New Jersey Rail Ridership Opportunities If ataxis Are. Available To Collect or Distribute Its Customers

Transcription

1 New Jersey Rail Ridership Opportunities If ataxis Are Available To Collect or Distribute Its Customers Department of Operations Research and Financial Engineering (ORFE) Princeton University Princeton, NJ Professor Alain L. Kornhauser *71 Chenyi Chen *16 Matthew Garvey 15 Tharald Fongaard 15 Kyle Douglas 15 Ed Walker 15 Hill Wyrough 14 March 8, 2014

2 Abstract Examined are the casual ridesharing opportunities if a fleet of autonomous taxis (ataxis) were available to serve typical mobility needs. A pragmatic, rather than theoretical, approach is used to model the mobility needs of those living and working within New Jersey on a typical weekday in the recent past, based on 2010 data. New Jersey was selected as an ideal test state because of its widely varied land use - from dense urban centers to rural areas. The spectrum of land use within the state is representative of land use across the United States. The model draws from critically important 2010 data on land use, demographics, and travel behavior to facilitate the correlation of the findings to some reality in the recent past, rather than some ambiguous future. A disaggregate trip synthesizer, using cumulative probability distributions drawn from said data, generated a simulated representation of each person living or working in New Jersey, and each of their trips taken on a typical weekday, resulting in over 32 million trips for 9 million individuals. Characteristics of the individuals, such as sex, age, income level, etc., informed their daily trip tour down to the precise geographic locations and arrival/departure times. Journey-to-Work Census, Employee-Patronage data, and School Enrollment data allowed for realistic reflections of the travel demand within the state, and guarenteed realistic reflections of the true spatial distribution of trips taken as of Candidates for causal ridesharing are those trips that happen to be originating from about the same place at about the same time and going to about the same place or somewhat along the way. Casual ridesharing potential is parameterized by how close trips are correlated with respect to time of origin, spatial separation of the origin, spatial separation of the destination and number of intermediate locations that could be accommodated in a shared vehicle. A pixelization technique is used to readily find all trips originating from the same location at about the same time that enables the analysis to be performed sequentially in linear time with number of trips analyzed. As expected, the statewide casual ridesharing potential is not sufficiently great to be able to incur the cost of labor that a conventional taxi system would have to endure in order to serve the demand. However, an autonomoustaxi system could conceivably offer the service at an acceptable price to individual users, without requiring unacceptably high public subsidies. If such service were indeed consumed by New Jersey s traveling population, the implications are substantial, particularly with respect to New Jersey Transit. Overall daily Average Vehicle Occupancy (AVO) could double (or more) over what exists today throughout New Jersey s roadways which means energy needs and environmental consequences of daily travel would be cut in half. Denser locations during peak hours have even more substantial ridesharing potential that would correspondingly decongest essentially all roadways while providing essentially all with high quality, demand-responsive mobility that is comparable and in some ways better than is offered by today s personal automobile. In addition, thorough analysis is performed on the effect that this model has on NJ Transit. It was found that under conservative estimates, this transportation system could yield a 1.38x increase in NJ Transit volume. These estimates were made to mimic a likely scenario, however, the ultimate scenario yielded an over 5x increase about current NJ Transit volume. The model shows NJ Transit can serve as an efficent complement to transportation systems like the one describe and it deserves to be kept intact and receive further investment.

3 Contents 1 Introduction 3 2 The ataxi Alternative The Mode-Split Process Technical Summary of Mode-Split Process Train Network Analysis Volume and Geographic Analysis Ridership Volume and Scenario Analysis Temporal Analysis Flow Charts for Train Network Visualization Mercer County Spot Light Princeton Station Study Impacts and Conclusion 36 A Likely and Ultimate Rail Ridership 41 2

4 Chapter 1 Introduction To ascertain the opportunities for autonomoustaxis (ataxis) to serve the mobility needs of a large region, such as the state of New Jersey, it is necessary to properly model: the needs and decision processes of the users (each individual that lives or works in NJ), and the choices, the range of mobility opportunities and alternatives (Walk, Cycle, Rail, ataxi) The opportunities for each of the alternatives are quantified by what is traditionally called a mode-split process. It quantifies the relative attractiveness of each mobility option to serve each trip. Each user s decision process selects an alternative. The opportunities for ataxis as well as the other modal alternatives is simply the sum of the choices made by all users for all of their trips. Census data 1 from 2010 was used to create a realistic representation of the more than 9 million current New Jersey residents and out-of-state workers. Call this a 3rd life 2 representation of each of those individuals when ensembled reflect the demographic characteristics of those individuals are reported in the Census data. Importantly, each of those individuals resides at a location whose United States Census 2 http : //en.wikipedia.org/wiki/second L ife 3

