ENGINEER

ENGINEER -- Vol.
Vol. LVI,
LVI, No.
No. 01,
01, pp.
pp. [51-62], 2023 2023
[page range],
©
© The
The Institution
Institution of
of Engineers,
Engineers, Sri
Sri Lanka
Lanka DOI: [Link]

Travel Behaviour Analysis and Exploration of Mode

Choice Variability of Short-Distance Travellers
P.M.G.D.M. Herath, W.R.S.S. Dharmarathna, W.M.V.S.K. Wickramasinghe and
A.G.H.J. Edirisinghe

Abstract: Commuters tend to shift from public to private transport modes due to various
reasons. This results in an increased traffic volume in the urban road network. The ultimate
consequence is traffic congestion which creates massive economic losses and adverse environmental
pollution. To provide a feasible solution for the above problems, this study is set to examine the
factors affecting transport mode choice. Factor analysis was used to identify the factors influencing the
mode choice variability of the commuters. Parameter estimation is done using the Multinomial Logit
(MNL) model based on the utility maximization theory. The adjusted likelihood ratio index is used to
show the model fits. Also, t-statistics and the respective sign of parameters are used to show the
validity of estimated parameters. The outcome of the factor analysis shows that age and occupation
significantly affect the public transport mode choice under personal characteristics. For private mode
choice, the contributing factors are gender, age, occupation and monthly income. Trip distance, in-
vehicle travel time, and travel cost are significantly influencing both public and private transport
mode choices. Results of the estimated parameters and elasticity analysis suggest that more
commuters can be attracted to public transport by reducing the total travel time of buses.

Keywords: Travel Behaviour, Mode Choice, Factor Analysis, Utility Maximization, MNL
Modelling

1. Introduction out in 2016 by the University of Moratuwa

and the University of Peradeniya [3]. The
Having an effective and efficient study revealed that public transport usage has
transportation system is one of the decisive decreased from 78.3% in 1999 to 62.4% in
factors that has a major impact on the [Link] categories of private vehicles carry
development of a country [1]. Public 19% of the passenger percentage that
transportation plays a major role to fulfil the contributes to 65% of the traffic flow entering
travel requirements of the commuters in a
country. Nevertheless, most of the commuters Eng.P.M.G.D.M. Herath, AMIE(SL), BScEng(Hons):
are converting their mode of transport to MPhil Candidate, University of Peradeniya, Sri Lanka
private modes for various reasons. According Email:dhanushkamadusanka1995@[Link]
[Link]
to the central bank report of Sri Lanka in 2019,
Eng.(Dr.)W.R.S.S. Dharmarathna, AMIE(SL), PhD,
99% of total registered vehicles in Sri Lanka University of Tokyo, Tokyo, Japan, MPhil, University of
are private vehicles [2]. Therefore, many Peradeniya, BScEng(Hons), University of Peradeniya, Sri
private vehicles are accumulated to the road Lanka
Senior Lecturer, Department of Civil Engineering,
network day by day, creating adverse effects University of Peradeniya.
on society. Among them, traffic congestion is Email:samal@[Link]
one of the main problems that make people [Link]
Eng.(Dr.)W.M.V.S.K. Wickramasinghe, AMIE(SL), PhD,
suffer a lot.
Hokkaido University, Japan, MEng, Hokkaido University,
Japan, BScEng(Hons), University of Peradeniya, Sri Lanka,
In Sri Lanka, there has been great concern Senior Lecturer, Department of Civil Engineering,
about traffic congestion and the government University of Peradeniya.
Email:vskw@[Link]
seeks possible remedies to control this [Link]
problem. Traffic congestion tends to grow as Eng.(Dr.)A.G.H.J. Edirisinghe, AMIE(SL), University of
many private vehicles are added to roads Ehime, Japan, Diploma: Road safety Lund – Sweden, MEng,
University of Ehime, Japan, BScEng(Hons), University of
every day. Other modes of transport services Peradeniya, Sri Lanka,
such as taxi and three-wheeler services by the Senior Lecturer, Department of Civil Engineering,
private sector are also expanding rapidly, University of Peradeniya.
encouraging passengers to leave the public Email:jayalath@[Link]
[Link]
transport modes. This has been well revealed
by the Kandy traffic study which was carried

