This item is only available as the following downloads:
1 THE FACTORS THAT AFFECT LONG DISTANCE TRAVEL MODE CHOICE DECISION S AND THEIR IMPLICATIONS F OR TRANSPORTATION P OLICY BY HEE DEOK CHO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 3
2 201 3 Hee Deok Cho
3 To my family
4 ACKNOWLEDGMENTS I would like to give a special thanks to my chair Dr. Ruth Steiner from the bottom of my heart for h er continued patience and guidance Ruth directed and guided me in a right direction whenever I was having a difficulty working on my disse rtation and my interest endeavor while studying in Urban and Regional Planning department. She helped me to gain better understanding for the bigger pictures. I would also appreciate Dr. Siva Srini vasan who provided me valuable instruction, and guidance throughout this entire process. H e is an admirable professor and from h im I have learned the importance of working with something that I feel passionate about. Additionally I want to thank Dr. Paul Zwick and Dr. Andres Blanco who give me continuous moral and academic support and advice I am very lucky to have wonderful people in my life. I give my special appreciation to Carthell Everett, who has reviewed my papers and project reports. I would also like to thank all the people who spent time with me in Urban and Regional Planning (faculty, staff, and colleagues). It was great for me to live South Korea who always liste n to my complaints patiently. Last but certainly not least, I would like to thank my family. It has been long and hard time for them to endure. In particular, I am deeply grateful to my parents in law for their prayer and sacrifice. I also give thanks to m y wife and two daughters. I could not come to here without their love patient, and endless support. They are the reason why I have to live with my best in this world. I cannot wait to give them a huge hug
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 9 LIST OF ABBREVIATIONS ................................ ................................ ........................... 10 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 Background ................................ ................................ ................................ ............. 13 Statement of Problems ................................ ................................ ........................... 16 Purpos e of Study ................................ ................................ ................................ .... 19 Scope of Study ................................ ................................ ................................ ....... 22 2 LITERATURE REVIEWS ................................ ................................ ........................ 24 3 ANALYSIS METHOD AND DATA ................................ ................................ ........... 30 Theoretical Framework of Mode Choice Models ................................ ..................... 30 Probabilistic Choice Theory ................................ ................................ .............. 30 The Multinomial Logit Model ................................ ................................ ............. 30 The Conditional Logit Model ................................ ................................ ............. 32 Model Specification ................................ ................................ ................................ 34 Data ................................ ................................ ................................ ........................ 36 The 2009 National Household Travel Survey ................................ ................... 36 Operationaliz ing Daily Trips in to Household Level Data ................................ ... 38 Descriptive Analysis of Household Level Long Distance Travel Data ..................... 40 4 ESTIMATION OF MODEL CHOICE MODELS ................................ ....................... 51 Explanatory Variables ................................ ................................ ............................. 51 Hypoth etical Relationships of Variables to Mode Choice Decisions ....................... 53 Estimating Synthetic Travel Times and Costs ................................ ......................... 57 Automobile Travel Times and Costs ................................ ................................ 57 Airline Travel Time and Costs ................................ ................................ .......... 59 Bus and Train Travel Time and Costs ................................ .............................. 61 Results of Model Estimations ................................ ................................ .................. 62
6 Travel Time and Costs Variables ................................ ................................ ..... 63 Individual Socio Demographic Variables ................................ .......................... 65 Geographical Characteristics of Household ................................ ..................... 66 5 POLICY IMPLICATIONS FOR FUTURE LONG DISTANCE TRNSPORTATION PLAN ................................ ................................ ................................ ...................... 71 Average Marginal Effects of Travel Time and Costs ................................ ............... 71 Estimations of the Probabilities Choosing a New Alternative Mode by Travel Time and Costs Scenario ................................ ................................ .................... 73 Estimations of the Probabilities of Choosing an I mproved T rain for the Whole of the US ................................ ................................ ............................ 74 Probability Estimations for the Major MSA Corridors ................................ ........ 77 Policy Recommendations ................................ ................................ ....................... 84 6 CONCLUSIONS AND FURTHER STUDY ................................ .............................. 88 Conclusions ................................ ................................ ................................ ............ 88 Further study ................................ ................................ ................................ ........... 92 APPENDIX: DISCRIPTIVE STATISTICS OF LONG DISTANCE TRIP BY STATE ...... 94 LIST OF REFERENCES ................................ ................................ ............................... 97 BIOGRAPHIC AL SKETCH ................................ ................................ .......................... 106
7 LIST OF TABLES Table page 3 1 Household, Individuals and Daily Trips in the 2009 NHTS ................................ 37 3 2 Data Operationalization Processes Obtaining Household Level Dataset ........... 39 3 4 Descriptive Statistics of Long distance Trips by Travel Mode ............................ 41 3 5 Share of Transportation Mode by Distance Group ................................ ............. 42 3 6 Descriptive Statistics of Long Distance Travel by Trip Purpose ......................... 44 3 7 Origins and Destinations of Long Distance Travel ................................ .............. 44 3 8 Share of Long Distance Travel by Trip Purpose ................................ ................. 45 3 9 Comparison of Travel Mode Share by Existence of Heavy Rail System ............ 46 3 10 Share of Travel Mode by MSA Category ................................ ............................ 46 3 11 Patterns of Long Distance Travel by Income Group ................................ ........... 47 3 12 Descriptive Statistics of Long Distance Travel for Traveling with Child .............. 49 3 13 Descriptive Statistics of Long Distance Travel by Existence of the Elderly ........ 49 4 1 Potential Explanatory Variables ................................ ................................ .......... 52 4 2 Average Distance to the Closest Airport by Census Division ............................. 60 4 3 Average Distances to Bus Terminal and Rail Station by Census Division .......... 62 4 2 Estimations of Mode Choice Models ................................ ................................ .. 69 4 2 Estimations of Mode Choice Models (Continued) ................................ ............... 70 5 1 Average Marginal Effects of Travel Time and Costs by Mode ............................ 72 5 2 The Probabilities of Choosing an improved train System by Travel Time and Cost Scenario ................................ ................................ ................................ ............. 76 5 3 Comparison of the Probabilities of Choosing an improved train System by Corridor ................................ ................................ ................................ .............. 81
8 A 1 Long Distance Trips and Average Trip Length by State ................................ ..... 94 A 2 Descriptive Statistics of Long Distance Travel b y Mode and b y Division ............ 96
9 LIST OF FIGURES Figure page 3 1 Share of Long Distance Travel Mode by Census Division ................................ .. 41 3 2 Distribution of Distance Group by Census Division ................................ ............ 42 3 3 Share of Travel Mode by Distance Group ................................ .......................... 43 3 4 Share of Long Distance Travel Mode by Income Group ................................ ..... 47 3 5 Share of Long Distance Travel Mode by Existence of Child ............................... 49 3 6 Share of Long Distance Travel Mode by Existence of the Elderly ...................... 50 4 1 ................. 56 5 1 The Probabilities of Choosing an improved train Mode by Travel Time and Cost Scenario ................................ ................................ ................................ ..... 77 5 2 The Probabilities of Choosing an improved train Mode by Corridor and by Travel Time and Cost Scenario ................................ ................................ .......... 82 5 2 The Probabilities of Choosing an improved train Mode by Travel Time and Cost Scenario and by Corridor (Continues) ................................ ........................ 83 5 3 Comparison of t he Probabilities of Choosing an improved train Mode with 1.5 times of Driving Costs ................................ ................................ ......................... 85
10 LIST OF ABBREVIATION S AAA American Automobile Association ARTBA American Road & Transportation Builders Association BTS Bureau of Transportation Statistic CL Conditional Logit CUTR Center for Urban Transportation Research at University of South Florida DOT U.S. Department of Transportation ENC East North Central division FAA Federal Aviation Administration FDOT Florida Department of Transportation FRA Federal Railroad Administration HSR High Speed Rail MNL Multinomial Logit MT Mountain division NEC North East Corridor NHTS Nationa l Household Travel Survey PAC Pacific division RP Revealed Preference SA South Atlantic division SP Stated Preference
11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THE FACTORS T HAT AFFECT LONG DISTANCE TRAVEL MODE CHOICE DECISIONS AND THEIR IMPLICATIONS FOR TRANSPORTATION POLICY By H ee Deok Cho May 201 3 Chair: Ruth L. Steiner Major: Design, Construction and Planning The overall goal s of this study are to enhance the fundamental understanding of long distance travel characteristics in the US, and to provide policy implications for long distance transportation planning in the future. This study uses the 2009 National Household Travel Survey (NHTS) and state add on data that provide a daily trip data base in both national and state perspectives. In particular, this study focuses on long distance trips defining long distance trip as a trip segment that is 50 or more miles from origin to destination. In order to achieve the research goal, this study first summarizes current patterns and characteristics of long distance travel in the United States. In addition, this study develops mode choice models for long distance travel that can explain the relationship between travele rs choices of transportation mode and a set of explanatory variables such as alternative specific attributes (travel time, travel cost s and access/egress time and cost s ), individual characteristics (income, age, and trip purpose) and geographical charact eristics of household (MSA category, and existence of heavy rail service) T his study e stimates the Multinomial Logit models based on the mathematical function of the Conditional Logit (CL) mode l
12 In order to develop logit models, this study estimates syn thetic travel time and costs of all available transportation modes. In estimating synthetic travel time and costs, this study used all possible sources of published data including average driving costs per mile by passenger car type, air passengers fare an d flight distance survey, bus and train fare and travel time tables. In addition, this study calculated shortest distance from each household to intercity terminals such as 422 commercial airports, 1482 bus terminals, and 533 train stations. Finally, this study predicted the probability of choosing a new alternative by applying travel time and cost s scenarios to the estimated model. For that, t his study assumes a new alternative mode as an improved train system because no other ground modes are able to provide a speed of 200 or more miles. Based on the findings, this study suggests viable options for planners and decision makers to plan for long distance transportation in the future.
13 CHAPTER 1 INTRODUCTION Background E xtensive population growth and economic development h ave caused continuous increase s in travel demand, average travel distance, and consequently congestion and delays. A verage daily person mile s traveled (PMT) in the US for example, increased from 83.1 miles to 90.4 miles between 2001 and 2009 Meanwhile, a verage vehicle mile s traveled (VMT) increased from 49.8 miles to 54.4 miles between 1990 and 2009 (Santos et al 2011 p. 10 ). I t should be noted that highway delays near urban areas caused travelers to waste an average of 34 hours resulting in 3.9 billion gallons of fuel as of 2009. Overall, roadway congestions are known to c ause a total of $115 billion congestion cost in 2 009 (Schrank et al. 2010 p.7 ) In addition, 30 percent of the domestic flights in the U S arrived late in 2007 which is up from 20 percent in 2003 (Whalen Carlton, Heyer & Richard 2008 p.31 ). The delays of air transportation are estimated to result in $32.9 billion of total cost s in 2007 of which $16.7 billion is direct costs to passengers (Ball et al, 2010 p.3 ). T hese problems are expected to be worse in the future because the U.S. population is expected to grow to 439 million and the GDP is expected to increase to 43 trillion over the next 40 years ( U.S. DOT, 2010 p.31 ) increasing travel demand on both highways and airports significantly. In order to address growing congestion and delay problems federal, state and local governments have traditio nally focused on the capacity expansion of existing transportation infrastructure through significant capital outlays However, even these capacity additions will not be able to f ully accommodate the predicted highway s and airports demand at an adequate se rvice level. Instead, h ighway congestion is rather
14 expected to spread to medium sized cities small cities and even suburban and rural fringe over the next 10 to 15 year s ( U.S. DOT, 2006 p.9 ). Meanwhile, providing additional capacity of highways near ur ban area s is increasingly difficult not only because highways are becoming increasingly expensive to build, but also because h ighway construction cost around urban area s 8 10 million per mile of 4 lane highway, is nearly double compare to that of rural an d suburban areas, 4 6 million per mile (ARTBA, 201 0 ). It is also estimated that a greater number of large hub airports and their associated metropolitan areas are expected to face capacity constraints by 2013 and 202 0. Besides, they will not be able to inc rease capacity beyond what is currently p rovided because some areas have limited geographical capability to increase runways (The MITRE Corporation, 2007 p.6 ) T hese concerns have motivated both policymakers and researchers to acknowledge the needs of a new approach that can address current congestion problems as well as growing travel demand in the future. A variety of alternative options have been evaluated over the past decade s and c onsiderable attention has been directed toward new ground transportat ion mode such as high speed rail ( HSR ) Maglev rail, and upgraded rail service s T he US Department of Transportation (DOT) has attempt ed to provide a full array of affordable and practical mode choice option s to reduce road travel ( U.S. DOT, 200 6 ; 2010). T he Federal Aviation Administration (FAA) has emphasize d the importance of exploring alternative means of transportation that can substitute short to medium distance air routes ( U.S. DOT, FRA, 1997; U.S. DOT, FAA 2001). In th e same context, t he FAA recommended the development of an
15 efficient and effective intermodal system to mitigate capacity problem s at airports ( U.S. DOT, FRA, 1997). In those provisions, the HSR option has been given top priority that is considered as a viable option to mi tigate both highway and airport congestions by diverting long distance travel demand to its system and to reduce the need of massive capital spending to pay for capacity expansions in other existing modes (Peterman Frittelli & Mallett 2009). The HSR system is also expected to improve attractiveness and the potential economic development of the regions by promoting accessibility and connectivity. These extensive interests and efforts resulted in the Obama s active leadership in a funding decision of $8 billion down payment on the HSR system in a 100 600 mile range intercity corridors that connect major metropolitan areas in the US ( U.S. DOT, FRA, 2009). However, it has been difficult to implement the high speed rail ( HSR ) proposals in the US because the s uccess of the HSR option has been in question Much of the criticism of HSR is based on the concerns about its cost effectiveness in the near to medium distance range. One of the major challenge s geography with lo w er population density in urban areas compared to urban areas in Asia and Europe (Peterman et al 2009). More importantly, against p ersonal automobiles puts the success of the system in question. Critics argue that HSR will not be able to compete with personal cars in the US because the 58 percent of the long distance trips are not that long, less than 250 miles (or 400 km ) in ro und trip distance ( U.S. DOT, FHWA, 2006) and personal cars account for more than 90 percent of trips tha t longer than 50 miles (U.S. DOT, BTS, 20 06 ). These travel
16 patterns transportation because previous studies in Europe and Asia have suggested that HSR can have strong potential at a one way s ervice range between 250 mile and 500 mile Given these promises and concerns, it is needed to evaluate whether or not a new alternative high speed train system would be a viable option in the US. In the same context, it is necessary to understand whethe r there is such a desired service qualit y for a new alternative mode to become an effective and efficient option. However, t here ha ve not been sufficient efforts to evaluate the potentiality of HSR system in the US even though such supports are essential. In particular, academic studies have not been active in developing a broad spectrum of measures to propose a viable option for the US. Statement of Problems When a new alternative mode of transportation enter s into an existing transportation market, it wi ll inevitably compete with other modes of transportation in the market, and consequently a portion of other mode use will shift to a new mode T h at portion of shift is critical to evaluate the adequacy and efficiency of a new transportation investment in a wide transportation system In this respect, the focus has to be o n the understanding of travelers mode choice behavior because travel demand of certain mode of transportation depends largely on the individuals choice decisions o f mode s However, previous studies of HSR system have been faced criticisms with regarding to methodologies, data, and reliability of ridership estimations. First of all, the method that was used to forecast the demand for HSR systems has raised several valid issues. Most ridership studies, for example: KPMG Peat Marwick LLP (1998) and SYSTRA (1998) for Florida corridor; Charles River Associates
17 (2000) and Cambridge Systematic (2006) for Bay Area / California High Speed Rail ; and Chen (201 1 ) for Richmond, VA Washington, D .C. HSR line have developed disaggregate demand models to forecast t he travel demand of HSR system s in a specific corridor. However, m ajority of these studies formulated mode choice as separate binary diversion models in which percentage of automobile, air, and bus passengers are diverted to HSR. It is the simplest specification to evaluate the impact of a new alternative mode on the com peting mode in the market, and to utilize the stated preference (SP) data that is collected for a HSR project in a given corridor. However, the binary diversion approach does not capture mode shift between the existing alternatives, and thus it can increas e the difficulty of evaluating level of service changes to existing modes ( Brownstone, Hansen, & Madanat, 2010) Secondly, most demand forecasting efforts have used imperfect data because there is lack of public accessible data sources Since HSR has uniq ue attributes and does not exist in the market yet, well designed and collected stated preference ( SP ), revealed preference (RP), or combined dataset is essential to evaluate such an option (Roth, 1998). However, m o st survey data sets collected for HSR proj ects are considered to include a large portion of samples from wrongly defined population s For example, SP data for California HSR surveyed largely from air transportation instead of personal car users that accounts for nearly 93 percent of total interreg ional travel Hence, the estimated models can produce distorted mode choice decision of the average intercity travelers (Brownstone, et al., 2010). In addition, the SP data is susceptible to serious problem s of survey respondent bias because respondents h ave the tendency to stick with their current mode choice regardless of the alternatives attributes (Roth, 1998).
18 Thus, the estimated models are likely to contain unobserved serial correlation making the parameter estimates in consistent. Third, most of pre vious ridership studies are commercially confidential studies, and thus demand forecasts are often significantly optimistic and different ridership forecasting methods have yielded uncertain and inconsistent results ( U.S. GAO, 2009). In addition, ridership studies have encountered criticism because parameters of the mathematical models were considerably adjusted from assigned values in the models, and thus the forecasts of demand have a strong likelihood of very large error bound from one level that can sup port the implementation of HSR investment to the other level that show significantly low demand (Brownstone et al. 2010). Th e reason is that most of ridership studies for HSR projects are so far, conducted by private consultants who are working under a v ariety of pressures and for different types of clients that are directly related to the implementation of HSR projects. Florida High Speed Rail Authority (FHSRA), for example, estimated Florida HSR line between Tampa and Orlando would attract more than 10 percent of automobile users on the interstate highway (AECOM Consulting and Wilbur Smith Associates 2002), but another study estimated that HSR would reduce traffic on the busiest sections of I 4 by less than 2 percent (Peterman et al, 2009). Similar results were produced in a study of HSR in the Northeast Corridor. The study noted that rail travel must be extremely competitive in factors such as speed or cost to attract automobile travelers because automobile travel is different from air and rail tra vel in that it involves door to door transportation, provides higher flexibility in time of departure and does not require individuals to share space with others ( U.S. DOT FRA 200 8 ).
