A Model for Evaluating the Connectivity of Multimodal Transit

Networks
Junsik Park
Associate Research Fellow
The Korea Transport Institute
Department of Green Growth Policy and Implementation
1160 Simindaero, Ilsanseo-gu, Goyang-si
Gyeonggi-do, 411-701, Korea
forejs@koti.re.kr
Phone: 82 31 910 3048
Fax: 82 31 910 3228
Seong-Cheol Kang
Associate Research Fellow
The Korea Transport Institute
Department of Logistics Research
1160 Simindaero, Ilsanseo-gu, Goyang-si
Gyeonggi-do, 411-701, Korea
jsckang@koti.re.kr
Phone: 82 31 910 3098
Fax: 82 31 910 3227
(Corresponding author)

Submitted for Presentation at the 90th Annual Meeting of the Transportation Research
Board and Publication in the Transportation Research Record Series
July 30, 2010
Word Count:
Text: 3869 (Abstract: 231)
Figures: 4 x 250 = 1000
Tables: 1 x 250 = 250
Total: 5119

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

ABSTRACT
This study aims to develop a model for evaluating the connectivity of a multimodal transit network. In
order for the model to well represent characteristics of a multimodal transit network, connectivity should
be analyzed based on nodes and transit lines of the network instead of nodes and links. The degree
centrality of a node, which is a commonly used concept in social network analysis, is employed with
appropriate modifications so that it is suitable for connectivity analysis for multimodal transit networks.
As the representative characteristics of a transit line, we select its daily capacity, average speed, and
length. We then define the connecting power of the line as the product of those quantities. Using the
degree centrality of a transit stop and the connecting powers of transit lines serving the stop, we develop
an index quantifying the level of connectivity of the stop. We also derive the connectivity index of a
transit line as well as an area within the network from the connectivity indexes of transit stops. We show
that the connectivity indexes of transit stops can be used to evaluate the connectivity of a transfer center.
The connectivity evaluation model developed in this study takes well into consideration characteristics of
multimodal transit networks, adequately measures the connectivity of transit stops, transit lines, and areas,
and furthermore can be used in determining the level of service of transfer centers.

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

1. INTRODUCTION
A multimodal transit network is a network interconnected by various public transportation modes with
different characteristics, such as buses, subways, express buses, and regional railways. With the
advancement of ITS technologies for public transportation, such a network has been gaining significant
attraction in recent days because if well designed and operated, it can enhance the performance of the
whole transportation system and also reduce greenhouse gas emissions. The efficiency and convenience,
among other things, of a multimodal transit network are crucial to inducing more people to use public
transportation instead of their cars. Thus the efficiency and convenience need be evaluated often and if
needed, should be improved by modifying or upgrading the network.
One of the concepts that could describe the efficiency and convenience of a multimodal transit
network is its connectivity, that is, how well or easily people can move from one point (or one area) of the
network to another point (or another area). This paper aims to develop a quantitative model for measuring
such connectivity of a multimodal transit network.
Multimodal transit networks, like road networks, consist of nodes and links. However, links in a
multimodal transit network have different characteristics from those in a road network. While a link of a
road network is a physical segment that connects one node to another, a link of a multimodal transit
network is part of a transit line that serves a sequence of transit stops (nodes). Since a stop can be served
by different transit lines, multiple links may exist between nodes in a multimodal transit network. In a
road network, on the other hand, usually a single link exists between nodes. Moreover, even if a node is
connected by incoming and outgoing links in a multimodal transit network, transfer cost may occur unless
those links belong to the same transit line. This is usually not the case in a road network. These
characteristics of multimodal transit networks suggest that connectivity analysis for a multimodal transit
network should be conducted based on nodes and transit lines instead of nodes and links.
The rest of the paper is organized as follows. In Section 2, we review relevant literature and
several centrality measures used in the area of social networks, one of which is to be employed in our
model. In Section 3, we develop an index quantifying the level of connectivity of a transit stop using a
centrality measure and several characteristics of a transit line. From the connectivity indexes of transit
stops, we derive the connectivity index of a transit line as well as an area within the network. In Section 4,
we show that the connectivity indexes of transit stops can be used to evaluate the connectivity of a
transfer center. We also demonstrate how the connectivity index of a transfer center can be exploited in
determining the level of service of the transfer center. Concluding remarks are given in Section 5.
2. LITERATURE REVIEW
The literature on analysis of the connectivity of a multimodal transit network is relatively scant.
Derrible and Kennedy [5] developed three transit network indicators, coverage, directness, connectivity
and investigated the relationship between the indicators and ridership, computed as the annual number of
boardings per capita. Coverage measures the percentage area covered by the transit network. Directness
relates to the maximum number of transfers necessary to go from one station to another. Connectivity
looks at the structure of the network and the affluence of transfer stations. Multiple regression analysis
conducted for 19 subway systems worldwide showed a strong relationship between those three indicators
and ridership. Derrible and Kennedy [6] applied the model of [5] to evaluate 15 and 25-year transit plans
produced by the Toronto regional transportation authority, and compared the results with other transit
systems in the world.
The connectivity indicator used by Derrible and Kennedy [5-6], however, has a limitation of not
taking into account the characteristics of transit lines in a network other than topological connections. As
we will show in Section 3, the concept of connectivity we put forth in this paper considers not only
topological connections between nodes of a transit network, but also the strength of those connections,
thereby producing more realistic results than the model of Derrible and Kennedy.