1
51 ENGINEER
ENGINEER
the Kandy city. Route buses contribute 8.5% solution is to improve the public transport
to the total traffic flow, but they carry 63.4% of system so that commuters will shift from
passenger movement [3]. private transport modes to public transport
modes. This solution seemed to be more
In Kandy, Sri Lanka, the main mode of public practical and efficient since a high-quality
transportation is the bus. There are 500 - 600 public transport system is capable of retaining
buses operating under Sri Lanka Transport existing customers as well as attracting private
Board (SLTB) and around 1900 buses are vehicle users.
operating under private operators. Nearly
300,300 passengers are using the public bus Improving the public transportation system is
transport service in Kandy city. Passenger an innovative approach to meet the challenges
queues can often be seen in Kandy city due to due to rapid growth in demand for private
the lack of adequate space, which creates transport modes. Therefore, it is very
significant safety risks for pedestrians [4]. important to identify the factors which are
affecting the mode choice behaviour of the
Limited road infrastructure, conflicting land commuters.
use, poor orientation of bus routing, terminal
and stops, inadequate by-pass roads for Talagala and Kalukottege, (2014) studied the
Kandy Municipal Area and Central Business factors affecting public transport ridership [7].
District, inappropriate traffic circulation and Even though the study was limited to the
parking demand management measures, capital Colombo, it revealed important factors
underutilisation of the railway network, and that affect passengers’ perception of the bus-
lack of management of para-transit modes transport service. It has been identified that
such as school vans and three wheelers are over 90% of the road space is used by private
identified as the the major transport related and hired vehicles which served only 35% of
issues in Kandy [5]. Therefore, to minimize total travel demand. Therefore, it is clear that
these problems, improving the public attracting commuters to the public transport
transport system in Kandy becomes more and system is crucial in reducing traffic
more important. congestion. Further, it is reported that the
existing bus transport system is not
To attract more commuters toward the public satisfactory for both existing commuters and
transport system, it is very important to private vehicle users. In addition, the major
identify the factors affecting the mode choice discouraging reasons to use public transport
behaviour of the commuters. These factors areinconvenience, lack of seating facility,
can be categorized into three main categories delay and lack of an alternative.
as socio-economic characteristics, transport
system variables and attitudinal variables[6]. Ashalatha et al. (2013) studied the mode
Among the three categories, socio-economic choice behaviour of commuters in
characteristics and transportation system Thiruvananthapuram city by using the
variables have a significant effect on the mode multinomial logistic regression [6]. The study
choice behaviour of commuters. was restricted only for the work trips in the
city. Three categories of variables were
Therefore, this research is focused on the considered for the study, socio-economic
effect of socio-economic characteristics and variables, transport system variables and
transportation system variables on the mode attitudinal variables. The choice behaviour of
choice behaviour of the commuters in Kandy, modes, car, two-wheeler and bus, was
Sri Lanka, by using factor analysis and evaluated and compared with each other. The
creating a discrete choice model based on a findings revealed that, when age increases, the
well establish utility function. preference for car increases and preference for
two-wheeler decrease compared to public
2. Literature Review transport. An increase in time per distance
and increase in cost per distance cause the
Traffic congestion is a timely recognized issue commuters to shift from public transport to
by researchers due to its impact on economic private modes. A similar kind of study was
growth and the quality of human life. Even done by Maduwanthi et al. (2015) to identify
though researchers have come up with the factors affecting the selection of the
different solutions over the years, traffic transport mode for the regular activities of
congestion continues to grow. One such commuters in the aspects of personal

ENGINEER
ENGINEER 2
52
characteristics and travel based characteristics explains the travel behaviour of the decision-
[8]. From factor analysis using SPSS software, makers when choosing public transport [12].
four main substantial factors which are Five latent variable sets were used to model
influenced to choose the travel mode were the public transport mode choice behaviour.
identified as income, vehicle ownership, The five variables were convenience, personal
safety and comfort. safety, modal comfort, service environment
and waiting feelings. The model formation
Ten significant factors which affect the mode was done in two stages as (1) the latent
choice behaviour of the commuters were variable structural equation model and the
identified by Maduwanthi et al. (2016) latent variable measurement equation model
considering the Colombo metropolitan area and (2) the mode choice model.
[9]. The factors are the number of earning
members in the family, vehicle ownership, As a first step to identify the factors affecting
education, age, gender, occupation, trip the mode choice behaviour of the commuters
distance, trip time, total cost and safety of the in Kandy, Sri Lanka, this study focuses on the
mode. The analysis was done by using the personal characteristics and travel-based
binary logistic method and SPSS software was characteristics which are affecting the mode
used to estimate the model through the choice behaviour of the commuters
maximum likelihood function.
3. Methodology
Ding and Zhang (2016) studied the travel
behaviour of individual travellers based on 3.1 Study Area
their personal characteristics in the central Kandy district contains 20 Divisional
business district (CBD) in Nanjing City of Secretariat Divisions (DSDs). Figure 1
China [10]. Two-mode choices were graphically illustrates the administrative
investigated as public transport and car. The boundaries of the Kandy district with DSDs.
personal information of the travellers was Five class-A roads are connected with Kandy
collected through a revealed preference district, Kandy – Colombo (A001) in the
survey and a stated preference survey. south-west direction, Peradeniya - Badulla –
Gender, occupation, income and car Chenkaladi (A005) in the south direction,
ownership were considered as the personal Kandy – Jaffna (A009) in the north direction,
information while mode choice, waiting time, Katugastota - Kurunegala – Puttalam (A010)
walking time, in-vehicle travel time, fare and from the north-west direction and Kandy -
comfort were considered as travel Mahiyangana – Padiyatalawa (A026) from the
information. The travel mode choices were south-east direction. 24 Class-B roads are
estimated using a discrete choice model and connecting to the Kandy district covering a
compared with the mode choices in the total length of 274.12 km. The population in
revealed preference survey and a stated the Kandy district is 1.38 million and the
preference survey. The study revealed that the population density is 715 per sq. km.
mode choice estimation using individual
grouping was remarkably higher than without
grouping.