5 spatial resolution at the centroid of a Census Block (CensusBlock) which is well within a very short walk of where a person actually resided at the time of the 2010 Census 3. The spatial resolution of the Home of every user is sufficiently precise to appropriately model the home-end access of each modal choice alternative. Travel enables users to do things that they can t or do not wish to do in their own homes. It was chosen to characterize those things as either School, Work or Other for every individual between the ages 5 through 85. Those under 5 and above 85 are assumed to travel only as personally correlated individuals and make no independent travel choices. If they travel, they only ride along. They don t contribute to the statistical analysis except as counts in the total population. For children of a certain age, it is known from Department of Education statistics how many go to school as well as public v private rates. Our data involved a file of all of the schools (SchoolFile), their precise location and the ages that they serve. Then, a bell time was added as to when school starts and ends. This enables each child to be assigned a school. Later in the process, absenteeism rates are used in the travel choice process to choose if the child goes to school on this particular day, as is done with each trip purpose. 4 For simplicity, every individual above the age of 18 and below 86 is assigned by default to be a homeworker. A subset of those homeworkers are probabilistically assigned to an out-ofhome workplace using the Journey-to-work census (J2WCensus) which provides for any county, the number of workers that work in any other county. This file provides the fundamental spatial distribution mechanism for each worker in each county choosing a workplace county. The actual workplace in that particular county is chosen according to relative availability of jobs having a salary opportunity to be commensurate with the income level of the individual worker 5. 3 There are 169,588 Census Blocks in New Jersey in the 2010 Census, where 70% of the blocks house at least one individual. A further disaggregation of the individuals in a census block was begun, but not completed by using publically available telephone directories listing the names and addresses of individuals. Those addresses were geocoded as a linear interpolation of address ranges contained in a digital map representation of New Jersey s streets and assigned to the nearest block centroid. Families from block centroids were disaggregated to the geocoded addresses assigned to that block centroid, if any. 4 Tala Mufti Princeton University Masters Thesis Tala Mufti Princeton University Masters Thesis

6 The employer characteristics, including a precisely geo-coded address of the workplace location are contained in what is believed to be a fairly accurate representation of 435,000 businesses in New Jersey (EmploymentPatronageFile). A specific workplace is assigned to each individual that chooses to become an out-of-home worker. The others remain as homeworkers. This completes the characterization of each individual. The next choice process is for each individual to choose a daily activity pastern. This specifies each individual s daily activity and resulting mobility demand. While more patterns could be considered, for simplicity only 8 are available as depicted in Figure 1 from no travel to a daily tour that involves 7 trip segments between 8 trip ends of 4 trip-end types, Home (H), School (S), Work (W) and Other(O). Home, work and school are each locations associated with the individual. Each is a precise spatial location that is within a very short walk of a real comparable physical location so that any difference between the modeling of the utility of various modal alternatives to and from actual locations as opposed to modeled locations is at most a very short walk difference. Other locations are chosen competitively by the relative attractiveness and accessibility competing alternatives for a compatible activity. Since activities take place at places that have employees, and every employer requires some number customers to visit their premises in order to generate the revenue that it takes to employ people, a daily patronage value, number of people per day, was estimated for each employer in the employment/patronage file. This was done by business classification on a patron per employee basis. Some businesses, such as a manufacturing plant, cater to other businesses and therefore have relatively few patrons per employee on a typical day. Others are consumer oriented, such as a supermarket, these require a large number of patrons/employee on a typical day in order to generate sufficient revenue to remain viable. For the most part, educated guesses were used and are deemed to be sufficiently accurate because the decision process is based on the normalized attractiveness and not its actual value. The attractiveness of a particular Other activity was taken to be the ratio: number of patrons divided by square of the Cartesian distance 5

7 between this trip s current location and this patron s location 6. The normalization is the sum of the attractivenesses of all of the competing activities. If Other is a noon hour lunch, then the competing locations are the subset of restaurants within a short distance (taken as 5 miles) from the work location. If the purpose is shopping, then the subset contains retail stores. Thus, the attractiveness of an activity is actually the probability that this activity will be chosen by the traveler. The actual activity is selected randomly from the probability distribution constructed from the individual probabilities of each of the competing activities. This process assigns specific spatial locations to each of the other locations. Specific departure times, in seconds from midnight, are assigned to trip segment based on timing conditions associated with one of the trip ends. A bell time is associated with each employer, school and store as is a probability distribution of when users tend to arrive. For example an assembly plant requires all of its shift workers to punch a clock and expect employees to arrive by 7am. Being late is discourage thus employees tend to arrive early rather than late. This skewedearly probability distribution Is used to select an actual expected arrival time for each individual. The actual home departure time is determined by doing a rough estimate of a nominal distancebased travel time to set a specific departure time for that work trip. Trips leaving workplaces are also distributed in time. Employees rarely leave early and few stay late. In more relaxed professional work environments, the distributions between early and late arrivals tends to flip towards late arrivals; however, many more stay much later than in manufacturing work places. For shopping, stores have fixed hours and tend to have time varying appearances and durations for its shoppers. Simplified probability distributions that encapsulate the essence of these variations were used to assign a precise departure time to each trip of each tour of each individual. This created a total of 34,062,249 trips. These trips are considered to be a accurate reflection of the demand for mobility on a typical weekday in New Jersey. 6 Distance is floored at 0.25 miles (5 minute walk) to make the level the attractiveness value for equally sized (number of employees) entities nearby the trip origins 6

