LUẬN văn THẠC sĩ impacts of BRT introduction on commuter behavior in hanoi

INTRODUCTION

Research background

In each country, especially in the developing stage, accompanying with the economic development is the rapid development of the number of people in major cities People are coming to big cities for job opportunities and better living conditions They are one of the major factors in the development of the country and exist in all countries of the world However, the concentration of the population in major cities has its disadvantages The higher the population, the greater the demand for transportation Therefore, the city traffic system development and traffic jam control become the top priority task in urban development

Bus Rapid Transit (BRT) has become a preferred worldwide transit mode

BRT is considered to be a cost efficiency transit mode to create a high-performance transport service due to its characteristics of service capacity, inexpensiveness, network synchronization and flexibility Many countries in South America, Latin America and Europe have BRTs existed and played an important role in their public transport system Some developing Asian cities also consider BRT in their public transport planning because of its advantages of low cost of construction, short preparation time and flexible deployments over rail systems Additionally, BRT is mentioned to reduce local air pollutant emission for developing cities since BRT would encourage people to shift from using private vehicle to public transport sector which emits smaller amount of CO2 However, there are different between the characteristics of Asian developing cities with those successful BRT implementing cities such as the level of urban development due to city planning and high private vehicle utilization due to lacking of public transport Some previous studies have indicated that it would be challenging to achieve high modal shift to BRT in these developing countries due to personal vehicle’s higher accessibility, mobility and cheapness comparing with existing public transport On the other hand, there are also several studies show that BRT has successful attracted car passengers and motorcycle users to change their mode choice towards public transport (BRT) in developing cities To summarize, there is good potential to encourage modal shift from private vehicle to BRT but it is also not easy due to current advantages of private vehicle over the poor service existing public transport in Asia developing cities

In Hanoi, public transport development is considered to be a fundamental solution to address prolonged congestion and is increasingly caused by too many personal vehicles There are many types of public transport currently under development priority in Hanoi Bus Rapid Transit (BRT) with its advantage of lower investment cost and adjustable deployments to respond to altered circumstances or conditions becomes an inevitable solution of Hanoi's public transport development

Hanoi has its own BRT (Figure 1.1) with 8 rapid bus routes and 3 transitional routes planned with the expectation of addressing traffic congestion in this rapidly growing population of this city Hanoi BRT is introduced by People's Committee of Hanoi and operated by Hanoi Express Bus Company (Hanoi bus) Currently only the first route - BRT route 01: Kim Ma bus station - Yen Nghia bus station (Figure 1.2) which is 14.7 km long has been put into operation since December 31, 2016

The first BRT route run at a frequency of 5-15 minutes per trip, capacity of 90 guests, 4 doors, travel speed of 22 km/h with ticket price of 7000 VND In addition to the goal of reducing traffic congestion in Hanoi city, another important and expected task of Hanoi BRT is to change the perception of traffic participants about the use of public transport through its advantages

However, unlike the world, the effect of BRT in Hanoi somehow remains unclear It can be observed that there are lacking of studies about Hanoi BRT in general and the Hanoi BRT effects in particular The BRT system is proved its undeniable and important impacts on improving transportation system as well as changing commuter behavior to use public transport through studies from all over the world However, Hanoi BRT is now also received the mix comments about its effectiveness and there is a lot of controversy surrounding the investment efficiency of the project This situation happened may be because the BRT has just been put into operation in early 2017 so the number of researches about it is still limited

Therefore, the desire of this research is to understand the commuter behavior after BRT implementation in Hanoi, to see if BRT implementation has change the commuter travel behavior (mode choice and walking behavior) and evaluate quantitavively the true impact of BRT introduction in Hanoi through data collected by surveying This is further reinforced in the literature review process and carefully indicated as in the next chapters to illustrate how the story of this study was developed

The research questions for my study are “Does Hanoi BRT advantages enough to encourage commuters to change their choice of transportation mode?” and “People that using Hanoi BRT, have their commute behavior been changed (from using personal vehicle and from normal bus to Bus Rapid Transit)?”

Figure 1.1 Deployment of Hanoi BRT

Research Framework

This dissertation consists of 8 chapters This chapter focuses on the background with a review of the area being researched as well as current information surrounding the possible issues for researching The following chapters are written as follows:

Chapter 2 literature reviews which overview previous researches in the world and also review the current status of researches in Vietnam about BRT This chapter aims to understand the situation of researching about BRT in Vietnam

Chapter 3 includes the problem statement, research question and research objective which are introduced respectively based on the literature reviews and understanding from it

Chapter 4 aims to develop the hypotheses in order to clarify the research objective which partly based on the key findings of some researches reviewed

Chapter 5 evaluates and provide appropriate methodology for the research as well as introduce the formula that used for analysis (the approach)

Chapter 6 describes how the survey was designed and conducted It also describes the descriptive statistic of the survey data

Chapter 7 includes all the analysis with chosen approach on the data Based on the analysis results, whether the hypotheses are supported or not is clearly declared through the discussion

Chapter 8 concludes the key findings obtained from the hypotheses though analysis

Finally, research limitations and future feasibility studies were discussed.