19 Therefore, it is needed to develop reasonable and applicable measures that can provide answers for p olicy makers and private industries to the questions: what are the existing patterns and trends of long distance travel in the US and what are the factors that affect people to choose one transportation mode or another one ? Furthermore, it is essential to lay the foundations of understanding: what are desired service qualit ies that may change long distance travelers mode choice behavior, and what are viable strategies for the future long distance transportation plans ? Purpose of Study The overall goal of this study is to enhance fundamental understanding of long distance travel p atterns and characteristics of the US. T his study also aims to evaluate the potentiality of an alternative option for long distance passenger s ervice in the US and provide strategic policy options for transportation planning in the future. For that t his study analyzes current long distance travel patterns and trends in the US using the 2009 National Household Travel Survey (NHTS) dataset estim ate mode choice models for long distance travel, and predict the probability of choosing a new alternative option. In order to estimate mode choice models, this study develops methods to estimate synthetic travel time and costs for all alternative modes in the market. Since the NHTS provide information of travel time of mode used, this process is essential to develop mode choice models. This study first conducts a descriptive analysis of long distance travel using the 2009 NHTS which comprises all interviews from the national sample households, and the 20 add on partners The 2009 NHTS provides fully disaggregated information of personal and household characteristics of individual, actual travel behavior in long distance trips, a nd main attributes of travel including trip purpose, number of people on
20 that trip and travel time (Koppelman & Hirsh, 1986). Therefore, the 2009 NHTS is useful in examining the patterns and trends of long distance travel. Previously conducted descriptive studies have focused on the fundamental questions: who is traveling, where are people traveling on long distance trips, what are the modes used, and why are people traveling long distance s ? (Bricka, 2001; Mallett, 2001; O Nelill & Brown, 2001; U S DOT, BT S, 2003; U S DOT, BTS, 2006) In addition to these traditional questions, this study generate multi dimensional table to structure, summarize and display how those basic patterns are related with mode choice and travel distance. M oreover, this study prese nts origins and destinations of long distance travel. The descriptive analysis is expected to enhance the understanding of c urrent patterns and trends of long distance travel from both national and regional perspectives Second, t his study develops the multinomial log it (MNL) model s adopting the mathematical function of the conditional logit (CL) model. The MNL model have re latively simple and closed form mathematical structure s and thus are straight forward to estimate and interpret the interactions of choice behaviors with the explanatory variables. In the mathematical function of the CL model the probability of choosing a particular alternative mode is expressed as a function of the alternative specific variables and thus provide s direct measures to evaluate the impacts of changes in a specific alternative mode on the probability of choosing that mode. This study uses the choices of individuals among four modes such as personal cars, bus, airplane and train as dependent variable. Meanwhile, explanatory variables include both alternative specific variables ( such as travel time travel cost or/and travel distance ) socio demographic attributes of individuals ( for example, income age, gender, number of
21 vehicles in the household, number of people on that trip, and trip purpose ) and dummy variables that represent spatial characteristics of the household in the models Among those factors, this study is, especially, interested in the role of the travel time and travel cost s on long distance travel mode choice decision s. T hese factors are considered as the key variables that enable a particular mode to gain a comp ara tive advantage over other alternative modes of transportation in the market. Therefore i t is needed to know travel time and costs of all available modes for each household. However, the 2009 NHTS dataset provides limited information of travel time of the mode used for that trip. Given these conditions, this study estimates synthetic travel ti me and costs for all alternative modes such as personal cars, bus, airplane, and train. Since this study can obtain census block group level spatial information on the households, it is possible to identify the intercity terminals that might be used by eac h household assuming people use the closest facilities, and in turn to calculate approximate distance and associated access time and costs to intercity terminals. The empirical models are expected to provide useful information to ex plain in the relationships with the unique characteristics of travel mode s and travel ers. In particular the results are expected to provide useable variation s among individuals and households and present the potential variations among MSAs (or regions ) This study tests various scenarios of travel time and cost combinations to predict the changes in the probability of choosing a specific mode of transportation. For that, this study assumes an improved train system that can provi de a speed of 150 or more miles per hour The estimated probabilities of choosing an improved train system under
22 scenarios are expected to provide valuable information to evaluate the possibility of an improved train system in the US. In addition, this study examines whether the probabilities of choosing an improved train system would vary across geographical locations. For that, this study uses regional dummy variables (such as northeast corridor (NEC), east north central (ENC), south atlantic (SA) div ision, and pacific (PAC) division) in addition to travel time and costs. T he results are expected to help policymakers and researchers to predict the potential role of an alternative mode in the transportation market It w ould also be possible to decide th e level of service improvement to existing public long distance travel modes to reduce the use of personal car s and airplanes. Scope of Study The first section of the Chapter 2 summarizes previous studies that developed mode choice models for long distance or intercity travel In Chapter 3, this study explains the theoretical frameworks of logistic regression models and then presents the specific model that will be developed in this study. This will be followed by the section about data in which the 2009 NHTS data and states add on data This study focuses on long distance trips among trips by definitions. The Chapter 4 describes the processes of choice model estimation comprising four sub sections. T he first se ction explains explanatory variables that might be included in this study, and this study presents hypothetical relationships of explanatory variables to the decisions on mode choices. T he third section presents the methods to calculate synthetic travel t ime and costs, and access time and costs. Since, the 2009 NHTS presents travel time of the used mode for that trips, this study estimates the travel costs of mode used and both travel time and costs of alternative modes for the given travel distance. T he l ast section of the Chapter
23 4 explains the results of estimated mode choice models. In the Chapter 5, this study expresses how travel time and costs of an improved train system affect to the choice decisions of households and individuals, and what appropri ate policy options would utilize these findings The Chapter 6 includes conclusions, recommendations, and suggested research es for the future
24 CHAPTER 2 LITERATURE REVIEWS Discrete mode choice models have been effectively employed to predict mode share either among existing alternative modes or the potential performance of a new alternative mode in a given transportation network. The former model s are necessary where demand forecasts are required for all modes or those modes which are of inter est to the stud ies These models have predicted the market share for each alternative mode using the estimated mode choice models, and then apply the estimated mode share of each alternative mode to identify total demand for intercity travel. Meanwhile th e latter models are a dopted where researchers are interested in how many travelers would shift from existing alternative mode to the new alternative mode. For example, many studies have actively measured the impacts of a new intercity mode on existing tran sportation market as many countries in both Europe and Asia have developed broad HSR networks By estimating discrete choice models, those studies have presented the potential competitiveness of HSR against air, passenger car bus and convention al rail ser vice. M any of mode choice models have, more importantly, attempted to identify critical points of travel time, cost, or distance where travelers may change their mode choice decisions. In e stimating the potential changes in long distance travelers mode choice decisions mode choice models have adopted various factors that represent travel characteristic ( such as travel time, cost, distance, trip purpose and frequency), travelers characteristics ( such as income, gender, age and group size), spatial characteristics in both origin and destination (such as population density, size of metropolitan area and
25 public transit service quality), and subjective factors (such as comfort, convenience, safety, reliability, and privacy). Am ong the range of variables that have been examined in the previous models, both travel time and cost consist of the key variables to all the models. Travel time is often split into in vehicle time and out of vehicle time of which the latter includes access and egress time, waiting time, terminal time, and transfer time. The access or egress time is the time taken from place of origin to the airport/the train station, or vice versa, respectively. Since public long distance modes such as bus, train and airpla ne are inherently a part of inter/multimodal transportation system, they are associated with both access and egress time. It should be noted that t ravelers who use public transportation mode are considered to be sensitive to these out of vehicle times (Bha t, 1995; Koppelman and Wen, 2000) Travel cost commonly means cost of driving personal cars or/and fare of public transportation modes in those mode choice models. However, some exceptional cases include parking cost (Hensher, 1991), differences of fare le vel by service class (Koppelman, 1989), or access/egress expenses (Kitagawa, Terabe, and Saratchai, 2005; Wardman, Toner and Whelan, 1997). Travel time and cost are also important variables to predict the impact of a new alternative mode on existing modes in the transportation system. In order to characterize individual preferences in relation to travel alternatives, many of these studies have the most commonly used RP and/or SP survey data because they can show either actual or hypothetical travel behavi or. The demand model estimates the mod e choices of passengers assuming that once a person decides to make the journey, the available alternatives and their characteristics such as travel time and cost will For example, m any co untries in both Europe and Asia have
26 present ed the impacts of a new intercity mode on existing transportation market and the potential competitiveness of HSR against air, passenger car bus and convention al rail service as they devel oped broad HSR networks Gonzalez Savignat (2004), Lopez Pita and Robuste (2005), Roman et al. (2007) analyzed the potential competition of high speed rail with the air transport between Madrid and Barcelona, Spain by adopting disaggregated mode choice models. Ivaldi and Vibes (2005) investigated intermodal competition between aviation and HSR in Europe travel market. Kim et al. (2003) and Park and Ha (2006) estimated the air travel demand changes in the Seoul Busan and Seoul Daegu routes. Meanwhile, Chang and Chang ( 2004), Zhang and Xiao Li (2007), and Ortuzar and Simonetti (2008) estimated the potential mode share of HSR system in competition with aviation, personal car, and conventional train. In those models, travel time and costs are commonly considered as variabl es that affect travelers decisions. The travel distance is taken into account of its potential influence on the unobserved perception of comfort and convenience of the ground transportation modes (Grayson, 1985; Koppelman 1989; Koppelman and Sethi, 2000; Abdelwahab Innes and Stevens 1992; Ashiavor Baik and Trani 2010, Wilson Damodaran and Innes 19 90 ). In these models, travel distance presents the likelihood of choosing surface modes (such as car, bus, or rail) relative to air. Travel dis tance is tested in regard of potential s on the same trip is another variable that is directly connected to travel cost, and thus many previous studies included it in the models (Morris on and Winston, 1985 ; Bhat, 199 7a ; Koppelman ait Sethi, 2000; Mandel et al, 1997; Swait, 2001; Wardman et al, 1997; LaMondia Snell
27 and Bhat 2009). This variable assumes that travelers who travel alone likely prefer expensive but fast and comfort mode mor e than travelers in a group. Moreover, purpose of trip is considered to have significant impacts on mode choice ( Morrison and Winston, 1985 ; Wardman et al, 1997; Carlsson 1999 ; Limtanakool Dijst, and Schwanen 2006 ). Since travelers preferences vary w ith the purpose of their trip, the different trip purpose could be an important issue for mode choice decisions. Previous studies have shown that business travelers and leisure travelers are expected to be different in their sensitiv ity to travel time and cost because business travelers have subsidization of travel cost while leisure travelers pay for themselves Service frequency, one of frequently employed variables, is mostly defined in terms of departures by time interval or headways (Algers, 1993; Mand el et al, 1997; Kitagawa et al 2005; Winzer Pidcock and Johnson 1990; Wardman et al, 1997; Vrtic and Awhausen, 2002). The effects of service frequency or number of transfer on the mode choice are investigated when the model included air and/or rail. But it is difficult to exactly measure service frequency because it varies depending on travel demand. In addition to travel attributes, attributes such as income, education car availability, age, gender and educati on are also employed in many mode choice models. Among various forms of traveler related variables, income has been the most widely used in the models (Bhat, 199 7a ; Grayson, 1985; Koppelman, 1989; Koppelman and Sethi, 2000; Swait, 2001; Limtanakool et al, 2006; Abdelwhab et al, 1992; LaMondia et al, 2009). Higher income travelers are generally assumed to choose an alternative mode that provides fast and convenient service even though it is more expensive. It should be noted that some studies focused
28 more on the impacts of these socioeconomic and demographic variables than travel attributes in order to explain mode choice behavior For example, Limtanakool, Dijst and Schwann (2006) examined the effects of age, gender, education, household type and car availab ility on mode choice decision in addition to income. Bhat (199 7b ) tested whether gender has impact on mode choice decisions, while McFadden (1973) predicted the potential impact of race, occupation and ratio of cars to workers in the household on mode choi ce decision for shopping trips. Notably, there are researches that attempted to identify the interrelations between spatial attributes and travel behavior by means of measuring differences of travel patterns in different types of urban form (size or dens ity) and supply of public transportation services (or infrastructure). With regard to spatial characteristics, some of these studies have suggested that the travelers in dense and compact cities with mixed land use use comparatively more public transportat ion for a large part of their daily trips (Frank and Pivo 1994; Timmermans et al., 2003; Schwanen and Mokhtarian, 200 7 ; Dargay and Hanly 2004), while other studies presented that people in larger city are more like to have better public modes such as air and train as well as service quality (Baht, 1995, 1997 a 1998 b ; Limtanakool et al, 2006 ). MSA size, as a large city indicator, identifies whether a trip originated or/and terminated in a large metropolitan area where there is a preference for the train or bus over air mode. In a similar vein higher population densities are expected to associate with higher demand for transport, and thus they likely facilitate well developed public transportation networks resulting in smaller shares for automobile and larger proportions of public transportati on trips. In this respect it is important to measure impacts of spatial
29 characteristics on travelers mode choice decisions. It is also worthy of note that spatial characteristics and travel behavior can mutually affect each other. T ravel behavior might b e a critical factor for individuals or households to make location decisions while urban form at the place of residence affect travel behavior (Scheiner and Holz Rau, 2007). Exceptionally, Sirinavasan, Bhat and Holguin Veras (2006) and Winzar et al. (199 0) concentrated on measuring impacts of perceptions on long distance travel mode the l atter investigated whether comfort, food quality, reliability and convenience have link s to long distance travel for pleasure. Comfort is expected to have impacts on whether or not travelers have higher probability to choose the luxury service class and to avoid bus a less comfortable alternative. Reliability is presented as the share of departure/arrival within a certain time from the predetermined service time.