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

In the area of social networks, the concept of centrality has been extensively studied as a measure
of the relative importance of a node in a network [1-4, 7, 9, 10, 12, 13]. Perhaps the simplest of centrality
measures is degree centrality. Let G ( N , L) be a network where N and L denote the set of nodes and the
set of links, respectively. The degree centrality of node n N , denoted by C d (n) , is defined as the
number of links incident upon node n , i.e.,
C d ( n) =

Ans

(1)

sN

where Ans takes 1 if node s is incident upon node n , 0 otherwise; that is, Ans is the (n, s ) -element of
the incident matrix A of the network. Though simple, the degree centrality is often a highly effective
measure of the importance of a node.
A more sophisticated version of the degree centrality is eigenvector centrality. Whereas the
degree centrality simply counts the number of connections a node has, the eigenvector centrality
acknowledges that not all connections are equal. It assigns relative scores to all nodes in the network
based on the principle that connections to high-scoring nodes contribute more to the score of the node in
question than equal connections to low-scoring nodes. The eigenvector centrality of node n N , denoted
by C e (n) , is defined as
C e ( n) =

Ans C e ( s)

(2)

sN

which can be rewritten in vector equation as

C e = AC e

(3)
where Ce is the vector of Ce ( n), n N . In linear algebra, scalar and vector Ce that satisfy Equation
(3) are known as an eigenvalue and eigenvector of matrix A (hence the name of the centrality).
Two other commonly used centrality measures are closeness centrality and betweenness
centrality. A geodesic path in a network is the shortest path, in terms of number of links traversed,
between a specified pair of nodes. The closeness centrality is defined as the mean geodesic distance (i.e.,
the mean length of a geodesic path) from a node to all other nodes reachable from it. The betweenness
centrality of a node is the fraction of geodesic paths between other nodes that the node falls on. That is,
we find the shortest paths between every pair of nodes and then ask on what fraction of those paths the
node in question lies. The betweenness centrality is a crude measure of the control the node in question
exerts over the flow of information (or any other commodity) between other nodes [10].
The concept of degree centrality will be used in measuring the connectivity of a multimodal
transit network below. The degree centrality, like other centrality measures, focuses only on the topology
of a network without taking into consideration the characteristics of links in the network. Hence we will
make appropriate modifications of the degree centrality so that it accommodates the necessary
characteristics of links in a multimodal transit network.

3. MODEL
3.1 Connecting Power of a Transit Line
Connectivity analysis of a multimodal transit network begins with measuring the connectivity of a transit
stop. In order to measure the connectivity of a transit stop, one needs to consider not only topological
connections of the stop, which will be handled by the degree centrality, but also the characteristics of
transit lines serving the stop. The characteristics of a transit line can be categorized into vehicle, operation,

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

and line characteristics. Vehicle characteristics include capacity, speed, and acceleration/deceleration
capability. Operation characteristics include, among other things, frequency and headway. As line
characteristics, one can think of the length of a transit line and the average distance between stops. To
take into account these characteristics combined, we use the daily capacity, which is the product of the
capacity per car and the frequency per day, the average speed, and the length of a transit line as the
representative characteristics of the transit line.
Intuitively, a transit line with large daily capacity would provide passengers with better
connectivity between transit stops. Furthermore, passengers who ride a fast transit line would enjoy better
connectivity between transit stops. To formalize these intuitions, we introduce the concept of connecting
power of a transit line at a certain transit stop. The outbound connecting power of transit line l at transit
stop v represents how well or easily passengers can access from stop v to other stops using transit line l
and is defined as follows:
Tlv + = Dl Vl Lvl+