LITERATURE REVIEWS

Studies on BRT effects in some countries

There are some countries has deployed the Bus Rapid Transit as a part of their public transport system As a result, there are many studies from all over the world about BRTs effectiveness and how BRT affects the transportation system

Krizek and El-Geneidy (2007) stated that faster transport services such as express, limited and Bus Rapid Transit (BRT) services were increased implementation by transit agencies Their study indicated that one of the main purpose of introducing better transport services in general and BRT in particular is to attract commuter from using personal vehicle The writer also mentioned the importance of operating environment understanding in introducing a new type of transportation Gutiérrez & Hidalgo (2012) summarized the Bus Rapid Transit (BRT) and Bus of High Level of Service (BHLS) growth around the world and its large positive impacts through a survey As indicated in this article, there were already 120 cities had their own BRT and BHLS system developed from all over the world by late 2011 The cities which had most successful BRT development can be listed as Mexico City, Curitiba, Bogota, Istanbul, Guangzhou and Ahmedabad

The BRT development in these cities was considered speedy implementation, cheap with high performance The increasing of BRTS’s popularity in the world was also reported by Deng and Nelson (2011) Darido (2006) reported development of BRTS and the cities growth where BRT is in operation in China The performance of BRT system in major cities in China was also evaluated by Li and Hino (2013) The performance of BRT system was also assessed by using probe data and signal timing data (Zou, Li, and Li, 2012) Godavarthi, Chalumuri, and Velmurugun

(2014) had also estimated this criterion but based on volume by capacity ratio

There are a large number of articles that show the positive impact of BRT on urban transport in general and commuter's modal shift in particular Hidalgo, et al.,

(2013) did an ex-post evaluation of TransMilenio BRT system in Bogota which confirmed the positive impact the implementation in reducing travel time and travel cost in public transport of the Bogota city Bajracharya (2008) evaluated the possible modal shift from the current mode to the proposed BRT system through a survey by combining the stated and revealed preference carried out The willingness to shift of people is calculated by using their choice after comparing the travel time and travel cost between their current transport mode and the proposed BRTS for their trip After data analysis, the results of overall figure pointed out about 34% of the respondents willing to shift from their normal mode to the BRTS Satiennam and Jaensirisak (2015) estimated the modal shift possibility of commuter from private vehicle to BRT in Khon Kaen city, in Thailand This study conducted a survey using Stated Preference (SP) Method and predicted the choice of personal vehicle user between minibus (MNB), BRT without P&R and feeder (BRT) and BRT with P&R and feeder (BRTS) by modeling The results showed that BRT had potential to attract notably personal vehicle users to change the mode choice The study also found that travel time and travel cost has importantly affect the modal shift to BRT When identifying the effectiveness of BRT, McDonnell and Zellner

(2011) found that the reserved bus lane addition improved the efficiency of a route by encouraging mode share and decreasing travel time for bus users However, these authors also indicated that BRT lane only become effective starting at larger population levels (medium and large size cities)

Further studies were reported to assess the effect of BRT on commuter travel behavior and level of perception Lemoine, et al., (2016) estimated the BRT effect on walking behavior by using agent based modeling Their study found that BRT’s speed as well as distance to BRT’s station had a big impact on the increasing walking for transportation It also more specifically mentioned that commuters walking behavior could be increase by enhancing BRT access although this effect might depend on the others modes of transport availability Okagbue, et al., (2015) used a short self-rated questionnaire to collect data of commuters using BRT in Lagos, Nigeria and then analyzed with multiple regression The regression analysis revealed the commuters’ satisfaction when using the BRT were affected in order respectively: the buses security, commuting time reduction, staff’s attitude towards commuters, drivers’ behavior, availability of buses, buses compactness, traffic congestion reduction, traffic regulations compliance of drivers, routes availability, tickets availability, roads conditions and seats quality Reading the studies about the factors affect the mode choice behavior also contributes to the understanding about commuter travel behavior Satiennam (2016) and Li Chang (2007) estimated the effect of psychological factors in Thailand and vehicle dependence factors in Taipei on this aspect, respectively.

Articles and studies about BRT effects in Vietnam

As operation began in December 31, 2016, limited studies have been reported about Hanoi BRT in Vietnam

There are some articles with findings that Hanoi BRT is not really effective

According to Urban Traffic Management and Administration Center statistics:

 In the first nine months of 2018, the BRT bus service only achieved 92,828 vehicles with 3.72 million passengers, an increase of 2.3% over the same period last year

 An average of 40.2 guests per turn while the standard capacity is 90 guests per turn This means that BRT buses are less than 50% efficient

In document no 1468/KL-TTCP of Prime Minister, there is statement about BRT service in Hanoi: “The investment in the project is not synchronized, has not created the benefits to encourage people to switch from personal vehicles to use public transport.”

Vu Anh Tuan (2015) studied about the mode choice behavior and modal shift to public transport with the case study of Hanoi city Exploring people’s travel behavior, Vu Anh Tuan (2015) claim that BRT not effective enough to attract large number of motorcycle and car user until MRT introduction However, this researcher’s evaluation was based on the Stated Preference method when BRT had not been implemented This could be highly affected the outcome of the study

However, Hanoi BRT has also received positive comments about it effectiveness Based on a survey of 5000 people using BRT in 2017, Mr Nguyen Hoang Hai – Director of Hanoi Urban Transport Management and Operation Center said that the 23% of the respondents said that they shifted from motorbike to use BRT In a conference of Hanoi city committee in June 2018, Mr Nguyen Cong Nhat, Vice General Manager of Hanoi transport corporation – Transerco (which control the operation of many bus routes in Hanoi) stated that many commuters shifted from their personal vehicles to use BRT because it saved 20-25% of travel time in compare with normal bus and was also more civilized at the same time According to him, it partly showed Hanoi BRT was working quite effectively.

