7). The outcomes are enlisted in Table 7. According to Fig. 7, the implementation of bike
lanes does not reduce bicycling stress level significantly on proposed streets, and only a small percent of the streets can achieve a low-
stress level LTS = 1. As a result, improvements in bicycling connectivity and bicycling access to public transit are relatively small,

64
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Fig. 7. Improvements in bicycling LTS with proposed new bike lanes.

Table 7
Performance measurements of proposed bike lanes.
Criterion xi1 xi2 xi3 xi4 xi5 (veh·h/day) xi6 (103 U.S. dollars)

Existing condition 58.27% 68.48% 59.75% 68.74% 2377.99 0


Bike improvement planes BL_1 58.29% 68.83% 59.75% 68.74% 2384.96 107.62
BL_2 58.54% 68.98% 59.80% 68.79% 2402.24 90.53
BL_3 58.73% 68.60% 60.03% 68.85% 2400.49 116.68
BL_4 61.95% 69.48% 60.83% 69.82% 2384.80 168.98
BL_5 58.37% 68.48% 59.75% 68.74% 2378.92 126.40
BL_6 58.87% 68.60% 59.75% 68.74% 2379.84 104.61
BL_7 59.65% 68.98% 59.75% 68.79% 2380.29 188.09
BL_8 58.93% 68.85% 61.32% 70.52% 2391.28 383.90
BL_9 58.34% 68.76% 61.10% 70.03% 2389.39 205.19
BL_10 58.34% 68.48% 59.75% 68.90% 2384.22 196.81

especially at LTS = 1. For the bike lanes, 5 feet is recommended. Without road widening, it’s suggested to implement the new bike
lanes through road dieting by either narrowing automobile lanes or eliminating automobile lanes with the conditions given in Eqs. (8)
and (9). The narrowed lane width or reduced number of lanes would impact the roadway capacity for automobile traffic. Average

65
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Fig. 8. Decision matrices of all bike lanes implementing plans.

ADT counts collected by OKI are used in calculating the automobile traffic delay with Eqs. (10) and (11). A unit cost of $6.35/ft
(Weigand et al., 2013) is used to estimate capital cost of each alternative with Eq. (12). Results of traffic delay and capital cost are
also presented in Table 7.
The estimated performance {x ij , i = 1, 2, ⋯, m , j = 1, 2, ⋯, n} is normalized into {v (x ij ), i = 1, 2, ⋯, m , j = 1, 2, ⋯, n} using the
method introduced in Eq. (15). With criteria weights {wj, j = 1, 2, ⋯, n} given in Table 5, the final preference score of each alternative
is computed by Eq. (16). Decision matrices under the four development strategies are given in Fig. 8. Under the Strategy I, II, and III,
bike plan 4 (BL_4) in which new bike lanes planned on Madison RD outperforms other alternatives. While under Strategy IV that is
primarily focused on improving bicycle-transit connection, the bike plan 8 (BL_8) proposed on Madison RD and Martin Luther King
JR has the highest preference score. The BL_4 ranks the second and BL_9 (on Calhoun ST and William H Taft) ranks the third. The
streets in BL_4, BL_8, and BL_9 are identified as critical corridors for implementing bike lanes.

5.3. Bike lanes vs. protected bike lanes

Based on the LTS criteria, protected bike lanes with physical barriers to separate bicyclists from automobile traffic has the lowest
bicycle stress level (LTS = 1). Improving the high stress-level streets with protected bike lanes seems to be a more promising way to
increase bike network connectivity and improve bicycling access to transit. To measure the advantage of protected bike lanes over
bike lanes, it’s assumed that all bike improvement plans adopt protected bike lanes, and a 1.5-feet width for the physical barrier and a
total width of 6.5 feet is suggested. The unit capital cost of $24.79/ft is applied in the estimation of cost (Weigand et al., 2013).
With protected bike lanes, significant improvements in the network LTS are observed as 40.98% of streets are with an LTS = 1
and 37.31% of streets are with LTS = 2. Compared with the existing condition, 15.76% more of the streets become suitable for
bicycling. It’s noticed that protected bike lanes contribute an increase of 18.03% in the fraction of streets at LTS = 1, but a reduction
of −2.27% in the fraction of streets at LTS = 2 due to the fact that the LTS of some streets being improved from 2 to 1 with protected
bike lanes.
Performance of protected bike lanes (PBL_1–PBL_10) are assessed, and the measurements are presented in Table 8. The results
show some significant improvements in low-stress bicycling connectivity and low-stress bicycling access to transit. For instance, in
PBL_8, the fraction of trips connected by bicycling at LTS ≤ 2 ( x 81) is increased from 58.27% to 85.11%, and the fraction of bike-
sharing stations connected to destinations at LTS ≤ 2 ( x 82 ) rises from 68.48% to 81.00%. The bike improvement plan PBL_6 ex-
periences the largest growth in the fraction of population with access to transit by bicycling at LTS ≤ 2 ( x 63 ) and the fraction of bike-
sharing station with access to transit by bicycling at LTS ≤ 2 ( x 64 ), which are 10.80% and 12.52%, respectively. Comparing between
the figures in Tables 7 and 8, on the same locations, greater improvements in low-stress bicycling connectivity and low-stress
bicycling access to public transit are brought by protected bike lanes than bike lanes. Though protected bike lanes provide better
bicycle connectivity and improve bicycle-transit connection, as the protected bike lanes usually require more space and physical

