Abstract:
The goal of this study is to propose a number of water transportation-related strategies which enhance the usage of the water transportation mode. In this study, Nonthaburi area, which is suburban of Bangkok, and Downtown Bangkok were selected as the study areas for the analysis. In contrast to the surface transportation, water transportation has a potential to provide a good service for passengers and goods movements. Particularly, in Bangkok there is a Chao Phraya river which runs through the center of city. It provides reliable, pollution-free, and scheduled transportation service. To obtain the model representing the trip characteristics and mode choice behavior, the Revealed Preference (RP) is employed. The Multinomial and Nested Logit Model were examined in order to explain the mode choice behaviors of travelers who travel between the study areas. The significant attributes in demand model were identified. Examples of significant attributes are in-vehicle travel time, out-of vehicle travel time, travel cost, number of car ownership, income level. The results showed that different trip purposes, work and non-work trips, had totally different mode choice behaviors. In the application stage, the developed model is employed to estimate the effects of policy decisions on travel behavior of overall patronage by changes in travel time and cost of water transportation mode. Several strategies that can improve the water transportation mode are discussed including service frequencies, information technology and multimodal transit center.
Key Words: Water Transportation, Mode Choice, Logit Model
1. INTRODUCTION
In contrast to the surface transportation, water transportation has a potential to provide a goodservice for passengers and goods movements. It provides reliable, pollution-free, and scheduled transportation service. The goal of this study is to propose a number of water transportation-related strategies which enhance the usage of the water transportation mode. In this study, Nonthaburi area, which is suburban of Bangkok, and Downtown Bangkok are selected as the study areas for the analysis. To obtain a model representing the trip characteristics and mode choice behavior, the Revealed Preference (RP) is employed. The Multinomial and Nested Logit Model are examined in order to explain the mode choice behaviors of travelers who travel between the study areas. After identifying significant attributes in mode choice behaviors, this studyestimates the effects of policy decisions on travel behavior of overall patronages caused by changes in significant attributes such as travel time and cost of water transportation system.
2. LITERATURE REVIEW
There are several approaches in forecasting the demand of water transportation. The traditional four-step travel demand forecasting process can be employed to predict traffic demand. However, smaller agencies often do not have either the data or the resources to develop and implement complex forecasting models for their systems. Also this approach requires much more data in each step to achieve the objective of study and consequently cause the high budget and time in conducting those data (Savage, 1998). For this reason, a variety of simplified forecasting methods are widely used with socioeconomic data from published sources.
Forecasting water transportation ridership is much more difficult than predicting roadway traffic volumes. The capacity and availability of a ferry depend on the size, frequency, and schedule of the vessels operating on a given day or given portion of a day. The existing studies of the demand analysis of water transportation are broadly classified into two: Timeseries analysis and Logit-based demand model.
There are several approaches in forecasting the demand of water transportation. The traditional four-step travel demand forecasting process can be employed to predict traffic demand. However, smaller agencies often do not have either the data or the resources to develop and implement complex forecasting models for their systems. Also this approach requires much more data in each step to achieve the objective of study and consequently cause the high budget and time in conducting those data (Savage, 1998). For this reason, a variety of simplified forecasting methods are widely used with socioeconomic data from published sources.
Forecasting water transportation ridership is much more difficult than predicting roadway traffic volumes. The capacity and availability of a ferry depend on the size, frequency, and schedule of the vessels operating on a given day or given portion of a day. The existing studies of the demand analysis of water transportation are broadly classified into two: Timeseries analysis and Logit-based demand model.
2.1 Time-Series Analysis Approach
The ability to predict travel demand requires that certain conditions be met. Of this method, first and foremost is that information is available about past travel volumes and characteristics for the types of demand needed to predict, and that the information is accurate and complete. More than any other factor, the quality of the historical data determines the accuracy of the resulting forecasts. A simple straight-line projection, estimated manually using graph paper and a straight edge, made with accurate historical data might be preferable to computer models derived from sparse and unreliable data that produce statistically complex but basically unsound results.