Chiu Chuen et al. (2014) studied the possible

mode shift of travellers in the Klang Valley
towards public transport based on the utility
function of available transport modes through
a review preference survey [11]. The results of
the study revealed that travellers' likelihood of
choosing the car mode depends on the
parking cost, car prices and fuel prices. The
factors affecting the public mode choice is
identified as, transport network, accessibility Figure 1 – Administrative boundaries of the
distance, transit time and in-vehicle travel Kandy district
time.

Jian and Shoujie (2017) have developed an

improved discrete choice model that better

3
53 ENGINEER
ENGINEER
3.2 Sample Size behaviour of the commuters as bus, train,
The sample size for the study was selected bicycle, motorbike, three-wheeler, car, van,
based on the population in the district jeep, lorry, taxi (three-wheeler), taxi (car) and
according to Solvin's formula as shown in other modes.
Eq.(1)
𝑁𝑁
𝑛𝑛 = …(1) The third part of the questionnaire considered
1+𝑁𝑁𝐸𝐸 2
where, the public opinion about the bus-based public
𝑛𝑛 = Sample size transportation system in Kandy district, Sri
𝑁𝑁 = Population size Lanka. It consisted of 12 questions with a five
𝐸𝐸 = Error tolerance point Likert scale to rate the answer between
never and always. The 12 questions were
The confidence interval was selected as 95% identified using a pilot survey to describe
which implies that the margin of error is 5%. three main latent variables, operator
The population of the Kandy district is behaviour, comfort level and service quality.
1,375,382. The comments from the respondents about the
1,375,382 bus-based public transportation system were
𝑛𝑛 = 2
1+1,375,382×(0.05) obtained in this section. The respondents for
= 399.88  the third section were selected from the
≈ 400 commuters who are using the bus-based
 public transport system daily, several times a
Hence, the minimum sample size should be week or several times a month. Responses
400 respondents. However, the sample size of from the commuters who use the public buses
the study was taken as 513 respondents who rarely and very rarely are omitted in the third
are more than 16 years old. section of the questionnaire.
3.3 Data Collection The secondary data for the study was
Data collection was done in two categories as
collected through published data. Train
primary data and secondary data. Individual
schedules, train fares, bus fares, fuel
mode choice of the commuters who live in
consumption for different vehicle categories,
Kandy district was collected through a
fuel prices, and fare rates of taxi (three-
questionnaire. Participants were randomly
wheelers) and taxi (cars) were obtained from
selected and all of them are more than or
published literature and authority websites as
equal to 16 years old.
secondary data.
The questionnaire included three main 3.4 Factor Analysis
sections. The first section was about the social Before the statistical analysis using the SPSS
demographic characteristics of the commuters software, collected responses were screened to
to gain a general view of the personal profile. ensure the reliability and completeness of the
It included gender, age, educational data. Responses of the trips within the Kandy
background, occupation, monthly income, district were selected for the analysis and the
number of family members, number of others were removed. After screening, the
earning members in the family, vehicle responses were coded into SPSS. Factor
ownership and driving licence ownership. analysis was used as the data analysis
method.
The second section of the questionnaire
considered the travel behaviour of the Factor analysis is a collection of a procedure
commuters which included travel mode, for analysing the relations among a set of
starting time of the trip, origin, destination, measured or observed variables for each
trip distance, in-vehicle travel time, and total individual of a group. Factor analysis
travel cost, commonly for both private and examines the interrelationships among a large
public transport users. Additionally, for the number of variables, then attempts to explain
commuters, who are using public transport them in terms of their common underlying
modes, access time, mode of access, egress dimensions.
time, and mode of egress were collected.
Moreover, from private transport users, There are a few assumptions that should be
parking mode, parking cost and the reasons to completed before starting the factor analysis.
choose private transport instead of a public The main three assumptions are, (1)
mode were obtained. 12 vehicle categories Correlation among the variables should be
were used when obtaining the travel

ENGINEER
ENGINEER 4
54
strong. (2) Variables should be spread travel mode. The probability of the decision-
normally. (3) Comparison index between the maker choosing an alternative can be
correlation coefficient and partial correlation calculated using Eq. (3).
coefficient should be small.
𝑒𝑒 𝑈𝑈𝑘𝑘
𝑝𝑝(𝑘𝑘) = ∑ 𝑈𝑈 … (3)
3.5 Parameter Estimation 𝑥𝑥 𝑒𝑒 𝑥𝑥

