against city bus services. Figure 1. A snapshot of low-service-quality buses with open-end carriages (City bus route 46) To cope with increased traffi

Determinants of Intention to Shift to a New High Quality Bus Service: A Mixed Logit Model Analysis for Ho Chi Minh City, Vietnam Hong Tan VAN a, Daisuke FUKUDA b a Department of Civil Engineering, Ho Chi Minh City University of Technology, Vietnam; E-mail: vhtan@hcmut.edu.vn b School of Environment and Society, Tokyo Institute of Technology, Tokyo 1528552, Japan; E-mail: fukuda@plan.cv.titech.ac.jp Abstract: This paper discusses the determinants of the intention to shift to a new high-quality Locally Adapted, Modified, and Advanced Transport (LAMAT) bus service in Ho Chi Minh City using a stated preference (SP) survey. Mixed logit models (MXL) are used to estimate the mode choice between motorcycles (MCs) and LAMAT in the SP choices. The use of a panel-mxl is shown to significantly improve the estimation results compared with the standard multinomial logit model. Accordingly, heterogeneity in the LAMAT choice is largely explained by citizens unobserved characteristics, whereas their preference for time and money might be homogenous. Additionally, four mode-specific factors, i.e., provision of information, air conditioning, seat availability (LAMAT), and the risk of traffic accidents (MCs) are found to be significant determinants of the intention to use the LAMAT bus service. Keywords: Bus Service, Stated Preference Data, Mode Choice, Mixed Logit Model, LAMAT 1. INTRODUCTION Buses are generally regarded as one of the most prioritized modes of transport in many cities around the world. However, regulating bus services to maintain an acceptable level of service is not easy for transport authorities. This may be because other policies such as land-use planning, vehicle fuel pricing, and road space allocation are not well synchronized with policies promoting bus use (Hanaoka and Achrya, 2008). Under few constraints, people tend to use cars and/or motorcycles because of their higher comfort, convenience, and independence; consequently, the volume of traffic overloads the road network, resulting in congestion. Such conditions make buses less attractive than private modes of transport in terms of travel time, travel cost, reliability, accessibility, and so on (e.g., Bar-Yosef et al., 2013). Ho Chi Minh City (HCMC), Vietnam, is faced with the familiar situation of a regular bus service that suffers from failures. Bus use has fallen significantly since 2013, decreasing by an average rate of more than 10% per year (Dau, 2016). Moreover, the rapid increase and overwhelming use of motorcycles (MCs) and cars has been recorded. Thus, the market share of buses in HCMC has switched to MCs and/or cars, despite considerable subsidies from the government that were expected to see an increase in demand of 20% by 2020 (HCMC Communist Party Committee, 2016). Regular bus services in the city are still considered unreliable with minimal comfort, security, and safety (HCMC Public Transportation Operation and Management Center, 2012). Furthermore, in suburban districts that have received less transportation investment from the government, some unsafe paratransit-like vehicles are still in service, as illustrated in Figure 1. This hardens the attitude of citizens 2284