8 Figure 1.1: Possible Trip Tours of Synthesized Individuals Today, this demand is served by the personal automobile ( 26,500k (83%)), walking ( 3,000k (10%), cycling (300k (1%), school buses ( 600k (2%)), NJ Transit buses ( 600k (2%)), NJ Transit rail ( 300k (1%), commuter buses (to/from NYC (150k (0.5%), light rail (75k (0.2%)), conventional taxis ( 1%). 7

9 Figure 1.2: What Each Trip Tour and Pattern Looks Like 8

10 Chapter 2 The ataxi Alternative The main objective is the analysis is to determine how well an ataxi system could serve New Jersey s mobility demands on a typical weekday. The service characteristics of the ataxi systems considered in this analysis are essentially the same as todays conventional taxi service except: they operate from designated taxi stands, thus requiring a short but less than 5 minute (0.25 mile walk access) to enter a waiting ataxi, a short wait (Departure Delay (DD)), up to 5 minutes for the first passenger to wait for the appearance of potential casual ride sharers, much as elevator doors remain open for a short while waiting for others who may going in the same direction, and as with elevators, may require a slight disruption in a non-stop trip to drop off casual ride sharers that are destined to their desired location. Like conventional taxi service, ataxis service does not require the user to have to deal with the ownership responsibilities associated with a personal automobile such as providing a place for the vehicle to sit and wait to be used, such as a parking place at home, work, school or anywhere else. Also there is no maintenance burden associated with ownership. Moreover, since there is essentially no labor cost associated with operating the vehicle and vehicles mare shared as well 9

11 as individual rides being shared, the cost per trip will be substantially less than that associated with personal private ownership of vehicles. Thus an ataxi fleet operator should be able to be profitable offering such service at a price that is at least a factor of 10 less than conventional taxis and substantially less than even the marginal cost of less than conventional used private automobiles. Thus, there should be little if any incentive for anyone to purchase an automobile as an instrument of daily mobility, except for possibly those living in the most rural areas where the likelihood of both shared ride and shared use is very small. As with express bus service to/from NYC this is considered as a secondary mode in our nested mode split process (NestedModeSplit). The primary modal characteristic is that of ataxi, which once determined might later be separated to conventional auto. 2.1 The Mode-Split Process Three distinctive modes are assumed to compete to serve the daily demand: Walk/biycle, Rail and ataxi. Since Cycle, Rail and ataxi are served from fixed facilities (bike racks/stations/ataxi stands) which are rarely actual trip origins or destinations or destinations there exists an access mode to each of these facilities. For Cycle and ataxi, it is assumed that those access facilities are distributed so ubiquitously that at most a very short walk whose disutility is negligible relative is necessary to complete the trip. For ataxis it is assumed that the walk access is at most 5 minutes (0.25 miles) at the departure end and basically front door at the destination. For rail, access to/from the rail station may actually be provided by ataxi in addition to walking, where it makes more sense to use the rail system. While some precision is necessary in the consideration of the measure: distance, it must be recognized that there is a limit to the precision of spatial locations. Moreover, cognitively, human decision processes tend to be qualitative rather than precisely quantitative so little, if anything, is gained by some level of precision. In this study, it was considered that walk distances of less than 10

12 5 minutes (0.25 miles) were all indistinguishable short walks and all incurred the same disutility. This assumption enabled the pixelization of spatial data to a 0.25 mile resolution. Trip origins and destinations and distance measures were granulized to that level of precision. In one dimension a 3 mile trip covers the range mile trip 3.5, a 0.5 mile trip: mile trip 1. In two dimensions, the Cartesian measure is even more pronounced along the diagonal. While like the personal automobile, ataxis could effectively serve all trips. Some people drive to visit their next door neighbor; however, the imposition of practical limits enables the travel market to be segmented into non-overlapping regions, each of which is dominated by one of these modes. This allows for an all-or-nothing assignment of mode to any trip. While ataxis could serve very short trips, all of these trips are assumed to be served by Walk/Cycle. A short trip was taken to be any trip that is no longer than between neighboring 0.5 mile pixels. ataxis were restricted to not operate into New York City (NYC) and Philadelphia (PHL) for policy reasons in order to steer to the rail system any trip that otherwise would not see the virtue of the NJ transit rail system and efficient interconnecting mass transit systems in those cities to serve those trips. While some of these trips have trip ends within a neighboring pixel of a New Jersey rail station and use walk/cycle to access that station, most trips use ataxis to access the NJ Transit rail station as is depicted in figure The rail system also is assigned to serve all trips that originate and terminate within any neighboring pixel of any rail station (walk/cycle access) that is along the same rail line. Furthermore, for trips longer than 3.5 miles that have one trip end within a neighboring pixel of a rail station (walk/cycle access) and that the total trip length of an ataxi trip to/from the nearest train station on the line serving the walk access station plus the train trip length plus the walk/cycle trip length is not more than 20% greater than the pixel2pixel ataxi Cartesian distance, then the trip is assigned to be a multi-modal trips involving a ataxi access to the nearest rails station and Rail trip to the destination, or visa versa. ataxis are assigned to serve all other trips. As listed in Table 2.1 below xx% trips are ataxi alone, zz% are walk/cycle alone yy% are rail alone and hh% are multi-modal ataxi-rail. Because rail trips tend to be longer and 11