There are two major types of forecasting methods in this category, i.e., time-series analysis (TSA) and explanatory of cause-and-effect models(Savage, 1998). Usually, time-series methods assume that historical patterns observed in ridership data will continue, and that future levels of ridership can be extrapolated from mathematical analysis of past trends. The basic assumption is that the condition growth, economic activity, and ferry services will continue without any significant changes relative to their historical trends. For explanatory models, they assume that there are definable relationships between ferry demand and one or more independent variables such as population, employment, household income, and retail sales. Statistical methods are applied to various combinations of ferry ridership and independent variables to define these relationships. The least squares regression techniques were applied to combine the ferry ridership and independent variables yielding explanatory or cause-and effect models. These methods are preferred for long-range ferry forecasts because they provide an estimate of the changes in ferry ridership in relation to expected changes in economic activity levels and conditions in the ferry system’s service area. Savage (1998) applied these two types of approach to forecast the long-range ridership of ferry in relation to expected changes in economic activity levels and conditions in the ferry system’s service area.
Baker and Deardorf (1998) developed the time series analysis models that are used to project ferry ridership at the individual route level by six fare categories. A combination of statistical approaches is used to develop near- to mid-range ridership and revenue forecasting models for Washington State Ferries for use in quarterly budget updates. Regression models with historical and forecast trends in state economic and demographic variables was used to predict system wide ridership by six fare categories for different fare scenarios. The sum of the time series route forecasts is calibrated to the econometric system wide totals to yield unconstrained ridership forecasts by route and fare category.
The ability to predict travel demand requires that certain conditions be met. Of this method, first and foremost is that information is available about past travel volumes and characteristics for the types of demand needed to predict, and that the information is accurate and complete. More than any other factor, the quality of the historical data determines the accuracy of the resulting forecasts. A simple straight-line projection, estimated manually using graph paper and a straight edge, made with accurate historical data might be preferable to computer models derived from sparse and unreliable data that produce statistically complex but basically unsound results.
There are two major types of forecasting methods in this category, i.e., time-series analysis (TSA) and explanatory of cause-and-effect models(Savage, 1998). Usually, time-series methods assume that historical patterns observed in ridership data will continue, and that future levels of ridership can be extrapolated from mathematical analysis of past trends. The basic assumption is that the condition growth, economic activity, and ferry services will continue without any significant changes relative to their historical trends. For explanatory models, they assume that there are definable relationships between ferry demand and one or more independent variables such as population, employment, household income, and retail sales. Statistical methods are applied to various combinations of ferry ridership and independent variables to define these relationships. The least squares regression techniques were applied to combine the ferry ridership and independent variables yielding explanatory or cause-and effect models. These methods are preferred for long-range ferry forecasts because they provide an estimate of the changes in ferry ridership in relation to expected changes in economic activity levels and conditions in the ferry system’s service area. Savage (1998) applied these two types of approach to forecast the long-range ridership of ferry in relation to expected changes in economic activity levels and conditions in the ferry system’s service area.
Baker and Deardorf (1998) developed the time series analysis models that are used to project ferry ridership at the individual route level by six fare categories. A combination of statistical approaches is used to develop near- to mid-range ridership and revenue forecasting models for Washington State Ferries for use in quarterly budget updates. Regression models with historical and forecast trends in state economic and demographic variables was used to predict system wide ridership by six fare categories for different fare scenarios. The sum of the time series route forecasts is calibrated to the econometric system wide totals to yield unconstrained ridership forecasts by route and fare category.