against city bus services. Figure 1. A snapshot of low-service-quality buses with open-end carriages (City bus route 46) To cope with increased traffic congestion, the city government is seeking to improve bus service quality with the aim of making it more competitive with private modes of travel, particularly MCs. Constructing new types of public transport is regarded as one possible countermeasure to reduce the use of MCs. However, introducing advanced modes like bus rapid transit may be unviable, as there is not enough road space for bus priority lanes (HCMC Public Transportation Operation and Management Center, 2012). Considering these constraints, it is important to understand how many MC users in HCMC would convert to a certain new mode of public transport. In other words, it is necessary to identify the characteristics of public transport modes that are suited to the road conditions and the preferences of HCMC citizens. An earlier study in HCMC showed that perceived bus service quality and moral concerns are significant factors behind the behavioral intention to use buses (Fujii and Van, 2009), but these findings were based on the subjective opinions of bus users rather than quantifiable attributes. It is also necessary to examine the impact of objective attributes of bus services, which may use environmentally friendly vehicles integrated with information and communication technology, without investing in road construction or expansion. Recently, local transport services (including paratransit vans) have been referred to as Locally Adapted, Modified, and Advanced Transport (LAMAT), as proposed by Phun and Yai (2016). The introduction of a new public transport mode might be considered as an alternative to the deep-rooted use of MCs, and this could take the form of an appropriately modified and upgraded bus service (e.g., Hoang and Okamura, 2015; Tuan and Son, 2015; Bando et al., 2015). Considering the above-mentioned aspects, it is important to investigate citizens daily travel choice behavior in areas where the bus service quality is quite low and to focus on low-income people who are just able or unable to own an MC. Thus, this paper explores measures to shift MC users in HCMC to LAMAT-type buses. We assume a so-called high-technology bus (hereafter referred to as a LAMAT bus ) as a specific new type of public transport, and then employ the stated preference (SP) survey technique to elicit whether MC users would change to use this new travel mode. This study has the following two objectives: (1) to identify the key features that determine people s intention to use the LAMAT bus service; (2) to validate the travel mode choice model that accounts for unobserved heterogeneity in the preference of HCMC citizens towards the use of the LAMAT bus. One novelty of this study is that we use a Mixed Multinomial Logit (MXL) model (McFadden and Train, 2000) to capture the inherent panel nature of repeated SP questions, whereby a serial correlation may be observed in the sequence of SP choices given by the same respondent. The findings of this study are expected to provide some insights into the introduction of bus lines with higher service quality that will increase the number of bus passengers and reduce the reliance on MC use. 2285

The remainder of this paper is structured as follows. Section 2 gives an outline of the survey. Section 3 describes the rationale for using the MXL model and its specifications. Section 4 presents the model estimation results, and Section 5 summarizes our conclusions. 2. SURVEY OUTLINE 2.1 Survey area In line with the above objectives, we selected a survey area that is geographically separated from other parts of HCMC. The target survey respondents are people who live in District No. 6 of HCMC. From this area, the only main route along to the city center is along the canal. As shown in Figure 2, this route has an existing bus service (i.e., Route 46, 5 6 km route length, four vehicles per hour during the daytime on weekdays) that uses the unsafe vehicles shown in Figure 1. Hence, for most people in District No. 6, the existing bus service or the upgraded LAMAT buses represent an alternative to MCs for reaching the city center. Figure 2. Survey area and bus route 46 (numbered squares indicate bus stops) 2.2 Stated choices and experimental design The experiment to investigate stated choices between MC and the LAMAT bus service involves several design and testing steps. First, after consulting experts and referring to another survey by the authors, we assume that the fare, travel time, seat availability, information system, and air conditioning are influential attributes specific to LAMAT, and that the risk of traffic accidents is a specific attribute of MCs. More detailed explanations are as follows: (1) To compensate for the longer travel time by bus and to promote a modal shift, the LAMAT bus fare suggested in the survey was set to be lower than the common bus fare in HCMC. Hence, we set three fare levels: 2,000, 3,000, and 5,000 VND/trip. (2) Three levels were used to depict the additional travel time of the LAMAT bus 2286