HYPOTHESES, METHODOLOGY AND APPROACH

Hypotheses

In order to clarify the efficiency of Hanoi BRT, some hypotheses are proposed which partly based on the the key findings of some researches (which shown in literature review)

When identifying the effectiveness of BRT, McDonnell and Zellner (2011) found that the reserved bus lane addition improved the efficiency of a route by encouraging mode share and decreasing travel time for bus users However, these authors also indicated that BRT lane only become effective starting at larger population levels (medium and large size cities)

Lemoine, et al., (2016) estimated the BRT effect on walking behavior by using agent based modeling Their study found that BRT’s speed as well as distance to BRT’s station had a big impact on the increasing walking for transportation It also more specifically mentioned that commuters walking behavior could be increase by enhancing BRT access although this effect might depend on the others modes of transport availability

Satiennam and Jaensirisak (2013) estimated the modal shift possibility of commuter from private vehicle to BRT in Khon Kaen city, in Thailand This study conducted a survey using Stated Preference (SP) Method and predicted the choice

Hypothesis 1: The exclusive bus lane has significantly decreased the travel time for bus user using BRT comparing with normal bus

Hypothesis 2: The advantages that benefits from BRT introduction have increased the walking for transportation of commuter of personal vehicle user between minibus (MNB), BRT without P&R and feeder (BRT) and BRT with P&R and feeder (BRTS) by modeling The results showed that BRT had potential to attract notably personal vehicle users to change the mode choice

Satiennam and Jaensirisak (2013) also detailed that travel time and travel cost has importantly affect the modal shift to BRT To be more specific, travel cost has bigger effect on motorbike users’ mode choice meanwhile travel time dominant in encouraging car users to change theirs.

Methodology

With the objective of evaluating the impact of BRT introduction, commuter behavior is observed at two different periods, before (Y| t=1) and after (Y| t=2) The change of commuter travel behavior during these times could be identified as: λ = (Y| t=2) - (Y| t=1)

However, besides the effect of BRT introduction, there are probably many factors could affect the commuter behavior between the two before and after periods Therefore, it is not reasonable to simply calculate the difference between before and after BRT implementation and consider that change is BRT effect on

Hypothesis 3: Hanoi BRT has attract a number of commuters to switch from their private vehicle to use the public transport (BRT) (changed their mode choice)

Hypothesis 4: The proportion of car users shift towards BRT is higher than the portion of motorcycle users that change their mode choice (due to travel time reduction) commuter travel behavior (which mean λ cannot be considered to be the BRT effect)

When it comes up to the task of evaluating the effect of a specific intervention or treatment, Difference-in-Difference (DID) Estimation is one of the best solution DID estimation could be able to obtain an appropriate counterfactual to estimate a causal effect by comparing the changes in outcomes over time between a population that is enrolled in a program (treatment or intervention group) and a population that is not (comparison or control group) For that reason, DID estimation is chosen to identify and obtain the effectiveness of BRT in this study

The influence of BRT implementation on commuter behavior will be identified by observing and analyzing the changes between the changes of BRT user (treatment group) and normal bus user (comparison group) before and after BRT introduction.

Difference-in-Difference (DID) Estimation Approach

In this section, we will point out why the results of the DID estimation could be considered to be the effect of a specific treatment (which is BRT effect in this study) It could be proved as explained below:

Figure 3.1 Graphical explanation of DID explanation

There is only one cell among those four cells that is truly treated: Y

 As DID estimation assumption, without the program, independent i’s outcome at time t is given by:

The selection bias related to determinate characteristics of independent i (which is not changing overtime)

Time trend ( which is same for the treatment and comparison groups)

These conditions of and is needed to represent of DID estimation’s assumption

Without the program, i’s outcome at the time τ is:

[ | = 0, = ] = + Outcomes in the comparison group:

Y = [ | = 0, = 2] = [ | = 0] + The difference between pre and post program of the comparison group is:

 Call is the program’s true impact, then we have:

= [ | = 1, = ] − [ | = 1, = ] which does not depend on the i’s characteristics or time trend

Outcomes in the intervention group:

Differences in outcomes pre-treatment versus post-treatment cannot be attributed to the program because the treatment (program) effect are conflated with time trend

If we calculate the difference between pre-treatment and post-treatment, we have:

If we determine the treatment group and comparison group difference, we have:

Substituting in the term from our model:

Therefore, DID estimation does give the true effect of the program on participants (with the constraint that the assumption conditions are not violated).

Difference-In-Difference in a regression framework

The DID estimation will be implemented using Stata Software It will be implemented as a regression model with an interaction term between time dummy variables and group dummy variables The formula of the regression model for DID analysis of my thesis:

, Outcome of the interest (representative function which capable of considering each investigated aspect through the relevant coefficients resonance) is the time dummy which indicated the data is pre/post treatment

( which = 0 if in 2016 and = 1 if in 2018) is the group dummy which shown the data in comparison or treatment group (which = 0 for bus commuter and = 1 for BRT commuter) is the coefficient of the interest (the program effect)

= [ | = 0] + pre-program mean in comparison group

∗ Interaction term (which use to indicate who is truly received the effect of BRT implementation)

In chapter 7, we will use the Structural Equation Modeling in Stata Software to create the model of this DID regression framework to acquire the results.

SURVEY DATA

Survey design

As mentioned in Section 5.1, normal bus users are selected as comparison group and BRT users will be the treatment group for the DID estimation However, conducting DID estimation with BRT route and only one normal bus route would not be sufficient enough to point out the BRT effect because the changing of bus users themselves and the context different existence which possibly also cause some effects (mainly or partly) Therefore, 3 different normal bus routes in Hanoi city were selected to address this problem The chosen bus routes ought to have the similar characteristics about the area it goes through (high demand area), road’s cross-section of the route (number of lane) and ticket fee in order to have the highest similarity possible.