2.2 Logit-Based Demand Model
Berkowitz (1990) evaluated the viability of implementing new waterborne passenger transportation systems. The analysis of passenger travel characteristics and ridership potential was conducted using a logit-based demand model. The model and data collection techniques included the essential elements of time, cost, comfort, convenience, special enjoyment, and validation. Special emphasis was given to the analysis of passenger travel characteristics and mode choice preference. In addition, the model was used to estimate the effects of policy decisions on travel behaviors. The effects of changes in various policies were represented in terms of overall patronage estimates by varying the values of one or two variables. Thecompetitive modes in his study area consist of passenger ferry, express bus, and automobile.
The mode shift from the Staten Island Ferry to express bus service was attributed to the express buses’ improved travel time, levels of comfort and convenience, and pricing structure. He found that the increase in ferry users is attributed to the population explosion on StatenIsland, increases in express bus fares, express bus system capacity restraints, and reductions in the general quality of express bus service. In addition, improvements to the express bus system, including special bus lanes and traffic control modifications, have not significantly improved the system’s operating characteristics.
Alijarad and Black (1995) analyzed intercity mode choices in the Saudi Arabia-Bahrain corridor using logit model. The study used water transportation, air transportation and land transportation as the main modes available for the commuters. The land transportation mode was further divided into public transit and others. It was found that variables such as frequency of service, immediate decision time, etc. played an important role in the mode choice. As the water transportation service was offered only twice a day, time of decision was considered as one of the major factors in mode choice.
3. STUDY AREA AND DATA COLLECTION
3.1 Water Transportation in Bangkok and Study Area In Bangkok, demand for transportation facilities and services far outweighs the supply. Although almost of the metropolitan areas face this challenge, the problem is especially acute in Bangkok because of rapid growth in population and employment. Analysis has shown that unless the government takes action to make improvements, transportation system performance will decline dramatically over the next 10 years. In the last decade, traffic engineers and practitioners had come to the conclusion that traffic problems can be mitigated through the appropriate Transportation Demand Management (TDM) measures. TDM is a measure covering a broad range of demand management schemes including a improvement of the public transportation system to encourage the usage of the mass transportation system.
The Chao Phraya river in Bangkok, Thailand, separating the land of Bangkok into two majorareas namely “Thonburi” and “Phra Na Khon” zone, is an important mode of transportation not only inside Bangkok but also the boundary of Bangkok. Ferry services on the Chao Phraya River which connect Bangkok with the northern neighboring province of Nonthaburi area are operated in different types of ferry classified by the color of flag . The water transportation of Chao Phraya river has played an important role in moving people from suburban to downtown for the last decades. However, the usage of the water transportation of Bangkok has been reported to be decreasing. In this study, two areas in Bangkok which are downtown of Bangkok and Nonthaburi area were selected for data collection and analysis.
Berkowitz (1990) evaluated the viability of implementing new waterborne passenger transportation systems. The analysis of passenger travel characteristics and ridership potential was conducted using a logit-based demand model. The model and data collection techniques included the essential elements of time, cost, comfort, convenience, special enjoyment, and validation. Special emphasis was given to the analysis of passenger travel characteristics and mode choice preference. In addition, the model was used to estimate the effects of policy decisions on travel behaviors. The effects of changes in various policies were represented in terms of overall patronage estimates by varying the values of one or two variables. Thecompetitive modes in his study area consist of passenger ferry, express bus, and automobile.
The mode shift from the Staten Island Ferry to express bus service was attributed to the express buses’ improved travel time, levels of comfort and convenience, and pricing structure. He found that the increase in ferry users is attributed to the population explosion on StatenIsland, increases in express bus fares, express bus system capacity restraints, and reductions in the general quality of express bus service. In addition, improvements to the express bus system, including special bus lanes and traffic control modifications, have not significantly improved the system’s operating characteristics.
Alijarad and Black (1995) analyzed intercity mode choices in the Saudi Arabia-Bahrain corridor using logit model. The study used water transportation, air transportation and land transportation as the main modes available for the commuters. The land transportation mode was further divided into public transit and others. It was found that variables such as frequency of service, immediate decision time, etc. played an important role in the mode choice. As the water transportation service was offered only twice a day, time of decision was considered as one of the major factors in mode choice.