13 walk/cycle trips are short, the percentages are slightly different when a societal impact measure such as person trip miles is used. Trip Statistic ataxi Alone Trips 29,002,346 Walk/Cycle Alone Trips 2,191,097 Rail Alone Trips 44,312 Multi-Modal Trips Involving Rail 1,406,656 Table 2.1: Statistics of Synthesized Trips by Mode Technical Summary of Mode-Split Process Restated, here are the general criterion for mode of transportation and mode splits: Walking/Biking: If the trip is within a pixel, or the trip is from a pixel to an adjacent pixel (i.e. if the trip was less than 1 mile), the trip is walking or biking Train Trips: If the trip is to/from NYC/PHL, go via walking/biking, or ataxi to nearest train station with direct service to NYC/PHL If the trip is longer than 3.5 miles and has only one end within walking/biking of nearest train station, the trip goes to/from non-walking/biking via ataxi from/to nearest train station and continues on a train the rest of the way as long as this mode is not greater that 1.2 times the Cartesian distance between the origin and destination of the original trip. If both the origin and destination are within walking/biking distance of train stations, the trip is a unimodal train trip 12

14 ataxi Trips: All trips that do not qualify for walking/biking or train trips as above are deemed ataxi trips. First, rail trips are not included in the dataset used to analyze ataxi trips. However, emphasized rail transit has two implications on ridesharing: (1) Train travel inherently aggregates travelers both spatially and temporally (they disembark a train at the same location and tame as it arrives at a station) (2) It reduces ataxi trips, by replacing them with rail trips in certain situations. The figure below shows an example of how a multimodal trip substitutes for a unimodal ataxi trip. Point A represents the origin. Point B represents the nearest station to Point A that has a line to NYC, where Point C represents the destination. The possible ataxi trip is A to C, but because A to B, then rail from B to C is both feasible and does not add too much time to the trip, the ataxi trip is replaced by a mode split where an ataxi takes the traveler from A to B and then a train takes him or her to C. The same substitution can take place in reverse, where C is the origin and A is the destination. Figure 2.1: A Demonstration of Rail Mode Split In order to achieve an accurate mode split modal, it was necessary to perform certain spatial modifications to correctly integrate the existing NJTransit static data and train lines into our system. What the above demonstration makes clear is that every pixel needs to be mapped to a nearest train station. Using the 155 stations within the NJTransit system, they were transformed into pixel locations. Next, each pixel in the system was mapped, using minimal Manhattan distance, to 3 different categories of stations. First, they were mapped to the closet train station, 13

15 without qualification. Secondly and thirdly, each was mapped to the closet station that serves NYC and Philadelphia, respectively. Below are figures that provide visualizations of the transformation done to the grid system to achieve this aim. 14

16 Figure 2.2: NJTransit Rail System Map 15

17 Figure 2.3: A Spray Chart Mapping Each Pixel To Nearest Train Station Figure 2.4: A Magnification of the Above Spray Chart 16

18 Chapter 3 Train Network Analysis 3.1 Volume and Geographic Analysis After synthesis, there were a total of 1,513,339 train trips taken in a typical day in New Jersey, as part of our simulated dataset. There were a total of 34,062,249 trips taken in total. Therefore, the train trips represent approximately 4.25% of the trips taken in total. Below is a quick, summary table of the train network to be analyzed. Train Statistics Total Train Trips 1,513,339 Train Passenger Miles million mi. Average Train Trip Length mi. Table 3.1: Train Network The number of train trips taken is the most striking result in our simulated result. In 2012, NJTransit released that on an average weekday, there are just 281,576 train trips taken. Our dataset calls for roughly 5 times as many train trips. The proposed autonomous taxi system would induce 5 times as much traffic on the rail network. Our only assumption was that there were more trains running than there are currently. The proposed train network would have to essentially 17

19 Figure 3.1: Cumulative Distribution of Train Trip Times run as many trains as it runs during peak hours all day. This is not entirely infeasible, and with 5 times as much operating revenue, not without clear benefit. States like New Jersey, with rail networks connecting metropolitan hubs, could represent other areas where such as a system as the one proposed might produce similar results. Because the train trips were taken on a local network, NJTransit, the geographic distribution of train trips was fairly limited. Indeed, only 100 pixels in the entire grid network represent approximately 95% of all train trips taken. Below is a percentage cumulative distribution of all the train trips by pixel, along side a bubble chart, where the size of the bubble is proportional to the volume at that pixel (at that station, assuming there is just one station in the pixel). For scale, the bubble over NYC is on the order of 450,000. Following is a table of the top 10 stations by passenger volume, that correlates with the geographic chart. 18

20 Figure 3.2: Cumulative Distribution of Train Trip Length Within New Jersey 19

21 Figure 3.3: Geographic Spread of Passenger Volume of Train Trips Figure 3.4: Cumulative Percentage of otrain Trips v Pixels 20