Questionnaire design

There are two types of questionnaire designed One type is for BRT route and the other one is for the 3 normal bus routes However, they have exactly the same questions, only the target respondent is different Most of the questions ask people about information in two period of time which is 2016 (before BRT introduction) and 2018 (after BRT introduction) There are also other types of question needed to further gather the information about travel behavior

Completed questionnaires were developed by Mr.Luu Duy, the student of the first intake of Master of Infrastructure Engineering Program of Vietnam Japan University The questions included in the questionnaire are categorized by part, information collected type and question type in detail as indicated in the table below:

Table 4.1 Questionnaire form of survey Information type Question Variable Question type

Vehicle using 2016 Travel time Travel distance Bus using frequency

Acceptable travel time Punctuality Security Safety Comfortable Suitable for children Satisfy BRT/Bus vs Motor (Faster, Higher security, More satisfy)

Likert Scale Question (from 1 – strongly disagreed,

2 – disagreed, 3- neither agreed or disagreed, 4 – agreed, 5- strongly agreed) Walking behavior

Walking distance acceptable Walking in hurry Walking prefer

Data collection

The survey has been done in June, 2018 with commuters of BRT and 3 common bus route Three common bus route selected are Route 01, 30 and 32 which have their characteristics and route alignments as shown in the table and figure below, respectively:

Table 4.2 Selected common bus routes for the survey

Route ends Yen Nghia Station –

My Dinh Station – Mai Dong

Ha Dong, Thanh Xuan, Dong Da, Hai

Ba Trung, Hoan Kiem, Long Bien

Nam Tu Liem, Cau Giay, Dong Da, Hai

Hoang Mai, Hai Ba Trung, Dong Da, Ba Dinh, Cau Giay, Nhon

Routes length 22.4 km 15.4 km 18.5 km

Roads cross section 6 lanes 4 - 6 lanes 6 lanes

Ticket Fee 7000VND 7000VND 7000VND

*The districts that connected by those routes have difference main function such as: administrative, entertainment and residential areas This has created a very high demand for travel between these districts

Figure 4.1 Route alignment of Bus 01

The questionnaires were distributed randomly irrespective of the sex (gender) of the respondents and also irrespective of the purpose The survey was carried out in a week during rush and normal hours No extra explanations were given to avoid the introduction of biases There were 200 samples of questionnaire distributed for each normal bus route and BRT route There were total 750 valid responded which is different for each route:

 182 samples responded for BRT route

 192 samples responded for Bus route 01

Table 4.3 The Socio-Demographic Profile of Sample Respondents

Factor % Sample Respondents % Hanoi Population Gender*

*Source: Forecasting the size and structure of Hanoi's population by 2020

According to the table of Socio-Demographic Profile of Sample Respondents, it is reasonable to state that the survey data is able to represent the Hanoi population characteristics for further analyzing.

Descriptive Statistic of survey data

The descriptive statistic is conducted for each data set of BRT and 3 common bus routes It is used to present quantitative descriptions in a manageable form and simplify large amounts of data in a sensible way thereby allowing to observe and assess the consistency of the data and check if the data is normally distributed or not

It is also a necessary step to assess whether data can be used for further analysis

Each route’s descriptive statistic is shown in the tables below, respectively:

Table 4.4 Descriptive Statistic results of BRT

Variable Unit Mean SD Min 25% 50% 75% Max

Bus using per day 2016 time

(s) 2.68 1.46 1.00 2.00 2.00 2.50 10.00 Bus using per day 2018 time

(s) 2.27 0.98 1.00 2.00 2.00 2.00 10.00 Day using bus per week 2016 time

Travel time 2016 min (s) 32.48 16.26 3.00 20.00 30.00 40.00 105.00 Travel time 2018 min (s) 26.82 13.15 2.00 15.00 25.00 37.25 60.00 LOS Perception

(1- 5) 3.53 0.68 2.00 3.00 4.00 4.00 5.00 Acceptable travel time 2018 rated 4.22 0.67 1.00 4.00 4.00 5.00 5.00 Punctuality 2016 rated 3.17 0.77 1.00 3.00 3.00 4.00 5.00 Punctuality 2018 rated 4.18 0.65 1.00 4.00 4.00 5.00 5.00 Security 2016 rated 3.39 0.75 1.00 3.00 3.00 4.00 5.00 Security 2018 rated 4.24 0.68 1.00 4.00 4.00 5.00 5.00 Safety 2016 rated 3.46 0.73 1.00 3.00 4.00 4.00 5.00

Table 4.4 Descriptive Statistic results of BRT - Continued

Suitable for children 2016 rated 2.96 0.81 1.00 2.00 3.00 3.00 5.00 Suitable for children 2018 rated 4.00 0.75 1.00 4.00 4.00 4.00 5.00 Satisfy 2016 rated 3.39 0.68 1.00 3.00 3.00 4.00 5.00

Satisfy 2018 rated 4.27 0.53 3.00 4.00 4.00 5.00 5.00 Faster than motor

Higher security than motor 2016 rated 3.75 0.64 2.00 4.00 4.00 4.00 5.00 Higher security than motor 2018 rated 4.19 0.63 1.00 4.00 4.00 5.00 5.00 More satisfy than motor 2016 rated 3.67 0.71 1.00 3.75 4.00 4.00 5.00 More satisfy than motor 2018 rated 4.21 0.67 1.00 4.00 4.00 5.00 5.00 Walking Behavior

In hurry 2018 rated 2.74 0.97 2.00 2.00 2.00 4.00 5.00 Like to walk

Table 4.5 Descriptive Statistic results of Bus 01

Table 4.5 Descriptive Statistic results of Bus 01 - Continued

Variable Unit Mean SD Min 25% 50% 75% Max Safety 2018 rated 4.05 0.67 2.00 4.00 4.00 4.00 5.00 Comfortable 2016 rated 3.73 0.75 2.00 3.00 4.00 4.00 5.00

Comfortable 2018 rated 3.83 0.83 1.00 3.00 4.00 4.00 5.00 Suitable for children 2016 rated 3.46 0.77 2.00 3.00 4.00 4.00 5.00 Suitable for children 2018 rated 3.51 0.90 1.00 3.00 4.00 4.00 5.00 Satisfy 2016 rated 3.87 0.63 1.00 4.00 4.00 4.00 5.00

In hurry 2018 rated 3.43 0.86 2.00 3.00 4.00 4.00 5.00 Like to walk 2016 rated 3.62 0.73 0.79 3.00 4.00 4.00 5.00 Like to walk 2018 rated 3.60 0.78 2.00 3.00 4.00 4.00 5.00