66
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Table 8
Performance measurements of protected bike lanes.
Criterion xi1 xi2 xi3 xi4 xi5 (veh·h/day) xi6 (103 U.S. dollars)

Existing condition 58.27% 68.48% 59.75% 68.74% 2377.99 0


Bike improvement planes PBL_1 59.27% 68.81% 61.66% 69.07% 2394.52 420.16
PBL_2 61.44% 70.48% 61.53% 70.74% 2414.37 353.41
PBL_3 60.18% 69.62% 60.34% 69.88% 2413.47 455.50
PBL_4 86.19% 70.78% 63.24% 71.04% 2448.31 659.69
PBL_5 66.25% 68.96% 61.54% 69.22% 2381.43 493.46
PBL_6 85.11% 81.00% 70.55% 81.26% 2381.81 408.38
PBL_7 73.61% 70.78% 60.50% 71.04% 2387.80 734.30
PBL_8 85.11% 81.00% 70.55% 81.26% 2434.59 1498.70
PBL_9 65.11% 73.40% 64.32% 73.66% 2408.41 801.05
PBL_10 71.22% 82.00% 66.29% 82.26% 2404.54 768.33

barriers, the congestion might be exacerbated, and higher cost is required than bike lanes. Trade-offs between those criteria need to
be carefully considered in the decision making process.
With performance measurements of all new bike lanes and protected bike lanes, the preference score of each plan can be esti-
mated. Under the four designated development strategies, decision matrices of all bikeway planes are presented in Fig. 9(a)–(d). It’s
noted that on the same locations, protected bike lanes are not always more (or significantly more) preferable than bike lanes if equal
attention paid on the traffic delay and the capital cost as on the bicycling connectivity and bicycle-transit connection, such as BL_1 vs.
PBL_1 and BL_3 vs. PBL_3 in Fig. 9(a). With more importance focused on the benefits of bicyclists and transit users, protected bike
lanes exhibit a more significant advantage over bike lanes, see Fig. 9(b)–(d). Among those plans and bike network development
strategies, Plan 6 with protected bike lanes (PBL_6) on Woodburn AV receives the highest preference score. Other protected bike lanes
proposed on Jefferson AV, Ludlow AV, Nixon ST, and Vine ST in PBL_10 and those on Madison RD and Martin Luther King JR in
PBL_8 also show high preferences regarding the overall performance of all criteria. Those locations are identified as the critical
bicycling corridors that should receive higher priorities in the bike network planning.