22 3.2 Ridership Volume and Scenario Analysis A fundamental assumption in the model is that all trips coming from or to New York City and Philadelphia were made by train. Ultimately, this inflates the number of train trips to and from both of these hubs. In a sense, the actual numbers yielding from the model represent the ultimate rail ridership achievable under the proposed ataxi system. However, in an effort to be as conservative as possible, a more moderate scenario was considered in which the trips to and from these two metropolitan areas were not entirely served by NJ Transit. If the ultimate scenario represents 100% of the trips taken served by the exisiting rail network (under the conditions and assumptions stated), the conservative, or likely, scenario was taken to be 25% of the trips taken served by the exisiting rail network. As stated, the ultimate scenario yields 1,513,339 trips throughout the rail network on a given day. When compared to the current NJ Transit levels, this represents an over 5x increase. However, the likely scenario, which assumes that 25% of potential rail trips actually use rail, to and from and NYC and PHL, yields a 1.38x increase over current volume. This translates to 378,335 daily train trips. Even under conservative estimates, this represents a significant increase. Below are two tables detailing the exact number of trips throughout the network, with a particular focus on New York City and Philadelphia, under each scenario for the top 10 stations by volume, including Princeton and Princeton Junction, as matters of local interest. Station Name To From To From To From To From Station Station NYC NYC PHL PHL Total Total 454, , , ,862 New York Penn Station Philadelphia Street Station 30th 98,761 98, ,761 98,761 Hoboken Terminal 75,942 82,436 50,180 41, ,942 82,436 21

23 Newark 54,284 50,476 13,777 18, ,284 50,476 International Airport Secaucus Junction 54,184 54,086 27,134 24, ,184 54,086 Lindenwold 38,860 32, ,990 36,748 38,860 32,722 Meadowlands Sports 38,468 37,138 19,981 19, ,468 37,138 Complex Station 33,179 23,506 12,540 22, ,179 23,506 Newark Penn Station Cherry Hill 25,973 27, ,341 23,997 25,973 27,971 Newark Broad Street 22,745 18,602 13,695 17, ,745 18,602 Station Princeton Junction 11,672 9,319 5,886 7,064 1,440 2,268 11,672 9,319 Princeton 7,763 7,228 3,434 3, ,763 7,228 Table 3.2: The Ultimate Scenario of Train Ridership in Top 10 Stations by Volume Station Name To From To From To From To From Station Station NYC NYC PHL PHL Total Total Philadelphia Street Station 30th Hoboken Terminal Newark New York Penn Station International Airport 22

24 Secaucus Junction Lindenwold Meadowlands Sports Complex Station Newark Penn Station Cherry Hill Newark Broad Street Station Princeton Junction Princeton Table 3.3: The Likely Scenario of Train Ridership in Top 10 Stations by Volume 3.3 Temporal Analysis As mentioned earlier, the train network in our proposed system carries much heavier traffic than the one in place currently. Trains on each line leave the originating station at 15 minute intervals, staggered throughout time (so not each line leaves at once). At first thought, one might think this were overkill - that this was too much supply for too little demand. However, below is a chart showing the number of passengers leaving on trains leaving New York City during an entire 24 hour period. During peak times, the number is incredibly high, with trains carrying three to four thousand passengers. The average train occupancy is between 1100 and 1200 passengers, and during non peak times the average seems to be just less than 500. With more passengers to alight and disembark during the train s journey over its line, these numbers show just how important 23

25 NJTransit could be, under the proposed system. Figure 3.5: Passengers Leaving NYC 24

26 Figure 3.6: Passengers Leaving PHL 25

27 3.4 Flow Charts for Train Network Visualization As part of the analysis of the train network, it was an objective to provide visuals of the flow of the train passengers over the entire network of our ataxi system. Inherent in a flow chart is the idea of a literal network of nodes and arcs. Real train networks are naturally translated into such data structures. A software suite was written to translate the real train network into a data structure so that you can create visuals of the flow of passengers on certain lines, for certain times of day, and for trips originating in specific counties (stations). The first hurdle was to map each station to a unique node in the network and then, by hand, create predecessor nodes corresponding to the line and to the heading of the train. After that, the data is read in, qualified by the given parameters of time and space, and a visual of the network is created, using Java s Standard Draw library. Below are representations of the visuals of the train network used to provide quick analysis of the network. Each color designates a different train line, which match the actual train lines of NJTransit. 1 A legend with relative flow values can be found at the end of the following 8 graphs each for a 3-hour period during the day. The scale for each of the 8 graphs can be found in Figure 10. Of course, passengers flow both ways on rail transit. Figure 14 is a side-by-side comparison of the volumes of passengers by heading. Heading 1 includes all trains flowing out of NYC, Hoboken, or Philadelphia. Heading 2 incldues all trains with NYC, Hoboken, or Philadelphia as their destination. There is a clear concentration of train volume during rush or peak hours, which correlates to the passenger volume by train throughout the whole day, graphed above. During the early morning and early evening, there is the highest concentration. The temporal distribution of the train trips is therefore quite realistic. The actual volume is not at all realistic, as mentioned earlier. Our train network volume far exceeds that which NJTransit currently experiences. 1 That the volume is split by train lines might be deceiving. For example, the true volume coming out of NY Penn Station is the sum of all of the lines depicted, which in total matches closely to what is seen out of Philadelphia. However, because only two lines come out of Philadelphia the proportions are tricky. 26