Multinomial Logit (MNL) Model is one of the

widely used methods to analyse mode choice where𝑝𝑝(𝑘𝑘) is the probability of the decision-
assessments. The mathematical form of a maker choosing alternative 𝑘𝑘 and 𝑈𝑈𝑘𝑘 is the
discrete choice model is determined by the utility of alternative 𝑘𝑘.
assumptions made regarding the error
components of the utility function. Three The parameter estimation is done by
assumptions lead to the MNL model as, (1) maximizing the log-likelihood function. The
the error components are extreme-value (or log-likelihood can be calculated using the
Gumbel) distributed, (2) the error components Eqns.(4), (5) and (6).
are identically and independently distributed
(IID) across alternatives and (3) the error If a sample of N decision-makers is obtained
components are IID across for estimation. The probability of a person n
observations/individuals. The most common choosing the alternative that he/she was
assumption for error distributions is the actually observed to choose can be expressed
normal distribution. However, in the case of using Eq.(4).
choice models, it creates numerical problems.
Hence, the Gumbel distribution is selected ∏𝑖𝑖(𝑃𝑃𝑛𝑛𝑛𝑛 ) 𝑦𝑦𝑛𝑛𝑛𝑛 …(4)
due to computational advantages and it
approximates the normal distribution. where𝑦𝑦𝑛𝑛𝑛𝑛 = 1, if person 𝑛𝑛 chooses𝑖𝑖 and zero
otherwise. It simply gives the probability of
Discrete choice models have usually been the chosen alternative.
derived under an assumption of utility-
maximizing behaviour by the decision-maker. Assuming that each decision-maker's choice is
The decision-maker would obtain a certain independent of others, the probability of each
level of utility (or profit) from each person in the sample choosing the mode that
alternative. This utility is known to the he/sheactually using is calculated from Eq.(5).
decision-maker but not to the researcher. The
decision-maker chooses the alternative that 𝐿𝐿(𝛽𝛽) = ∏𝑁𝑁
𝑛𝑛=1 ∏𝑖𝑖 (𝑃𝑃𝑛𝑛𝑛𝑛 )
𝑦𝑦𝑛𝑛𝑛𝑛
…(5)
provides the greatest utility. The researcher
does not observe the utility of the decision- where𝛽𝛽 is a vector containing the parameters
maker, but instead observes some attributes of of the model. The log-likelihood function is
the alternatives such as travel time, access expressed as in Eq.(6).
time, waiting time, egress time, travel cost etc.
Some of the socio-economic characteristics of 𝐿𝐿𝐿𝐿(𝛽𝛽) = ∏𝑁𝑁
𝑛𝑛=1 ∏𝑖𝑖 𝑦𝑦𝑛𝑛𝑛𝑛 ln(𝑃𝑃𝑛𝑛𝑛𝑛 ) …(6)
the decision-maker such as, gender, age,
monthly income, education level, etc. are also The goodness of fit for the model is measured
observed by the researcher. Since there are using the likelihood ratio index (𝜌𝜌2 ) value.
aspects of utility that the researcher does not The likelihood ratio index value is calculated
or cannot observe, the utility is decomposed using Eq.(7).
by including a random error term.
̂)
𝐿𝐿𝐿𝐿(𝛽𝛽
𝜌𝜌2 = 1 − …(7)
𝐿𝐿𝐿𝐿(0)
Including all these factors, the utility function
can be expressed as in Eq.(2). where𝐿𝐿𝐿𝐿(𝛽𝛽̂ ) is the value of the log-likelihood
function at the estimated parameters and
𝑈𝑈 = 𝛽𝛽0 + 𝛽𝛽1 𝑋𝑋1 + 𝛽𝛽2 𝑋𝑋2 + ⋯ + 𝛽𝛽𝑛𝑛 𝑋𝑋𝑛𝑛 + 𝜀𝜀(2) 𝐿𝐿𝐿𝐿(0) is the value when all the parameters are
set equal to zero.
where𝑈𝑈 is the utility, 𝑋𝑋1 , 𝑋𝑋2 , 𝑋𝑋3 , … are attributes
and 𝛽𝛽0 , 𝛽𝛽1 , 𝛽𝛽2 , … are parameters, and𝜀𝜀 The problem with the likelihood ratio index is
represents the random error term. that it improves with the increase of the
variables, even though the variables do not
The logit model is used to calculate the have any relationship with the output results.
proportion of trips that will select a specific Therefore, the likelihood ratio index is