(1- 5) 3.62 0.73 1.00 3.00 4.00 4.00 5.00 Acceptable travel time 2018 rated 3.77 0.72 1.00 4.00 4.00 4.00 5.00 Punctuality 2016 rated 3.48 0.87 1.00 3.00 4.00 4.00 5.00 Punctuality 2018 rated 3.56 0.86 1.00 3.00 4.00 4.00 5.00 Security 2016 rated 3.64 0.80 1.00 3.00 4.00 4.00 5.00 Security 2018 rated 3.67 0.81 1.00 3.00 4.00 4.00 5.00

Variable Unit Mean SD Min 25% 50% 75% Max Safety 2016 rated 3.88 0.57 1.00 4.00 4.00 4.00 5.00

The descriptive statistic on the survey data in this study is also called univariate analysis Univariate analysis involves the examination across cases of one variable at a time There are three major characteristics of a single variable which are the distribution, the central tendency and the dispersion

In the tables above, only the central tendency and dispersion for each of the variables were shown The central tendency of a distribution is an estimate of the

"center" of a distribution of values The estimation of central tendency could be seen through the Mean (average) and the Median (which is the 50% in the Quartiles 25%-50%-75%) values Dispersion refers to the spread of the values around the central tendency The standard deviation (SD) is a measure of dispersion which shows the relation that set of scores has to the mean of the sample

As can be seen in and across the tables, most of the variable have the values Mean and Median similar to each other The same variable of different table also has the value which is not abnormal (significantly) difference with each other Even though there should be Skewness and Kurtosis value to check the consistency of the data set, it is reasonable to stated that these descriptive statistic values or the survey data is normally distributed.

DATA ANALYSIS AND DISCUSSION

Simple analysis

The simple analysis is introduced and conducted to check the reasonable of each hypothesis before continuing further analysis (DID analysis) There is one simple analysis for each hypothesis, respectively from hypothesis 1

Using the survey data about the travel time coordinated with travel distance of bus commuters in 2016 and bus/BRT commuters in 2018, the average travel time over travel distance of each route at two period of time is calculated

Table 5.1 Average travel time/ travel distance of bus/BRT users

Average travel time/ travel distance Bus users in 2016

(mins/km) Bus and BRT users in 2018

As can be seen from the table 7.1, commuters using normal bus has their average travel time/travel distance in 2018 higher than the average in 2016, which means that with the same bus route they use from 2016 to 2018, their average travel time of their trip is increased However, commuters shifting from normal bus (2016) to use BRT (2018) have their average travel time of their trip decreased

Therefore, it is reasonable to consider that BRT does improve the commuter’s trip by reducing the travel time Hypothesis 1 will be analyzed further by DID estimation to assess the BRT effect

In the questionnaire, BRT/Bus commuters had been asked 2 questions which were: Q1: “For the quality of your current BRT/bus service (in 2018), will you accept to walk longer distances than when using bus in 2016?”

Q2: “In 2016 and now (in 2018) when using a BUS / BRT how long will you accept your maximum walking distance to the waiting point (station)?

From question one, the proportion of commuter which accepted or not accepted to have longer walking distance to the station are illustrated in the figure 7.1 below: a

Figure 5.1 Proportion of commuter accepted walking longer for transportation

It is obvious from the chart that majority of commuters of normal bus (Route 01,30 and 32) will not accept longer walking distance to the station Meanwhile the majority of BRT commuters accept to walk further to reach the station

Route 01 Route 30 Route 32 BRT Route

Consequently, considering that BRT effect does have increased the walking for transportation of commuter is reasonable DID analysis will be carry out for hypothesis 2 to detect the effect of BRT on increasing the walking for transportation of commuters The difference between maximum acceptable walking distance to the station of commuters which is calculated through the data gathered from Question 2 shown above will be used for DID estimation

In the survey, bus/BRT commuters also asked what vehicle that they used for daily travel in 2016 With the data collected, the proportion between bus, motorbike, car and others vehicle used in 2016 of each route is statistically listed through the following chart:

Figure 5.2 Respondents’ used rates of different types of transportation in 2016

The amount of people used personal vehicle in 2016 is also considered to be the amount of people that changed their mode choice and shifted to use public transport because all these respondents are now use bus/BRT in 2018 (with high frequency)

According to the chart, BRT have attracted significantly number of commuters to shift from personal vehicle to use public vehicle from 2016 to 2018

Although it is less than BRT, all 3 common bus routes also attract people to change mode choice Thus, hypothesis 3 is considered reasonable Furthermore, DID analysis is needed to perform to see the true effects of BRT on this aspect

In section 7.1.3, the proportion between bus, motorbike, car and others vehicle used in 2016 of each route is illustrated With the same data, the chart below shows the details proportion of personal vehicle (between motorbike, car and others) used by the survey’s commuters in 2016

Figure 5.3 Respondents’ used rates between different types of private vehicle in 2016

As can be noted from the bar chart, the number of motorbike users attracted to change mode choice is significantly higher than the car users for all the routes and BRT apparently have the ability to attract more car users than the common bus

However, it is not sufficient to define this hypothesis fully reasonable There is an information that can't be overlooked in the figure The proportion of people using others type of personal vehicle is also really high As illustrated, it is even higher than motorbike percentage For the BRT route, it somehow remarkably lower than motorcycle Unfortunately, the survey data did not instruct the respondent to indicate clearly what the vehicle was when they select the “others” option

Therefore, it is really hard to figure it out what the “others” contains for further analysis

For that reason, this hypothesis is declared to be partly reasonable but it will not be able to analyze further in the DID analysis section.