alternative over MCs for individual trips: same as an MC, 5 min longer than an MC, and 10 min longer than an MC. (3) Seat availability rates were set as follows: highly occupied (seats available on 33% of buses), fairly occupied (seats available on 50% of buses), sparsely occupied (seats available on 75% of buses). (4) The attributes of an information system (i.e., informing passengers about the arrival times of LAMAT buses at stops) and air conditioning on board had two levels (yes/no). (5) Whereas LAMAT buses are assumed to be absolutely safe, MCs are assumed to be involved in accidents at certain levels. The risk of an accident when using an MC was set to 0.02, 0.01, and 0.005, corresponding to an accident every 50, 100, and 200 trips, respectively. These settings are summarized in Table 1. For MCs, some observed trip attributes obtained from the survey can also be incorporated in mode choice modeling. Table 1. Attributes to be considered in mode choice modeling LAMAT bus MC - Fare (three levels) - Travel time (three levels) - Risk of traffic accident (three levels) - Air condition (two levels) - Seat availability (two levels) - Information provision (two levels) - Observed trip attributes [obtained from the questionnaire] In SP design, to minimize respondent confusion, important words are given in bold font and capitalized in the questionnaire. To ensure the suitability of the questionnaire, we implemented a simple testing survey on 24 people. Their feedback showed that three choices with five LAMAT attributes and one MC attribute are not overly complicated. To increase the sampling efficiency, a sequential orthogonal design was employed (e.g., Caussade et al., 2005). Through this design, 36 choices for this problem were given (see Table 2). However, as it is unreasonable to present 36 choices to a single respondent, we divided the orthogonal design into 12 blocks with three SP questions in each, which means that there are 12 types of questionnaire sheets. Blocking ensures that the attribute level is balanced within each block, such that the respondents do not face only low or high levels for a certain attribute. Finally, each choice profile was reviewed and tested by some respondents to ensure that the choice scenarios were realistic. Each respondent then undertook three choice tasks, thus enhancing the accuracy of the survey. A snapshot of the SP questions is illustrated in Figure 3. 2.3 Layout of survey questionnaire A concise four-page questionnaire was designed to increase the possibility of collecting comprehensive information from each respondent. The four parts of the survey were as follows: - Part 1 was designed to collect information about recent trips to the city using MCs as the main mode of transport. The information collected includes the travel time and cost, trip purpose, rate of punctuality for that trip, other family members involved in the trip, combination with other trips, and the destination in the city center. - Part 2 supplied some pictures of LAMAT buses (electric vehicles that use various technologies to control their operation and provide comfort and reliable service). We 2287

stated that LAMAT buses would replace the current unsafe buses, and then asked the respondents how they felt about the LAMAT service in terms of travel mode and travel time. 2288

Block No. Fare (VND) Time difference compared with MCs (min.) Information system Table 2. SP profiles created with sequential orthogonal design Time difference Seat MC risk Air Block Fare compared Information availability of conditioning No. (VND) with system rate (%) accident MCs (min.) Air conditioning Seat availability rate (%) 1 2000 15 1 0 50 0.01 7 3000 0 1 1 50 0.01 1 5000 10 0 1 75 0.005 7 3000 15 1 0 50 0.01 1 5000 10 1 1 50 0.005 7 3000 0 0 0 75 0.005 2 3000 0 0 1 33 0.02 8 3000 10 0 1 75 0.02 2 5000 15 1 1 75 0.01 8 2000 15 1 1 33 0.005 2 2000 10 0 0 33 0.02 8 2000 0 0 1 50 0.01 3 2000 0 0 1 50 0.01 9 2000 0 1 0 75 0.005 3 3000 10 1 1 50 0.01 9 3000 0 0 0 75 0.02 3 5000 0 1 0 75 0.005 9 5000 0 1 1 33 0.01 4 2000 15 1 1 33 0.02 10 5000 10 1 1 75 0.005 4 3000 10 1 0 33 0.02 10 2000 10 0 1 50 0.01 4 5000 0 1 1 50 0.02 10 3000 15 1 1 33 0.02 5 2000 10 1 0 50 0.005 11 3000 0 0 0 75 0.005 5 2000 0 0 0 33 0.02 11 3000 15 1 1 33 0.005 5 5000 10 1 0 75 0.02 11 5000 10 1 1 50 0.02 6 5000 0 1 1 33 0.02 12 5000 10 1 0 50 0.01 6 3000 15 1 0 75 0.005 12 3000 15 0 1 50 0.01 6 2000 15 0 0 75 0.005 12 2000 10 0 0 33 0.02 MC risk of accident 2289