5
55 ENGINEER
ENGINEER
modified as an adjusted likelihood ratio index The cross elasticity of a selected mode can be
(𝜌𝜌̅ 2 ) as shown in Eq.(8). calculated using Eq.(11).
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 = −𝛽𝛽𝑘𝑘 𝑋𝑋𝑖𝑖𝑖𝑖 𝑃𝑃𝑖𝑖 … (11)
̂ )−𝑘𝑘
𝐿𝐿𝐿𝐿(𝛽𝛽
𝜌𝜌̅ 2 = 1 − … (8)
𝐿𝐿𝐿𝐿(0)
where 𝛽𝛽𝑘𝑘 is the parameter of the 𝑘𝑘 𝑡𝑡ℎ attribute,
𝑋𝑋𝑖𝑖𝑖𝑖 is the 𝑘𝑘 𝑡𝑡ℎ attribute of the alternative 𝑖𝑖 and
where 𝑘𝑘 is the number of degrees of freedom
𝑃𝑃𝑖𝑖 is the probability of choosing the alternative
(parameters) used in the model. 𝑖𝑖.
The validity of the estimated parameters was
tested using the t - statistic value of each 4. Results
parameter. The t- statistic values were
4.1 Factors Influencing Mode Choice
calculated using Eq.(9).
Behaviours of Commuters
̂𝑘𝑘 −𝛽𝛽 ∗
𝛽𝛽 The factors influencing the mode choice
𝑡𝑡 − 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 = 𝑘𝑘
…(9)
𝑆𝑆𝑘𝑘 behaviour of the commuters are analysed
where 𝛽𝛽̂𝑘𝑘 is the estimate for the 𝑘𝑘 𝑡𝑡ℎ parameter. using the factor analysis under two sections.
𝛽𝛽 ∗ 𝑘𝑘 is the hypothesized value for the 𝑘𝑘 𝑡𝑡ℎ The first section contains the Personal
Characteristics (PC) of the commuter while
parameter and𝑆𝑆𝑘𝑘 is the standard error of the
the second section contains the Travel based
estimate.
Characteristics (TC) of the commuters.
The null hypothesis is rejected with a higher Although twelve vehicle categories were
absolute value of the t-statistic which implies considered in the questionnaire, bus, car and
that the parameter is equal to the motorbike show the highest percentage
hypothesized value. The low absolute value of among all modes which is 93% of the total
the t-statistics implies that the variable does share. Therefore those three modes were
not contribute significantly to the explanatory considered for the data analysis.
power of the model and can be considered for
exclusion. The accuracy of the factor analysis is checked
by using the KMO (Kaiser-Meyer-Olkin) score
3.6 Elasticity Analysis and Bartlett's significant score. KMO statistic
Choice probabilities in logit models are a is a value that measures sampling adequacy. It
function of the values of the attributes that takes a value from zero to one. Values close to
define the utility of the alternatives. Therefore zero are considered not suitable, while values
it is useful to know the extent to which the close to one are acceptable. The KMO score
probabilities change in response to changes in and Bartlett's significant score for each mode
the values of associated attributes. One are shown in Table 1.
measure for evaluating the response to Table 1 – Model Estimation Results
changes is to calculate the elasticity of the Travel KMO and Bartlett's Value Value
choice probabilities of each alternative for the Mode Test for PC for TC
variable in question. Kaiser-Meyer-Olkin
Measure of 0.494 0.676
Elasticity can be identified in two sections as Sampling Adequacy.
Public
(1) direct elasticity and (2) cross elasticity. Bus
Approx. Chi-Square 316.267 760.137
Bartlett's Test of
36 21
Direct elasticity measures the percentage Sphericity df
change in the probability of choosing a sig. 0.000 0.000
particular alternative in the choice set with Kaiser-Meyer-Olkin
Measure of 0.526 0.690
respect to a given percentage change in an
Sampling Adequacy.
attribute of that same alternative. The direct Motor
Approx. Chi-Square 240.731 355.193
elasticity of a selected mode can be calculated Bike
Bartlett's Test of
using Eq.(10). 36 15
Sphericity df
sig. 0.000 0.000
𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 = 𝛽𝛽𝑘𝑘 𝑋𝑋𝑖𝑖𝑖𝑖 (1 − 𝑃𝑃𝑖𝑖 ) ...(10) Kaiser-Meyer-Olkin
Measure of 0.619 0.617
Cross elasticity measures the change in the Sampling Adequacy.
probability of choosing a particular alternative Car Approx. Chi-Square 264.392 436.038
in the choice set with respect to a given Bartlett's Test of
36 15
percentage change in a competing alternative. Sphericity df
sig. 0.000 0.000