3. STUDY AREA AND DATA COLLECTION
3.1 Water Transportation in Bangkok and Study Area In Bangkok, demand for transportation facilities and services far outweighs the supply. Although almost of the metropolitan areas face this challenge, the problem is especially acute in Bangkok because of rapid growth in population and employment. Analysis has shown that unless the government takes action to make improvements, transportation system performance will decline dramatically over the next 10 years. In the last decade, traffic engineers and practitioners had come to the conclusion that traffic problems can be mitigated through the appropriate Transportation Demand Management (TDM) measures. TDM is a measure covering a broad range of demand management schemes including a improvement of the public transportation system to encourage the usage of the mass transportation system.
The Chao Phraya river in Bangkok, Thailand, separating the land of Bangkok into two majorareas namely “Thonburi” and “Phra Na Khon” zone, is an important mode of transportation not only inside Bangkok but also the boundary of Bangkok. Ferry services on the Chao Phraya River which connect Bangkok with the northern neighboring province of Nonthaburi area are operated in different types of ferry classified by the color of flag . The water transportation of Chao Phraya river has played an important role in moving people from suburban to downtown for the last decades. However, the usage of the water transportation of Bangkok has been reported to be decreasing. In this study, two areas in Bangkok which are downtown of Bangkok and Nonthaburi area were selected for data collection and analysis.
3.2 Data Collection and Preliminary Analysis
Sample is defined as “a collection of units which has been especially selected to represent a larger population with certain attributes of interest”. From the above definition, three aspects have particular importance. First, which population the sample seeks to represent; second, how large the sample should be; and third, what is meant by especially selected (Ortuzar and Willumsen, 1994).
The questionnaire design by applying Revealed Preference(RP) is done prior to distribution to the respondents. The questionnaire has to be clear and ready to respond all necessary variables employed to develop the model. In order to measure how much the respondents experienced to comfort, a set of multiple choices such as very poor, poor, fair, good, and very good was provided to respondents. The respondents had to select only one choice among them to represent their experienced to comfort. The survey was conducted in November 2002 in Nonthaburi and Bangkok’s downtown areas and the total number of data used in the model development was 1,060 samples. In this study, in order to measure how much the respondents experienced to comfort, a set of multiple choices such as very poor, poor, fair, good, and very good is provided to respondents. The respondents have to select only one choice among them to represent their experienced to comfort. This choice is then replaced by number from one to five, respectively, and will be used to calibrate the model.
4. MODEL ESTIMATION
4.1 Market Segmentation
At first a model was tried with the assumption that there was no taste difference between trip purposes. The value of travel time for different trip purpose was assumed to be the same. Moreover those travelers with different trip purpose were assumed to have the same structure of modal choice. For this, a market segmentation test by trip purpose was done to verify the different taste between work and non-work trip. The assumption was that two trip purpose groups have totally different mode choice behavior and each group has its own utility function that does not relate with the pooled model. The coefficients of segmented model using the same model structure and specification to the best model from the pooled data set were seen to have the same sign except number of car ownership In order to judge whether the segmented model is better than the pooled model, the total absolute deviation was calculated. The total absolute deviation for the pooled model was obtained by summing the absolute deviation from all modes. For the segmented model, the weighted average deviation of each mode was first carried out. Then the total absolute deviation of segmented model was obtained by summing the weighted average deviation from all modes. The total absolute deviation of pooled and segmented model in this study was found to be 119.78 and 142.90, respectively. Thus, from the perspective of model calibration, the pooled model was considerably better than the segmented model.
4.2 Nested Logit Model and Multinomial Logit Model
Six different types of nested logit model were developed and compared with multinomial logit model. Based on the log-likelihood values of the nest logit model and the results of t-test of each independent variables, multinomial logit model was found to be better than nested logit model.