Difference-in- difference analysis on the data

As mentioned in section 5.3, Stata software will be applied to define the true impact of BRT using the regression formula of DID analysis:

, Outcome of the interest (representative function which capable of considering each investigated aspect through the relevant coefficients resonance) is the time dummy which indicated the data is pre/post treatment

( which = 0 if in 2016 and = 1 if in 2018) is the group dummy which shown the data in comparison or treatment group (which = 0 for bus commuter and = 1 for BRT commuter) is the coefficient of the interest (the program effect) pre-program mean in comparison group selection bias time trend

∗ interaction term (which use to indicate who is truly received the effect of BRT implementation)

DID estimation, as a comparison of pre-post program assessment, needs one control group and one treatment group to create a set of data We have BRT users (treatment group) and 3 common bus users (comparison groups) Hence, there are 3 possible treatment-comparison group generated:

In addition, as indicated in the section 7.1, there are three out of four hypotheses set in chapter 4 are needed to analyze further using the DID estimation

Each hypothesis points out a set of “Outcome of the interest” (Yi,t) for the formula above Furthermore, based on the information about these sets, we can easily define the time dummy (D2018), the group dummy (DBRT) and the interaction term between them (D2018*DBRT)

After understood and carefully defined the parameters, the survey data is imported into Stata through the excel form Figure 7.5 is an example of Stata input data On the contrary, based on the form of indicated data, we could track backwards about its’ “pre-post treatment” and “treatment-comparison group” characteristics For example, in the line number 1 of figure 7.5 input data, there is answer of Mr Quang Huy about the maximum acceptable walking distance (Distance) The DBRT = 1 and D2018 = 0 means that the answer is for BRT route in

Figure 5.4 Import survey data into Stata using excel form

Figure 5.5 Example of Stata input data

5.2.2 Sample analysis and Structural Equation Modeling (SEM)

After completing input data, the true impact of BRT ( ) could be determine by performing regression calculations This task could be done by the simplest and quickest way by directly using the command to perform the linear regression calculation

Conducting the regression calculation of data set related to hypothesis 1 (reducing the travel time/travel distance of commuters) of Group 1, the results defined is shown in Table 7.2

Table 5.2 Simple linear regression calculation results on hypothesis 1, group 1

Time Distance Coefficient SD t-Stat P-value Standard Coefficient

The coefficient of the interaction term which also known as the BRT effect :

= -0,5925879 < 0 Thus, it can be concluded that BRT has effect on reducing travel time/travel distance of commuters (or reducing travel time in short)

Hypothesis 1 is proven true with the BRT and bus 01 The same analysis will be carried out with groups 2, 3 of hypothesis 1 and with other hypotheses with Number of obs = 560 F(3, 556) = 23.60 Prob > F = 0.0000 R-squared = 0.1130 Adj R-squared = 0.1082 Root MSE = 1.1328

However, there are few unusual points noted to be different from regular regression conditions that needed to be review before continuing analysis The R- squared and Adjust R-squared with a regression calculation, which show how many percentage of the variation of the dependent variable explained by the independent variables, normally needed to be at least larger than 0,50 then the model could be declared to be fit In the analysis above, both R-squared and Adjust R-squared are smaller than 0,50 The cause of this situation may be due to the following reasons

Firstly, DID estimation with regression framework is quite different from normal regression formula Regular regression formula normally has many independent variables which directly affect the value of R-squared and Adjust R-squared The closer of these values to 1, the more model of those variables are appropriate On the other hand, DID estimation with regression framework with its assumption generates only 2 independent values (DBRT and D2018) with one interaction between them (DBRT*D2018) Secondly, the characteristics of independent variables is also different The time dummy, group dummy and interaction term value is only 0 or 1

The another unusual point is about the p-value Normally, if p-value of an independent variable more than 0,05, this variable will be considered not significantly affect the dependent variable and sometimes could be drop out of the regression formula In the analysis above, D 2018 is not satisfy this condition (0,373

>0,05) It happens to the other cases as the analysis carried on (which shown in the next section) However, there are two things secure this problem Firstly, the most important variable DBRT*D2018 always has its p-value satisfy the condition in all cases Secondly, although sometimes not satisfy the condition p-value, theoretically

DBRT and D2018 does have indirectly affect dependent variable Yi,t through the interaction term (DBRT*D2018) Unfortunately, it is not recognizable with simple regression analysis using commands because it does not show the covariance between independent variables Therefore, the Structural Equation Modeling (SEM) will be used to create the model of DID regression framework to define and reveal the covariance of independent variables

Figure 5.6 Regression model in SEM

Performing estimation with the same case of data set (hypothesis 1 of Group

1), SEM gives exactly the same results with Table 7.2 and the covariance between

DBRT, D2018 and DBRT*D2018 These covariance is listed in Table 7.3

Table 5.3 Covariance between variables generated by SEM

Time Distance Coefficient SD t-Stat P-value Standardized Coefficient cov (DBRT , D2018) -7.61e-17 0.0105512 -0.00 1.000 -3.05e-16 cov (DBRT , DBRTXD2018) 0.1246875 0.0104118 11.98 0.000 0.5867387 cov (D2018 , DBRTXD2018) 0.11875 0.0102969 11.53 0.000 0.5580998

According to table 7.3, there is covariance between DBRT and DBRT*D2018,

D2018 and DBRT*D2018 with significantly high influence which accordance with DID estimation assumption In other words, it confirms that DBRT and D2018 have a indirectly affect to the dependent value through interaction term DBRT*D2018 Therefore, they could not be drop out of the formula even though they might not satisfy the p-value condition of regular regression One more thing, table 7.3 also shows that there is no covariance between DBRT and D2018 which is rational and consistent with the explanation because time dummy and group dummy are truly independent variable

With all the observed possible unusual parameters are explained as above, the next section 7.3 will only inform the results of DID analysis on all hypotheses with all 3 treatment-comparison groups and some discussions SEM is used to analyze all the cases to show both coefficient of independent variables (normal and standardized) and covariance between them.