30 CHAPTER 3 ANALYSIS METHOD AND DATA Theoretical Framework of Mode Choice Models Prob abilistic Choice Theory This study develops logistic regression choice models that are based on the probabilistic choice theory in which the individual is assumed to choose an alternative if it utility is greater than that of any other alternative ( Algers 1993 ; Forinash & Koppelman, 1993 ; Koppelman & Bhat, 2006). Each utility that decision maker n obtains from alternatives allows researchers to rank a series of alternatives and identify the alternative that has the highest utility. Therefore, the indivi dual, n chooses an alternative if and only if: ( 3 1 ) In probabilistic choice theory, the utility function for the individual n to choos e mode t includes two components: the deterministic or observable portions that represent the portion of utility observed by the analyst (V n ,t ) and the error or the portion of the u nobserved u tility to the analyst ( n ,t ) (t = i, j, ..) ( 3 2) The unobserved utility term ( n ,t ) makes the deterministic choice process as probabilistic, and thus leads to a random utility model (RUM) in which the highest observed utility has the highest probability of being chosen (Hess, 2005). The Multinomial Log it Model The e rror term ( n ,t) has been considered as important to determine the mathematical form of choice model because traveler s mode choice decision are not completely and correctly measured or specified (Koppelman & Bhat, 2006). Thus, it has
31 motivat ed researchers to develop numerous mathematical model structure s by applying a different set of assumptions to the distribution of the error components of the utility function for each alternative. Among a wide range of assumptions, three specific assumpt ions, such as 1) the error components are distributed with a Gumbel distribution, 2) the error components are identically and independently distributed across alternatives and 3) the error components are identically and independently distributed across obs ervations/ individuals lead to the multinomial logit (MNL) model structure ( Forinash & Koppelman, 1993; Koppelman & Bhat, 2006). The MNL model gives the choice probabilities of alternative as a function of the deterministic portion of the utility of all the alternatives. In the MNL model, t he deterministic or observ ed portion (V n ,t ) of the model is represented by a linear additive function that parameters, t and explanatory variables, X. The parameters, t may be interpreted as reflecting the effects of the covariates on the odds of making a given choice, while explanatory variables X are characteristics of individuals n This observ ed portion can be presented as: ( 3 3) Therefore, t he utility of an individual n to cho o se alternative t can be stated as: ( 3 4) In this utility function, the probability that individual n chooses an alternative mode is simply: ( 3 5)
32 The choice probability for decision maker n to choose alternative i is given by: ( 3 6 ) Since the MNL model assumes that the error components are identically and independently distributed across alternatives the choice probabilities do not involve the error term n,t Therefore the choice probability f or decision maker n to choose alternative i can be expressed as: ( 3 7 ) This equation implies that t he probability of choosing an alternative increases as the deterministic utility of that alternative increases while decreases if t he deterministic utility of each of the other alternatives increase T he MNL model has been widely used in many previous studies because it ha s a relatively simple and closed mathematical structure and thus it is easy to estimate and explain the results. In estimating the MNL models, this study employs mathematical function of the conditional logit (CL) model in which the probability of choosing a particular alternative mode is expressed as a function of the alternative specific variables. The Conditional Logit Model According to Rodriguez (2012), McFadden (1974) proposed the CL model in which the expected utilities U n t in terms of characteristics of the alternatives rather than attributes of the individuals. In the CL model, the observed portion of the utility of an individual n choosing alternative t is presented as: ( 3 8 ) Where, Z n ,t is the attributes of the alternatives t (t = i, j, )
33 In the CL models, explanatory variables, Z, are assumed to have different values for each choice alternative in the model, and a single coefficient is estimated for each alternative specific variable Z. In other words, the estimated model can present a sep arate coefficient on each independent variable for each possible outcome. Consequently, the impact of a unit change of explanatory variable is assumed to be constant across alternatives, and a variable Z is appeared to have no impact on choice probability if it has with no variation across alternatives (Hoffman and Duncan, 1988). In the CL model, t he error terms follow independently and identically an extreme value distribution. Therefore, the difference of two error terms follows a logistic distribution a s in the MNL model. Given these assumptions the choice probability of the individual n to choose the alternative i can be expressed as equation ( 3 9 ): ( 3 9 ) Since the CL model assumes that the choice of mode depends only on the differences of variables in the utility function, it is appropriate to measure the impacts of a unit change in each explanatory variable on the probability of choosing a particular alte rnative. Therefore, the CL model is useful to evaluate how government polic y affects to the attractiveness of an alternative mode ( Hoffman and Duncan, 1988) In theory, the CL model is assumed to estimate the probability of choosing an alternative mode us ing only the differences in the value of characteristics of the alternatives (for example, travel time and costs) Thus, it may be different from the MNL model which depends on individual characteristics to estimate the probability of choosing a specific m ode. Yet in reality, many studies have d eveloped the choice models to examine how both the characteristics of an alternative mode and the
34 characteristics of individual affect the probability of choosing specific alternative mode (Abdekwahab Innes and Stevens, 1992; Koppelman and Bhat, 2006; Koppelman and Sethi, 2000; LaMondia, Snell and Bhat, 2009; Swait, 2001; Wardman and Toner, 1997; Winzer, Pidock and Johnson, 1990). Moreover, there are studies that have developed the methods to use both alternative specific attributes and the characteristics of individuals in the m odeling framework of the CL model indicating that the CL model is just slightly different form of the exact same model as the MNL model (Hoffman and Duncan, 1988; Rodriguez, 20 12 ; So and Kuhfeld, 2010 ). Model Specification Thi s study estimates the utility of choosing an alternative mode as a function of the alternative specific attributes and the characteristics of individual s. In estimating mode choice model, this study phases in these explanatory variables T his study, first, enters only travel time and travel cost s of each travel mode into the mode choice model These attributes are unique service attributes that are different among travel modes, and thus are expected to affect differently people s choice of travel mode. With these alternative specific variables, the utility can be expressed as: ( 3 10 ) Here, t = travel modes (1: personal cars, 2: bus, 3: airplane and 4: train) 0 1 and 2 = constant for personal cars, bus, and airplane 1 6 = coefficients for each alternative specific variable
35 In addition to these alternative specific variables, this study incorporates characteristics of individuals such as age, income, purpose of travel, and attributes of residence location into the CL mo del. Since different characteristics of among individuals may have different effects on mode choice decisions, it is also valuable to understand the impact of individual characteristics on the probability of choosing travel mode. With these explanatory var iables, the utility of an alternative is expressed as function of the attributes of the alternative modes and the characteristics of the traveler. ( 3 1 1 ) Where, t = travel modes (1: personal cars, 2: bus, 3: airplane and 4: train), 0 1 and 2 = constant for car, bus and air, X i represents individual characteristics such as age, income, purpose of travel and attributes of residence location. i = coefficient associated with individual variable X i Since equation (10) is based on the mathematical function of the (CL) model this study transforms individual characteristics into alternative specific variables by attaching the individual characteristics to each mode of transportation. For example, this study constructs personal car users age variable by applying the each age of an individual n to each personal car user if the person used car and zero otherwise.
36 Data The 2009 National Household Travel Survey In analyzing travel patterns of long distance travel, t his study u se s the 2009 NHTS that includes all interviews from the national sample of 26,000 households and the 20 add on partners The 2009 NHTS updated information gathered in the 2001 NHTS and in prior Nationwide Personal Transportation Survey s (NPTS) conducted in 1969, 1977, 1983, 1990 and 1995. The 2009 NHTS included more samples in both number of household and number of person terms than the 2001 NHTS The 2009 NHTS is therefore, expected to increase the availability of a data base with a n ational coverage of long distance travel I t is also expected to allow develop ing mode choice models that can test the potential variation among various geographic sectors. The 2009 NHTS officially provides household, person, vehicle and daily trip level d atasets for public use ( US DOT FHWA, 2011) of which this study focuses on the daily trip level dataset The daily trip dataset provides fully disaggregated information of daily trips for each person of sample household such as purpose of trip, mode of transportation, travel time, number of people in the vehicle and the most importantly travel miles taken for a given trip The dataset also provide s information of traveler s socio demographic characteristics such as age, income, gender, MSA category, and existence of heavy rail service. More importantly, it is possible to use the spatial information of sample households in a census block group scale. Thus, this study can calculate the shortest distance to each intercity terminal, and in turn estimate acce ss time and costs In analyzing the daily trip datasets t his study focus es on the long distance tr avel among various travels by definition. Until 1995 American Travel Survey (ATS), US
37 Department of Transportation (DOT) defined long distance travel as trip s that are taken away 100 miles or more from home but 2001 National Household Travel Survey (NHTS) redefined it as trips of 50 miles or more from home to the farthest destination traveled and include the return component of the trip as well as any overni ght stops or stops ( US DOT BTS 20 03, p.1 ) Under th is definition any daily trips could be long distance trip as long as total trip length is longer than 50 miles regardless of number of trip segments made in a daily survey This seems to be too broad and vague to identify a true long distance travel. Thus, this study n arrow s the standard of long distance tr ip by counting a trip as long distance trip only if a single trip segment is 50 or more miles This confines the long distance trips mostly to intercity trips. Table 3 1 Household, Individuals and Daily Trips in the 2009 NHTS Index Measures Number of Households 150,147 Number of Individuals 324,184 Number of Daily Trips 1,148,852 Daily trips per household Daily trips per person 7.7 3.5 Number of Long Distance Trips (% in total daily trips) 28,420 2.5 Long distance trips per household Long distance trips per person 0.19 0.09 Source: U.S. Department of Transportation, Federal Highway Administration, 2009 National Household Travel Survey. URL: http://nhts.ornl.gov As shown in Table 3 1, t he daily trip data set contains about 1 ,149 thousand trip segments that were made by 150,147 households or 324,184 individuals E ach household made an average of 7.7 trips in a given survey day, while each individual
38 made an average of 3.5 trips. By the narrowed standard of long distance trip in this study, t he 2009 NHTS includes 28 420 segments that are 50 or more miles. These trip s account for 2.5 percent in total daily trips of 1.2 million. Each household generated an average of 0.19 long distance trips. Operationaliz ing Daily Trips in to Household Level Data This study operationalizes the daily trips in the 2009 NHTS into house hold level data by applying four steps: 1) transform 1,148,852 daily trips into 324,184 individual level data, 2) sort out 17,316 individuals who have one or more trip segments that are 50 or more miles 3) identify trip segments that are involved in actual long distance trips, and 4) transform 17,316 individual level data into 12,846 household level data. For step 2, this study counts a trip as long distance trip if at least one segment in all tri p segments is 50 or more miles. In step 3, this study ident ifies actual trip segments that comprise long distance trips among daily trips segments. In other words, this study identifies true origin destination or origin intermediate stop (s) destination from all trip segments made by an individual in a given survey date For example, four trip segments out of ten daily trip segments could be involved in making a single long distance trip. In most cases, intermediate destination (s) includes stop (s) for gas, rest or meal on the way to destination. In step 4, this study separately counts each individual in the same household if each (or any) household member made long distance trip (s) with different transportation mode, trip purpose, and/or t ravel destination. As results of the data conversions this study obtains 12,846 long distance travel samples for the whole US of which South Atlantic census division comprises the largest share of 4,702 samples and it was follows by West South Central ( WSC) Pacific (PAC) and Middle Atlantic (MA) census divisions at 2,170, 1,853, and 1,502 samples,
39 respectively. Table 3 2 shows the results from data operationalization processes, and table 3 3 presents number of samples by census division. Table 3 2 Data Operationalization Processes Obtaining Household Level Dataset Index Samples Number of Individuals in the 2009 NHTS 324,184 Individuals who have one or more of trip segment that is 50 or more mile s (A) 17,316 Household level long distance trip samples (B) 12,846 B / A (%) 74.2 Table 3 3 Household Level Long Distance Trip Samples by Census Division C ensus Division Number of Samples Share (%) New England 296 2.3 Middle Atlantic 1 502 11.7 East North Central 638 5.0 West North Central 749 5.8 South At l antic 4 702 36.6 East South Central 317 2.5 West South Central 2 170 16.9 Mountain 619 4.8 Pacific 1 853 14.4 Among states, Texas comprises the largest samples at 2,085, and it was followed by California, Virginia, New York, Florida, North Carolina Georgia, and Arizona with 1,769, 1,473, 1,372, 1,072, 9 27 709, 414 samples respectively (see table A 1 in Appendix A for detailed information of samples by state). These states cons titute major portion of datasets for census divisions suc h as WSC, P AC SA, MA and Mountain ( MT ) These 12,846 household level long distance samples are the basis of both descriptive analysis and developments of mode choice models.
40 Descriptive Analysis o f Household Level Long Distance Travel Data As shown in Figure 3 1 n early 90 percent of long distance travelers used p ersonal cars, while about 6.0 percent of travelers chose public intercity transportation modes such as bus, airplane, and train. Both East North Central (ENC) and East South Central (ESC) divis ions show relatively higher share of personal cars at 90 percent, while Mountain division s share of personal cars was the least among divisions at 83.7 percent. It should be noted that airplane accounts relatively large shares in MT WSC, and P AC while N ew England (NE) and MA divisions were high in the share of bus. Train account for the largest share in MA, and it was followed by NE, ENC, P AC and SA divisions. It is also worthy to note that WNC, ESC, WSC, and MT divisions show n early no record of train use possibly because people in these areas live far away from train stations compared to other divisions In detail the average distances to train stations in these divisions of 79.7 75.0, and 43. 0 miles were much longer than the average distance of t he whole US at 32.8 miles. Buses are appeared across the divisions even though there are variations among states. The average distance to bus terminals at 11.8 miles may explain these patterns (see table A 2 in A ppendix A for more information). Airplane s hows the longest average travel distance about 1, 262 miles with relatively large standard deviation among samples. This implies that air transportation covers various service distance range s. In addition, there are large variations in average distances of air travel. For example, the average trip distances of both WNC (852.4 miles) and ESC (962.8 miles) divisions are much shorter than that of P AC ENC and MT divisions at 1,608.0, 1,308.0 and 1,212.0 miles, respectively. Interestingly, the two
41 former divisi ons have no large scale airports, while the three latter divisions have one or more of airports that provide nationwide services. Table 3 4 presents descriptive statistics of long distance travel by mode, and Table A 2 in Appendix A shows the same feature s comparatively among divisions Figure 3 1. Share of Long Distance Travel Mode by Census Division Table 3 4 Descriptive Statistics of Long distance Trips by Travel Mode Mode Number of Sample Share (%) Mean Minimum Maximum Std Dev Car 11 370 88.5 141.0 40.2 5 634.0 182.9 Bus 201 1.6 204.6 50.0 3 899.0 370.6 Air plane 489 3.8 1 261.5 58.1 9 113.0 1 062.0 Train 84 0.7 148.0 50.0 1 200.0 217.2 Other 682 5.3 204.6 50.0 3 000.0 247.6 About 55 percent of long distance trips were made in a less than 100 mile range, and another 24 percent of destinations were within 200 mile range. In all, nearly 80 75% 80% 85% 90% 95% 100% Other Train Air Bus Perspnal Cars
42 percent of long distance trips were made within 200 miles range. MA and P AC divisions show relative large share of trip distance between 50 and 99 miles, w hile MT division records the smallest share at 45 percent in the same distance range. These patterns may be associated with the share of travel mode used. For example, MT division was the lowest for personal cars use, while airplane s share was the largest among census divisions. Table 3 5 and Figure 3 2 present these patterns of mode share by distance group. Table 3 5 Share of Transportation Mode by Distance Group Distance Range (miles) 50 99.9 100 199 200 299 300 499 500 749 750 999 1000 1499 1500 2499 2500 Number of Trips 6 999 3 023 1 233 805 324 132 128 128 74 Share (%) 54.5 23.5 9.6 6.3 2.5 1 .0 1 .0 1 .0 0.6 Figure 3 2.Distribution of Distance Group by Census Division 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 25001500-2499 1000-1499 750-999 500-749 300-499 200-299 100-199 50-99.9
43 Over 93 percent of long distance travelers use personal cars in less than 100 mile distance range and personal cars share decreases as travel distance increases. In contrast, a irplane actively appeared over 300 miles travel distance range and its share increases as travel distance increases. Airplane accounts for higher than personal cars from 750 miles travel distance. Both bus and train s shares are less than 3 percent in all distance ranges. These patterns are expected to appear across the census divisions even though there would be variations. Figure 4 1 illustrates the share s of mode s used for long distance travel by distance. Figure 3 3 Share of Travel Mode by Distance Group As shown in table 3 6, people make long distance travel as they have social and recreational demand among purposes. It accounted for 32.5 perce nt of total long distance trips, and business and returning to home trips followed by taking 24.6 percent and 12.9 percent. Interestingly, returning home and social/recreation trips have 0 10 20 30 40 50 60 70 80 90 100 Personal Cars Bus AirPlain Train
44 relatively large standard variations at 492.4 miles and 334.2 miles, respectively. Meanwhile, medical and dental trips show the shorted standard deviation of 54.7 miles, possibly implying that people tend to use hospitals near from their home. Table 3 6. Descriptive Statistics of Long Distance Travel by Trip Purpose Purpose Number of trips Share (%) Mean Minimum Maximum Std. Dev. Home 1 648 12.9 267.6 50 9113 492.4 Work 3 150 24.6 153.9 50 5634 328.1 School/Daycare 241 1.9 139.9 50 2904 242.1 Medical/Dental 448 3.5 90.93 50 429 54.7 Shopping 1 097 8.6 124.4 50 2533 176.7 Social/Recreation 4 153 32.5 207.6 50 5015 334.2 Family/Personal 796 6.2 182.2 50 3018 303.5 Transport Someone 530 4.1 120.4 50 1700 131.2 Meals 479 3.7 142.7 5 0 1576 153.5 Other 240 1.9 527.7 50 3612 659.9 As presented in table 3 7, more than 73 percent of long distance trips are based on ho me, and another 42 percent and 3 1 percent are related to social/recreation places and workplaces, respectively. In detail, m ore than 60 percent of long distance trips have home as their origins, while nearly 13 percent ended at home. Work place s accounted for 6.7 percent of all origins and 2 4 .6 percent of all destinations, while social and recreation trips took 9.5 percent of origins and 32.5 percent of destinations. Table 3 7. Origins and Destinations o f Long Distance Travel Places Origin Destination Number of samples Share (%) Number of samples Share (%) Home 7751 60.4 1 648 12.9 Work place 853 6.7 3 150 24.6 School/Religious activity places 117 0.9 241 1.9 Hospitals/ Dental service 69 0.5 448 3.5 Shopping centers 541 4.2 1 097 8.6 Social/Recreation places 1220 9.5 4 153 32.5 Family/Personal Business locations 216 1.7 796 6.2 Transport Someone 243 1.9 530 4.1 Restaurants/Cafes 150 1.2 479 3.7 Other places 146 1.1 240 1.9
45 Personal cars accounts for more than 95 percent for the trips to medical/dental service, shopping, and transport someone, possibly because those destinations are less accessible by public transportation modes. In contrast, both work and school/daycare trips show re latively low share of personal cars at around 81 and 83 percent, respectively. Instead, airplane accounts for nearly 4 percent of work trips, while bus accounts for over 8 percent of school/daycare trips. Assuming that business travelers are less sensitiv e to travel cost but more sensitive to travel time, these mode shares are acceptable. High share of bus for school/daycare trips also are reasonable if we counts that travelers aged between 16 and 25 are major travelers with this purpose, and they likely c hoose travel mode that is less cost but take longer time. Table 3 8 present s these shares of long distance trips by travel purpose Table 3 8 Share of Long Distance Travel by Trip Purpose Purpose Car Bus Air Train Other Modes Home 87.8 0.6 7.7 0.5 3.5 Work 83.1 1.1 3.8 0.9 10.9 School/Daycare 81.3 8.3 0.8 0.4 9.1 Medical/Dental 95.1 0.7 0.5 3.6 Shopping 94.7 1.3 1.4 0.2 2.5 Social/Recreation 91.2 1.9 2.7 0.4 3.8 Family/Personal 91.8 1.5 3.3 0.4 2.6 Transport Someone 95.9 1.5 0.8 0.2 1.5 Meals 92.7 2.3 0.6 0.8 3.1 Other Purpose 51.7 2.9 32.9 7.5 5.0 As shown in Table 3 9 people in area with heavy rail system are likely to use public travel mode compared to the people in area with no heavy rain system. In particular, train s share of 2.7 percent in the area with heavy rail system appears much higher than that of 0.2 percent in the area of without heavy rail services. Both airplane and bus s shares are similar with train. These patterns result in the relatively low share of personal cars in location s with heavy rail services.