6. Conclusion

The paper contributes to developing an approach based on MCDA to prioritize bike improvement plans. In the developed ap-
proach, the trade-offs among criteria are explicitly considered through the pair-wise comparison to determine weights of criteria.
With the growing consensus on the benefits of bicycling, proving a connected bike network is crucial to encourage more people to
bike. Bicycle can be used as a transportation mean connecting ODs directly and providing access to transit in the first-and-last mile.
As the conflicts between the rights-of-way of bicyclists and automobiles, it’s preferable to estimate the possible impacts of im-
plementing bike facilities on traffic congestion if there is no road widening. For many communities, there is a budget limitation.
Besides bicyclist, users of other transportation modes including public transit and automobile and investors are considered. With the
four considered stakeholders, the developed approach is more flexible and applicable to various conditions of land use, traffic, transit
provision, and so forth. For example, in downtown areas with heavy traffic and high construction costs, in addition to low-stress
bicycling connectivity and bicycle-transit connection, the traffic delay and cost should also be set with a high priority. On some local
streets with little concerns of traffic congestion, low priority could be assigned to the criterion of traffic delay.
In addition, the flexibility of the MCDA-based approach and involvement of multimodal transportation users allow the bike
network planning process to coordinate with other urban and transportation planning processes. For example, the developed ap-
proach provides a tool for facilitating bicycle-transit integration and advancing Transit Oriented Development (TOD). The concept of
TOD generally refers to compactness, mixed-use, and pedestrian and bicycle friendly development oriented to promoting transit
(Cervero et al., 2002). TOD involves very well the number of population and jobs within transit service coverage areas. The poorly
connected bicycle network would discourage the bicycling distance to access transit and reduce transit service coverage area. In the
bike network design, it’s imperative to identify the locations on which bikeways are implemented to improve bicycling connectivity
and provide low-stress bicycling paths to transit. Bike share, emerged as one of the latest innovations in urban mobility, provides easy
access to bikes. It permits bikes to be picked up and returned at different docking stations, which facilitates one-way travel. The
marriage between bike sharing programs and public transit is a win-win situation (DeMaio, 2009). In the bike-sharing system design,
the bike network is one of the key determinants (Lin et al., 2013) and affects the service coverage. In the MCDA for new bikeways, the
connectivity between bike sharing stations, destinations and public transit are included in the performance measures. The perfor-
mance measures are associated with the attraction of bike share and could be applied in the design of bike-sharing systems.
The outputs of the proposed approach including a decision matrix and preference scores of alternatives assist decision makers in
identifying the critical corridors and selecting the most valuable locations that maximize the benefits in improving bicycle network
connectivity and/or bicycling access to transit and minimize traffic delay and budget from all alternatives. In addition, priorities of all
bike improvement alternatives can be determined with calculated preference scores. This provides a reference that allows decision
makers to make an informative city-wide strategic bike network planning and determine the implementation phases. In the case
study, the bikeways with higher performances could be considered to be included in the short-term implementation phase.
The biggest limitation of the study lies in that impacts of new bike facilities on mode choices of bicycling and bicycle-transit are

67
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Fig. 9. Decision matrices of bike lanes and protected bike lanes implementing plans.

not investigated. Increases in the modal share of bike and bike-n-ride should reduce automobile traffic and bring a positive effect on
the delay reduction. This consideration will be addressed in the future work. In the future study, we plan to examine the relationship
between the choice of bicycling or bicycle-transit with the presence of bikeways by developing behavior models describing how
bikeways affect choices on bicycling and bike-and-ride.

Appendix A. Supplementary material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.tra.2019.08.003.

References

AASHTO, 2011. A Policy on Geometric Design of HighWays and Streets, 6th ed. American Association of State Highway and Transportation Officials, Washington, DC.
Aguarón, J., Moreno-Jiménez, J.M., 2003. The geometric consistency index: approximated thresholds. Eur. J. Oper. Res. 147, 137–145. https://doi.org/10.1016/
S0377-2217(02)00255-2.
Ali, A.T., Cristei, C.M., Flannery, A., 2012. Using cumulative logistic regression model for evaluating bicycle facilities on urban arterials. In: TRB 91st Annual Meeting
Compendium of Papers DVD. Transportation Research Board, Washington, DC, pp. 16.
Alliance for Biking & Walking, 2018. Bicycling and Walking in the United States 2016 Benchmarking Report. Washington, DC.
Balmat, J.-F., Lafont, F., Maifret, R., Pessel, N., 2011. A decision-making system to maritime risk assessment. Ocean Eng. 38, 171–176. https://doi.org/10.1016/J.