Figure 3. Sample of the SP questionnaire (originally written in Vietnamese) - Part 3 examined the possibility of switching to LAMAT buses for the trip described in Part 1 using the three SP questions described in the previous subsection. After revealing the time and money cost of using MCs and LAMAT buses, respondent were asked to trade-off among the six attributes mentioned in the previous subsection. - Part 4 concerned some attitudinal factors, but the responses were not used in this study. 2.4 Recruiting survey respondents Paper-based face-to-face interviews were conducted from January 8 10, 2016, by a team from the Technology University of HCMC. The interviewers approached households in the survey area to request data. Of around 1,500 requests, 1,000 individuals were willing to answer the questionnaire. Respondents were first asked whether they had recently made a trip into the city center of HCMC using an MC. If a full description of such a trip was forthcoming, the surveyor explained the LAMAT bus service, which would supposedly replace the current bus service; otherwise, the interview was terminated. Respondents were then asked how they would use the LAMAT service instead of MCs for the reported trips, including basic questions relating to access/egress times and modes. Once the respondents fully understood the travel time and cost of MCs and LAMAT buses for the trip to the city center, three SP scenarios in combination with different LAMAT bus attributes were presented. The respondents were 2290

asked to consider whether they should change to use LAMAT buses in the form of a binary choice (Yes or No). Note that one of the 12 types of questionnaires (based on the blocking of SP questions) was randomly assigned to each respondent. After the interview, a thank-you gift containing a 10,000 VND note ( 0.45 USD) was given to each respondent as an incentive. 2.5 Respondent characteristics The basic characteristics of the interviewees are reported in Table 3. The respondents had a mean age of 34, with a higher number of males (57.4%) than females (42.6%). Most interviewees worked in offices. Approximately half of them reported a monthly income of less than 5 million VND ( 220 USD), indicating that the sample earned slightly below the average for HCMC. Such a figure is reasonable, as the above-mentioned area is poorer than other parts of HCMC. The large majority claimed that they own MCs (82.1%). As most people could afford an MC even though their salary is quite low, we may infer that MCs are a basic need of many people in HCMC, regardless of income. The majority of trips to the city center were related to recreation (38.9%). As most of the trips were unchained (65.3%), we expected this sample to present a high probability of converting to LAMAT bus use. Regarding the relations among the basic trip characteristics, work and schooling trips are mostly unchained, whereas recreation trips are mostly chained (Table 4). Additionally, trips are more likely to be chained when there is a need to carry other people (e.g., family members). These facts may be considered when constructing the choice models in the next session. Table 3. Fundamental characteristics of the respondents (N = 1,000) Variable Percentage Variable Percentage Gender Motorcycle ownership Male 57.4 Having 82.1 Female 42.6 Not having 17.9 Occupation Trip purpose Student 13.2 Work 27.4 Worker 18.2 School 9.0 Office worker 25.6 Visiting 6.6 Retailer 17.4 Shopping 11.1 Other 25.6 Recreation 38.9 Others 7.0 Monthly income (VND) <5 million 49.7 Trip chain 5 10 million 46.5 Yes 34.7 >10 million 3.7 No 65.3 2291