46 Table 3 9 Comparison of Travel Mode Share by Existence of Heavy Rail System Index Car Bus Air Train Other With Heavy Rail system Number of Trips 1 941 56 109 61 85 Share (%) 86.0 2.5 4.8 2.7 3.8 No Heavy Rail System Number of Trips 9 429 145 380 23 597 Share (%) 89.1 1.4 3.6 0.2 5.6 Table 3 10 presents the effects of heavy rail system on the mode share for long distance travel. T ravelers in MSA s of 1 million or more with heavy rail service are favorable to train, while residents who live in MSA s of 1 million or more but ha ve no heavy rail service are likely to be similar in choosing train. Instead, these MSAs show higher share of airplane in making long distance trips. Personal cars account for more than 90 percent in non MSA areas These patterns would be interesting to examine in the mode choice model. Table 3 10 Share of Travel Mode by MSA Category MSA Category Car Bus Air Train Other MSA over 1million with Rail 1 941 56 109 61 85 86.0 2.5 4.8 2.7 3.8 MSA over 1million no Rail 2 928 45 224 13 139 87.3 1.3 6.7 0.4 4.1 MSA less than 1 Million 3 279 57 108 5 236 88.9 1.6 2.9 0.1 6.4 Not in MSA 3 222 43 48 5 222 90.9 1.2 1.4 0.1 6.3 As shown in Table 3 11 income seems to affect to the both frequency and a verage travel distance of travel because high income group comprises more shares of long distance trips, and average trip distances increase as income increases. In addition, s tandard deviation of 399.1 miles is considerably higher than other income groups of 183.6 miles for low income group and 237.9 miles for medium income group. The large standard deviation of high income group may be associated with individual
47 traveler s capability to make long distance tr ips and to choose comfortable and convenient modes Table 3 11 Patterns of Long Distance Travel by Income Group Income Group Number of Trips Share (%) Mean Minimum Maximum Std Dev Low Income 1 849 14.4 142.3 50 2 555 183.6 Mid Income 3 078 24.0 158 .0 50 4 325 237.9 Hi gh Income 7 185 55.9 210.5 50 9 113 399.1 Among income groups, high income group accounts for the largest share in airplane, while low income group rely relatively large portion of t heir long distance trips on bus. Interestingly, train appears slightly higher at high income group. Therefore, it would be interesting to see whether these differences can be proved by the choice model. Figure 3 4 Share of Long Distance Travel Mode by Income Group E xistence of child seems to have no effect on the average travel distance and the capability of traveling distance range. The average trip distances of traveling without 80% 82% 84% 86% 88% 90% 92% 94% 96% 98% 100% Low Income Mid-Income High Income Other Train Air Bus Car
48 child at 188.4 miles and traveling with child at 183.2 miles are similar each other. In addition, the standard deviations of 348.2 miles and 282.1 miles show no significant difference However, relatively large portion, 4.0 percent, of people is expected to take airplanes when they travel without child. This is double of airplane s share for travelers traveling with child. The average number of people traveling together might be responsible for this gap. People traveling without child have an average of 1.1 person s on air travel, while the average number of people increases to 2.4 persons if people travel with child. It should be noted that t his gap between groups is similar with the average number of travelers for personal cars at 1.3 and 2.7 for the group traveling without child and the group traveling with child, respectively. However, tota l travel costs are different because personal cars costs are the same regardless of the number of travelers on that trip, while air travelers costs are expected to increase as number of people increase Table 3 12 and Figure 3 5 display descriptive statistics of long distance travel depending on existence of child. Similar with existence of child, traveling with the elderly (this includes travels made by the elderly) seems to have no effect on traveling patterns such as average travel distance and the range of traveling d istance. As shown in table 3 13 and figure 3 6, the average distances of 189.0 and 183.3 miles for both groups of traveling with elderly and traveling without elderly are not significantly different. However, pe ople show different mode choice behavior, if they are traveling in company with the elderly. Travelers are likely to choose more personal cars when they travel with the elderly in the group, while people traveling without the elderly show relatively higher share of airplane.
49 Table 3 1 2 Descriptive Statistics of Long Distance Travel for Traveling with Child Traveling With Child N umber of Trips Share (%) Mean Minimum Maximum Std Dev No 11 633 90.6 188.4 50 9 113 348.3 Yes 1213 9.4 183.2 50 3 612 282.1 Figure 3 5. Share of Long Distance Travel Mode by Existence of Child Table 3 1 3 Descriptive Statistics of Long Distance Travel by Existence of the Elderly Traveling With the Elderly N umber of Trips Share (%) Mean Minimum Maximum Std Dev No 10 351 80.6 189.0 50 5 634 346.2 Yes 2 495 19.4 183.3 50 9 113 327.0 82% 84% 86% 88% 90% 92% 94% 96% 98% 100% Without Child With Child Other Train Air Bus Personal Cars
50 Figure 3 6. Share of Long Distance Travel Mode by Existence of the Elderly 80% 82% 84% 86% 88% 90% 92% 94% 96% 98% 100% Without Elderly With Elderly Other Train Air Bus Personal Cars
51 CHAPTER 4 ESTIMATION OF MODEL CHOICE MODELS Explanatory V ariables As discussed in previous chapter, this study develops the MNL models adopting mathematical function from the conditional logit (CL) model. In order to simplify model development process this s tudy concentrate s on the attributes that can be quantitatively measured among various set of potential explanatory variables In estimating the model, t his study is particularly, interested in the impact of travel time and cost on mode choice decision s for long distance travel In case of public transportation modes, both travel time and costs include access time and cost s to t he closest intercity terminal. Then, this study attempts to test the impacts of traveler s characteristi cs on mode choice decisions by adding individual ch aracteristics such as age related attributes, purpose of trips and income group to explanatory variables For that, this study transforms individual attributes into alternative specific variables. In addition, this study applies dummy variables that repre sent the geographical characteristics of residence location. T hese variables include MSA categories that classify MSA into four types such as MSA of 1 million or more with heavy rail service, MSA of 1 million or more without heavy rain service, MSA less th an 1 million, and not in MSA. Furthermore, this study attempts to examine whether the impacts may vary across smaller geographical locations because these MSA categories are large. These sub categories include state, combination of census division, MSA sta tus and presence of a heavy rail system, and major transportation corridors. Table 4 1 presents potential explanatory variables that may be applied to the mode choice models.
52 Table 4 1 Potential Explanatory Variables Variables Description Measurement Travel time Travel time of each mode for that long distance trip Minute Travel cost Automobile: driving cost Bus/train/airplane: fare level Dollar Travel distance Distance between origin and destination pairs Mile Access distance (or time/cost ) Shortest distance to intercity terminals such as airport, bus terminal, train station Miles (minutes) Number of travelers Total number of people on that trip This variable may be incorporated into total travel cost Person Trip purpose D ummy variable that represents the purposes to make the trip : Three categories Work, social/recreational, and others Age indicator Traveling with child Traveling with the elderly (age 66 or more) Yes or no Income group Category of income L ow (less than 30000, medium (30000 60000), high (more than 60,000) MSA Category MSA category for the household home address MSA of 1 million or more with rail MSA of 1 million or more but no rail MSA less than 1 million N on MSA Census Division and MSA status Grouping of household by combination of census division, MSA status, and presence of rail system 32 groups depending on population size and existence of heavy rail service Urban area Home address in urbanized area or rural area Urban or Rural
53 Hypothetical Relationships of Variables to Mode Choice Decisions Among the range of variables that have been examined in the previous studies both travel time and cost co mprise the key variables that are expected to have negative signs As travel time and cost increase, traveler s utility from a choice of the alternative decrease s Travel time is often split into in vehicle time and out of vehicle time of which the latter includes access / egress time, waiting time, terminal time, and transfer time. Meanwhile, t ravel cost commonly m eans cost of driving automobiles or fare level of public modes such as air, train and bus. In case of public long distance travel mode, travel costs also can be divided into two subcategories such as fare and access/egress costs. All these subcategories of travel time and costs are also expected to have negative sign in the estimated models. It should be noted that the ratio of coefficient of travel cost over travel time implies the monetary value of travel time b y which traveler makes trade offs between v arious travel modes. In general, it is known that public intercity mode (bus or train) users are more sensitive to changes in travel fares tha n travel time, while air passengers a re not very sensitive to travel cost and are highly sensitive to travel time having the highest value of time (Ashiabor, Baik, and Trani, 2010; Bhat, 1997b; Carlsson, 2001) The conditional models are expected to provide the travel time and cost combinations to make a new alternative intercity mode attractive. In fact, most studie s of the HSR in Europe and Asia have e valuated the potential demand of a new HSR system by applying various scenarios of travel time and cost combinations to the estimated mode choice models (Beherens and Pels, 2009; Gonzalez Savigant, 2004; Kim, Seo, and Kim, 2003; Lopez Pita and Robuste, 2005, Ortuzar and Simonetti, 2008;
54 Park and Ha, 2003; Roman and Martin, 2007; Wardsman and Whelan, 1997; Zhang and Xiao, 2007). Travel distance can be used as an alternative indicator of travel time and cost because both time and cost are highly related to travel distance. T rip distance is expected to show negative sign and this implies that longer trip distances are associated with more time, more expense and less frequency. Therefore, this is also negatively related to individual s utility. In particular, distance variable is expected to interact negatively with automobile users utility while, airplane user s utility is expected to increase as distance increase. Many previous studies have suggested that there exist thre mode choices vary (Bel, 1997; Hensher, 2001; Jorensen and Preston, 2007; Kitagawa, Terabe, and Sarachai, 2005; Wardman, 2001). In addition to the attributes of the alternatives characteristics are also frequently emplo yed in many mode choice models Thus, this study includes th es e variables into the model by transforming personal characteristics into alternative specific variables. In specific, t his study include s income trip purpose, and age related attributes i n the model. Higher income travelers are generally assumed to choose an alternative mode that provides fast and convenient service even though it is more expensive. Hence, it is expected to have a positive sign for air travel, while a negative sign for pub lic modes such as train and bus. Trip purpose s such as work related trips and social and recreational trips are employed in the models as dummy variable s that present the reasons why people travel. Among trip purposes, work related variable is expected to positively affect to business travelers, particularly those who use airplane, since it is traditionally known
55 that business passengers prefer air travel while non business passengers tend to choose tr ain or automobile for intercity travel. This is because most business passengers have no burden to pay for their trips. In addition social and recreational trips are expected to have positive and significant effects on personal car and ground intercity mode users utilities. Social and recreational trips are known to be less sensitive to travel time, and large in the average number of travelers. In regard of age effects on mode choice decisions, this study uses dummy variables that reflect existence of child or the elderly in the travel group. Age is considered to affect to the mode choice decisions However, t his study cannot directly examine the impacts of age on mode choice decisions because household level samples can include many travelers, and thus it is not possible to apply a single age into the model. Given this condition, this study examines whether people traveling with child or the elderly are different in their mode choice decisions. Existence of child implies increase in number of travelers on that trip, and thus increases in trav el costs except personal cars. Meanwhile, existence of the elderly may imply both increase in number of traveler and more demands on comfortable and convenient modes With these variables, this study expect s that traveling without child have positive effec ts on the probability of choosing airplane, while traveling with the elderly could increase the probability of choosing personal cars. Notably, there are studies that emphasize the impacts of spatial attributes on travelers mode choice decisions. In cons ideration of spatial characteristics, this study includes dummy variables that represent the characteristics of residence locations such as large MSA indicator and urban/rural area. MSA size, as a n indicator of large
56 population and more public transportati on service is expected to have a positive sign for train and bus because a large metropolitan area. Existence of heavy rail service is also understood in the same context as MSA size. Higher demand of transport ation will facilitate public transportation n etwork, and thus large population will be positive ly related with the probability of choosing bus and train. T he number of traveler s on the trip is another variable that is expected to interact negatively with utility of certain modes including air. As number of travelers increase to tal travel cost increases and thus travelers may choose personal automobile instead of airplane because the larger party size the less a pers on is able to afford an expensive alternative (Capon et al., 2003). Figure 4 1 presents these relations in graphic. Figure 4 1 Hypothetical relationship of explanatory variables to travelers utility
57 Estimating Synthetic Travel Times and Costs In order to develop the mode choice models, travel time and costs for each alternative mode is essential. T he 2009 NHTS data only provides actual travel time for each trip segment, and thus this study can calculate total travel time for a given long distance travel made by the mode used. However, the 2009 NHTS does not provide travel costs of the mode used, and travel times and costs for other alternative modes that can be potentially used for each trip. In addition, the 2009 NHTS has no information about acce ss time and costs to/from intercity terminals such as airport, bus terminal, and train station. Therefore, this study develops synthetic travel time and costs for all alternative modes using average driving costs per mile by passenger car type, air passeng ers fare and flight distance survey, bus and train fare and travel time tables. In estimating synthetic travel time and costs, this study tak es into account one way trips assum ing that people use the same mode of transportation for their returning trips. Automobile Travel Times and Costs Automobile travel time mainly includes driving time and rest time for night. In order to calculate d riving times this study applies average travel speed of 60 miles per hour to travel distance and then inflates calculated travel time by distance ranges: T ravel distance less than 100 miles: 10 percent T ravel distance between 100 and 400 miles: 7 percent Travel distance over 400 miles: 5 percent This method is simple, but effective to calculate driving time. In addition to driving time this study applies time for intermediate overnight stay after every 10 hour driving. According to Ashiabor, Baik and Trani ( 2010) the Virginia Tech travel surveys re ported that travelers make overnight stay after 8 hours and 10 hours trips for business and
58 non business trips, respectively. However, this study simple applies non business standard because business trips accounted for less than 25 percent of total long distance trips. The total travel times are sum of driving time a nd overnight time. The driving costs are calculated similar to the procedure of driving time. The driving costs are calculated by multiplying an average driving cost per mile to travel distance reported in the 2009 NHTS. In calculat ing driving cost of the personal cars, this study use s the average driving costs per mile that has been issued by the American Automobile Association (AAA) in every year (AAA, 2010) The 2009 NHTS data does not present exact model and maker of personal cars used, but it provides information about modes such as car, mini van, sports utility vehicle ( SUV ) and pickup truck and other truck. By matching these classifications to the vehicle categorie s of the AAA such as small sedan, medium sedan, large sedan, SUV, and minivan, this study calculates an approximate driving cost of each long distance trip. The average driving costs are given as: Average of Sedan: 16.74 cents per mile Small sedan: 14.10 cents per mile Medium sedan: 17.30 cents per mile Large sedan: 18.82 cents per mile Sport utility vehicle: 22.31 cents per mile Minivan: 19.31 cents per mile The lodging costs are calculated assuming a $100 lodging cost per overnight stay. It would be ide al to increase the lodging costs depending on number of people on that trip, but this is not considered in this study.
59 Airline Travel Time and Costs Air fares and flight times percent sample ticket survey data (DB1B dat a). This data present s commercial airlines airport to airport f are, distance flight time, number of scheduled service, available seats, and origin / destination airport pairs From the dataset, this study can find average fare for 442 airport s in regard of distance groups such as less than 500 miles, 500 to 1000 miles, 1000 to 1500 miles, 1500 to 2000 miles, 2000 to 2500 miles, and more than 2500 miles. In addition, these fare and flight time data can be separately obtained for origin and desti nation flights. So, it is possible to appl y different fares depending on incoming or out going travel. In theory, air travelers prefer airports with low fares, high departure frequencies, and a large number of connections to other airports (Ashiabor, Baik and Trani, 2010). This may be true because air passengers for example, pay 55 cents per mile at Orlando International Airport traveling to destination within 500 miles, while people pay 90 cents per mile to make the similar flight from Gainesville Regiona l Airport. Moreover, Orlando International Airport provides wide range of flight destinations with frequent schedule than Gainesville Regional Airport. However, this study simply assumes that air travelers c hoose nearest airport from home. In order to ide ntify the cl osest airport from each sample, this study uses the location information of census block group where each household is located. T his study associates the average fare and flight time from the DB1B data with the nearest airport to each sample, and then calculates air fare and flight time for each sample at given travel distance.
60 In addition to air fare and flight time, this study adds access costs and time to airport using the distance from sample household to the neares t airport As shown in table 4 2, the average distance s to airport are relatively short at 1 4.8 miles in the census division of MT and it was followed by PAC of 16.9 miles and Northeast Corridor (NEC) (that comprises New England and Middle Atlantic) of 18 .3 miles. In contrast, ESC records the longest average distance of 28.3 miles. It should be noted that WSC, WNC, and MT show the largest standard deviations over 19 miles and WSC includes the sample household which is located the farthest from airport at 157.4 miles. Interestingly, ENC is the second longest in average distance to airport at 26.5 miles, but it has the smallest maximum distance of 67.6 miles. In estimating access time and costs to airport, this study applies the same procedure with driving t ime and cost calculation. T his study assumes that people uses personal cars to access airport Table 4 2. Average Distance to the Closest Airport by Census Division C ensus Division N umber of samples Mean Minimum Maximum Std Dev Northeast Corridor 1,798 18.3 0.2 82.6 1 4.9 E ast N orth C entral 638 26.5 1.7 67.6 17.4 W est North Central 749 21.8 1.2 97.4 19.6 S outh Atlantic 4 702 22.2 0.5 78.6 16.1 E ast S outh C entral 317 28.3 0.7 81.5 18.6 W est South Central 2 170 24.4 0.3 157.4 19.7 M ountain 619 14.8 0.3 104.9 19.1 P acific 1 853 16.9 0.1 102.9 16.3 US 12 846 21.3 0.1 157.4 17.4 Final travel time s for air travel are made up of the f light time and access time, while travel costs are computed as sum of air fare and access costs. It should be noted that total travel costs can be calculated by multiplying air fares with number of people on that trip. Unlike personal cars, air fares increase depending on number of travelers, and thus using total travel costs is reasonable. Individual s who are traveling alone may have
61 greater tendency to choose common carriers while, people travel ing with family members may prefer automobile to lower the travel cost per person. Bus and Train Travel Time and Costs B oth travel time and costs for bus and train are obtained through similar procedure as that of air travel time and costs. In both modes, travel time consists of include in vehicle time and access time, while travel costs comprises fares and access costs. Therefore, the total travel time and cos ts are sum of these two components. First, this study builds up travel time and fares data for both modes in reference to Amtrak and Greyhound s real service operation s In order to identify patterns of travel time and fare for bus and train, this study collects actual fare and travel time from randomly selected origin destination pairs in regard of distances and location characteristics of terminals Using these times and fares, this study estimates the travel distance per minute (mile per minute) and travel costs per mile (cents per mile) by distance group. In reality, b oth bus and train s fares do not monotonically increase as travel distance increase s but increase stepwise Yet, this study allows them to increase proportionally within distance group Secondly, this study estimates the shortest distance from each household to bus terminal and train station like as this study did for airport. In calculating the shortest distances, this study uses 1,482 Greyhound terminals and 533 train stations. Table 4 3 presents the average distances by census division. NEC, WS C and MT show relatively short average distance to bus terminals, while ESC records the longest average distance to bus terminal. In case of train station, people in both WNC and ESC divisions are the farthest in an average distances at 75 miles and 79.7 miles, while P AC and NEC show relatively short average distances of 11.4 miles, and 2 0.6 miles, respectively.