4.3 Value of Travel Time
Another perspective of the suitability of the model can be examined by evaluating the value of travel time. For the pooled model, the implied value of total travel time was computed using the estimated total journey time and total cost coefficients. For work and non-work trip, the implied value of travel time was computed differently according to the travel time variables used in the model. Table 3 shows the result of value of travel time for pooled, work trip, and non-work trip model, respectively. It was observed that the value of travel time for work trip was higher than that of the pooled model while the non-work trip had the lowest value of travel time. This result was also consistent with the priori expectation.
The result from the developed model is rather higher than that of the previous study, particularly in case of car and taxi. However, the previous study considered the value of time for all trip purposes (pooled) rather than specific to any trip purpose. It is expected that the value of time for work trip should be higher than that of the non-work trip and pooled model. In this study, the values of time of ferry, bus, and van for both pooled and segmented model were found to be in the same range as obtained from the previous study. However, the values of travel time of private car and taxi for both the pooled and segmented model were comparatively higher than those of the previous study. The segmented model, however, was selected and will be further applied in this study. It is because the segmented model includesthe more detail of travel time variables, and therefore it would be more useful when applied to the policy sensitivity analysis
5. CONCLUDING REMARKS
Bangkok has a good potential to rely on water transportation in order to improve the performance of the overall transportation system. It is because Bangkok has Chao Phraya river which runs through the center of city, and water transportation provides reliable, environment friendly, and scheduled transportation service. However, it is known that patronage of water transportation using Chao Phraya river has been decreasing. In this context, by developing logit model, this study proposed a number of water transportationrelated strategies which enhances the usage of the water transportation mode. Nonthaburi area, which is suburban of Bangkok, and Downtown Bangkok were selected as the study areas for the analysis. To develop a logit model representing the trip characteristics and mode choice behavior, the revealed preference based survey was employed. The Multinomial and Nested Logit Model were examined in order to explain the mode choice behaviors of travelers who travel between the study areas. The results showed that different trip purposes, work and non-work trips, had totally different mode choice behaviors.
In the application stage, the developed model is employed to estimate the effects of policy decisions on travel behavior of overall patronage by changes in service characteristics of water transportation mode. Based on the sensitivity analysis, following strategies are suggested:
Sample is defined as “a collection of units which has been especially selected to represent a larger population with certain attributes of interest”. From the above definition, three aspects have particular importance. First, which population the sample seeks to represent; second, how large the sample should be; and third, what is meant by especially selected (Ortuzar and Willumsen, 1994).
The questionnaire design by applying Revealed Preference(RP) is done prior to distribution to the respondents. The questionnaire has to be clear and ready to respond all necessary variables employed to develop the model. In order to measure how much the respondents experienced to comfort, a set of multiple choices such as very poor, poor, fair, good, and very good was provided to respondents. The respondents had to select only one choice among them to represent their experienced to comfort. The survey was conducted in November 2002 in Nonthaburi and Bangkok’s downtown areas and the total number of data used in the model development was 1,060 samples. In this study, in order to measure how much the respondents experienced to comfort, a set of multiple choices such as very poor, poor, fair, good, and very good is provided to respondents. The respondents have to select only one choice among them to represent their experienced to comfort. This choice is then replaced by number from one to five, respectively, and will be used to calibrate the model.
4. MODEL ESTIMATION
4.1 Market Segmentation
At first a model was tried with the assumption that there was no taste difference between trip purposes. The value of travel time for different trip purpose was assumed to be the same. Moreover those travelers with different trip purpose were assumed to have the same structure of modal choice. For this, a market segmentation test by trip purpose was done to verify the different taste between work and non-work trip. The assumption was that two trip purpose groups have totally different mode choice behavior and each group has its own utility function that does not relate with the pooled model. The coefficients of segmented model using the same model structure and specification to the best model from the pooled data set were seen to have the same sign except number of car ownership In order to judge whether the segmented model is better than the pooled model, the total absolute deviation was calculated. The total absolute deviation for the pooled model was obtained by summing the absolute deviation from all modes. For the segmented model, the weighted average deviation of each mode was first carried out. Then the total absolute deviation of segmented model was obtained by summing the weighted average deviation from all modes. The total absolute deviation of pooled and segmented model in this study was found to be 119.78 and 142.90, respectively. Thus, from the perspective of model calibration, the pooled model was considerably better than the segmented model.