ENGINEER
ENGINEER 6
56
The KMO score is almost greater than 0.5 for public transportation. Therefore, vehicle
each travel mode which implies that the ownership and driving licence ownership
analysis can be continued for all the travel come under one category. When the number
modes. Bartlett's score is 0.000 for each mode of family members is increasing, although the
which indicates a strong correlation among family owns a private vehicle, some of the
the variables. members have to use the public transport
service to fulfil their daily travel requirements.
The main public transportation system inside Therefore, when the number of family
the Kandy district is a bus-based public members is increasing, public transport usage
transportation system. Although there are also increases. When the number of earning
train services, a very limited number of trains members is high, the number of members
can be seen within the district. Therefore, only leaving the home is high. Therefore, some of
public bus transportation is considered for the the members tend to use the public
analysis. The factors influencing public bus- transportation service. Most of the ladies are
based transportation are divided into two using public transportation than males and
main segments for the analysis as PC and TC. most of the highly educated people are using
their private transport modes in Sri Lanka.
After the analysis, nine variables for the PC
are reduced into four latent variables and Table 3 represents the rotated component
eight variables for the TC are reduced into loadings of bus transport for TC.
three latent variables. Tables 2 to 7 show the
components which contribute a significant Table 3 – Rotated Component Loadings of
value of 50% for the total variance and their Bus Transport for TC
Eigenvalues are greater than one. Tables 2 to 7
show the final scores of each variable after the Travel Component
Direct Oblimin rotation. Characteristics 1 2 3
Trip Distance 0.969
Table 2 represents the rotated component
loadings of bus transport for PC. In-Vehicle Travel
0.935
Time
Table 2 – Rotated Component Loadings of Travel Cost 0.879
Bus Transport for PC Waiting Time 0.664
Egress Time 0.622
Personal Component
Characteristics 1 2 3 4 Access Time 0.607
Occupation 0.886 Trip Purpose 0.879
Age of the
0.879 Total Variance 69.61
Respondent
Vehicle
0.771
Ownership According to Table 3, most of the commuters
Licence
Ownership
0.764 are considering the trip distance, travel time
Number of and travel cost when selecting public
0.727 transportation as the mode of transportation.
Family Members
Number of Most of the people who own a motorbike or
Earning
0.711 three-wheeler tend to use public transport if
Members in the the trip distance is high. Since the cost of
Family
private transport is higher than public
Gender of the
Respondent
0.715 transport, most of the commuters are using
Educational public transport when the trip distance is high
-0.518
Background to reduce the total cost. Commuters consider
Total Variance 63.55 the time reliability and the waiting time when
choosing public transport. Therefore, access
According to Table 2, the occupation and age time, waiting time and egress time are
of the respondent come under one factor. important parameters for the mode choice
Most of the young people tend to use public variability of the commuters. The trip purpose
transportation as they do not own a private is also a considerable factor in the mode shift
vehicle or they are students. The people who of the commuters.
owned a private vehicle and have a driving
licence tend to use a private vehicle than use

577 ENGINEER
ENGINEER
Table 4 and Table 5 represent the rotated commuters. Parking facilities within the city
component loadings of the motorbike for PC limits are very poor. Therefore, people are
and TC, respectively. thinking about the parking mode and the
parking cost when choosing the motorbike as
Table 4 – Rotated Component Loadings of the mode of transport.
the Motorbike for PC
Personal Component Table 6 and Table 7 represent the rotated
Characteristics 1 2 3 4 component loadings of Car for PC and TC,
Licence respectively.
0.840
Ownership
Gender of the
0.776 Table 6 – Rotated Component Loadings of
Respondent
Car for PC
Vehicle
0.675
Ownership
Number of Component
Personal Characteristics
Earning 1 2 3
0.792
Members in the Occupation 0.881
Family Age of the Respondent 0.803
Educational Monthly Income 0.740
0.703
Background Number of Family
0.851
Age of the Members
-0.817
Respondent Number of Earning
0.833
Monthly Income 0.730 Members in the Family
Occupation 0.729 Licence Ownership 0.575
Number of Family Gender of the
0.936 0.763
Members Respondent
Total Variance 72.44 Vehicle Ownership 0.661
Total Variance 62.22
Table 5 – Rotated Component Loadings of
the Motorbike for TC
Table 7 – Rotated component loadings of car
Travel Component for TC
Characteristics 1 2 3
Trip Distance 0.966 Travel Component
In-Vehicle Travel Characteristics 1 2 3
0.948
Time Trip Distance 0.975
Travel Cost 0.918 In-Vehicle Travel
0.944
Parking Cost 0.867 Time
Parking Mode 0.619 Travel Cost 0.891
Trip Purpose 0.899 Trip Purpose -0.827
Total Variance 83.82 Parking Mode 0.729
Parking Cost 0.908
In Sri Lanka, most of the males tend to use the Total Variance 83.61
motorbike as their mode of transportation.
With the lower cost and the lower travel time Many high-income households are using the
in motorbikes, many commuters are shifting car as their mode of transport. Therefore,
to motorbikes instead of public transport. occupation and income significantly affect
Majority of educated people and elderly choosing the mode of car. Main TC factors
people are not willing to use motorbikes due affecting choice of the car are trip distance, in-
to low comfort level. Occupation is directly vehicle travel time and travel cost. As the
affecting the monthly income and mostly the comfort level of the car is high, many
middle-class households are using motorbikes commuters are choosing the car for long-
as their mode of transportation. distance trips with a higher travel time.

Commuters are considering the trip distance Figure 2 shows the factors affecting the mode
and the travel time when choosing the choice behaviour of the commuters for the
motorbike as their mode of transport as the three modes, bus, motorbike and car under
comfort level of the motorbike is low. Parking the two categories PC and TC.
mode and the parking cost also significantly
affect the private mode choice of the