28 (a) 12:00AM - 3:00AM (b) 3:00AM - 6:00AM Figure 3.7: Early Morning Train Volume (a) 6:00AM - 9:00AM (b) 9:00AM - 12:00PM Figure 3.8: Early Afternoon Train Volume 27

29 Figure 3.9: Flow Chart Legend: Applicable to all Time Interval Flow Charts (a) 12:00PM - 3:00PM (b) 3:00PM - 6:00PM Figure 3.10: Late Afternoon Train Volume 28

30 (a) 6:00PM - 9:00PM (b) 9:00PM - 12:00AM Figure 3.11: Evening Train Volume Figure 3.12: All Train Trips Flowing out of New York Penn Station 29

31 Figure 3.13: All Train Trips in the Network by Heading 30

32 Chapter 4 Mercer County Spot Light Mercer County train travel is of particular interest because (1) it is where Princeton University is located, and Princeton University is currently very involved in NJTransit and (2) because Princeton Junction, Princeton, Hamilton, and Trenton are extremely important rail stations on the Northeast Corridor Line. This section will provide an in-depth analysis of the rail travel behavior in Mercer County, with a specific focus on Princeton Station. There are two lines that serve Mercer County directly: the Northeast Corridor Line, with terminals at New York City and Trenton, and a Northeast Corridor extended line with service to Philadelphia in our network. First, here is the summary of the train trips that originate in Mercer County (at one of the 4 stations aforementioned). Below are two graphs providing cumulative distribution functions of all of the train trips originating in Mercer County for an overview of both the temporal distribution of the trips, and the length distribution of the trips taken. Another point of interest regarding any particular train is where all the passengers are going. Table 5 details all of the destinations of passengers originating in Mercer County. It ranks all of the destination stations of those trips by the frequency throughout a typical day in New Jersey. It is geographical very diverse, yet concentrated on trips to New York City and Philadelphia, which 31

33 Station Number of Trips Trip Miles (millions) Median Trip Length (miles) Average Trip Length (miles) Princeton Junction 9, Princeton Station 7, Hamilton 6, Trenton 20, Mercer 43, Table 4.1: Mercer County Originating Train Trips Figure 4.1: Cumulative Distribution Function of Percentage of Total Trips as a Function of Trip Time correspond to the train lines available in Mercer County. 32

34 Figure 4.2: Cumulative Distribution Function of Percentage of Total Trips as a Function of Trip Length 33

35 Destination Station Number of Trips ( % ) New York Penn Station 21,743 (50.3%) Philadelphia 30th Street Station 9,178 (21.2%) New Brunswick 4,874 (11.3%) Trenton 4,874 (2.9%) Princeton 986 (2.3%) Jersey Avenue 873 (2.0%) Metuchen 844 (2.0%) Metropark 762 (1.8%) Edison 687 (1.6%) Princeton Junction 605 (1.4%) Newark International Airport 388 (0.8%) Elizabeth 255 (0.6%) Hamilton 213 (0.5%) Rahway 212 (0.5%) Linden 209 (0.5%) Newark Penn Station 109 (0.3%) North Elizabeth 70 (0.2%) Secaucus Junction 11 ( 0.1%) Meadowlands Sports Complex 11 ( 0.1%) Station Total 43,249 (100%) Table 4.2: All Mercer County Originating Train Trip By Destination Station 34

36 4.1 Princeton Station NJTransit claims that at Princeton Junction there are 6,816 average weekday boardings. For comparison, our simulated data has 9,319 trips leaving from Princeton Junction directly. However, this does not include all the trips that leave from Princeton Station, which must go straight to Princeton Junction before continuing on. Therefore, the true Princeton Junction boarding for an average day is 7,228 in addition to the 9,319, which is 16,367 in total. This is roughly 2.5x the current number. Lastly, as the fight over the importance of the dinky, the train that runs from Princeton Station (on Princeton University Campus) to Princeton Junction, continues on, the simulated travel data can provide some reasonable point of reference for the travel potential of the dinky. The University is moving the station 500 yards further away from Campus in order to make room for new buildings, called the Arts and Transit Neighborhood. Currently, there are approximately 1,000 daily passengers who alight at Princeton Station, on the dinky. That counts for 2,000 trips (assuming all return, on average). Our simulated travel set calls for 7,228 daily passengers who alight at Princeton Station, counting for approximately 14,000 trips. This is 7x the current number. This should give us pause when we consider moving the dinky further away from the more central location it used to occupy in Princeton Township. 35