68
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

OCEANENG.2010.10.012.
Belton, V., Stewart, T.J., 2002. Outranking Methods, in: Multiple Criteria Decision Analysis. Springer US, Boston, MA, pp. 233–259 https://doi.org/10.1007/978-1-
4615-1495-4_8.
Biton, A., Daddio, D., Andrew, J., 2014. Statewide Pedestrian and Bicycle Planning Handbook. Washington, DC.
Brans, J.P., De Smet, Y., 2016. PROMETHEE methods. In: Greco, S., Figueira, J., Ehrgott, M. (Eds.), Multiple Criteria Decision Analysis. Springer, New York, pp.
187–219.
Broach, J., Dill, J., Gliebe, J., 2012. Where do cyclists ride? A route choice model developed with revealed preference GPS data. Transp. Res. Part A Policy Pract. 46,
1730–1740. https://doi.org/10.1016/J.TRA.2012.07.005.
Budescu, D.V., Zwick, R., Rapoport, A., 1986. A comparison of the eigenvalue method and the geometric mean procedure for ratio scaling. Appl. Psychol. Meas. 10,
69–78. https://doi.org/10.1177/014662168601000106.
Buehler, R., 2012. Determinants of bicycle commuting in the Washington, DC region: the role of bicycle parking, cyclist showers, and free car parking at work. Transp.
Res. Part D Transp. Environ. 17, 525–531. https://doi.org/10.1016/J.TRD.2012.06.003.
Callister, D., Lowry, M., 2013. Tools and Strategies for wide-scale bicycle level-of-service analysis. J. Urban Plan. Dev. 139, 250–257. https://doi.org/10.1061/(ASCE)
UP.1943-5444.0000159.
Cervero, R., Ferrell, C., Murphy, S., 2002. Transit-Oriented Development and Joint Development in the United States: A Literature Review. Washington, DC.
Charnes, A., Cooper, W.W., 1959. Chance-constrained programming. Manage. Sci. 6, 73–79. https://doi.org/10.1287/mnsc.6.1.73.
Chin, S.M., Franzese, O., Greene, D.L., Hwang, H.L., Gibson, R.C., 2004. Temporary Losses of Highway Capacity and Impacts on Performance: Phase 2. Oak Ridge, TN.
Cinelli, M., Coles, S.R., Kirwan, K., 2014. Analysis of the potentials of multi criteria decision analysis methods to conduct sustainability assessment. Ecol. Indic. 46,
138–148. https://doi.org/10.1016/J.ECOLIND.2014.06.011.
Crawford, G., 1987. The geometric mean procedure for estimating the scale of a judgement matrix. Math. Model. 9, 327–334. https://doi.org/10.1016/0270-0255(87)
90489-1.
Crawford, G., Williams, C., 1985. A note on the analysis of subjective judgment matrices. J. Math. Psychol. 29, 387–405. https://doi.org/10.1016/0022-2496(85)
90002-1.
Davis, W.J., 1995. Bicycle test route evaluation for urban road conditions. In: Transportation Congress, Volumes 1 and 2: Civil Engineers—Key to the World’s
Infrastructure. ASCE, New York, pp. 1063–1076.
Davis, W.J., 1987. Bicycle Safety Evaluation. Auburn University.
Deb, K., 1999. Solving goal programming problems using multi-objective genetic algorithms. In: Proceedings of the 1999 Congress on Evolutionary Computation-
CEC99 (Cat. No. 99TH8406). IEEE, Washington, DC, pp. 77–84. https://doi.org/10.1109/CEC.1999.781910.
DeMaio, P., 2009. Bike-sharing: history, impacts, models of provision, and future. J. Public Transp. 12, 41–56. https://doi.org/10.5038/2375-0901.12.4.3.
Dong, Y., Xu, Y., Li, H., Dai, M., 2008. A comparative study of the numerical scales and the prioritization methods in AHP. Eur. J. Oper. Res. 186, 229–242. https://doi.
org/10.1016/J.EJOR.2007.01.044.
Duthie, J., Unnikrishnan, A., 2014. Optimization framework for bicycle network design. J. Transp. Eng. 140, 04014028. https://doi.org/10.1061/(ASCE)TE.1943-
5436.0000690.
El-Assi, W., Salah Mahmoud, M., Nurul Habib, K., 2017. Effects of built environment and weather on bike sharing demand: a station level analysis of commercial bike
sharing in Toronto. Transportation (Amsterdam) 44, 589–613. https://doi.org/10.1007/s11116-015-9669-z.
El-Geneidy, A., Grimsrud, M., Wasfi, R., Tétreault, P., Surprenant-Legault, J., 2014. New evidence on walking distances to transit stops: Identifying redundancies and
gaps using variable service areas. Transportation (Amsterdam) 41, 193–210. https://doi.org/10.1007/s11116-013-9508-z.
Federal Highway Administration, 1998. Implementing Bicycle Improvements at the Local Level. Washington, DC.
Figueira, J.R., Mousseau, V., Roy, B., 2016. ELECTRE methods. In: Greco, S., Figueira, J., Ehrgott, M. (Eds.), Multiple Criteria Decision Analysis. Springer, New York,
pp. 155–185.