Table 4. Relations among trip chaining, trip purpose, and carrying other people Trip chain Trip purpose Carry others No Yes Total Work 180 34 214 School 50 26 76 Visiting 13 27 40 No (65.3%) Shopping 40 27 67 Recreation 70 142 212 Others 22 20 42 Total 375 276 651 % of total 57.6% 42.4% Work 35 24 59 School 2 12 14 Visiting 5 21 26 Yes (34.7%) Shopping 4 40 44 Recreation 20 158 178 Others 8 20 28 Total 74 275 349 % of total 21.2% 78.8% Total 449 551 1000 3. DEVLOPMENT OF TRAVEL MODE CHOICE MODEL 3.1 Model specification The use of the standard Multinomial Logit (MNL) model (Ben-Akiva and Lerman, 1985) can be problematic when applied to choice modeling using repeated SP questions, because MNL assumes that the error terms in the utilities follow an independent and identically distributed (IID) Gumbel distribution across the choice situations, respondents, and alternatives. That is, MNL ignores the possible correlations of error terms across choice situations by the same individual. As noted in the previous section, three different (but consecutive) SP questions were presented to each respondent. Hence, when there is some potential for correlated responses across choice observations by the same respondent, there might be a possibility that the IID assumption may not hold within the sequence of SP choices by the same individual. The sequencing choice may result in a mixture of learning and inertia effects (Hensher and Greene, 2002). The MXL model (McFadden and Train, 2000) can eliminate the biases caused by serial correlations by explicitly introducing individual-specific unobserved factors into the utility functions. Thus, in this study, we employed an MXL with panel effects (Train 2003) to analyze the travel mode choice and appropriately capture such serial correlations by the same individuals. Note that it is also possible to introduce taste heterogeneity into some attributes by employing random-parameter-type MXL, but we found the estimated heterogeneity parameters to be insignificant. Therefore, in the model, all utility parameters were fixed. The total utilities for the MC and LAMAT bus choices were formulated as the following sum of a deterministic term consisting of some explanatory variables (e.g., travel time, travel cost, attributes specific to each mode and socioeconomic variables) and unobserved factors: (1) 2292

where and are the total utilities of MCs or LAMAT, respectively, chosen by individual facing SP question ; and denote the unobserved heterogeneity, which captures panel effects intrinsic to individual (these values remain the same for the same individual); and and are error terms that follow an IID Gumbel distribution across choice alternatives, individuals, and choice situations. As there are only two alternatives (i.e., binary choice situation), Equations (1) and (2) can be reduced to the following single equation by taking their difference: where ; ; and. We further assume that follows the normal distribution, where is the standard deviation of the panel effects. Note that takes different values in different individuals, even though the following distribution is common. Consider the sequence of choices for three SP questions to be. Then, individual s joint choice probability under the Panel-MXL model is given by: (2) (3) The sample log-likelihood for all individuals can be written as: (4) where is the total number of individuals and is the parameter vector in the deterministic utility term. The optimal parameters can be estimated from the objective function (5) with respect to using the simulated log-likelihood maximization (Train, 2003). In our application, the deterministic utilities are specified as follows: (5) (6) where is an alternative specific constant for LAMAT buses and the terms are parameters associated with each explanatory variable. The explanatory variables are defined in Table 4. For the simulated maximum likelihood estimation, the parameters are stacked into a vector as follows: (7). 2293

Table 4. Descriptive statistics of variables used in the models Variables Definition Mean SD Personal characteristics Age Age 34.3 12.4 Gender 1: male, 0: female 0.57 MCown 1: having, 0: otherwise 0.83 Income Monthly income level (1: <5 mil. VND, 2: 5-10 mil. 1.54 0.57 VND, 3: >10 mil. VND) Characteristics of trips Cost MC (VND) MC travel cost = Cost of (fuel, parking fee) 21,292 8,543 Time MC (minute) MC travel time = (access/egress time, 50.56 16.3 in-vehicle time) Cost LM (VND) LAMAT bus travel cost = cost of (fare, 9,359 7,850 access/egress mode) Time LM (minute) LAMAT bus travel time = (access/egress time, 113.7 25.5 waiting time, in-vehicle time) TripChain 1: The trip chained with others, 0: otherwise 0.35 Carry 1: Carry others, 0: otherwise 0.56 Characteristics of modes Info LAMAT service could provide information and communication system, 1: having, 0: otherwise Aircon LAMAT bus having air conditioners, 1: having, 0: Seat otherwise LAMAT bus seat availability rate, 33%, 50%, and 75% Risk Risk of MC use, 0.005, 0.01, and 0.02 3.2 Variables in the models Table 4 presents descriptive statistics of the variables used in the models, as well as their definition. In the model, age is assumed to be a continuous variable, with dummy variables defined for gender, MC ownership, chained trips, trips carrying other people, and the availability of information and air conditioning on LAMAT buses. We used an ordinal variable to measure three income levels (<5 mil. VND, 5 10 mil. VND, and >10 mil. VND). The risk factor of using MCs was quantified in ordinal variables according to the calculated probability of the three suggested risk levels. For the travel cost and travel time of MCs and LAMAT buses, we summed all times and costs expended for the trip. The results show that the average travel cost of LAMAT buses hypothesized in the SP survey was approximately half of the MC cost for the reported trip. However, the travel time using LAMAT buses was assumed to be double that of MCs. Such balanced schemes require the respondents to trade-off between the two modes. 4. ESTIMATION RESULTS OF MODE CHOICE MODELS The BIOGEME software package was used to maximize the simulated likelihood over 500 Halton draws (Bierlaire, 2003). We estimated three panel MXL models for the respondents trip purposes, as summarized in Table 5. The first model used the pooled dataset for all trips. The second model was estimated using data from private trips, including those for recreation, visiting friends, and so on, and the third model was estimated from the data for work/school-related trips. The models were found to be statistically significant with a 2 statistic ranging from 166.3 580.7, which is above the critical value of 23.69 (with 14 degrees 2294