37 Chapter 5 Study Impacts and Conclusion The major impact of this study is that it has found that a great many trips could easily be using the NJ Transit rail system that don t choose to use it today. The increase in daily ridership is 515.3% One reason is that there is no prohibition on the use of personal cars for trips to/from NYC/PHL. According to NJ Transit s 2012 Facts At-A-Glance, the average weekday boardings at New York Penn Station is 79,616 another 8989 that use express bus service to the PABT 1. This still leaves 366,257 (80.5%) trips that were added because of the accessibility provided by ataxis to stations along train lines originating in NYC. 2 Thus, the restriction on the accessibility of ataxis to NYC added a significant number of trips to the rail system, and the same situation exists due to the similar restriction to PHL. Accessibility that exists today by the private automobile is unused because of the cost/unavailability of parking or the marginal use aspects of: I m in my car: should I drive to the station and park and take a train or should I just go ahead and drive and park near my destination. As opposed to: a short walk to the ataxi stand, it drops me off at the train station, there is frequent train service, then a short walk to where I m going; plus, when I come home, ataxis will be waiting to take me to my doorstep. That is a very compelling service proposition because I don t have to deal with parking, plus the price can be made attractive. What is really 1 http : // actsataglance.pdf 2 There are 454,862 train trips that originate in New York Penn Station (NYC) in our system 36

38 of interest are the societal impacts of the ataxi system, especially as might result if users could be encouraged to casually rides. Not surprisingly, the ataxi analysis suggests that ataxis could deliver about the same mobility that is offered by today s conventional automobile. This is highly probable if the ataxis provided personal origin-to-destination service. Since such services could be offered to all, irrespective of their age and at a price that is likely to be cheaper than the cost of today s automobile because of the shared-use capabilities. Unfortunately, except for its broader availability, such a system offers no other societal benefits when operated for personal use only as are most of today s conventional taxi systems. Any benefit that is derived from eliminating any chauffeuring that is going on in the use of today s cars is lost due to the need for empty ataxi repositioning so as to realize its shared use opportunities. In the end, a personal ataxi operation would have an AverageV ehicleoccupancy(av O) = P assengerm iles/at axim iles 1.0 (5.1) therefore, the energy used qnd the environmental impacts would remain unchanged. Also, the roads driven at the times they are used would also be essentially unchanged, therefore, there is no expected congestion relief. However, it is not necessary that ataxis be operated in a door2door personal use only basis which afford no opportunity for casual ridesharing, the kind of ride sharing that happens in elevators every day simply because un-otherwise-related individuals just happen to find themselves at the same elevator bank on the same floor going in the same direction to some group of floors. Each user could refuse to share the elevator and wait until there was no aother casual rider sharer around or the building designer could have required that sufficient elevators were built so that no none would ever have to share a ride and the operation of the elevator was such that only one destination floor could be selected each of which would have been much more expensive than the expected benefits that might have be delivered to users such that none were even ever contemplated, let alone implemented. The accumulated disutility associated with having some users share the ride 37

39 plus the disutility o having to make some intermediate stops, plus the wait time experienced by some riders while to doors stayed open in case someone else showed up to share the ride was so small in comparison to the benefits derived by the building owner associated with building fewer elevator shafts, and less up/down travel or in comparison with deciding to use the steps (although this choice is selected by some going for short trips) or not frequenting the services available in the building (staying home). The basic question is how do societal benefits accrue as the ataxi level of service diminishes or as the disutility of ataxi service increases? In general, allowing casual ridesharing decreases the attractiveness of ataxi service (increases its disutility). For some, this change is substantial; however, there are existing situations, for example elevators and airport automated people movers, where the change is insignificant because appropriate design considerations have been made to ensure personal safety and privacy. It is assumed that the ataxi stands and vehicles will be designed and equipped with elements that users will perceive no significant disutility associated with casual ridesharing. Also, a surcharge could be placed on the price of a guaranteed-personal ataxi ride. This charge could be set at a level that would more than compensate for the loss in potential societal benefits associated with the decrease in expected societal benefits associated with this restriction. In general this surcharge could be small in regions and times when casual ride sharing is expected to be low, such as in the middle of the night when the expected disutility of casual ridesharing is very large. Even if implemented, the AVO is so close to 1.0 that it really can t be reduced much more, therefore the effect is minimal. The surcharge should be high at times when ridesharing expectations are large such as during peak hours in peak directions. This is when the perceived disutility of ride sharing is smallest because travelers at those times in those directions have become comfortable with the concept and ride sharing often involves a crowd rather than just one individual. Thus the user is likely to forego the surcharge in favor of absorbing the additional disutility, resulting in no significant change. The attractiveness of ataxi service is also diminished if its departure is delayed in order to 38

40 wait for the potential of a later arriving ride sharer. For short periods of time, this disutility might increase linearly with Departure Delay (DD), but at some point increase much faster and quickly become intolerable. The crossover point must be related to the expected ataxi ride time. In conventional transit, wait time is often weighted by a factor of 2.5. Since the median ataxi trip length is about 12.5 miles, a departure delay of 5 minutes has the equivalent disutility as the eventual median trip. While a 2 minute DD might be considered on the border of insignificant, a DD of 5 minutes borders on intolerable. Since the DD in an ataxi operation is certain rather than the traditional wait for conventional transit service is traditionally very uncertain, the 2.5 factor may be high when applied to DD. Also, there are technological ways to obtain the knowledge equivalency of DD without actually having the user experience the full extent of DD physically. For example, if service is guaranteed and started while the first rider is still approaching the ataxi stand through the use of smartphone enroute (nroutecommerce) services, then the actual wait and the perceived disutility will be reduced. In our analysis, the effects of values of DD up to 5 minutes were computed, realizing that a DD=5 may well impose a disutility that substantially detracts the ataxi level-of service such that its underlying modal share is becoming questionable. The incursion of extra travel time associated with having to drop off casual ride sharers who aren t traveling as far also distracts from the personal ataxi ride. However, since the additional circuity added by any shorter ride sharer is limited to small increase (< 20%) in travel time (trip length), then this increased disutility is considered to be minimal and insignificant, especially since each traveler is dropped off essentially at the doorstep of their destination, which is better than the level of service offered by conventional automobiles today. Consequently, the modes are designed to serve well defined markets where at most a short walk is assumed. Basically each of these modes serves a well-defined trip market segment with little overlapping grey areas. While ataxis might be expected to become the dominant mode serving the demand and would be the primary user of the existing road infrastructure, it would not be the only mode. Walking and cycling would remain to compete to serve and would probably capture the short distant trips. 39