Analysis results and discussions

All of the analysis cases in this are also checked for goodness of fit with SEM Two important index to determine the goodness of fit are RMSEA (Root mean squared error of approximation) and CFI (Comparative fit index) All the cases of analysis, the fit statistics have the RMSEA = 0 and CFI = 1 which is indicated very good model fit

The results of DID analysis of hypothesis 1 on all 3 groups are shown in Table 7.4, Table 7.5 and Table 7.6

Table 5.4 DID analysis results on hypothesis 1, group 1

Time Distance Coefficient SD t-Stat P-value Standardized Coefficient

_cons 4.423102 0.0931015 47.51 0.000 3.690493 cov (DBRT , D2018) -7.61e-17 0.0105512 -0.00 1.000 -3.05e-16 cov (DBRT , DBRTXD2018) 0.1246875 0.0104118 11.98 0.000 0.5867387 cov (D2018 , DBRTXD2018) 0.11875 0.0102969 11.53 0.000 0.5580998

_cons 4.036038 0.1009527 39.98 0.000 3.522394 cov (DBRT , D2018) 2.70e-34 0.0110203 0.00 1.000 1.08e-33 cov (DBRT , DBRTXD2018) 0.1248467 0.011113 11.23 0.000 0.5704904 cov (D2018 ,

Hypothesis 1 is about assessing the effects of BRT on the average travel time/travel distance reduction of the commuters Coefficient of DBRT*D2018 ( ) < 0 in all 3 groups show that BRT have significantly decreased the average time/distance The coefficient in each group are also satisfy the p-value condition

Therefore, BRT does have effect on the travel time reduction of commuters

Table 7.7, 7.8 and 7.9 are illustrated the DID analysis results of 3 groups with hypothesis 2, respectively

_cons 431.1189 18.53034 23.27 0.000 1.921702 cov (DBRT , D2018) 3.72e-17 0.0112115 0.00 1.000 1.52e-16 cov (DBRT , DBRTXD2018) 0.120166 0.0105312 11.41 0.000 0.6118378 cov (D2018 , DBRTXD2018) 0.1004184 0.0102489 9.80 0.000 0.5013072

_cons 486.6935 22.19863 21.92 0.000 1.950819 cov (DBRT , D2018) -9.51e-34 0.0118214 -0.00 1.000 -3.83e-33 cov (DBRT , DBRTXD2018) 0.1229752 0.0113895 10.80 0.000 0.6003875 cov (D2018 ,

D2018) 2.32e-34 0.010996 0.00 1.000 9.49e-34 cov (DBRT , DBRTXD2018) 0.1189768 0.0102336 11.63 0.000 0.6154575 cov (D2018 , DBRTXD2018) 0.097561 0.009573 9.80 0.000 0.492366

Hypothesis 2 suggests that there is influence of BRT on increasing the maximum acceptable walking distance to station of commuters The coefficient of

DBRT*D2018 ( ) > 0 in all 3 groups give evidence of the BRT's positive impact on the acceptable walking distance increasing of commuters from 2016 to 2018 The coefficient in each group are also satisfy the p-value condition Thus, BRT’s effect on this aspect is determined Hypothesis 2 is supported

Hypothesis 3 on the 3 treatment-comparison groups is analyzed by using DID estimation and had its results listed in Table 7.10, 7.11 and 7.12

_cons 1.711957 0.0533521 32.09 0.000 2.091019 cov (DBRT , D2018) 6.44e-33 0.0092401 0.00 1.000 2.58e-32 cov (DBRT , DBRTXD2018) 0.1249963 0.0092274 13.55 0.000 0.578399 cov (D2018 ,

_cons 1.703125 0.0545735 31.21 0.000 2.012073 cov (DBRT , D2018) 4.55e-33 0.0091376 0.00 1.000 1.82e-32 cov (DBRT , DBRTXD2018) 0.1249106 0.0090747 13.76 0.000 0.5824282 cov (D2018 , DBRTXD2018) 0.1216578 0.0090179 13.49 0.000 0.567058

Hypothesis 3 predicts that BRT with its advantages has positive impacts on encouraging commuters to change their mode choice In the survey data, private vehicle (motorbike, car, others) is marked as 2,3,4 and bus/BRT is marked as 1

Therefore, the negative values of the coefficient will point out the shifting from personal vehicle towards public transport In all 3 comparison – treatment groups, the coefficient of DBRT*D2018 ( ) < 0 which verify BRT’s influence on attracting private vehicle users to switch to public transports Hypothesis 3 is declared to be supported.

Further discussion

Normally, different locations possibly bring some differences in various aspect such as people’s income, technical infrastructure, development strategy… which could affect the effect of introducing a new type of public transport In the methodology section, a location dummy was intended to put into the regression formula besides the time dummy and group dummy However, this idea has encountered some obstacles and has not been developed Firstly, the locations difference is likely included in the difference between group (different treatment- comparison group) Secondly, the 3 bus routes are located in the same city, and not quite far from each other They likely will share almost the similar characteristics instead of the differences that different locations would bring Therefore, the effect caused by the different between them is considered to be hard to observe even though it exists

However, the correlation between BRT and bus routes about route direction is noticeably different (which could possibly lead to some comments about the location effect):

 The route direction of BRT and bus route 01 start from the same station and have its major part parallel with each other They are also located near each other It is reasonable to consider that who use BRT is not using route 01 and vice versa (competitive relations)

 BRT and bus route 30 have different route direction but have connection (share a part with each other) People using BRT and bus route 30 could transit between the two routes (support relations)