(4)

where Dl is the daily capacity of transit line l , Vl is the average speed of transit line l , Lvl+ is the
length (distance) from stop v to the destination of transit line l , and , , and are scaling
coefficients.
As evident in Equation (4), high daily capacity or average speed of a transit line results in a high
outbound connecting power. Moreover, stops close to the origin of a transit line have a higher outbound
connecting power than stops close to the destination of the transit line because the former have more
reachable stops than the latter. Indeed, given Dl and Vl , the outbound connecting power decreases
linearly as moving from the origin of a transit line to its destination as illustrated in Figure 1.
Capacity
(D)
Length
(L)

A/L

B/L

Speed
(V)

Tl v+

C km

B km

A km

L km(Length)
Origin

C/L

A/L

B/L

C/L

Destin
ation

FIGURE 1. Outbound Connecting Power

We also characterize the inbound connecting power of transit line l at transit stop v as how well
or easily passengers can access to stop v from other stops using transit line l and define it as
Tlv = Dl Vl Lvl

(5)

where Lvl is the length (distance) from the origin of transit line l to stop v .
While the outbound connecting power of a transit line at a certain transit stop represents
connectivity from the stop to the downstream stops of the transit line, the inbound connecting power
measures connectivity from the upstream stops of the transit line to the stop under consideration. Figure 2

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

depicts how the inbound connecting power changes as moving from the origin of a transit line to its
destination. Contrary to the outbound connecting power, the inbound connecting power increases linearly
as moving towards the destination of a transit line.
Capacity
(D)

A/L
A km

B/L
B km

C/L

Length
(L)

Speed
(V)

C km
L km(Length)

Origin

A/L

Destin
ation

C/L

B/L

Tl v

FIGURE 2. Inbound Connecting Power

Having defined the outbound and inbound connecting powers, the connecting power of transit
line l at transit stop v is then given by
Tlv =

Tlv + + Tlv
2

(6)

To illustrate the computation of connecting powers, consider the bus stop in Figure 3 which is
served by two bus lines. Bus line 1, depicted by a blue line, provides 100 bus operations per day with
each bus having a capacity of 50 passengers. The average operating speed of this bus line is 20km/hr. The
distances from the bus stop to the origin and destination of the bus line are 5km and 15km, respectively.
Bus line 2, shown in a red line, has 80 operations per day with each bus having a capacity of 50
passengers. The average speed of bus line 2 is 25km/hr, and the distances from the bus stop to the origin
and destination of bus line 2 are 10km and 20km, respectively. Let us set to 1/(50 passengers/car 50
operations/day), to 1/(20km/hr), and to 1/(20km). Then the outbound (inbound) connecting powers
of bus line 1 and 2 at the bus stop are 1.0 and 2.0 (0.5 and 1.5) respectively, as shown in Figure 3. Thus
the connecting powers of bus line 1 and 2 are 0.75 and 1.75 respectively. This shows that bus line 2
provides better connectivity than bus line 1 in this example.

1
1

T =

50100
2500

20
20

5
20

= 0.5
5km
10km

80 25 10
T21 = 502500
20 20 = 1.5

Line1: 50passenger/car
100operations/day
average speed 20kph
Line2: 50passenger/car
80operations/day
average speed 25kph

100 20 15
T11+ = 502500
20 20 = 1.0

15km

20km
80 25 20
T21+ = 502500
20 20 = 2.0

FIGURE 3. Connecting Powers of Bus Lines

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

3.2 Connectivity Indexes

Based on the connecting powers of transit lines, we quantify the connectivity of a transit stop by summing
up the connecting powers of transit lines serving the stop. We call this quantity the connectivity index of
the stop. In other words, the connectivity index of transit stop v , denoted by CI (v) , is defined as
CI (v) = Tlv lv

(7)

lL

where L is the set of transit lines, Tlv is the connecting power of transit line l at transit stop v , and lv
is an indicator function that takes 1 if transit line l serves stop v , 0 otherwise. The connectivity index of
bus stop in Figure 3, for instance, is 2.5 (0.75+1.75).
The connectivity index in (7) is, in terms of its concept, similar to the degree centrality (1).
However unlike the degree centrality that takes into consideration only the topology of a network, the
connectivity index of a transit stop reflects not only the network topology but also the characteristics of
transit lines that passing through the stop.
Since a transit line is a sequence transit stops, we can also define the connectivity index of a
transit line using the connectivity indexes of transit stops that the transit line serves. The connectivity
index of transit line l is defined as
CI (l) =

1
CI (v)
S l 1 vS l ,v vo

(8)

where S l is the set of stops that transit line l serves and vo is the first stop (i.e., the origin) of transit
line l . The reason for the scaling factor 1 / ( S l 1) in (8) is that it allows comparing connectivity indexes
of transit lines from different modes, e.g., a bus line with many stops and a subway line with fewer stops.
The connectivity index of a transit stop can be also used to quantify the level of connectivity of a
certain area in a transit network. Let R be a some area within the network and S R be the set of transit
stops inside area R . Then the connectivity index of area R can be defined as
CI ( R) =

1
KR

CI (v)

(9)

vS R

where K R is a scaling factor. One possible choice for K R would be the population of the area. When R
equals to the entire network, (9) becomes the connectivity index of the transit network.