ENGINEER
ENGINEER 8
58
4.2 Parameter Estimation than 1.96 for each parameter other than the
A well-developed discrete choice model is travel cost and that implies the parameters are
used to analyse the mode choice behaviour of significant with more than a 95% significant
the commuters within the Kandy district. The threshold. For the travel cost parameter,
parameter estimation was done using RStudio absolute value of the t-statistics is greater than
statistical software for the parameters, travel 1.645 which implies that the parameter is
time, travel cost, access time, waiting time and significant with more than a 90% significant
egress time. Bus, car and motor bicycle modes threshold. The respective sign of the
were used for the parameter estimation as parameters is also used to show the accuracy
they are the governing transport modes of the of the model along with the 𝜌𝜌 2 and t-statistic
city. The respective estimation results are values. The positive sign of a parameter
shown in Table 8. implies that the option has a beneficial effect
on the choice and the option is likely to be
Table 8 – Model Estimation Results chosen. The negative sign of the parameter
implies that the option is not likely to be
The absolute value of the t-statistics is greater chosen. The likelihood ratio index of the
model is 0.236 and the adjusted likelihood
Estimate
t-statistic ratio index is 0.212 which represents a decent
value model fit for a discrete choice model.
Travel Time -0.0193 -2.55
Parameters, access time, waiting time, travel
Travel Cost -0.0012 -1.74 time, and egress time show negative signs.
Access Time -0.0741 -2.49 Commuters are always trying to reduce these
parameters in their trips to save time.
Waiting Time -0.0589 -3.26 Therefore, when the total travel time
Egress Time -0.1554 -5.53
increases, the likelihood of the choice of that
mode will decrease. The next considering
Sample 513 factor is the travel cost. Commuters are more
likely to use a mode with the lowest travel
Initial LL -289.74
cost. When the travel cost increases, the
Final LL -221.23 likelihood of the mode choice decreases. The
negative sign of the travel cost parameter in
𝜌𝜌2 0.236
Table 8 implies this phenomenon.
𝜌𝜌̅ 2 0.212

Figure 2 – Factors Influencing the Mode Choice Variability of the Commuters

9
59 ENGINEER
ENGINEER
 The parameters, access time, waiting time and system is for the travel cost where 0.01% of the
egress time are related only to public transport. commuters are leaving the bus transport
But travel time and travel cost are related to system for an additional 1 LKR in bus fare.
both public and private transport modes. With the results, it can be seen that the
Therefore, more attention should be paid to commuters are considering the travel time
above three parameters to attract more rather than the cost when choosing the public
commuters towards public transport modes. bus-based transport system.
Route coverage of public transportation can be
increased to reduce the access time and egress When considering the private transport modes
time of the commuters. More public transport which are motorbikes and cars, the highest
services can be implemented in rural areas to elasticity can be seen in travel time than travel
increase route coverage [1]. With that, people cost. 0.26% of commuters are leaving the
may get the chance to use alternative routes for motorbike mode for an additional minute of
their trips with public transport modes. The travel time where it is 0.22% for cars. The
waiting time of the passengers can reduce by highest percentage of commuters, which is
increasing the frequency of the public transport 0.12% of commuters, are leaving the car for an
service. additional 1 LKR in car. This implies that most
car users are willing to use buses or motorbikes
4.3 Elasticity Analysis with the increase in the cost.
Elasticity analysis for the choice set was
performed to measure the percentage change in According to the direct analysis results, it can
the probability of choosing a particular be identified that the total travel time of the bus
alternative with respect to a given percentage transport system should be reduced to attract
change in the same or in a competing more commuters. Route coverage can be
alternative. Elasticity analysis was done for all improved to reduce access time and egress time
the five attributes discussed above for the three while bus frequency can be increased to reduce
alternative modes,bus, motorbike and car. the waiting time of the passengers.

Table 9 shows the percentage values of direct Cross elasticity can be used to measure the
elasticity analysis for each alternative. change in the probability of choosing a
particular alternative in the choice set with
Table 9 - Direct Elasticity Values respect to a given percentage change in a
competing alternative. Table 10 shows the cross
Travel Travel Access Waiting Egress elasticity values for each alternative.
Alternat
Time Cost Time Time Time
ive
(%) (%) (%) (%) (%)
Table 10 - Cross Elasticity Values
Bus -0.20 -0.01 -0.18 -0.20 -0.54
Motor Alternative Travel Time (%) Travel cost (%)
-0.26 -0.05 - - -
Bike
Car -0.22 -0.12 - - - Bus 0.65 0.04
Motor Bike 0.41 0.07
According to Table 9, all the elasticity values
are obtained as negative values which says that Car 0.46 0.33
the increase of each of the alternatives is
reducing the attraction of the commuters to the According to Table 10, all the elasticity values
relevant mode. The value of -0.20% for the are obtained as positive values which says that
travel time of the bus represents that for an the increase of each of the alternatives in a
additional one minute of travel time in bus, mode is increasing the attraction of the
0.20% of commuters are leaving the bus commuters towards another mode. The highest
transport mode. The highest elasticity value for cross elasticity value for the travel time can be
the bus-based public transport system can be seen in the mode bus which is 0.65. This
seen in the egress time. 0.54% of commuters are indicates that, when the travel time of the
leaving the bus transport system for an motorbike or car is increased from one minute,
additional minute of the egress time. This 0.65% of more commuters are attracted to the
implies that the route coverage within the bus-based public transport system. Therefore
Kandy district can be improved to attract more by reducing the total travel time of the bus
commuters to the public transport system. The transport system, more commutes can be
lowest elasticity value for the bus transport attracted to use the public transport. When
considering the travel cost, the highest elasticity

ENGINEER
ENGINEER 10
60
value can be seen in the mode car which is are significant and have achieved the expected
0.33%. Therefore, the increase of 1 LKR in bus negative sign which implies that increasing the
fares can move some more passengers towards value of each parameter decreases the
the private modes. likelihood of choosing the relevant mode.