of freedom at an alpha level equal to 0.05). The overall goodness-of-fit values are not high but tolerable (adjusted rho square ranging from 0.092 0.156), possibly because of low variance in the explanatory variables. In spite of this, almost all explanatory variables are statistically significant at the 95% confidence level and have intuitive signs. In all three models, the panel effect parameter ( ) was found to be significantly different from zero at the 99% confidence level, implying that there would exist strong serial correlations among sequential SP choices made by the same individual. We separately estimated MNL models corresponding to each panel MXL, and found that the goodness-of-fit of the MNL models was very low. This confirms that the observations in the panel data can be considered to be dependent. Note that heterogeneity exists in all models, as the alternative specific constant for LAMAT buses had a low level of significance, possibly because of the inclusion of heterogeneity in the utility functions (Cherchi and Cirillo, 2008). First, as expected, the travel time and travel cost has a negative effect on the likelihood of travel mode use. From these two coefficients, the computed value of time for people in this area is 1.40 USD/h, which is a slightly lower than the average of HCMC residents ( 2.1 USD/h, equivalent to GDP per capita of 5,318 USD) (Savills, 2016). The value of time spent on school/work trips (3.18 USD/h) is higher than for private trips (0.87 USD/h). This is probably because work- and school-related trips are more time-dependent than private trips. Additionally, there are four highly significant alternative specific factors, i.e., Info, Aircon, Seat, and Risk. The significant coefficients of Info and Aircon indicate that people are more likely to choose to use the LAMAT bus service if it is equipped with air conditioning and an information system at the bus stops. Further, it is not surprising that people are more likely to use the LAMAT buses if there is a higher possibility of being seated. This indicates that applying information technology and improving the quality of the vehicle interiors could help attract more passengers. It is worth noting that the Risk factor had a significant negative impact on MC use. This indicates that MC users may shift to LAMAT buses if they perceived the higher risk of being involved in a traffic accident when using MCs. However, risk seems to have little effect on MC trips to school and work. This may be related to the fact that the convenience of daily MC use overrides the threat of accidents. Furthermore, it can be inferred that, under the current high risk of MC accidents in HCMC, there is more chance of persuading people to use LAMAT buses for trips other than to work and school. It is not surprising that the probability of choosing MCs is significantly higher for males than females for private trips and overall. This probability is also higher for those who own an MC and those who have higher levels of income. The results suggest that age has a significant negative impact on MC use for nearly all trip types, implying that elderly people are more likely to shift to LAMAT buses than younger residents. Owning an MC has a positive effect on the choice of MC for school/work trips, but this is less significant for private trips. This implies that the LAMAT service should target people making infrequent trips, rather than daily trips. Finally, regarding some trip characteristics, we found that the effect of chained trips was significantly negatively related to the probability of using LAMAT buses in the models using the pooled and private trip datasets, rather than in the model using the school/work trips. Whether a school/work MC trip carries an additional person was also found to have a significant negative effect on the likelihood of changing to use LAMAT buses. However, we were unable to identify any statistical significance in the other two trip types. Further descriptive analysis of the dataset shows that only 27.3% of the observed MC trips are private and chained. Moreover, only 9.7% of MC trips to school/work carried another person. This implies that a policy promoting the use of LAMAT buses would have a high chance of 2295