4.2 Nested Logit Model and Multinomial Logit Model
Six different types of nested logit model were developed and compared with multinomial logit model. Based on the log-likelihood values of the nest logit model and the results of t-test of each independent variables, multinomial logit model was found to be better than nested logit model.
4.3 Value of Travel Time
Another perspective of the suitability of the model can be examined by evaluating the value of travel time. For the pooled model, the implied value of total travel time was computed using the estimated total journey time and total cost coefficients. For work and non-work trip, the implied value of travel time was computed differently according to the travel time variables used in the model. Table 3 shows the result of value of travel time for pooled, work trip, and non-work trip model, respectively. It was observed that the value of travel time for work trip was higher than that of the pooled model while the non-work trip had the lowest value of travel time. This result was also consistent with the priori expectation.
The result from the developed model is rather higher than that of the previous study, particularly in case of car and taxi. However, the previous study considered the value of time for all trip purposes (pooled) rather than specific to any trip purpose. It is expected that the value of time for work trip should be higher than that of the non-work trip and pooled model. In this study, the values of time of ferry, bus, and van for both pooled and segmented model were found to be in the same range as obtained from the previous study. However, the values of travel time of private car and taxi for both the pooled and segmented model were comparatively higher than those of the previous study. The segmented model, however, was selected and will be further applied in this study. It is because the segmented model includesthe more detail of travel time variables, and therefore it would be more useful when applied to the policy sensitivity analysis
5. CONCLUDING REMARKS
Bangkok has a good potential to rely on water transportation in order to improve the performance of the overall transportation system. It is because Bangkok has Chao Phraya river which runs through the center of city, and water transportation provides reliable, environment friendly, and scheduled transportation service. However, it is known that patronage of water transportation using Chao Phraya river has been decreasing. In this context, by developing logit model, this study proposed a number of water transportationrelated strategies which enhances the usage of the water transportation mode. Nonthaburi area, which is suburban of Bangkok, and Downtown Bangkok were selected as the study areas for the analysis. To develop a logit model representing the trip characteristics and mode choice behavior, the revealed preference based survey was employed. The Multinomial and Nested Logit Model were examined in order to explain the mode choice behaviors of travelers who travel between the study areas. The results showed that different trip purposes, work and non-work trips, had totally different mode choice behaviors.
In the application stage, the developed model is employed to estimate the effects of policy decisions on travel behavior of overall patronage by changes in service characteristics of water transportation mode. Based on the sensitivity analysis, following strategies are suggested:
1) In terms of gaining more ferry riders of work trip, the performance of ferry service such as operating speed and adjusting skip stop strategy should be enhanced to reduce in-vehicle travel time of ferry. Adjusting fare structure and/or providing free feeder buses can be adopted in this case. To gain more ferry riders of the non-work trip, since the non-work trip makers were more sensitive in in-vehicle travel time than out-of-vehicle travel time, the same policy as of work trip should be adopted 2) Another issue that reflects the inconvenience of using ferry is the long walking distance on access and/or egress to/from pier. The existing condition at the Nonthaburi pier is a very good example for this case. It is because most of the bus coming from Ngam Wong Wan did not provide the services up to Nonthaburi pier. If the travelers want to go to Nonthaburi pier, they have to change the bus or further walk to Nonthaburi pier. The coordination with bus agency should be taking into consideration. 3) The lack of ferry services information is another issue that can be solved. The information such as the nearest get-in and get-off pier for various places, and other public transportation that link to each pier should be provided to travelers.
Tidak ada komentar:
Posting Komentar