Fishburn, P.C., 1967. Conjoint measurement in utility theory with incomplete product sets. J. Math. Psychol. 4, 104–119. https://doi.org/10.1016/0022-2496(67)
90043-0.
Flamm, B., Rivasplata, C., 2014. Perceptions of Bicycle-Friendly Policy Impacts on Accessibility to Transit Services: The First and Last Mile Bridge. San Jose, CA.
https://transweb.sjsu.edu/PDFs/research/1104-bicycle-policy-transit-accessibility-first-last-mile.pdf.
Foster, N., Monsere, C.M., Dill, J., Clifton, K., 2015. Level-of-service model for protected bike lanes. Transp. Res. Rec. J. Transp. Res. Board 2520, 90–99. https://doi.
org/10.3141/2520-11.
Franek, J., Kresta, A., 2014. Judgment scales and consistency measure in AHP. Proc. Econ. Financ. 12, 164–173. https://doi.org/10.1016/S2212-5671(14)00332-3.
Furth, P.G., 2012. Bicycling infrastructure for mass cycling: a transatlantic comparison. In: Pucher, J., Buehler, R. (Eds.), City Cycling. MIT Press, Cambridge, MA, pp.
105–140.
Furth, P.G., Mekuria, M.C., Nixon, H., 2016. Network connectivity for low-stress bicycling. Transp. Res. Rec. J. Transp. Res. Board 2587, 41–49. https://doi.org/10.
3141/2587-06.
Govindan, K., Rajendran, S., Sarkis, J., Murugesan, P., 2015. Multi criteria decision making approaches for green supplier evaluation and selection: a literature review.
J. Clean. Prod. 98, 66–83. https://doi.org/10.1016/J.JCLEPRO.2013.06.046.
Greene, R., Devillers, R., Luther, J.E., Eddy, B.G., 2011. GIS-based multiple-criteria decision analysis. Geogr. Compass 5, 412–432. https://doi.org/10.1111/j.1749-
8198.2011.00431.x.
Griffin, G.P., Sener, I.N., 2016. Planning for bike share connectivity to rail transit. J. Publ. Transp. 19, 1–22. https://doi.org/10.5038/2375-0901.19.2.1.
Guerreiro, T.de C.M., Kirner Providelo, J., Pitombo, C.S., Antonio Rodrigues Ramos, R., Rodrigues da Silva, A.N., 2018. Data-mining, GIS and multicriteria analysis in a
comprehensive method for bicycle network planning and design. In: Int. J. Sustain. Transp. 12. pp. 179–191. https://doi.org/10.1080/15568318.2017.1342156.
Haezendonck, E., 2008. Transport project evaluation: Extending the social cost benefit approach. Marit. Econ. Logist. 10, 322–324. https://doi.org/10.1057/mel.
2008.7.
Harkey, D.L., Reinfurt, D.W., Sorton, A., 1998. The Bicycle Compatibility Index: A Level of Service Concept, Implementation Manual. Federal Highway Administration,
Washington, DC.
Heinen, E., van Wee, B., Maat, K., 2010. Commuting by bicycle: an overview of the literature. Transp. Rev. 30, 59–96. https://doi.org/10.1080/01441640903187001.
Hochmair, H.H., 2015. Assessment of bicycle service areas around transit stations. Int. J. Sustain. Transp. 9, 15–29. https://doi.org/10.1080/15568318.2012.719998.
Hood, J., Sall, E., Charlton, B., 2011. A GPS-based bicycle route choice model for San Francisco, California. Transp. Lett. Int. J. Transp. Res. 3, 63–75. https://doi.org/
10.3328/TL.2011.03.01.63-75.
Huang, I.B., Keisler, J., Linkov, I., 2011. Multi-criteria decision analysis in environmental sciences: ten years of applications and trends. Sci. Total Environ. 409,
3578–3594. https://doi.org/10.1016/J.SCITOTENV.2011.06.022.
Hwang, C.-L., Yoon, K., 1981. Multiple Attribute Decision Making: Methods and Applications: A State of the Art Survey. Springer Berlin Heidelberg, Berlin, Heidelberg.
https://doi.org/10.1007/978-3-642-48318-9.
Ilie, A., Oprea, C., Costescu, D., Roşca, E., Dinu, O., Ghionea, F., 2016. The use of the bicycle compatibility index in identifying gaps and deficiencies in bicycle
networks. IOP Conf. Ser. Mater. Sci. Eng. 161, 012097. https://doi.org/10.1088/1757-899X/161/1/012097.
Ishizaka, A., Balkenborg, D., Kaplan, T., 2011. Influence of aggregation and measurement scale on ranking a compromise alternative in AHP. J. Oper. Res. Soc. 62,
700–710. https://doi.org/10.1057/jors.2010.23.
Josselin, J.-M., Le Maux, B., 2017. Multi-criteria decision analysis. In: Statistical Tools for Program Evaluation. Springer International Publishing, Cham, pp. 385–416.
https://doi.org/10.1007/978-3-319-52827-4_11.
Kabak, M., Erbaş, M., Çetinkaya, C., Özceylan, E., 2018. A GIS-based MCDM approach for the evaluation of bike-share stations. J. Clean. Prod. 201, 49–60. https://doi.
org/10.1016/J.JCLEPRO.2018.08.033.
Kabir, G., Sadiq, R., Tesfamariam, S., 2014. A review of multi-criteria decision-making methods for infrastructure management. Struct. Infrastruct. Eng. 10,