success. Table 5. Estimation results of panel MXL models for mode choice according to trip purpose (1) All trips (2) Private trips (3) School/work trips Variables (Alternatives) Parameter t-value Parameter t-value Parameter t-value ASC (LM) 0.0974 0.23 0.143 0.24 0.792 1.27 Time [Generic] 0.0213 6.08 0.0166 3.61 0.0259 4.74 Cost [Generic] 4.00 10-5 6.45 5.04 10-5 5.78 2.14 10-5 2.55 Info (LM) 0.672 6.27 0.716 5.08 0.611 3.74 AirCon (LM) 1.33 11.22 1.48 9.45 1.13 6.1 Seat (LM) 0.0179 5.64 0.019 4.54 0.0146 2.96 Carry (LM) 0.171 1.28 0.269 1.33 0.445 2 TripChain (LM) 0.387 2.79 0.582 3.33 0.231 0.93 Risk (MC) 17.8 2.17 24.7 2.28 7.53 0.59 Age (MC) 0.0175 3.19 0.0129 1.82 0.02 2.09 Gender (MC) 0.475 3.67 0.572 3.3 0.325 1.69 MCown (MC) 0.361 2.2 0.24 0.98 0.518 2.35 Income (MC) 0.633 5.34 0.652 3.97 0.615 3.49 (Panel effects) 1.36 13.8 1.48 11.5 1.12 7.41 Sample size (N) 3,000 1,911 1,089 Initial log likelihood 2,079 1,324 754 Final log likelihood 1,789 1,104 671 2 580.7 440.4 166.3 Adjusted rho-square 0.133 0.156 0.092 Value of time (in USD/hr.) 1.40 0.87 3.18 5. CONCLUSIONS This paper has examined the determinants affecting the intention to use a high-quality LAMAT bus service in HCMC using an SP survey questionnaire. The methodology was straightforward, but we employed mixed logit models of mode choice between MC and LAMAT buses to account for the panel effects in SP choices. Accordingly, we found that a panel MXL model significantly changes the estimation results compared with the standard multinomial logit model. The estimation results also suggest that the preference for LAMAT buses varies across individuals. It can therefore be inferred that heterogeneity in individual behavior regarding the probability of using LAMAT buses depends on certain characteristics of the people. In this study, the random effects of travel time and travel cost were found to have weak significance, meaning that people s attitudes toward time and money might be homogenous. We did not investigate the random effects of other parameters; this will be considered in future work. Based on the findings of the model estimation, four factors are specific to the modes of transport, i.e., provision of information, air conditioning, seat availability on LAMAT buses, and the risk of traffic accidents for MCs are significant determinants of the probability of using LAMAT buses. This means that, when commencing the LAMAT bus lines, the operator should enhance the reliability of the service by supplying information such as the waiting time for the next bus, improve the quality of the vehicle interiors by installing air conditioning, and ensuring a high possibility of passengers finding a seat. Such measures could help to attract and retain potential passengers. Socioeconomic factors such as age, gender, MC ownership, and income have significant effects on the probability of shifting to high-quality LAMAT buses. Together with the taste variation found in modeling the mode choice, this implies that push-and-pull measures should 2296