62 Table 4 3. Average Distances to Bus Terminal and Rail Station by Census Division Terminal C ensus Division N umber of samples Mean Minimum Maximum Std Dev Bus Northeast Corridor 1,798 7.9 0.1 113.2 8.6 E ast N orth C entral 638 11.7 0.3 48.9 10.5 W est North Central 749 13.6 0.0 89.8 14.2 S outh Atlantic 4 702 13.9 0.0 66.5 12.7 E ast S outh C entral 317 16.5 0.6 54.8 13.1 W est South Central 2 170 9.8 0.1 78.0 11.7 M ountain 619 10.3 0.3 120.3 16.0 P acific 1 853 11.4 0.5 86.3 8.0 US 12 846 11.8 0.0 120.3 11.8 Train N ew England 1,798 20.6 0.1 173.7 20.9 E ast N orth C entral 638 33.9 0.2 213.9 33.2 W est North Central 749 75.0 0.6 245.9 61.0 S outh Atlantic 4 702 29.7 0.1 219.4 35.9 E ast S outh C entral 317 79.7 0.1 180.3 43.1 W est South Central 2 170 43.0 0.7 246.3 52.3 M ountain 619 43.8 0.3 271.9 58.4 P acific 1 853 11.4 0.4 153.1 17.5 US 12 846 32.8 0.1 271.9 42.3 It is true that there are other intercity bus services by region, but this study does not count them because many of them provide limited service within certain geographical locations. Like air fares, t his study reflects number of travelers to the total fares of both bus and train. Results of Model Estimations As explained above, this study intends to test the impacts of alternative specific characteristics o n the probability of choosing a specific mode. At the same time, this study aims to examine the potential impacts of an individual attributes on the mode choice decisions for long distance travel. In this regards, this study estimates multiple models that include alternative specific characteristics, indi vidual attributes, or both variables at the same model. For that this study develops a mode choice model that includes only travel time and costs of each alternative (the first column of Table 4 4 ).
63 Then, this study adds variables that represent travelers characteristics such as income, travel purpose and age related attributes (the second column of Table 4 4 ). Furthermore, this study expands the explanatory variables into geographical attributes of the household such as MSA categories, census divisions, and states. These models are expected to explain whether certain area has higher probability of choosing a specific alternative mode. These modes are presented from column 3 to column 6 in the Table 4 4 Travel Time and Costs Variables The estimated coeffi cients of both travel time and travel costs are negative in all models developed in this study. These negative signs indicate that travelers utility decrease as travel time and costs increase as expected in hypothetical model. Thus, people may make less l ong distance trips as travel time and costs increase. These estimated coefficients of personal cars are acceptable considering that personal cars account for large shares in shopping, social and recreation, and family or personal business trips, and people more likely travel with other family members for these trips. It should be noted that the absolute values of the estimated coefficients for both travel times and costs are similar across predicted models even though the y vary slightly depending on the explanatory variables added. For example, personal cars coefficients of travel time at 0.013 6 and travel costs at 0.0 189 are similar with the coefficients of 0.014 3 and 0.019 3 in the model with individual characteristics These coefficients imply tha t personal car users are less likely affected by exogenous factors in making mode choice decisions. In contrast, the estimated coefficients of travel time and cost s show that a ir travelers are more sensitive to travel time than travel costs. The absolute values of
64 coefficients for air travel times are larger than the estimated coefficients for air travel costs meaning one unit change in travel time reduce more air travelers utility These results share the same view with m any previous studies These patterns are reasonable because people traveling with work related purposes are expected to use more airplane than travelers with other purpose. In addition, airplane users with work related purpose likely travel alone and thus they are likely free from additional burden on travel costs. I nterestingly, airplane shows considerable differences among the estimated coefficients of travel time and costs as individual variables are included in the model. In other words, airplane s coefficients of travel ti me at 0.0118 and costs at 0.0014 in the time cost only model are different from that of 0.0050 and 0.0006 in the model with individual characteristics T hese changes imply that individual characteristics such as status of employment and income also affect to the probability of choosing airplane compared to other modes of transportation. Gr ound intercity modes such as bus and train show similar patterns as personal cars in terms of the variations of coefficients on travel time and costs across the models In other words, t he estimated coefficients are relatively stable across the estimated models even though the estimated models show that individual characteristics affect more on the bus users choice decisions than train. This may imply that both bus and train are not affected by individual characteristics compared to airplane. The higher absolute values of coefficients for travel costs show that both bus and train users are more sensitive to travel costs than travel time. These estimated coefficients support hypothetical assumptions on the impacts of travel time and costs on choice decisions for these modes.
65 Individual Socio Demographic Variables In addition to alternative specif ic attributes, some traveler s individual characteristics are appeared to have impacts on mode choice decisions. T he estimated model s with individual characteristics show that trip purpose, income and age have significant and positive influences on the choice of transportation mode. First of all, the model s show that people with social and recreational purpose are positive to choose personal cars or bus. In case of social and recreational purpose personal car users travel in a group of 1. 6 persons on average which are relatively large than other mode This implies that personal car users can benefit from total travel costs. In addition, bus users on travel time justifies these results. Thus, the se positive signs of soc ial and recreational purpose on personal cars and bus are reasonable. It is worthy of note that long distance travelers are likely to choose airplane when they have the needs for business trips. T he positive signs of airplane for work related t rips are as might be expected. Business travelers are expected to be paid for their trips, and thus they are less sensitive to travel costs than travel time. As explained in previous section, this study indirectly examines the effects of age on the mode choice decisi ons by including dummy variables of people traveling with/without child and people traveling with/without the elderly. The results show that existence of child has no statistically significant impacts on mode choice decisions for all alternative modes. For example, people traveling with child and air travelers without child have positive sign on personal cars and airplane, respectively, but it was not statistically meaningful. In contrast, personal car users utility is expected to increase if they travel w ith the elderly. Since existence of the elderly also includes the cases of traveling elderly without other age group, this may imply that people rely more on
66 personal cars as they are aged. This is as expected. However, there are no other statistically sig nificant impacts in regards of the elderly. As expected, income is expected to be positively related to fast and convenient transportation service The estimated coefficients show that the high income group is more favorable to use airplanes while people in low income group have higher utility as they choose buses. These results imply that higher income group is more sensitive to travel time, and this they will pay more if they can reduce travel time. In contrast, low income group is likely willing to red uce travel costs rather than travel time. In summary, personal cars users have relatively higher utility if they are traveling with the elderly and have social and recreational purpose. Meanwhile, travelers are likely to increase the probability of choos ing airplane as they are traveling for business purpose or e arn more than $60,000 annually. The probabilities of choosing bus increase when travelers have social and recreational purpose or they earn less than $30,000 annually. Geographical Characteris tics of Household This study uses dummy variables that represent geographical characteristics of household In a pplying dummy variables, this study narrows geographical scale from MSA categories to state and major transportation corridor levels. The third column in Table 4 4 shows the results of how different MSA categories may increase or decrease the utility of an alternative mode. The estimated model show s that people in both MSA of less than 1 million and non MSA may have relatively higher probability to drive personal car to make long distance trips compared to people living in MSAs of 1 million or more Since these area s are expected to have less accessibility to airport interc ity bus services and rail stations, the positive coefficient s of personal car s are reasonable.
67 In fact, average distances to rail stations of 41.8 miles for MSA of less than 1 million and 54.6 miles for non MSA are much longer than that of MSA of 1 million with heavy rail system at 11.0 miles and MSA of 1 million without heavy r ail system at and 14.5 miles The average distances to airports and bus terminals are similar with rail stations. Thus, people are not able to benefit from public transportation modes. T he MSA s of 1 million or more without heavy rail system is positive to the utility of people traveling with airplane For example, the MSAs of 1 million or more with heavy rail system in both Middle Atlantic and South Atlantic divisions show much lower share of airplane compared to the MSAs of 1 million or more w ithout heavy rail system in the same divisions. This result makes sense because air line companies can provide better services of frequent flight, more destinations and low fares in large MSAs, and in turn these services attract more travelers as they hav e no competition with rail systems In contrast, both train and bus users are able to increase their utilities as they live in MSA of 1 million w ith heavy rail system. Among MSAs of 1 million with heavy rail system, the MSAs in NE C ENC SA and PAC are po sitive to increase t he probabilities of choosing train as shown in the fourth and fifth columns of the Table 4 4 These MSAs are known to have relatively developed rail service systems. Interestingly, these sub MSA categories are matched with federal gover nment s proposals of high speed rail development. Meanwhile, travelers have higher probabilities of choosing buses if they live in the MSAs of 1 million or more that are located in the N EC (which comprises New England or Middle Atlantic divisions). It should be noted that MSAs of 1 million or more in NEC are known to provide better intercity bus services than other areas in the US. Thus, this result is highly reasonable.
68 This study, finally tests whether people s utility choosing an alternative mode may va ry across the states in the US. The estimated model shows that airplane users have higher utility as they live in Arizona, California and Florida, while bus users in New York can increase their utility of choosing bus. In similar context, travelers can increase their utility of choosing train by residing in California or New York. However, both coefficients of airplanes for California and Florida are statistically significant only at 70 pe rcent significance interval, so it cannot be fully proved in this study.
69 Table 4 2 Estimations of Mode Choice Model s Variables Time cost effect Socio D emographic A ttributes E ffects of M SA C ategor ies Estimate t Value Estimate t Value Estimate t Value Car 4.8998 32.29 4.9882 29.97 5.8974 26.18 Bus 0.7865 4.11 0.7466 3.54 1.6704 6.56 Car Time 0.0136 20.28 0.0143 20.69 0.0136 19.30 Air Time 0.0118 4.55 0.0159 5.72 0.0050 1.64 Bus Time 0.0024 3.59 0.0028 4.02 0.0018 2.65 Train Time 0.0040 5.52 0.0042 6.09 0.0034 4.68 Car Cost 0.0189 11.45 0.0193 11.41 0.0219 12.52 Air Cost 0.0014 3.57 0.0009 2.46 0.0006 1.67 Bus Cost 0.0187 5.70 0.0193 5.68 0.0186 5.62 Train Cost 0.0097 3.18 0.0068 2.48 0.0072 2.33 Car with Elderly 0.2920 2.27 0.2263 1.74 Car Social 0.8303 5.50 0.7795 5.13 Bus Social 1.0894 5.21 1.0123 4.83 Air Work 0.7485 4.79 0.8858 5.44 Air H igh Income 0.4522 3.26 0.5865 3.90 Bus L ow Income 0.5779 3.32 0.6594 3.75 Car Non MSA 0.5149 3.17 Car MSA of less than 1 million 0.2638 1.90 Air w ithout H eavy Rail 0.2678 1.69 Train w ith H eavy Rail 2.4920 10.42 Bus w ith H eavy Rail 0.4833 2.73 Bus NEC MSA of 1Mil or more Train NEC w ith H eavy Rail Train ENC w ith H eavy Rail Train SA w ith H eavy Rail Train PAC w ith H eavy Rail Air A rizona Air C alifornia Air F lorida Bus N ew Y ork Train N ew Y ork Train C alifornia Log Likelihood at convergence 2389 2 324 2237 Log Likelihood at constant only 3496 3496 3496 R2 0.3166 0.3 352 0.3 601
70 Table 4 2. Estimations of Mode Choice Models ( Continue d) Parameter Mixed Effects Pure Corridor E ffect State E ffects Estimate t Value Estimate t Value Estimate t Value Car 5.8970 26.40 5.5948 31.88 5.5559 26.62 Bus 1.7105 6.81 1.3394 6.28 1.5231 6.19 Car Time 0.0137 19.38 0.0134 20.01 0.0137 19.57 Air Time 0.0050 1.65 0.0040 1.50 0.0065 2.15 Bus Time 0.0018 2.67 0.0017 2.68 0.0019 2.83 Train Time 0.0032 4.41 0.0034 4.36 0.0032 4.48 Car Cost 0.0220 12.59 0.0213 12.62 0.0217 12.45 Air Cost 0.0006 1.66 0.0013 3.20 0.0006 1.56 Bus Cost 0.0186 5.63 0.0184 5.67 0.0186 5.61 Train Cost 0.0076 2.44 0.0122 3.45 0.0080 2.59 Car with Elderly 0.2265 1.74 0.2481 1.90 Car Social 0.7859 5.17 0.7978 5.25 Bus Social 1.0041 4.78 1.0210 4.87 Air Work 0.8924 5.48 0.8506 5.26 Air H igh Income 0.5862 3.90 0.5045 3.42 Bus L ow Income 0.6647 3.78 0.6792 3.87 Car Non MSA 0.5436 3.46 0.8067 5.67 Car MSA of less than 1 million 0.2909 2.18 0.6154 5.36 Air w ithout H eavy Rail 0.2586 1.66 Train w ith H eavy Rail Bus w ith H eavy Rail Bus NEC MSA of 1Mil or more 0.8230 3.81 0.9881 4.78 Train NEC w ith H eavy Rail 3.0668 11.43 2.8195 11.90 Train ENC w ith H eavy Rail 3.2503 5.87 3.0033 5.57 Train SA w ith H eavy Rail 2.3014 6.82 2.1236 6.79 Train PAC w ith H eavy Rail 1.6265 4.47 1.4099 4.14 Air A rizona 0.5224 1.68 Air C alifornia 0.2208 1.18 Air F lorida 0.2628 1.16 Bus N ew Y ork 0.5127 2.59 Train N ew Y ork 2.5615 10.62 Train C alifornia 1.1576 3.73 Log Likelihood at convergence 2 224 23 12 2 243 Log Likelihood at constant only 3496 3496 3496 R2 0.3 638 0.33 87 0.3 584
71 CHAPTER 5 POLICY IMPLICATIONS FOR FUTURE LONG DISTANCE TR NSPORTATION PLAN Average Marginal Effects of Travel Time and Cost s The marginal effects measure the effect of a one unit change in explanatory variable on the probability of choosing an alternative model They are different from the elasticity that measure s the effect of one percent change in explanatory variables on the dependent variable. The marginal effects are informative means to provide what is the change in the probability of choosing an alternative t for a decision maker i because of small c hange in the attribute k of alternative t The direct marginal effects for continuous variables are expressed as: By applying estimated coefficient s to the individual traveler s travel time and travel cost s by m ode this study calculate s each individual s marginal effect of each average change in the probability of choosing an alternative mode by alternative specific attribute. As shown in Table 5 1, one hour increase in travel time is expected to decrease the probability of choosing personal car by 0.0 35 percent while t he probabilit ies of choosing bus air and train are also expected to decrease by 0.00 3 percent 0.0 15 percent and 0.00 2 percent, respectively Meanwhile, $100 dollar i ncrease s in travel costs are also expected to decrease the probability of choosing personal cars, bus, air and train by 0.0 79 0.0 3 8 0.00 4 and 0.00 8 respectively.
72 It should be noted that personal car users are the most sensitive to additional 1 unit of travel time and costs, while train users are less likely affected by the changes in both travel time and cost Bus users are relatively sensitive to travel time compared to travel costs. In contrast, air travelers are like ly willing to decrease travel time rather than decrease travel costs. This implies that service speed would be a key point to make a new alternative mode to compete with airplane. Table 5 1. Average Marginal Effects of Travel Time and Costs by Mode Index Car Bus Airplane Train Travel Time Travel Cost Travel Time Travel Cost Travel Time Travel Cost Travel Time Travel Cost US Average 0.035 0.079 0.003 0.038 0.015 0.004 0.002 0.008 NEC 0.031 0.072 0.002 0.035 0.012 0.003 0.002 0.008 ENC 0.033 0.075 0.002 0.033 0.015 0.004 0.002 0.007 SA 0.034 0.077 0.003 0.037 0.014 0.004 0.002 0.009 PAC 0.034 0.077 0.002 0.036 0.014 0.004 0.002 0.009 Major MSAs of 1 million or more with heavy rail system show similar measures of marginal effects for travel time and costs, although there are variations. For example, people in NEC are less responsive to the change of air travel time, while PAC region sho ws higher response in travel costs of train Meanwhile, travelers in SA region are more sensitive to both bus travel time and train travel time compared to the average of the US. Interestingly, ENC is expected to experience less decrease in probabilities in every unit changes of travel time and cost for all modes. These variations may imply that policymakers and researchers are required to apply different strategies in different regions.