69
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

1176–1210. https://doi.org/10.1080/15732479.2013.795978.
Kangas, A., Kangas, J., Pykäläinen, J., 2001. Outranking methods as tools in strategic natural resources planning. Silva Fenn. 35, 215–227.
Karim, A., Adeli, H., 2003. Radial basis function neural network for work zone capacity and queue estimation. J. Transp. Eng. 129, 494–503. https://doi.org/10.1061/
(ASCE)0733-947X(2003) 129:5(494).
Kim, Jumsan, Kim, Jongmin, Jun, M., Kho, S., 2005. Determination of a bus Service coverage area reflecting passenger attributes. J. East. Asia Soc. Transp. Stud. 6,
529–543.
Kittelson & Associates, I., Brinckerhoff, P., KFH Group, I., Texas A&M Transportation Institute, Arup, 2013. Transit Capacity and Quality of Service Manual, third ed.
TCRP Report, Washington, DC. https://doi.org/10.17226/24766.
Koopmans, T.C., 1951. An analysis of production as an efficient combination of activities. In: Koopmans, T.C. (Ed.), Activity Analysis of Production and Allocation.
John Wiley and Sons Inc, London, pp. 33–97.
Landis, B., Vattikuti, V., Brannick, M., 1997. Real-time human perceptions: toward a bicycle level of service. Transp. Res. Rec. J. Transp. Res. Board 1578, 119–126.
https://doi.org/10.3141/1578-15.
Lee, J., Choi, K., Leem, Y., 2016. Bicycle-based transit-oriented development as an alternative to overcome the criticisms of the conventional transit-oriented de-
velopment. Int. J. Sustain. Transp. 10, 975–984. https://doi.org/10.1080/15568318.2014.923547.
Lin, J.-J., Yu, C.-J., 2013. A bikeway network design model for urban areas. Transportation (Amsterdam) 40, 45–68. https://doi.org/10.1007/s11116-012-9409-6.
Lin, J.-R., Yang, T.-H., Chang, Y.-C., 2013. A hub location inventory model for bicycle sharing system design: formulation and solution. Comput. Ind. Eng. 65, 77–86.
https://doi.org/10.1016/J.CIE.2011.12.006.
Lootsma, F.A., 1989. Conflict resolution via pairwise comparison of concessions. Eur. J. Oper. Res. 40, 109–116. https://doi.org/10.1016/0377-2217(89)90278-6.
Lowry, M., Loh, T.H., 2017. Quantifying bicycle network connectivity. Prev. Med. (Baltim) 95, S134–S140. https://doi.org/10.1016/J.YPMED.2016.12.007.
Lowry, M.B., Callister, D., Gresham, M., Moore, B., 2012. Assessment of communitywide bikeability with bicycle level of service. Transp. Res. Rec. J. Transp. Res.
Board 2314, 41–48. https://doi.org/10.3141/2314-06.
Lowry, M.B., Furth, P., Hadden-Loh, T., 2016. Prioritizing new bicycle facilities to improve low-stress network connectivity. Transp. Res. Part A Policy Pract. 86,
124–140. https://doi.org/10.1016/J.TRA.2016.02.003.
Ma, D., Zheng, X., 1991. 9/9-9/1 Scale method of AHP. In: Proceedings of the Second International Symposium on the AHP, pp. 197–202.
Macharis, C., Bernardini, A., 2015. Reviewing the use of Multi-Criteria Decision Analysis for the evaluation of transport projects: time for a multi-actor approach.
Transp. Policy 37, 177–186. https://doi.org/10.1016/J.TRANPOL.2014.11.002.
Macharis, C., Nijkamp, P., 2013. Stakeholder bias in multi-actor multi-criteria transportation evaluation: issues and solutions. In: Vanoutrive, T., Verhetsel, A. (Eds.),
Smart Transport Networks–Market Structure, Sustainability and Decision Making. Edward Elgar, pp. 248–268.
Macharis, C., Nijkamp, P., 2011. Possible Bias in Multi-Actor Multi-Criteria Transportation Evaluation: Issues and Solutions. Amsterdam, Netherlands.
Macharis, C., Turcksin, L., Lebeau, K., 2012. Multi actor multi criteria analysis (MAMCA) as a tool to support sustainable decisions: state of use. Decis. Support Syst. 54,
610–620. https://doi.org/10.1016/J.DSS.2012.08.008.
Martens, K., 2004. The bicycle as a feedering mode: experiences from three European countries. Transp. Res. Part D Transp. Environ. 9, 281–294. https://doi.org/10.
1016/J.TRD.2004.02.005.
Marttunen, M., Lienert, J., Belton, V., 2017. Structuring problems for Multi-Criteria Decision Analysis in practice: a literature review of method combinations. Eur. J.
Oper. Res. 263, 1–17. https://doi.org/10.1016/J.EJOR.2017.04.041.
Mauttone, A., Mercadante, G., Rabaza, M., Toledo, F., 2017. Bicycle network design: model and solution algorithm. Transp. Res. Proc. 27, 969–976. https://doi.org/
10.1016/J.TRPRO.2017.12.119.
Mekuria, M., Furth, P., Nixon, H., 2012. Low-Stress Bicycling and Network Connectivity. Mineta Transportation Institute, San José, CA.