4. CONNECTIVITY OF TRANSFER CENTERS

A transfer center usually consists of several transit stops through which passengers change their mode of
transportation. Analysis of the connectivity of a transfer center is fundamental to assessing the level of
service of the transfer center. However, to the best of our knowledge, methods for measuring the
connectivity of a transfer center in the literature are scarce. In this section, we propose an index that
quantifies the connectivity of a transfer center using the connectivity index of a transit stop developed in
the previous section.

4.1 Passenger Acceptance Rate

The passenger acceptance rate for a transit stop is defined as the percentage of passengers who would be
willing to use public transportation at the stop as a function of the walking distance (or time) to the stop.

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

Intuitively as the walking distance (or time) to a stop gets longer, the percentage of passengers who would
use the stop becomes smaller. Kim and Kwon [8] estimated the following passenger acceptance rate
function

y = 1.13189 exp( 0.0872 x )

(10)

where x is the walking time (in minutes) to a transit stop and y is the passenger acceptance rate.
Equation (10) implies that if the time to walk to a transit stop is less than 3 minutes, almost all passengers
would use the stop. If the walking time is 10 minutes, only 55% of passengers would use the stop and
45% of passengers would forgo trips or use other transit stops.
Given a walking distance (or time), comparison of rates from (10) with those in TCRP Report
100 [11] reveals that the passenger acceptance rate of North America is smaller than that of Korea. This is
because trips in North America are more car-oriented compared to those in Korea.

4.2 Connectivity Index of a Transfer Center

Since a transfer center is a collection of transit stops, one may think that the connectivity index of a
transfer center is simply the sum of the connectivity indexes of the transit stops within the transfer center.
However, as exemplified above, some passengers may give up transferring to another mode if the
distance between stops is perceived to be too long to walk. This implies that the connectivity index of a
transfer center should be less than the sum of the connectivity indexes of the stops. Thus, we define the
connectivity index of a transfer center as the sum of the connectivity indexes of the stops weighted by
passenger acceptance rates. Specifically, the connectivity index of a transfer center, denoted by CI TC , is
given by
CI TC =

CI (v) y wv
S TC 1 wSTC vSTC ,v w

(11)

where S TC is the set of transit stops in the transfer center, CI (v) is the connectivity index of stop v , and
y wv is the passenger acceptance rate from stop w to stop v , i.e., the percentage of passengers who are
willing to transfer from stop w to stop v given the distance between the two stops. The reason of
dividing by S TC 1 in (11) is that the connectivity index of each stop is counted STC 1 times in
calculation.
Consider, for example, the transfer center comprising of four transit stops in Figure 4.

CI(1)=5
CI(4)=2
CI(2)=1 7min
10min 3min
4min 5min 3min
CI(3)=1.5
FIGURE 4. Transfer Center

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

Table 1 summarizes the calculation of the connectivity index of the transfer center in Figure 4,
where the function in (10) is used to compute passenger acceptance rates. The resulting connectivity
index, 8.11, is smaller than the sum of the connectivity indexes of individual stops, 9.5 (5+1+1.5+2).
The ratio of the connectivity index of a transfer center to the sum of the connectivity indexes of
individual stops within the transfer center can be employed to assess the level of service of the transfer
center. As this ratio is close to 1, it can be said that the transfer center provides a better level of service
because less passengers are giving up transfer due to longer walking distance or time between stops. Let
r denote this ratio. If, for instance, one sets the level of service of a transfer center with r 0.9 ,
0.8 r < 0.9 , 0.7 r < 0.8 , 0.6 r < 0.7 , and r < 0.6 to A, B, C, D, and E respectively, the level of
service of the transfer center in Figure 4 is B ( r =8.11/9.5= 0.85).