73 Estimations o f t he Probabilities Choosing a New Alternative Mode b y Travel Time a nd Costs Scenario Using the coefficients of the travel time and cost from the CL model, this study attempts to identify the potential service qua lity of a new alternative mode. This study presumes that a n alternative mode replaces current train s ervice s by improving its speed. In specific, this study assumes a new high speed rail system as a new alternative mode because other ground transportation modes such as personal cars and bus cannot travel with such a speed. In order to calculate tra vel time of the alternative train system, this study assumes three speed levels such as 1 5 0, 200 and 3 00 miles per hour that are currently available for high speed rail around the world. B y applying these speeds to the incremental travel distances from 10 0 miles to 600 miles, this study calculates in vehicle times for the new alternative train system. This study also estimates travel times for personal cars, bus, airplane and train in the same travel distances These distances represent the service range where a high speed ground transportation mode, including HSR system, is empirically expected to have competitiveness against air and personal cars. In addition to in vehicle time, this study applies access time and costs to public transportation modes inc luding the new alternative mode. For that this study, first, assum es an average distances of 29.1 miles to airport, 14.4 miles to bus terminal and 21.2 miles to train station Second, this study assumes personal cars for all access travel. This study first, estimates the probability of choosing an improved train system for the whole of the US. Then, this study predicts the probabilities of major regions such as NEC, ENC, SA, and PAC. Since, these regions show relatively higher share of train, they mi ght have different patterns in the probability estimations.
74 Estimations of the Probabilities of Choosing an I mproved T rain f or t he Whole o f t he US As shown in Figure 5 1 and Table 5 1, an improved train service is expected to ha ve less than 1 percent of probability in a 100 mile travel distance range regardless of speed level s and travel costs. In both 200 mile and 300 mile service range s the probabilities of choosing an improved train are expected to increase up to 2.5 percent These shares are higher t han current average share of train in the US, but the new train system is expected to suffer from insufficient demand. It should be noted that these estimated shares are still lower than current mode share of train at 2.6 percent in the NEC where Amtrak is experiencing serious revenue shortfall. In addition, it may be difficult for an improved train system to have price competitiveness against personal cars and airplane. F or example, it is required to set a $100 fare with a speed of 300 miles per hour to r etain 2.5 percent mode share, but, people already pay $149 for about 230 mile trip from Penn Station in New York to Union Station in Washington DC. It would be difficult for an improved train system to charge less than current fare level. T herefore, an imp roved train may not be an attractive policy option targeting 200 to 300 mile transportation market of the US A 400 mile seems to be a frontier for a new alternative rail system to draw long distance travelers from other competing modes such as personal c ars and airplane. A new alternative rail system has space to adjust its fare level by 1.5 times of driving costs while it retains relative less constraint on service speed. For example an improved train system can account for 1.8 percent at a speed of 200 miles per hour and a fare level of 1.5 times of driving costs. This is a significant rise of the probability, but the new train system is not free from constraints on travel time and costs. The new tra in system is
75 required to maintain fare level as close to driving costs to maintain its competitive power. In general, construction costs increase as speed limit increases, thus a new rail system may suffer from revenue shortfall, and in turn this will esca late the concerns on cost effectiveness In both 500 mile and 600 mile service distance ranges, a new rail system is expected to draw more than 4 percent of long distance travel demand if it could hold its fare level as close to driving costs. These proba bilities are significantly higher than current average share of train in the US. However, it should be noted that the probabilities of choosing the new train system are expected to decrease to less than 3.6 percent with a fare level of 1.5 times of driving costs, and to less than 1.5 percent with a fare level of 2 times of driving costs. T herefore, an improved train system is likely to be restrained by travel time and costs constraints. Moreover, the shares of 5 to 8 percent may not sufficient to sustain ma ssive construction and operation costs of high speed rail system. In summary, a n improved train is expected not to be competitive against personal cars in a distance range between 100 and 300 miles. In a 400 mile distance range, a combination of low fare level as close as to driving costs and a speed of 200 or more miles per hour is essential to assure the least probability of choosing a new alternative rail system. In contrast, it would have a relatively stron g competitive ness in a distance range of 500 or more mile s but it is required to transport people with a speed of 200 or more miles per hour and a fare level of less than 1.5 times of driving costs
76 Table 5 2 The Probabilities of C hoosing an improved train System by Trav el Time and Cost Scenario Distance Car Air Bus Improved Train Time Cost Time Cost Time Cost Speed Time Probability of Cost (%) 100 110 30 80 90 150 40 40 50 70 150 80 0.91 0.82 0.68 200 70 0.94 0.86 0.71 3 00 60 0.98 0.89 0.74 200 220 65 115 135 300 90 70 100 130 150 120 1.43 1.08 0.81 200 100 1.55 1.17 0.88 3 00 80 1.68 1.26 0.95 300 330 95 120 200 430 135 100 140 190 150 160 2.00 1.37 0.85 200 130 2.25 1.54 0.95 3 00 100 2.53 1.73 1.07 400 445 125 135 210 560 175 130 190 250 150 200 2.75 1.56 0.88 200 160 3.22 1.83 1.03 3 00 120 3.75 2.14 1.21 500 550 155 150 260 745 190 160 230 310 150 240 4.02 2.09 0.98 200 190 4.86 2.54 1.19 3 00 140 5.87 3.08 1.45 600 670 185 170 310 890 230 190 280 370 150 280 5.18 2.24 0.95 200 220 6.49 2.83 1.21 3 00 160 8.11 3.57 1.53
77 Figure 5 1. The Probabilities of Choosing an improved train Mode by Travel Time and Cost Scenario Probability Estimations f or t he Major MSA Corridors This study predicts the probabilities of MSAs of 1 million or more with heavy rail system in NEC ENC SA and PAC regions where the probabilities of choosing train appeared positive and significant. By estimating the probabilities of choosing an improved train system in these regions, this study indirectly evaluates the potentiality of HSR systems in the US. The estimated probabilities are presented in Table 5 2 and Figure 5 2. In the NEC, the new high speed train system is expected to share around 6 to 18 per cent of long distance travel assuming the same level of fare level as the driving costs. However, the probabilities are decreased to 5 to 6 percent if fare level goes up to 1.5 times of driving costs and then decreased to 2 to 4 percent as fare level is set to 2 times of driving costs More importantly, the shares of an improved train system at 600 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 The same with driving costs 1.5 times of driving costs 2 times of driving costs
78 mile distance are lower than that of 500 mile distance if fare levels are higher than 1.5 time of driving costs. Th ese results seem to imply that an improved train is appropriate option for up to 500 mile distance ranges premising a fare level of less than 1.5 times of driving costs. Overall, t he predicted probabilities of ENC show s imilar patterns with that of NEC An improved train system is likely to be com petitive in ENC when it provides passenger services up to 500 mile range with a fare level of 1.5 times or more of driving costs. However, it should be noted that t he probabilities of ENC are slightly higher than NEC by 400 mile distance range while the p robabilities of NEC are larger than ENC if travel distance is 400 or more miles, but less than 600 miles. This may imply that an improved train system would be more viable policy option for a relatively short distance range (up to 400 miles) in ENC. The estimated probabilities of SA are higher than the average probabilities of the US in all distances and travel time scenarios as long as its fare level can be marked by 1.5 times of the driving costs However, the probabilities are less than 4 percent a s fare level increases to 1.5 times or 2 times of driving cost. In particular, the probabilities decrease continuously if fare level is set to 2 times of driving costs. These patterns may occur because train s marginal effect on travel cost is larger than NEC and ENC, and thus long distance travelers in SA shift to other modes easily. Given these conditions, it is expected that a fare level of less than 1.5 times of driving costs is critical to SA for developing an improved train system. In addition, it is appropriate for SA to provide services as fast as possible because higher speed increases the probabilities of choosing the new alternative train system. This may imply that a new high speed rail
79 system could be a viable policy option for SA in less than 5 00 mile range, but it may have much strict constraints in both travel time and costs compared to NEC and ENC. The estimated probabilities for PAC are interesting compared to other corridors. First of all, the probability of choosing an improved train syste m is higher than the US at all service speed levels if it can set the fare level as lower than driving costs. However, this may not be a possible option for an improved train system because this will be cause serious financial problems. Interestingly, the probabilities of choosing the new train system are higher than the average measure of the US if travel distance is 500 or more mils and the fare level is less than 1.5 times of driving costs. This simply implies that an improved train system might be a vi able option in PAC for over 400 mile distance ranges. The results present that an improved train system would be a policy option for PAC with higher limitations on both travel distance and fare level. The probabilities of choosing a new high speed rail system are higher than the average of the US if travel distance is less than 600 miles and fare level is similar as the driving costs. However, t he probabilities are lower than national average if fare level is higher than 1.5 times of the driving costs and travel distance is 400 or more miles. Thus, the new train system is expected to have s tronger restrictions on both travel distance and fare level. It should be noted that an improved train is relatively appropriate for short distance service in the PAC region The results also show that the probabilities of choosing an improved train is much lower than the average measures of the US if fare level is 2 times or more of driv ing costs rega rd le s s of speed.
80 In summary, an improved train is expected to have advantage of serving long distance travelers in less than 500 mile distance range s. Among corridors, both NEC and ENC are expected to have the highest potential to develop an improved train system in this travel distance range while SA and PAC are expected to face stronger travel time and costs constraints. In particular, PAC is expected to have less than 2 percent of probabilities if it provides passenger services less than 2 00 miles per hour. It should be noted that it is required to retain fare level of less than 1.5 times of the driving costs.
81 Table 5 3. Comparison of the Probabilities of Choosing an improved train System by Corridor Distance An improved train Probability by Cost (%) Time Speed US NEC ENC S A PAC 100 40 50 70 40 50 70 40 50 70 40 50 70 40 50 70 80 150 0.9 0.8 0.7 5.8 5.1 3.9 6.6 5.8 4.3 3.1 2.8 2.1 2.0 1.7 1.3 70 200 0.9 0.9 0.7 6.0 5.2 4.0 6.9 6.0 4.5 3.2 2.8 2.2 2.1 1.8 1.4 60 300 1.0 0.9 0.7 6.1 5.4 4.2 7.1 6.1 4.6 3.3 2.9 2.3 2.1 1.9 1.4 200 70 100 130 70 100 130 70 100 130 70 100 130 70 100 130 120 150 1.4 1.1 0.8 7.5 5.1 3.5 8.3 5.4 3.5 4.2 2.9 2.0 2.7 1.8 1.2 100 200 1.6 1.2 0.9 8.0 5.5 3.7 8.8 5.8 3.8 4.5 3.1 2.1 2.8 1.9 1.3 80 300 1.7 1.3 0.9 8.5 5.8 4.0 9.4 6.2 4.0 4.8 3.3 2.3 3.0 2.1 1.4 300 100 140 190 100 140 190 100 140 190 100 140 190 100 140 190 160 150 2.0 1.4 0.8 8.7 5.2 2.7 9.2 5.2 2.5 5.0 3.0 1.6 3.1 1.9 1.0 130 200 2.2 1.5 1.0 9.5 5.7 3.0 10.1 5.8 2.8 5.5 3.3 1.8 3.5 2.1 1.1 100 300 2.5 1.7 1.1 10.4 6.3 3.3 11.0 6.3 3.1 6.0 3.7 2.0 3.8 2.3 1.2 400 130 190 250 130 190 250 130 190 250 130 190 250 130 190 250 200 150 2.7 1.6 0.9 9.9 4.6 2.1 10.1 4.3 1.8 5.8 2.8 1.3 3.6 1.7 0.8 160 200 3.2 1.8 1.0 11.2 5.3 2.4 11.4 4.9 2.1 6.6 3.1 1.5 4.2 1.9 0.9 100 300 3.8 2.1 1.2 12.6 6.0 2.7 12.8 5.6 2.3 7.5 3.6 1.7 4.7 2.2 1.0 500 160 230 310 160 230 310 160 230 310 160 230 310 160 230 310 240 150 4.0 2.1 1.0 12.1 5.0 1.8 11.9 4.5 1.4 7.3 3.1 1.1 4.6 1.9 0.7 190 200 4.9 2.5 1.2 14.0 5.9 2.1 13.8 5.3 1.6 8.5 3.6 1.3 5.4 2.2 0.8 140 300 5.9 3.1 1.4 16.1 6.9 2.4 16.0 6.2 1.9 9.9 4.3 1.6 6.3 2.6 0.9 600 190 280 370 190 280 370 190 280 370 190 280 370 190 280 370 280 150 5.2 2.2 1.0 13.1 4.2 1.3 12.6 3.6 0.9 8.1 2.7 0.8 5.1 1.6 0.5 220 200 6.5 2.8 1.2 15.5 5.1 1.6 15.0 4.3 1.1 9.7 3.3 1.0 6.1 2.0 0.6 160 300 8.1 3.6 1.5 18.4 6.2 1.9 17.7 5.2 1.4 11.7 4.0 1.3 7.4 2.4 0.7
82 Figure 5 2 The Probabilities of Choosing an improved train Mode by Corridor and by Travel Time and Cost Scenario -1.0 1.0 3.0 5.0 7.0 9.0 11.0 13.0 15.0 17.0 19.0 Northeast Corridor The same with driving costs 1.5 times of driving costs 2 times of driving costs -1.0 1.0 3.0 5.0 7.0 9.0 11.0 13.0 15.0 17.0 19.0 East North Central The same with driving costs 1.5 times of driving costs 2 times of driving costs
83 Figure 5 2 The Probabilities of Choosing an improved train Mode by Travel Time and Cost Scenario and by Corridor ( Continues ) -1.0 1.0 3.0 5.0 7.0 9.0 11.0 13.0 15.0 17.0 19.0 South Atlantic The same with driving costs 1.5 times of driving costs 2 times of driving costs -1.0 1.0 3.0 5.0 7.0 9.0 11.0 13.0 15.0 17.0 19.0 Pacific The same with driving costs 1.5 times of driving costs 2 times of driving costs
84 Policy R ecommendations Based on the finding s o f long distance travel patterns, this study suggests several policy implications. First nearly 80 percent of long distance trips are made less than a 200 mile range with an averag e of 130 miles and thus it is recomme nded to find a new alternative travel mode or improvement of current service that can cover up to a 200 mile travel distance In addition, nearly 9 0 percent of long distance trips are made by personal cars and a new alternative mode is needed to have the capability to d raw people from personal cars. Secondly, it is known that public ground transportation modes such as bus and train are not popular alternative options to the people in the US. However, it may not true considering that both bus and train a ccount for relatively large portion of total daily trips where these services are provided. For example, MSAs of 1 million or more show relatively large shares of bus at N EC (4 percent), MT (2.8 percent), SA (2.3 percent) and P AC (1.8 percent) divisions. S imilarly, train accounts for about 4.9 percent, 5.1 percent, 2.5 percent, and 1.2 percent at N EC ENC, S A and P AC divisions where heavy rail services are provided. These patterns seem to imply that people may be willing to use a ne w alternative mode if th ey have such services with easy access and satisfaction of their service quality needs. Based on the estimated empirical model s it is required for an improved train system to set its fare level less than 1.5 times of driving costs providing services at a speed of 200 or more miles per hour to retain its competitiveness power. In addition, an improved train system is expected to draw relatively higher share of passengers within 500 mile distance range of NEC, ENC, SA and PAC regions B oth NEC and ENC are especially considered as appropriate locations to construc t an improved train system.
85 Assuming a fare level of 1.5 times of driving costs, ENC is expected to have higher probabilities in less than 300 miles distance compared to NEC, while NEC seems to be more favorable in the distance range of 300 to 500 miles. PAC is expected to draw the least probabilities of choosing an improved train system, while SA is expected to rely considerably on the scenario of travel time and costs. Figure 5 3 illus trates the probabilities of choosing an improved train system presuming a fare level at 1.5 times of driving costs. Figure 5 3 Comparison of t he Probabilities of Choosing an improved train Mode with 1.5 times of Driving Costs Depending on the estimated model and prediction of the probabilities of choosing an improved train system, PAC is expected to have relatively the low est potentiality to implement high speed rail policy. However, these low shares of estimated probabilities 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 NEC-1.5 times of driving costs ENC-1.5 times of driving costs SA-1.5 times of driving costs PAC-1.5 times of driving costs
86 may be complemented by air travel between major cities in the region. For example, there are nearly 14.8 million air passengers within 400 miles connecting major cities in California such as Los Angeles San Francisco San Jose, Oakland, San Diego, and Sac ramento as of the end of 2012. Compared to the air passengers of NEC at 5.8 million and ENC at 4 million in the same distance range, these air travel demand is considered as large enough to form a strong ground to support an improved train system. It shoul d be noted that the examples of the air rail market in Europe have presented that high speed rail can account for more than 70 percent within 2 and a half hour journey. Therefore, an improved train system might draw more than 10 million of air passengers o nto its system in PAC. Si milar patterns are witnessed in both northern and southern SA regions. For example, there are 11.5 million air passengers in northern corridor connecting major cities such as Atlanta, GA, Charlotte, NC, Raleigh Durham, NC, Washington DC, and Baltimore, while airlines transported 13.5 million passengers among cities in southern SA such as Atlanta, GA, Orlando, FL, Miami, FL, Jacksonville, FL, and Fort Lauderdale, FL. In the same regards of air rail market share experiences in Europe, 8 million and 9.5 million air passengers are expected to shift to an improved train system in northern and southern SA regions, respectively Th ese figures support current efforts of high speed rail system in the US, but they also re call the impor tance of target mode that can compete with high speed rail by distance range. In summary, it is required for a n improved rail system needs to provide fast and affordable service to compete with both airplane s and personal cars in a distance range of 500 or less miles. In specific, less than 1.5 times of driving costs and 200 or more
87 miles per hour are considered as essential requirement for an improved train system to compete with personal cars and airline services Among census divisions in the US, NEC, ENC, SA and PAC are considered to have higher potential to develop high speed rail system. In consideration of the estimated model, both NEC and ENC are higher in their probabilities of choosing an improved train system, while both PAC and SA are able to i ncrease their possibilities of successful high speed rail implementation s by providing services targeting on air passengers within 500 mile distance range. For that, it might be necessary to establish plans of high speed rail network from wide spatial pers pective.