 BRT and bus route 32 have different route direction and no connections

It is considered that BRT and bus route 32 have no effect on each other

Figure 5.7 Route alignment interaction between BRT and 3 bus routes

The comparison between the coefficient of each aspect of these routes is carried out using treatment-comparison groups to observe the interaction of BRT with each other bus route (location effect) However, normal coefficient of each aspect has a big gap between each other, which makes it difficult to compare because they are regressed from quantities with different units Maximum acceptable walking distance, travel time over travel distance and mode choice have different units On the other hand, standardized coefficients' advocates note that the coefficients ignore the independent variable's scale of units, which makes with BRT route is the red line Bus route 01 is the blue line Bus route 30 is the green line Bus route 32 is the pink line comparisons easy Therefore, the standardized coefficient of each aspect are used for comparison This comparison is indicated in Table 7.13 The negative and positive signs, which indicated the trend of aspect (negative is decrease and positive is increase), is exclude in this table

Table 5.13 BRT effects comparison between treatment-comparison groups

BRT effects on BRT & Bus

Increase the maximum accepted walking distance 0.186 0.193 0.169

Decrease the average travel time/travel distance 0.210 0.179 0.189

Attract commuters to change their mode choice 0.120 0.053 0.056

According to the table, it is reasonable to consider the interaction between BRT and different bus route has some affects to BRT effects The group with support relation (BRT & Bus 30) shows the strongest BRT effects on increase the maximum accepted walking distance The group with competitive relation (BRT &

Bus 01) shows the strongest BRT effects on decrease the average travel time/travel distance and attract commuters to change their mode choice BRT and Bus 32 do not have any interaction to each other and therefore BRT effects are average in all 3 aspects considered It could be concluded that the BRT effects on commuter travel behavior are higher with bus route users that have location relation to BRT (which is considerable as a type of location effect).

CONCLUSION

This study has assessed the impacts of BRT introduction on commuter travel behavior in Hanoi using DID estimation It was found that BRT implementation in Hanoi did have positive impact to commuters The positive impacts are proved and shown through some aspects Firstly, BRT introduction with the advantage of its exclusive bus lane has improved efficiency of public transportation by reducing travel time over travel distance of normal bus commuters It means that with the same travel distance, commuter using BRT will have their travel time smaller than using normal bus Secondly, the advantages that benefits from BRT introduction have increased the walking for transportation of commuter More specifically, it is proved in this study that BRT implementation has the positive influence on increasing the maximum acceptable walking distance to the station (the willing to walk longer distance for public transportation) Finally, the last proved impact of BRT introduction in this thesis is the ability to attract commuters to change the mode choice Besides the normal bus commuter, Hanoi BRT has attracted a number of private vehicle users to switch from their mode towards BRT These impacts on reducing travel time, increasing the walking for transportation and mode choice towards BRT are noticeably related to each other in many ways and through that the rationale when confirming the influence of BRT on commuter travel behavior could be further reinforced

Furthermore, the impact of location to Hanoi BRT introduction’s effectiveness is also discussed and partly confirmed when comparing the effect of Hanoi BRT with normal bus routed which have the type of interaction of their routes and BRT route differently Discussing the impact of the location on the BRT effect may be a possible idea for future feasibility studies Additionally, the survey actually also provides good information and data but has not been used completely which possibly use for further feasibility studies For example, the level of service perception data can be used to evaluate the commuter satisfaction with BRT compared to normal bus or other kind of transport In addition, it can be used to rank criteria that affect the satisfaction of BRT users and then compare with others type of transport

This research, however, is subject to several limitations The first possible limitation in this study is about the surveying The survey was conducted with commuter of BRT and 3 normal bus routes Although the Socio-Demographic Profile of Sample Respondents show the similarity of the survey respondents to the population of Hanoi city, there still might be a chance that sample does not reflect the general population or appropriate population concerned well enough This results in limitations for the study known as “sample bias” or “selection bias” The next limitation is also about the survey but related to the methodology The treatment and comparison group definition in DID estimation should be “people who live within the chosen radius of BRT/Bus Station” which is normally used

However, the information gathered by the survey is not included the address or location of respondents Therefore, their distance to the station are unknown and the regular definition is impracticable Then the definition of the treatment and comparison will be defined by what respondents use (BRT/bus) The final limitation mentioned is few unusual values noted to be different from regular regression conditions of DID estimation under regression framework results This limitation has been discussed and explained carefully in Section 7.2.2 which suggested that these difference come from the DID estimation’s specific assumption and it is acceptable

In conclusion, the key findings of this study verify the positive impact of BRT introduction to the commuter travel behavior in Hanoi Thus, further studies and development for Hanoi BRT in general and public transport in particular is suggested to be further researched

Bajracharya A.R (2008) The impact of modal shift on the transport ecological footprint, A case study of the proposed Bus Rapid Transit System in Ahmedabad, India

Darido, G (2006) Perspectives on bus rapid transit (BRT) developments in China

Darío H and Luis G (2012) BRT and BHLS around the world: Explosive growth, large positive impacts and many issues outstanding

Darío H., Liliana P., Nicolás E & Pedro L J (2013) TransMilenio BRT system in Bogota, high performance and positive impact - Main results of an ex-post evaluation

Deng, T., & Nelson, J D (2011) Recent developments in bus rapid transit: A review of the literature

Godavarthi, G R., Chalumuri, R S., & Velmurugun, S (2014) Measuring the performance of bus rapid-transit transitways based on volume by capacity ratio

H.I Okagbue, et al., (2015) On the Motivations and Challenges Faced by Commuters Using Bus Rapid Transit in Lagos, Nigeria

Hsin- L C., Shun- C W (2007) Exploring the vehicle dependence behind mode choice: Evidence of motorcycle dependence in Taipei.

Tiêu đề	Impacts Of BRT Introduction On Commuter Travel Behavior In Hanoi
Tác giả	Tran The Huy
Người hướng dẫn	Dr. Nguyen Hoang Tung
Trường học	Vietnam National University, Hanoi
Chuyên ngành	Infrastructure Engineering
Thể loại	master's thesis
Năm xuất bản	2019
Thành phố	Hanoi

Định dạng
Số trang	65
Dung lượng	2,19 MB