41 More importantly, the rail system should be kept intact to determine the extent to which it can complement the ataxi system to best serve mobility that is from or to locations in NYC, PHL as well as near any other rail station. 3 It was found that even under conservative estimates, the ridership on NJ Transit can be greatly increased as new transportation systems like the one proposed come to bear. This emphasizes the point that NJ Transit is underutilized, but when the full potential is reached, it in turn emphasizes that NJ Transit is an important complement to our transportation system within New Jersey. 3 While this analysis did not include express bus service from park & ride lots to the Port Authority bus terminal via the Lincoln tunnel and the XBL, some of the rail traffic to NYC could be readily diverted to such a service without imposing a substantial effect on the ataxi services that would deliver the users to from the nearest park 7 ride lot instead of the nearest rail station. 40

42 Appendix A Likely and Ultimate Rail Ridership Below are the two extended tables that detail the ridership under both mentioned scenarios at every train station within NJ Transit. Station Name To From To From To From To From NYC NYC PHL PHL Total Total Station Station New York Penn Station 454, , , ,862 Secaucus Junction 54,184 54,086 27,134 24, ,184 54,086 Newark Penn Station 33,179 23,506 12,540 22, ,179 23,506 Newark International 54,284 50,476 13,777 18, ,284 50,476 Airport North Elizabeth 11,596 7,788 5,015 8, ,596 7,788 Elizabeth 21,660 19,219 8,061 10, ,660 19,219 Linden 19,461 16,516 6,106 8, ,461 16,516 Rahway 13,014 10,929 4,029 5, ,014 10,929 Metropark 9,991 9,579 5,340 5, ,991 9,579 Metuchen 7,764 7,880 3,948 4, ,764 7,880 Edison 7,412 7,726 5,446 5, ,412 7,726 New Brunswick 16,903 15,854 6,854 8, ,903 15,854 Jersey Avenue 15,138 14,952 11,268 11, ,011 15,138 14,952 Princeton Junction 11,672 9,319 5,886 7,064 1,440 2,268 11,672 9,319 Princeton 7,763 7,228 3,434 3, ,763 7,228 Hamilton 6,579 6,116 3,089 3, ,579 6,116 41

43 Trenton 21,118 20,586 8,947 11,678 6,061 4,016 21,118 20,586 Hoboken Terminal 75,942 82,436 50,180 41, ,942 82,436 Avenel 3,148 3,362 2,004 1, ,148 3,362 Woodbridge 7,358 6,610 2,390 3, ,358 6,610 Perth Amboy 13,835 13,872 3,249 3, ,835 13,872 South Amboy 9,273 9,446 4,210 3, ,273 9,446 Aberdeen-Matawan 10,732 13,388 9,081 6, ,732 13,388 Hazlet 3,833 5,205 3,446 1, ,833 5,205 Middletown 2,960 4,515 3,285 1, ,960 4,515 Red Bank 5,940 7,503 2,878 1, ,940 7,503 Little Silver 1,813 2,425 1, ,813 2,425 Monmouth Park 1,313 1,648 1, ,313 1,648 Long Branch 2,888 3,594 1, ,888 3,594 Elberon 1,122 1, ,122 1,470 Allenhurst 1,887 2,562 1, ,887 2,562 Asbury Park 4,062 4,899 1,807 1, ,062 4,899 Bradley Beach 3,111 3, ,111 3,533 Belmar 4,164 4,964 1,796 1, ,164 4,964 Spring Lake 2,254 2,628 1, ,254 2,628 Manasquan 3,700 4,483 3,420 2, ,700 4,483 Point Pleasant 1,144 1, ,144 1,564 Beach Bay Head 12,599 14,256 4,699 2, ,599 14,256 Union 1,181 1, ,181 1,200 Roselle Park Cranford 1,598 1, ,598 1,497 Garwood 1,945 1, ,945 1,920 Westfield 1,603 1, ,603 1,526 Fanwood-Scotch Plains Netherwood Plainfield 3,260 2, ,260 2,881 Dunellen 2,101 2, ,101 2,116 Bound Brook 1,745 1, ,745 1,761 Bridgewater Somerville 2,762 2, ,762 2,900 Raritan 2,065 2, ,065 2,095 North Branch White House Lebanon Annandale