be implemented to shift people from using MCs to LAMAT buses in HCMC. That is, besides improving and maintaining facilities, any promotion schemes should be customized to target groups of people whose characteristics are most affected by those measures. We believe that, as long as there is a long-term strategy for such efforts, a modal shift towards high-quality LAMAT bus services could help to reduce traffic congestion in HCMC. ACKNOWLEDGEMENTS Toshiba Corporation and a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (A) #15H02632 supported this study. The authors thank Professors Tetsuo Yai, Yasuo Asakura, Yasunori Muromachi, and Shinya Hanaoka of Tokyo Institute of Technology for their valuable comments. REFERENCES Bando, T., Fukuda, D., Wicaksono, A., Wardani, L.K. (2015) Stated preference analysis for new public transport in a medium-sized Asian city: A case study in Malang, Indonesia. Journal of the Eastern Asia Society for Transportation Studies, 11, 1451-1466. Bar-Yosef, A., Martens, K., Benenson, I. (2013) A model of the vicious cycle of a bus line. Transportation Research Part B: Methodological, 54, 37-50. Ben-Akiva, M., Lerman, S. (1985) Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press, Cambridge MA, USA. Bierlaire, M. (2003) BIOGEME: A free package for the estimation of discrete choice models. Proceedings of the 3rd Swiss Transportation Research Conference, Ascona, Switzerland. Caussade, S., Ortúzar, J.D., Rizzi, L.I., Hensher, D.A. (2005) Assessing the influence of design dimensions on stated choice experiment estimates, Transportation Research Part B: Methodological, 39, 621-640. Cherchi, E., Cirillo, C. (2008) A modal mixed logit choice model on panel data: Accounting for systematic and random variations on responses and preferences. Proceedings of the 87th Annual Meeting of the Transportation Research Board, Washington DC, USA. Dau, A.P. (2016) Summarizing the public transportation operation in HCMC. A Report Presented at the Workshop on Summarizing Public Transportation Operation and Bus Subsidy in HCMC in the Period of 2002 2015. Department for Transport UK (2014) Supplementary Guidance: Mixed Logit Model. Available online at https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/427138/ webtag-tag-supplementary-mixed-logit-models.pdf. Hanaoka, S., Acharya, S.R. (2008) Characteristics and issues of transport policy in East Asian megacities. Transportation and Economy, 68(11), 14-20. HCMC Communist Party Committee (2016) Document No. 14-CTrHĐ/TU: Action program to implement the Enactment of the 10th Congress of the Communist Party of HCMC about the Program to Alleviate Traffic Congestion and Accidents in the Period of 2016-2020. Available online at http://www.sggp.org.vn/pdfanddoc/document14602_14ctrtu.doc. 2297

HCMC Public Transportation Operation and Management Center (2012) A Report on 10 Years of Consolidating and Developing the System of Public Transportation by Bus in HCMC (the period of 2002 2012). Fujii, S., Van, H.T. (2009) Psychological determinants of the intention to use the bus in Ho Chi Minh City. Journal of Public Transportation, 12(1), 97-110. Hensher, D.A., Greene, W.H. (2002) The mixed logit model: The state of practice and warnings for the unwary. Institute of Transport Studies, the University of Sydney and Monash University. Hoang, L.Q., Okamura, T. (2015) Influences of motorcycle use on travel intentions in developing countries: A case of Ho Chi Minh City, Vietnam. Journal of the Eastern Asia Society for Transportation Studies, 11, 890-905. McFadden, D., Train, K. (2000) Mixed MNL models of discrete response. Journal of Applied Econometrics, 15, 447-470. Phun, V.K., Yai, T. (2016) State of the art of paratransit literatures in Asian developing countries. Asian Transport Studies, 4, 57-77. Savills (2016) Vietnam Chart Book 2016. Available online at http://pdf.savills.asia/asia-pacific-research/vietnam-research/hcmc/vietnam-chartbook-2 016.pdf. Train, K. (2003) Discrete Choice Methods with Simulation. Cambridge University Press, CA, USA. Tuan, V.A., Son, D.T. (2015) Accessibility to public transport systems in developing countries - An empirical study in Ho Chi Minh City, Vietnam. Journal of the Eastern Asia Society for Transportation Studies, 11, 1240-1258. 2298