Mesbah, M., Thompson, R., Moridpour, S., 2012. Bilevel optimization approach to design of network of bike lanes. Transp. Res. Rec. J. Transp. Res. Board 2284, 21–28.
https://doi.org/10.3141/2284-03.
Mistretta, M., Goodwill, J.A., Gregg, R., De Annuntis, C., 2009. Best Practices in Transit Service Planning, Report No. 21177772000. https://doi.org/10.1016/j.ejor.
2006.07.038.
Murray, A.T., Davis, R., 2001. Equity in regional service provision. J. Reg. Sci. 41, 577–600. https://doi.org/10.1111/0022-4146.00233.
NACTO, 2016. Equitable Bike Share Means Building Better Places for People to Ride. New York.
NACTO, 2011. Urban Bikeway Design Guide, second ed. Island Press, Washington, DC.
Noël, N., Leclerc, C., Lee-Gosselin, M., 2003. CRC index: compatibility of roads for cyclists in rural and urban fringe areas. In: 82nd Annual Meeting of the
Transportation Research Board CD-ROM, Washington, DC, pp. 1–20.
Opricovic, S., Tzeng, G.-H., 2007. Extended VIKOR method in comparison with outranking methods. Eur. J. Oper. Res. 178, 514–529. https://doi.org/10.1016/J.
EJOR.2006.01.020.
Opricovic, S., Tzeng, G.-H., 2004. Compromise solution by MCDM methods: a comparative analysis of VIKOR and TOPSIS. Eur. J. Oper. Res. 156, 445–455. https://doi.
org/10.1016/S0377-2217(03)00020-1.
Parkin, John, Wardman, Mark, Page, Matthew, Parkin, J., Wardman, M., Page, Á.M., Page, M., 2008. Estimation of the determinants of bicycle mode share for the
journey to work using census data. Transportation (Amsterdam) 35, 93–109. https://doi.org/10.1007/s11116-007-9137-5.
Pucher, J., Dijkstra, L., 2000. Making walking and cycling safer: lessons from Europe. Transp. Q. 54, 25–50.
Rietveld, P., 2000. The accessibility of railway stations: The role of the bicycle in The Netherlands. Transp. Res. Part D Transp. Environ. 5, 71–75. https://doi.org/10.
1016/S1361-9209(99)00019-X.
Rybarczyk, G., Wu, C., 2010. Bicycle facility planning using GIS and multi-criteria decision analysis. Appl. Geogr. 30, 282–293. https://doi.org/10.1016/J.APGEOG.
2009.08.005.
Saaty, T.L., 1994. Fundamentals of Decision Making, sutlib2.sut.ac.th. RWS Publications, Pittsburgh, PA.
Saaty, T.L., 1990. The Analytic Hierarchy Process: Planning, Priority Setting, Resource Allocation. RWS Publications.
Saaty, T.L., 1980. The Analytic Hierarchy Process. McGraw-Hill, New York.
Salo, A.A., Hämäläinen, R.P., 1997. On the measurement of preferences in the analytic hierarchy process. J. Multi-Criteria Decis. Anal. 6, 309–319. https://doi.org/10.
1002/(SICI)1099-1360(199711)6:6<309::AID-MCDA163>3.0.CO;2-2.
Sanayei, A., Farid Mousavi, S., Abdi, M.R., Mohaghar, A., 2008. An integrated group decision-making process for supplier selection and order allocation using multi-
attribute utility theory and linear programming. J. Franklin Inst. 345, 731–747. https://doi.org/10.1016/J.JFRANKLIN.2008.03.005.
Saplıoğlu, M., Aydın, M.M., 2018. Choosing safe and suitable bicycle routes to integrate cycling and public transport systems. J. Transp. Heal. 10, 236–252. https://
doi.org/10.1016/J.JTH.2018.05.011.
Schniederjans, M.J., Wilson, R.L., 1991. Using the analytic hierarchy process and goal programming for information system project selection. Inf. Manage. 20,
333–342. https://doi.org/10.1016/0378-7206(91)90032-W.
Schrank, D., Eisele, B., Lomax, T., Bak, J., 2015. 2015 Urban Mobility Scorecard.
Sirikijpanichkul, A., Winyoopadit, S., Jenpanitsub, A., 2017. A multi-actor multi-criteria transit system selection model: a case study of Bangkok feeder system. Transp.
Res. Proc. 25, 3736–3755. https://doi.org/10.1016/j.trpro.2017.05.228.
Sorton, A., Walsh, T., 1994. Bicycle stress level as a tool to evaluate urban and suburban bicycle compatibility. Transp. Res. Rec. 1438, 17–24.
Stinson, M., Bhat, C., 2003. Commuter bicyclist route choice: analysis using a stated preference survey. Transp. Res. Rec. J. Transp. Res. Board 1828, 107–115. https://
doi.org/10.3141/1828-13.
Terh, S.H., Cao, K., 2018. GIS-MCDA based cycling paths planning: a case study in Singapore. Appl. Geogr. 94, 107–118. https://doi.org/10.1016/J.APGEOG.2018.03.
007.
Toole Design Group, 2010. Cincinnati Bicycle Transportation Plan. Cincinnati, OH.
Transportation Research Board, 2010. Highway Capacity Manual 2010. Transportation Research Board.