TABLE 1. Calculation of Connectivity Index of Transfer Center

w
1

CI (v )

2
3
4
1
3
4
1
2
4
1
2
3

1
1.5
2
5
1.5
2
5
1
2
5
1
1.5
CI TC

walking time
passenger acceptance
(minutes)
rate (%)
7
72
5
85
3
100
7
72
4
93
10
55
5
85
4
93
3
100
3
100
10
55
3
100
1
=
CI (v) y wv = 8.11
S TC 1 wSTC vSTC ,v w

CI (v) y wv

0.72
1.28
2.00
3.58
1.40
1.10
4.26
0.93
2.00
5.00
0.55
1.50

5. CONCLUSION
As transit networks are becoming more multimodal, the concept of connectivity of a transit network
becomes important. However, methods for measuring the connectivity of a multimodal transit network
are scant in the literature. This study aims to develop a quantitative model for measuring the connectivity
of a multimodal transit network. To that end, we first define the outbound and inbound connecting powers
of a transit line based on the daily capacity, average operating speed, and length of the transit line. We
then develop an index quantifying the connectivity of a transit stop using the connecting powers of transit
lines serving the stop and the degree centrality of the stop. We also derive the connectivity index of a
transit line as well as an area within the network from the connectivity indexes of transit stops. We further
show that the connectivity indexes of transit stops can be used in evaluating the connectivity of a transfer
center.
The connectivity evaluation model developed in this study takes well into consideration
characteristics of multimodal transit networks, adequately evaluates the connectivity of transit stops,
transit lines, and areas, and furthermore can be useful in determining the level of service of transfer
centers. Therefore we believe that the model contributes to theoretical analysis of connectivity of

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.

Junsik Park, Seong-Cheol Kang

10

multimodal transit networks. On a practical side, we expect the model to be efficaciously used in
designing better multimodal transit networks in conjunction with ITS technologies for public
transportation systems.
Connectivity is an important measure to assess the effectiveness of road networks as well, and the
concept of connectivity index considered in this paper can be extended, in principle, for road networks.
However due to different characteristics of transit and road networks, the ingredients for the connectivity
index for road networks can be different from those for transit networks and should incorporate
components such as the level of circuity and the number of paths between nodes. We will present the
results of connectivity analysis in road networks in a separate paper.

ACKNOWLEDGEMENTS
This research was supported by a grant (06-A-02) from Intermodal Connectivity and Transfer System
Technology Development Program funded by the Ministry of Land, Transport and Maritime Affairs of
Korea.

REFERENCES
1.
2.
3.
4.
5.
6.

7.
8.
9.
10.
11.
12.
13.

Bonacich, A., Factoring and weighting approaches to status scores and clique identification,
Journal of Mathematical Sociology, vol. 2, 1972, pp. 113-120.
Borgatti, S. P., Centrality and network flow, Social Networks, vol. 27, 2005, pp. 55-71.
Borgatti, S. P. and Everett, M. G., Network analysis of 2-mode data, Social Networks, vol. 19,
1997, pp. 243-269.
Borgatti, S. P. and Everett, M. G., A Graph-theoretic perspective on centrality, Social Networks,
vol. 28, 2006, pp. 466-484.
Derrible, S. and Kennedy, C., A network analysis of subway systems in the world using updated
graph theory, Transportation Research Record, vol. 2112, 2009, pp. 17-25.
Derrible, S. and Kennedy, C., Evaluating, comparing, and improving metro networks: an application
to the Toronto plans, Proceedings of 82th Transportation Research Board Annual Meeting,
Washington, D.C., 2010.
Freeman, L.C., Centrality in networks: I. Conceptual clarification, Social Networks, vol. 1, 1979,
pp. 215-239.
Kim, J. S. and Kwon, Y. S., Determination of a bus service coverage area reflecting passenger
attributes, Journal of Korean Society of Transportation, vol. 23(3), 2005, pp. 149-159 (in Korean).
Newman, M. E. J., Analysis of weighted networks, Physical Review E, vol. 70, 056131, 2004.
Newman, M. E. J., Mathematics of networks, In The New Palgrave Encyclopedia of Economics,
2nd edition. Palgrave Macmillan, Basingstoke, 2007.
TCRP Report 100, Transit Capacity and Quality of Service Manual-2nd Edition, Transportation
Research Board. National Academy Press, Washington, D.C., 2003.
Wuchty, S. and Stadler, P. F., Centers of complex networks, Journal of Theoretical Biology, vol.
223, 2003, pp. 45-53.
Yang, S. and Knoke, D., Optimal connections: strength and distance in valued graphs, Social
Networks, vol. 23, 2001, pp. 285-295.

TRB 2011 Annual Meeting

Original paper submittal - not revised by author.