8 8 CHAPTER 6 CONCLUSIONS AND FURTHER STUDY Conclusions This study attempted to enhance fundamental understanding of long distance travel patterns in the US and provide policy options for long distance transportation planning in the future. In achieving these research objectives, this study focused on four tasks as: 1) d escriptive analysis of the 2009 NHTS and the state add on data sets 2) estimation of logistic regression models that are sensitive to both travel mode specific variables and t ravelers characteristics, 3) prediction of the probabilities of choosing an improved rail system by applying scenarios of travel time and cost s and 4) implications for viable alternative options for long distance transportation planning in the US Among various trips by definition, this study focused on long distance trips that are defined as trips of 50 or more miles T his study counted a trip as long distance trip if at least one segment of the daily trip is 50 or more miles This narrowed the long dist ance trips mostly into intercity trips. It should be noted that this study explicitly attempted to develop a sound method to estimate synthetic travel time and costs of all alternative modes. Since the 2009 NHTS provides only information of mode used for a given long distance trip and travel time, it is essential to estimate travel time and costs of all available modes. However, only couple of studies have presented the travel time and costs for the all available alternative modes. In estimating synthetic t ravel time and costs, this study used all possible sources of published data including average driving costs per mile by passenger car type, air passengers fare and flight distance survey, bus and train fare and travel time tables. In addition, this study calculated shortest distance from each
89 household to intercity terminals such as airport, bus terminal, and train station. For that, this study used spatial information of census block group level household locations, 422 commercial airports, 1,482 greyhoun d terminals, and 533 train stations across the US. In estimating synthetic travel time and costs, this study tak es into account one way trips assum ing that people use the same mode of transportation for their returning trips This study descriptively examin e d current patterns and characteristics of long distance travel in the US focusing on the modes used travel distance s purpose of trips, and potential regional variations. Long distance travel has certain patterns even though they are made as a part of daily activities. First, long distance trave l is likely to be related with home how many trips are made by an individual in a given day, or prior to a long distance trip. Consequently, home accounts for about 7 3 percent of total trip s are made either from or to home. In detail, about 60 percent of trips were begun from home, while about 13 percent of long distance trips were end at home Secondly, about 80 percent f of long distance trips are made in less than a 200 mile range from origins, possibly using personal cars. Eighty percent of car users travel less than 200 miles, and airplane overtakes personal cars if the travel distance is over 800 miles. Third, people seem to locate themselves close to certain service facilities such as medical/dental servic e s and school s Meanwhile, age seems to make no significant difference in generating long distance trips although t ravelers aged 40 to 49 and 50 to 59 tend to travel slightly longer than other age cohorts. Fo urth, the shares of buses and trains are relative ly high where 1 million or more people reside or heavy rail system exists. This may show that people may change choice of mode if they have such a service within their boundary of life.
90 The estimated CL model showed that both travel time and travel cost decrease d indicating that people may shift to other mode s as travel time and travel costs increase. The ratio of the coefficients of travel time and travel cost indicate that car users are willing to pay about 72 to reduce 1 minute of tra vel time. This is equivalent to a $ 43 17 per hour valu e for personal car users. The estimated coefficients of both travel time and travel cost for airplane are statistically significant at a 95 percent confidence interval T he negative sign s of travel time and costs are as expected thus the model is considered to be acceptable In addition to those alternative specific attributes, travelers characteristics such as age, income, and location of residence appeared to affect to the mode choice decisions. The estimated m odel s showed that age has positive impacts on car users mode choice decision but it is not statistically significant at 80 perc ent confidence interval. However, certain age groups are positive and significant toward certain transportation mode. For example, people aged between 40 and 49 have higher probability of choosing airplane, while age group of 19 or less is likely to choose public modes such as buses and trains. Income seems to be related positively to the choice of airplane, while low income family is positive to bus among alternatives. Interestingly, personal cars are likely used in MSA of less than 1 million or non MSAs. In contrast, bus and train are positively and significantly related with population size and existence of heavy rail system. In addition, regional variations exist in the choice of public intercity modes. For example, MSAs of 1 million or more without hea vy rail system are positive toward airplane, while both bus and train increase traveler utilities in the MSAs of 1 million or more with heavy rail system. It should be noted that
91 both air and train have regional variations, and thus they are positive and significant in certain regions. People have higher probability of choosing bus if they live in northeast corridor that comprises MSAs of 1 million or more with heavy rail in Middle Atlantic and New England divisions. Meanwhile, train increases its users utilities as they reside northeast corridor, South Atlantic, East North Central, and Pacific divisions. Among states, Arizona, California, and Florida show positive signs of choosing airplane, but California and Florida s coefficients are not statistically significant at 80 percent confidence interval. Overall the independent variables used in this final model were statistically meaningful in predicting the probability of mode choice for long distance trips. The r squared value s of 0.3166 to 0.3 638 are acc eptable considering that the data is cross sectional data. The measures of marginal effects show that potential changes in the probability of choosing an alternative depending on the changes in travel time and travel cost. The measures show that bus users are more sensitive to the changes in travel time and cost, while air travelers are less responsive to the changes of travel time and cost. The probability of choosing personal car decreases by 0.0 35 percent and 0.0 79 percent as travel time and cost increa se 1 hour and $ 100 respectively. This shows that personal car users are more sensitive to costs than travel time. Among other variables, train users are relatively less sensitive to both travel time and costs, while bus users are cost sensitive and airplan e users are travel time sensitive, relatively. Finally, this study identifie d the potential service quality of a new alternative mode using the coefficients of the travel time and cost from the CL model. The results show ed that a new mode will be able to attract long distance travelers if it has a speed
92 of 200 or more miles per hour in a service distance of less than 5 00 miles In particular, NEC and ENC seem to have relatively higher potential among regions in the US, whil e SA and PAC are considered to have strong constraints on both travel time and costs. A speed of 200 or more miles per hour and a fare level of less than 1.5 times of driving costs are required for both SA and PAC. Further study The descriptive analysis has several limitations. First of all the 2009 NHTS and the Florida add on data have limitation s to represent complex long distance traveler patterns and trends. For example, many trips provide no information about either the origin or the destination, s o this study was not able to confirm their trips accurately. Secondly, a larger sample of trips would improve the accuracy of results. Since the dataset reflects daily travel behavior of ordinary Americans, it can represent certain portion s of long distanc e travel. However, long distance travel is not a common activity that happens in a day, a week, or even in a month. So, any data collection effort should collect more travel information focusing on long distance travel to enhance the current level s of ana lysis. For this, MSAs such as Orlando, Tampa, Miami, and Fort Lauderdale can be good place s to collect additional data because now we know these areas are important for long distance transportation plan s in Florida. Third, descriptive analysis results are not sufficient to explain complex inter/multimodal systems of long distance travel. Therefore, comprehensive studies ( for example, a supportive connection system between stations and other local destinations ) are of critical importance to produce the desir ed transportation plans and policy options. Cases studies for different states can advance the discussion of this study.
93 T h e mode choice model also has several limitations. First, this study reflected access distance to intercity terminals such as bus and airplane, and thus was able to improve the accuracy of travel time and cost. However, this study assumed that there is no difference in access mode. Therefore, the model could be i mproved by consider ing difference s in access mode. Secondly, the estimated results are based on broad assumptions on missing information such as fare level s of public intercity modes, in terminal waiting time, and service frequency of public modes. These variables would be more accurately represented in future studies. Third, the structure of the equations should be tested to include nonlinear functions to determine if they are more suitable to explain complex mode choice behavior. Finally, analysis for de mand forecasts can follow this study. The information on marginal effects and service quality of a new alternative mode will be able to enhance the study of demand forecast in Florida. For that, it is required to collect stated preference (SP) data from in dividuals who use intercity terminals. The SP data will reinforce the results of potential service quality for a new alternative mode.
94 APPENDIX A DISCRIPTIVE STATISTICS OF LONG DISTANCE TRIP BY STATE Table A 1. Long Distance Trip s and Average Trip Length by State S tate Number of Trips Percent Mean Minimum Maximum Std Dev AK 20 0.2 512.9 51 4520 1160.0 AL 31 0.2 156.6 50 1126 198.8 AR 21 0.2 89.9 50 255 47.2 AZ 414 3.2 272.6 50 3034 473.9 CA 1769 13.8 215.6 50 9113 507.3 CO 29 0.2 287.7 52 1500 345.6 CT 34 0.3 134.0 50 704 155.2 DC 6 0.1 345.9 50 1663 646.2 DE 26 0.2 162.0 52 1617 307.7 FL 1072 8.4 208.7 50 3216 363.6 GA 709 5.5 162.4 50 3022 195.8 HI 5 0.0 109.9 50 216 64.6 IA 350 2.7 150.6 50 2146 192.4 ID 23 0.2 118.6 50 402 92.2 IL 68 0.5 198.6 50 2150 368.7 IN 264 2.1 180.7 50 4500 385.8 KS 35 0.3 166.4 50 541 140.7 KY 22 0.2 166.9 50 627 145.1 LA 29 0.2 154.1 50 956 179.7 MA 33 0.3 121.8 50 407 82.4 MD 39 0.3 158.6 50 805 170.3 ME 35 0.3 108.3 50 332 76.8 MI 57 0.4 148.3 50 1082 185.7 MN 38 0.3 177.2 50 1470 234.0 MO 39 0.3 176.7 50 1537 265.9 MS 37 0.3 152.7 50 1296 227.1
95 Table A 1 (Cont.) S tate Number of Trips Percent Mean Minimum Maximum Std Dev MT 30 0.2 194.6 50 689 181.0 NC 927 7.2 168.9 50 3612 249.0 ND 43 0.3 106.0 50 437 76.0 NE 95 0.7 227.4 50 1595 289.0 NH 23 0.2 119.5 50 551 114.5 NJ 61 0.5 145.8 50 945 178.7 NM 30 0.2 263.5 50 1650 359.9 NV 24 0.2 200.7 50 1207 242.8 NY 1372 10.7 157.6 50 3899 276.0 OH 54 0.4 322.1 50 5634 803.2 OK 35 0.3 159.0 50 1200 196.7 OR 26 0.2 218.7 50 2517 478.0 PA 69 0.5 195.8 50 2302 320.2 RI 17 0.1 209.9 50 1825 424.0 SC 415 3.2 160.5 50 2042 173.7 SD 149 1.2 178.8 50 1381 203.2 TN 227 1.8 154.5 50 1216 154.3 TX 2085 16.2 201.4 50 4325 309.8 UT 23 0.2 262.4 50 2054 496.1 VA 1473 11.5 172.3 50 4091 308.0 VT 154 1.2 219.2 50 3000 393.3 WA 33 0.3 404.0 50 5004 1044.0 WI 195 1.5 150.0 50 2000 192.9 WV 35 0.3 160.0 50 735 148.5 WY 46 0.4 142.0 50 585 118.0 US Total 12846 100.0 187.9 50 9113 342.6
96 Table A 2 Descriptive Statistics of Long Distance Travel b y Mode and b y Division Mode C ensus Division Mean Minimum Maximum Std Dev Cars New England 121.8 50.3 704.2 96.9 Middle Atlantic 123.3 50.2 1811.0 132.3 East North Central 148.0 50.3 5634.0 290.7 West North Central 143.3 50.2 2146.0 149.3 South Atlantic 140.6 50.2 3018.0 153.8 East South Central 140.5 50.2 1296.0 136.6 West South Central 150.2 50.2 2557.0 170.6 Mountain 165.2 50.2 2113.0 217.2 Pacific 137.7 50.2 4240.0 245.7 Bus New England 161.7 54.0 550.5 152.0 Middle Atlantic 236.8 50.0 3899.0 631.5 East North Central 177.7 55.0 660.0 168.7 West North Central 215.1 51.0 556.0 165.0 South Atlantic 206.6 50.0 2017.0 302.9 East South Central 193.0 52.0 430.0 168.0 West South Central 265.7 50.0 987.5 254.5 Mountain 248.7 50.0 1508.0 403.5 Pacific 104.3 50.0 408.6 85.9 Airplane New England 1084.0 58.0 2119.0 675.1 Middle Atlantic 1194.0 250.0 3000.0 740.2 East North Central 1308.0 70.0 4500.0 1075.0 West North Central 962.8 170.1 1595.0 517.5 South Atlantic 1199.0 75.0 4091.0 860.6 East South Central 852.4 539.1 1216.0 302.9 West South Central 1124.0 112.0 4325.0 720.2 Mountain 1212.0 90.0 3034.0 800.9 Pacific 1608.0 106.7 9113.0 1429.0 Train New England 179.5 60.0 415.5 204.4 Middle Atlantic 111.1 50.0 1028.0 160.8 East North Central 74.4 50.0 101.8 23.0 West North Central 65.0 65.0 65.0 South Atlantic 155.0 50.0 561.3 142.6 West South Central 300.0 300.0 300.0 Pacific 129.9 50.0 1053.0 246.7
97 LIST OF REFERENCES Abdelwahab, W. M., Innes, J. D., & Stevens, A. M (1992) Development of Disaggregate Mode Choice Models of Intercity Travel in Canada, Canadian Journal of Civil Engineering 19, 965 974 AECOM Consulting & Wilbur Smith Associates (2002) Investment Grade Ridership Study: Summary Report Florida High Speed Rail Authority Algers, S (1993) Integrated Structure of Long Distance Travel Behavior Models in Sweden, Transportation Research Record 1413 141 150 American Automobile Association (2010) Your Driving Costs, Behind the Numbers, 2010 Edition, Retrieved from http://www.colorado.aaa.com/community/ driving_costs_2010.pdf American Road & Transportation Builders Association (2011) FAQS: Transportation & General Public Retrieved October 20, 2010, from http://www.artba.org/about/ faqs transportation general public/faqs/ Ashiabor, S Baik, H & Trani, A (2010) Logit Models for Forecasting Nationwide Intercity Travel Demand in the United States, Transportation Research Record #2007, 1 12 Bhat, C. R. (1995) A Heteroscedastic Extreme Value Model of Intercity Mode Choice, Transportation Research Part B 29(6), 471 483. Bh a t, C R. (1997 a ) Covariance heterogeneity in nested logit models: econometric structure and application to intercity travel, Transportation Research Part B 31 (1) 11 24. Bhat, C. R. (1997b) An Endogenous Segmentation Mode Choice Model with an Application to Intercity Travel, Transportation Science 31(1), 34 48. Retrieved from http://www.caee.utexas.edu/ prof/bhat/ABSTRACTS/Endogenous_ Segment.pdf Ball, M Barnhart, C., Dresner, M., Hansen, M., Neels K., Odoni, A., Peterson, E., Sherry, L., Trani, A., & Zou, B. (2010), Total Delay Impact Study: A Comprehensive Assessment of the Costs and Impacts of Flight Delay in the United States The National Center for Excellence, Retrieved from http://www.isr.umd.edu/NEXTOR/pubs/TDI_Report_Final_10_18_10_V3.pdf Bricka, Stacey (2 001) Variations in Long Distance Travel, Conference Proceedings : Personal Travel: The Long and Short of It Transportation Research E Circular, Number E C026 197 206. Retrieved from http://onlinepubs.trb.org/ onlinepubs/circulars/ec026/06_bricka.pdf
98 Brownstone, D., Hansen, M., & Madanat, S. (2010) Review of Bay Area/California High Speed Rail Ridership and Revenue Forecasting Study Institute of Transportation Studies, University of California, Berkeley, Research Report UCB ITS RR 2010 1, Retrieved from http://www.its.berkeley.edu/publicati ons/UCB/ 2010/RR/UCB ITS RR 2010 1.pdf Cambridge Systematics (2006) Bay Area/California High Speed Rail Ridership and Revenue Forecasting Study: Interregional Model System Development Draft Report, Metropolitan Transportation Commission and the Californ ia High Speed Rail Authority. Retrieved from http://cdm15025.contentdm.oclc.org/cdm/ singleitem/collection/p266401coll4/id/1315/rec/2 Cambridge Syst ematics (2009) The Florida Rail System Plan: Policy Element Prepared for the Florida Department of Transportation Retrieved from http:/ /www.dot.state.fl.us/rail/PlanDevel/Documents/2009PolicyElementoftheRail SystemPlan webfinal.pdf Carlsson, F. (1999) Private vs. Business and Rail vs. Air Passengers: Willingness to Pay for Transport Attributes Department of Economics of Goteborg University, Working Paper in Economics No. 14. Retrieved from https://gupea.ub.gu.se/ handle/2077/2679 Center for Urban Transportation Research (CUTR) (2010 ) Trends and Conditions: Special Report on Commuting Trends in Florida Florida Department of Transportation, Retrieved from http://www.dot.state.fl.us/planning/trends/special/ acs121410.pdf Charles River Association Incorporated (2000) Independent Ridership and Passenger Revenue Projections for High Speed Rail Alternatives in California CRA Project No. 1680 04, The California High Speed Rail Authority Chen (2011) C alibrating the Intercity High Speed Rail (HSR) Choice Model for the Richmond Washington, D.C. Corridor Theoretical and Empirical Researches in Urban Management, 6(2), 35 53. Retrieved from http://um.ase.ro/no62/3.pdf Dar gay, J and Hanly M. (2004, July) Land use and mobility Presented at the World Conference on Transportation Research, Istanbul, Turkey, July 4 8, 2004. Florida Department of Transportation (2002) Florida Interstate Highway Sy stem (FIHS) Status Report State of Florida, Retrieved from http://edocs.dlis.state.fl.us/ fldocs/dot/fihs/status02.pdf Florida Department of Transportation (2011) 2060 Florida Transportation Plan : HORIZON 2060 a new era for transportation in Florida R etrieved from http:// www.2060ftp.org/images/uploads/home/2060FTPlanbook7%2004152011.pdf
99 Florida Department of Transportation (2010, November) Florida Highway Data CD ROM Retrieved fro m http://www.dot.state.fl.us/planning/statistics/ hwydata/hwydata cd.shtm Forina sh, C V. & Koppelman, F S. (1993) Application and Integration of Nested Logit Models of Intercity Mode Choice, Transportation Research Record 1413 98 106 Frank, L D. & Pivo, G (1994) Impacts of mixed use and density on utilization of three modes of travel: Single occupant vehicle, transit, and walking, Transportation Research R ecord 1466 44 52 Gonzalez Savigant, M (2004) Competition in air transport : the case of the high speed train Journal of Transport Economics and Policy 38 (1) 77 108. Grayson, A (1985) Disaggregate Model for Mode Choice in Intercity Travel, Transport ation Research Record 835 36 42 Hensher, D. A. (199 1 ) Efficient Estimation of Hierarchical Logit Mode Choice Models, Proceedings of the Japanese Society of Civil Engineering 425(IV 14), 17 28. Retrieved from http://library.jsce.or.jp/jsce/open/00037/425/425 120617.pdf Hensher, D. A. & Greene, W. H. (2001) The Mixed Logit Model: The State of Practice and Warnings for the Unwary, Workin g Paper, University of Sydney, Retrieved from http://people.stern.nyu.edu/wgreene/MixedLogitSOP.pdf Hess, S. (2005) Advanced Discrete Choice Models with Applications to Transport Deman d Thesis for the degree of Doctor of Philosophy of the University of London, Retrieved from http://www.cts.cv.ic.ac.uk/documents/theses/ HessPhD.pdf Hoffman, S D. & Duncan, G J. (1998) Multinomial and Conditional Logit Discrete Choice Models in Demography, Demography 25 (3), 415 427 Ivaldi, M & Vibes, C (2005) Intermodal and intramodal competition in passenger rail transport IDEI Working Paper, 345 trielle Retrieved from http://idei.fr/doc/wp/2005/ivaldi_vibes.pdf Kim, K.W, Seo, H.Y. & Kim, Y (2003) Forecast of domestic air travel demand change by opening the high speed rail, KSCE Journa l of Civil Engineering 7 (5) 603 609 Kitagawa, T Terabe, S & Sarachai, O. (200 5 ) Various Factors Affecting Modal Choice Behavior of the Intercity Passenger between Keihanshin and Fukuoka, Proceedings of the Eastern Asia Society for Transportation Studies 5, 199 208.