According to the results in Table 10, it can be Elasticity analysis was done for the three modes
seen that the best way to attract commuters to using the estimated parameters for each
the public transport system is reducing the attribute. From the elasticity analysis, it was
travel time of the bus transport system rather identified that more commuters can be
than reducing the bus fare. The travel time of attracted towards the public transport system
the bus transport system can be done by by reducing the total travel time of busses.
reducing the access time, waiting time and Route coverage of the bus service and the
egress time by increasing the route coverage frequency of the bus service can be improved to
and the bus frequency. reduce the travel time of the commuters [13].

5. Conclusions Accordingly, the relevant authorities can use

these results to identify the improvements
The study was conducted to identify the mode which have to be done in public transport
choice behaviour of the commuters in Kandy services to attract more commuters towards
district, Sri Lanka. Trip data was collected by public transportation. Some studies can be done
using online and face to face interviews to identify the service intensity of public
through a structured questionnaire. According transport service in the Kandy district and to
to the findings, the highest percentage of identify the measures which can be taken to
people are using buses, cars and motorbicycles provide a better public transport service to the
as their mode of transportation. About one- people. Furthermore, the results can be
third of the population is using public transport improved with more trips and some more
modes for their travel purposes within the modes can be used for the estimation.
district. Kandy is attracting more work and
educational trips compared to other purposes. References
513 respondents were selected for the study
within the Kandy district. 1. Lles, R., 2005 Public Transport in Developing
Countries. Oxford: Elsevier.
Factor analysis was used to analyse the data.
2. Central Bank of Sri Lanka (2019) Annual Report,
Two main categories of factors were considered
pp. 112-116.
as Personal Characteristics (PC) and Travel
Characteristics (TC). Three vehicle categories 3. University of Moratuwa in association of
were considered for the study as public bus, University of Peradeniya (2016) - Kandy City
motorbike and car. Transport Study.

According to the results, age and occupation 4. Wang, W., 2020. Project Information Document-
are the considerable factors that significantly Kandy Multimodal Transport Terminal
Development Project-P172342.
affect the public mode choice under the PC
category, while for the TC category, it is trip 5. Kumarage, A.S., 2001. Strategic Plan for Improving
distance, in-vehicle travel time and travel cost. Transport & Traffic in Kandy.
Gender, age, occupation and monthly income
are significantly influencing the choice of a 6. Ashalatha, R., Manju, V.S. and Zacharia, A.B.,
private transport mode under the PC category. 2013. Mode Choice Behaviour of Commuters in
Same as public transport, trip distance, in- Thiruvananthapuram City. Journal of
vehicle travel time and travel cost are the Transportation Engineering, 139(5), pp.494-502.
factors that are significantly affecting the
7. Talagala, P. & Kalukottege, C., 2014. Factors
private mode choice under the TC category.
Affecting Public Bus Transport Ridership: A
Case of Capital Colombo. In International Forum
The multinomial logit model was used for the for Mathematical Modeling.
parameter estimation with the utility
maximization theory. Travel time, travel cost, 8. Madhuwanthi, R.A.M., Marasinghe, A., RPC, J.,
access time, waiting time and egress time were Dharmawansa, A.D. and Nomura, S., 2015.
identified as the key parameters for the mode Factors Influencing Travel Behavior on
choice behaviour. All the estimated parameters Transport Mode Choice-A Case of Colombo

11
61 ENGINEER
ENGINEER
 Metropolitan Area in Sri Lanka. International
Journal of Affective Engineering, [Link]-D.

9. Madhuwanthi, M., Marasinghe, A. and Nomura,

S., 2016. Appraising the Strongly Associated
Impacts to Choose the Mode of Public
Transport: A Case Study of Colombo
Metropolitan Area in Sri Lanka. International
Journal for Traffic & Transport Engineering, 6(1).

10. Ding, L. and Zhang, N., 2016. A Travel Mode

Choice Model Using Individual Grouping Based
on Cluster Analysis. Procedia Engineering, 137,
pp.786-795.

11. Chiu Chuen, O., Karim, M.R. and Yusoff, S.,

2014. Mode Choice Between Private and Public
Transport in Klang Valley, Malaysia. The
Scientific World Journal, 2014.

12. Chen, J. and Li, S., 2017. Mode Choice Model for
Public Transport with Categorized Latent
Variables. Mathematical Problems in Engineering,
2017.

13. Lowe, W.U.A., Herath, P.M.G.D.M., Edirisinghe,

A.G.H.J., Dharmarathna, W.R.S.S. &
Wickramasinghe V, (2021) Analysis of Network
Features and Travel Characteristics for
Enhancing Bus - Based Public Transport System
– Case Study in Kandy, Sri Lanka. Journal of
Eastern Asia Society for Transport Studies (EASTS),
2022.

ENGINEER
ENGINEER 12
62