70
T. Zuo and H. Wei Transportation Research Part A 129 (2019) 52–71

Triantaphyllou, E., 2000. Multi-criteria decision making methods. In: Applied Optimization, vol. 44. Springer, Boston, MA, pp. 5–21. https://doi.org/10.1007/978-1-
4757-3157-6_2.
Turner, S., Shafer, C.S., Stewart, W., 1997. Bicycle Suitability Criteria for State Roadways in Texas. College Station, TX.
U.S. Department of Transportation, Federal Highway Administration, 2018. 2017 National Household Travel Survey [WWW Document]. U.S. Dep. Transp. Fed.
Highw. Adm. URL http://nhts.ornl.gov (accessed 10.29.18).
Velasquez, M., Hester, P.T., 2013. An analysis of multi-criteria decision making methods. Int. J. Operat. Res. 10 (2), 56–66.
Von Neumann, J., Morgenstern, O., 2007. Theory of Games and Economic Behavior, 60th anniv. ed. Princeton University Press, Princeton, NJ.
Wang, H., Palm, M., Chen, C., Vogt, R., Wang, Y., 2016. Does bicycle network level of traffic stress (LTS) explain bicycle travel behavior? Mixed results from an Oregon
case study. J. Transp. Geogr. 57, 8–18. https://doi.org/10.1016/J.JTRANGEO.2016.08.016.
Wang, J.-J., Jing, Y.-Y., Zhang, C.-F., Zhao, J.-H., 2009. Review on multi-criteria decision analysis aid in sustainable energy decision-making. Renew. Sustain. Energy
Rev. 13, 2263–2278. https://doi.org/10.1016/J.RSER.2009.06.021.
Wargelin, L., Stopher, P., Minser, J., Tierney, K., Rhindress, M., O’Connor, S., 2012. GPS-Based Household Interview Survey for the Cincinnati Ohio Region.
Columbus, OH.
Weigand, L., McNeil, N., Jennifer Dill, M., 2013. Cost Analysis of Bicycle Facilities: Cases from Cities in the Portland, OR region. Portland, OR.
Williams, C., Crawford, G., 1980. Analysis of Subjective Judgment Matrices. Santa Monica, CA.
Winters, M., Brauer, M., Setton, E.M., Teschke, K., 2013. Mapping bikeability: a spatial tool to support sustainable travel. Environ. Plan. B Plan. Des. 40, 865–883.
https://doi.org/10.1068/b38185.
Winters, M., Davidson, G., Kao, D., Teschke, K., 2011. Motivators and deterrents of bicycling: comparing influences on decisions to ride. Transportation (Amsterdam)
38, 153–168. https://doi.org/10.1007/s11116-010-9284-y.
Winters, M., Teschke, K., Brauer, M., Fuller, D., 2016. Bike Score®: Associations between urban bikeability and cycling behavior in 24 cities. Int. J. Behav. Nutr. Phys.
Act. 13, 18. https://doi.org/10.1186/s12966-016-0339-0.
Zavadskas, E.K., Turskis, Z., Kildienė, S., 2014. State of art surveys of overviews on MCDM/MADM methods. Technol. Econ. Dev. Econ. 20, 165–179. https://doi.org/
10.3846/20294913.2014.892037.
Ziari, H., Khabiri, M.M., 2010. Applied GIS software for improving pedestrian & bicycle safety. Transport 20, 160–164. https://doi.org/10.1080/16484142.2005.
9638014.
Zionts, S., 1992. Some thoughts on research in multiple criteria decision making. Comput. Oper. Res. 19, 567–570. https://doi.org/10.1016/0305-0548(92)90028-4.
Zuo, T., Wei, H., Rohne, A., 2018. Determining transit service coverage by non-motorized accessibility to transit: case study of applying GPS data in Cincinnati
metropolitan area. J. Transp. Geogr. 67, 1–11. https://doi.org/10.1016/J.JTRANGEO.2018.01.002.

71

You might also like