100 Koppelman, F S. & Bhat, C (2006) A Self Instructing Course in Mode Choice Modeling: Multinomial and Nested Logit Models U.S. Department of Transportation Federal Transit Administration Retrieved from http://www.ce.utexas.edu/prof/bhat/COURSES/LM_Draft_060131Final 060630.pdf Koppelman, F. S. & Hirsh, M. (1986). Intercity Passenger Decision Making: Conceptual Structure and Data Implications, Transportation Research Record 1085, 70 75 Koppelman, F S. & Sethi V. (2000) Incorporating Complex Substitution Patterns and Variance Scaling in Long distance Travel Choice Behavior Paper Presented at the 9th International Association on Travel Behavi or Research Conference, July 2 7, 2000, Goldcoast, Gueensland Koppelman, F. S. & Wen, C. H. (2000) The paired combinational logit model: properties, estimation and application, Transportation Research Part B 34, 75 89 Koppelman, F S. (1989) Multidimensional Model System for Intercity Travel Choice Behavior, Transportation Research Record 1241, 1 8. KPMG Peat Marwick (1998) Florida Overland Express High Speed Rail Study: Final Ridership and Revenue Report Florida Department of Transportati on LaMondia, J., Snell, T., & Bhat, C. R. (20 10 ) Traveler Behavior and Values Analysis on the Context of Vacation Destination and Travel Mode Choices: European Union Case Study, Transportation Research Record 2156, 140 149. Retrieved from http://trb.metapress.com/content/t55748084440t5hw/fulltext.pdf LaMondia, J., Bhat, C., & Hensher, D. (2008) An annual time use model for domestic vacation travel, Journal of Choice Modeling 1 (1) 70 97 Retrieved from http://www.jocm.org.uk/index.php/JOCM/article/viewFile/38/15 Lee, S. W., Min, S. H., Park, J. Y. & Yoon, S. D. (2008) The Practice on Logit & Probit Model Seoul, Korea, Park Young Sa. (In Korean) Limtanakool, N Dijst, M., & Schwan e n, T. (2006) The Influence of Socioeconomic Characteristics, Land Use and Travel Time Considerations on Mode Choice for Medium and Longer Distance Trips, Journal of Transport Geography 14 327 341 Lopez Pita, A & Robuste, F. (2005) Impact of h igh speed lines in relation to very high frequency air services, Journal of Public Transportation 8 (2), 17 35. Retrieved from http://www.nctr.usf.edu/jpt/pdf/JPT%208 2%20Lo pez Pita_Robuste.pdf Mallett, W (2001) Long Distance Travel by Low Income Households Conference Proceedings : Personal Travel : The Long and Short of It Transportation Research E Circular, Number E C026, 169 177. Retrieved from http://onlinepubs.trb.org/onlinepubs/circulars/ec026/04_mallet.pdf
101 Mandel, B Gaudry M. & Rothengatter W. (1997) A disaggregate Box Cox Logit mode choice model of interc ity passenger travel in Germany and its implications for high speed rail demand forecast, The Annals of Regional Science 31, 99 120. Retrieved from htt p://www.mkm.de/content_files/publications/ 1997_Mandel_Gaudry_Rothengatter_ARS.pdf McFadden, D (1973) Conditional logit analysis of qualitative choic behavior, I n P Zarembka (Ed.), Frontiers in Econometrics (pp.105 142) Academic Press, New York McFadden, D. & Train, K. (2000) Mixed MNL Models for Discrete Response, Journal of Applied Econometrics 15, 447 470. Monzon, A & Rodriguez Dapena, A. (2006) Choice of mode of transport for long distance trips: Solving the problem of sparse data, Transp ortation Research Part A 40, 587 601. Morrison, S A. & Winston, C. (1985) An Econometric Analysis of the Demand for Intercity Passenger Transportation, In T. E. Keeler (Ed.) Research in Transportation Economics : A Research Annual Volume 2 ( pp. 213 237 ), JAI Press Inc., Greenwich, Connecticut. NHTB Co rporation (2003) Florida High Speed Rail Authority: Orlando Miami Planning Study Flor ida High Speed Rail Authority O Neill, W & Brown, E. (2001) Long Distance Trip Generation Modeling Using ATS, Conference Proceedings: Personal Travel : The Long and Short of It, Transportation Research E Circular, Number E C026, 271 286. Retrieved from http://onlinepubs.trb.org/onlin epubs/circulars/ec026/11_oneill.pdf Ortuzar, J D & Simonetti, C. (2008) Modeling the demand for medium distance air travel with the mixed data estimation method, Journal of Air Transport Management, 14, 2 79 303 Park, Y.H. & Ha H. K. (2003) Analysis of the impact of high speed railroad service on air transport demand, Transportation Research Part E l 42, 95 104 Peterman, D. R., Frittelli, J. & Mallett, W. (2009) High Speed Rail (HSR) in the United States CRS Report for Congress 7 5700, Congressional Research Service. Retrieved from http://www.fas.org/sgp/crs/misc/R40973.pdf Rasmidatta, I (2006) Mode Choice Models for Long distance Travel in USA The University of Texas at Arli ngton, Dissertation for the Degree of Doctor of Philosophy Retrieved from http://dspace.uta.edu/bitstream/handle/10106/49/umi uta 1429.pdf?sequence=1
102 Resource Sys tems Group Inc. (2010) Innovative approaches to addressing aviation capacity issues in coastal mega regions Airport Cooperative Research Program, ACRP report 31, Transportation Research Board, Washington D.C. Retrieved from http://onlinepubs.trb.org/onlinepubs/acrp/acrp_rpt_031.pdf Rodriguez, G. (201 2 ) 6.3 The Conditional Logit Model Lecture note in Generalized Linear Models, Princeton University, Retrieved from http://data.princeton.edu/ wws509/notes/c6s3.html Roman, C Espino, R. & Martin J. C. (2007) Competition of high speed train with air Barcelona, Journal of Air Transport Management 13, 277 284 Roth, D. L. (1998) State of the Art Practices in High Speed Rail Ridership Forecasting: A Review of Recent Modeling Methodology, In T. Lynch (Ed.), High Speed Rail in the U.S. Super Trains for the Millennium (pp. 52 78), Gordon and Breach Science Publishers. Santos, A., McGuckin, N., Nakamoto, H.Y., Gray D., & Liss, S. (2011) Summary of Travel Trends: 2009 National Household Travel Survey U.S. Department of Transportation, Federal Highway Administration, FHWA PL 11 022, Retrieved from http://nhts.ornl.gov/2009/pub/stt.pdf Scheiner, J & Holz Rau, C. (2007) Travel mode choice: affected by objective or subjective determinants?, Transportation 34, 487 511 Schmidheiny, K. (2007) Multinomial Choice (Basic Model) Lecture note in Microeconometrics, University of Pompeu Fabra, Retrieved from http://kurt.schmidheiny.name/teaching/multinomialchoice2up.pdf Schrank, D Lomax, T., & Turner, S. (2010) Texas Transportation Institute, The Texas A&M Univers ity System. Retrieved from http://mobility.tamu.edu/ums/report/ Schwanen, T. & Mokhtarian, P. L. (2007) Attitudes toward Travel and Land Use and Choice of Residential Neighborhood Type: Evidence from the San Francisco Bay Area, Housing Policy Debate 18, 171 207. So, Y. & Kuhfeld, W. F. (2010) Multinomial Logit Models Online available SAS technical suppor t document Retrieved from http://support.sas.com/techsup/ technote/mr2010g.pdf Srinivasan, S Bhat, C., & Holguin Veras, J. (2006) An Empirical Analysis of the Impact of Security Perception on Intercity Mode Choice Using a Panel Ordered Mixed Logit Model, Transportation Research Board 1942, 9 15 Retrieved from http://pubsindex.trb.org/view.aspx?id=777494
103 Swait, J. (200 1) Choice set generation within the generalized extreme value family of discrete choice models, Transportation Research Part B 35, 643 666 SYSTRA (1998) Florida Overland Express Ridership and Revenue Study: Final Report Florida Department of Transporta tion. The MITRE Corporation, Center for Advanced Aviation System Development (2007), Capacity Needs in the National Airspace System 2007 2025: An Analysis of Airports and Metropolitan Area Demand and Operational Capacity in the Future Federal Aviation Adm inistration, Retrieved from http://www.faa.gov/airports/ resources/publications/reports/media/fact_2.pdf Timmermans, H., Waerden, P., Alves, M., Polak J., Ellis, S., Harvey, A. S., Kurose, S., & Zandee, R. (2003) S patial context and the complexity of daily travel patterns: and international comparison Journal of Transport Geography 11, 37 46 Tindale Oliver & Associate, Inc. (2009) Florida Intercity Bus Service Needs Assessment and Action Plan: Technical Memorandum Number One Baseline Conditions and Policy Review Florida Department of Transportation (FDOT) Central Transit Office Retrieved from http://www.dot.state.fl.us/transit/ pages/FloridaIntercityBusBaselineConditions.pdf U.S. Government Accounting Office (2009) High Speed Rail: Future Development Will Depend on Addressing Financial and Other C hallenges and Establishing a Clear Federal Role Report to Congressional Requesters, GAO 09 917. Retrieved from http://www.gao.gov/new.items/d09317.pdf U.S. Department of Commerce Bureau of Economic Analysis (2012) Regional Economic Accounts: Gross Domestic Product by State Retrieved from http://www.bea.gov/regional/index.htm U.S. Department of Transportation (2006) National Strategy to Reduce Congestion on Retrieved from http://isddc.dot.gov/ OLPFiles/OST/ 012988.pdf U.S. Department of Transportation ( 2010, April ) DOT Strategic Plan 2010 2015 : Transportation for the New Generation Draft for Public Comments, Retrieved from https://www.uab.edu/utc/PDF%20Files/Dot_strategic_plan_2010 15.pdf U.S. Department of Tra nsportation Bureau of Transportation Statistics (2003) NHTS: Highlights of the 2001 National Household Travel Survey BTS03 05, Washington, DC Retrieved from http://www.rita.dot.gov/bts/sites/ rita.dot.gov.bts/files/publications/highlights_of_the_2001_nati onal_household_tra vel_survey/pdf/entire.pdf
104 U.S. Department of Transportation Bureau of Transportation Statistics (2006, May) Findings from the National Household Travel Survey: Long distance Transportation Patterns Mode Choice America on the Go Ret rieved from http://www.rita.dot.gov/bts/sites/rita.dot.gov.bts/files/publications/america_on_the _go /long_distance_transportation_patterns/pdf/entire.pdf U.S. Department of Transportation, Bureau of Transportation Statistics (2010, November 1) Airline On Time Performance and Causes of Flight Delays Retrieved from http://www.bts.gov/help/aviation/ U.S. Department of Transportation, Bureau of Transportation Statistics (2011, May 5) Long Distance Travel Retrieved from http://www.bts.gov/publications/ highlights_of_the_2001_national_household_ travel_survey/html/section_03.html U.S. Department of Transportation, Federal Aviation Administration (2001) Aviation Capacity Enhancement Plan 2001 Retrieved from http://www.faa.gov/about/ office_org/headquarters_offices/ato/publications/bench/media/01_ACE.PDF U.S. Department of Transportation Federal Highway Administration (2011) 2009 National Household Travel Survey: User Guide (Version 1), Retrieved from http://nhts.ornl.gov/publications.shtml U.S. Department of Transportation, Federal Highway Administration (2011, June 10) Data Center, N ational Household Travel Survey: Our Nation s Travel Retrieved from http://nhts.ornl.gov/download.shtml U.S. Department of Transportation, Federal Highway Administration (2006, March) NPTS Brief, 2006 Retrieved from http://nhts.ornl.gov/briefs/ Long%20Distance%20Travel.pdf U.S. Department of Transportation, Federal Highway Administration (2010) 2009 National Household Travel Survey Retrieved from http://nhts.ornl.gov U.S. Department of Transportation, Federal Highway Administration (2006) NPTS Brief Retrieved from http://nhts.ornl.gov/briefs/Long%20Distance%20Travel.pdf U S. Department of Transportation Federal Railroad Administration and Florida High Speed Rail Authority (2005) Final Environmental Impact Statement: Florida High Speed Rail Tampa to Orlando Retrieved from http://www.fra.dot.gov/Downloads/ RRDev/florida_tampa orlando_feis.pdf U.S. Department of Transportation Federal Railroad Administration (2002) Five Year Strategic Plan for Railroad Research, Development, and Demonstrations U.S. Department of Transportation Federal Railroad Administration (200 2 ) Analysis of the Benefits of High Speed Rail on the Northeast Corridor Retrieved from http://www.oig.dot.gov/library item/4521
105 U.S. Department of Transportation Federal Railroad Administration ( 1997) High Speed Ground Transportation for America Retrieved from htt p://www.fra.dot.gov/ Downloads/RRDev/cfs0997all2.pdf U.S. Department of Transportation Federal Railroad Administration (2009) Vision for High Speed Rail in America Retrieved from http://www.fra.dot.gov/ downloads/rrdev/ hsrstrategicplan.pdf Vrtic, M. & Axhausen, K. W. (2002) The impact of tilting trains in Switzerland: a route choice model for regional and long distance public transport trips Paper submitted to the 82n d Annual Meeting of the Transportation Research Board, Retrieved from http://e collection.library.ethz.ch/eserv/eth:25719/eth 25719 01.pdf Wardman, M., Toner, J. P., & Whelan, G. A. (1997) Interactions between Rail and Car in the Inter Urban Leisure Travel Market in Great Britain, Journal of Transport Economics and Policy 31(2), 163 181 Wen, C H & Koppelman F. S. (2001) The Generalized Nested Logit Model, Transport ation Research Part B 35, 627 641 Whalen, W. T Carlton, D. W., Heyer, K, & Richard, O. (2008) A Solution to Airport Delays, Regulation, Spring 2008, 30 36. Retrieved from http://www.cato.org/ sites/cato.org/files/serials/files/regulation/2008/2/v31n1 5.pdf Wilson, F. R., Damadaran, S., & Innes, J. D. (1990) Disaggregate mode choice models for intercity passenger travel in Canada, Canadian Journal of Engineering 17, 184 191 Winzer, H Pidock P., & Johnson, L. (1990) Modeling long distance pleasure travel mode using perceived modal attributes School of Business Discussion Papers 12 1 1990 Bond University School of Business. Retrieved from http://epublications.bond.edu.au/discussion_papers/20/
106 BIOGRAPHICAL SKETCH The author received Bachelor of Science majoring architectural engineering at Yonsei University in Seoul South Korea. Then, he earned his Master of Engineering degree concentrating on urban planning. His thesis, titled as An Effect of Transportation Investment on Economic Product ivity, analyzed how transportation investments are related with total factor productivity of manufacturing industry in Korea. H e began his professional carrier at Transportation and Logistics Institute of Keumho Engineering Corporation in October, 1994 a nd worked for about two and 8 months focusing on analysis of the impacts of urban development on transportation congestions and delays. In June, 1997, he moved to the Korea Transport Institute, the transportation policy and planning oriented government res earch division. For about five and a half year, h e served as researcher of Aviation Research Division and Transportation Economics Division, and manager of Division of Research Planning and Budget H e conducted various research projects dealing with financ ing strategies for transportation infrastructure investment, cost benefit analysis for transportation projects, the impacts of transportation on r egional economies, and the role of the private sector in transportation and their cost recovery In spring 2003, h e entered to Indiana University Bloomington and he transferred to University of Texas at Austin in fall, 2005. He earned Master of Public Affairs After from both programs. He began his doctoral study at University of Florida in fall, 2009, and he e arned Doctor of Philosophy in May 2013 with dissertation, titled as The Factors that Affect Long Distance Travel Mode Choice Decisions and Their Implications for Transportation Policy.