Integrating Aviation and Passenger Rail Planning (2015)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

64 C H A P T E R 5 Introduction and Structure Chapter 5 now focuses on the Northeast Corridor (NEC) of the United States. First, new data comparing the rail share of the air plus rail market in the NEC is compared with simi- larly defined data from major HSR corridors in Europe. Sec- ond, the Chapter presents an in-depth analysis of the change in air travel demand in the NEC—in terms of segment volumes, O-D volumes, number of flights, and the size of the aircraft used in the corridor—following increased competition from other modes. Initial airline response research for the ACRP project’s model building is presented in terms of a description of cost per operation, cost per seat, and an early cross-sectional analysis of change in supply. The analysis of data needs and the adequacy of planning tools continues with an examination of several recent analysis efforts concerning the modeling of demand on the corridor, including a major multimodal plan- ning effort undertaken for the PANYNJ by the Regional Plan Association (RPA), and a discussion of the structures of mod- els currently being applied in the area. The Chapter closes with a review of key questions concerning the adequacy of models and tools being applied in the NEC today. Part One: Diversion to Rail and its Effect on Air Service The NEC is the ideal corridor in the United States for learn- ing about the substitutability of HSR trips for air trips. There are both observed data stemming from the fact that this cor- ridor is the only corridor served with HSR (Acela), and stated preference data from several studies of existing and improved rail service along the entire corridor. However, it is difficult to separate the effect on air travel from observed data on the gradual introduction of Acela between 1999 and 2005; this is because additional events since 1999 affected air travel in the corridor (e.g., September 11, increased airline screening delays, economic volatility, and the introduction of frequent and very cheap bus service, etc.). This can be seen in Figure 5-1, which shows a time series of Amtrak’s share of the air plus rail market between Boston and New York between 2000 and 2010. There are, however, several recent studies of the impact of improved rail service on NEC air travel, including some studies that use stated preference data. This Chapter will dis- cuss the results of recent studies that include this impact and discuss how these studies were conducted. An important recent RPA study of ways to improve the operation of New York Region airports is included; this study evaluated HSR’s ability to free up capacity at the airports. This study illus- trates the limitations of available data on corridor travel by all modes and their effect on conducting such a study. The previous Chapter presented the rail portions of the air plus rail markets for major European city pairs. Consistent graphics were presented for summary observations about mar- kets in France, Spain, and the United Kingdom. As noted in the text, these calculations were based on the air passenger volumes for the corridor, specifically including those using air to connect with other flights. Comparing United States Market Behavior with that of Europe Mode share data in the United States, shown in Figure 5-1, has been designed to help the analyst understand rail’s success in capturing a given market between Region A and Region B. In the NEC, this analysis uses station-to-station rail flows for a group of stations at the origin end of the trip and a group of stations at the destination end of the trip; these are then compared with U.S. DOT data that specifically excludes air passengers transferring to other final terminals. In order to make a comparison of the American experience with rail diversion from air with that of the Europeans, a smaller database of European corridor data estimated on an O-D basis has been created—this specifically excludes air passengers des- tined for network connections. Table 5-1 presents the data from the two continents, which is graphed in Figure 5-2. Rail Diversion from Air in the United States: Data and Methods

65 mode share to result from good in-vehicle travel time. When the European data is organized in this manner, the 3.5 hour observation is supported. The Madrid-Barcelona corridor has an estimated 55% market share, when it is defined in terms of O-D markets only; this reflects its very low rail travel times of 2 hours 40 minutes. In order to gain a majority of the air plus rail market, having a travel time of better than 3.5 hours appears to be a necessary, but not sufficient, condition for achieving a majority market share. Effects of Improved Rail Service on Diversions from Air: Boston to New York City In 2000, Amtrak’s regional NEC rail service captured 20% of the rail plus air passenger trips between Boston and New York City. The Acela service that Amtrak introduced in Decem- ber 2000 offered travel times between Boston and New York of approximately 3.5 hours for the 200-mile trip, representing a reduction of between 30–50 minutes compared to the com- parable regional trains. The net result of this new service has been a significant increase in the rail share, to 54% of the air/ rail market. Over the corresponding period, the total O-D air passenger volume in the Boston to New York City market declined by almost half, as shown in Figure 5-3. There were also shifts in the distribution across New York airports and a shift to JFK following the introduction of JetBlue service. There were also some changes in patterns at the regional airports in the Boston area (MHT and PVD), but the overall pattern is clear: introduction of Amtrak’s Acela service resulted in a significant increase in rail passenger volumes between Boston and New York City, with corresponding large decreases in air passenger volumes in this market. In total, the air passenger volumes in the Boston to New York City origin-to-destination market decreased by almost half between 1999 and 2010. During this same period, the Similarities and Differences Arguably, the most visible characteristic revealed by the chart (Figure 5-2) is the similarity of the data points from the two separate continents; however, the linear equation for the United States shows a slightly lower propensity for high NEC Data Minutes Rail Share of O D Market Albany New York 60 97% New York Philadelphia 70 95% Philadelphia –Washington 130 89% Providence New York 150 90% New York Washington 190 63% Boston New York 200 55% Boston Philadelphia 280 17% Boston Washington 420 7% European Data Frankfurt Cologne 70 99% London Paris 150 90% London Brussels 125 90% Paris Marseille 200 82% Madrid Seville 150 92% Madrid–Barcelona (post 160 52% Madrid Barcelona (pre) 420 10% London Edinburgh 270 25% Milan Rome 270 43% London Manchester 127 80% Sources: SDG 2006, Amtrak 2009 Table 5-1. Comparison of U.S. city pairs with European city pairs—O-D market only. 60 Figure 5-1. Change in rail share of air plus rail market between Boston and NYC airports. Source: Amtrak 2014.

66 In this market, the average plane size decreased by 24% between 1999 and 2010. This down-gauging allowed airlines to offer only slight reductions in flight frequency in response to much larger reductions in air passenger volume, as shown in Figure 5-6, and in table format in Table 5-2. In summary, HSR services, such as Acela, can significantly affect air passenger volumes in markets like Boston to New York City. However, reductions in the air passenger market do not necessarily translate directly into reductions in flight vol- umes. In particular, O-D passenger volumes between any two major airports represent only a portion of the total segment passenger volumes; as a result, reductions in O-D passenger number of air passengers who used Boston to New York City flights to connect to flights going elsewhere increased by 13%. However, the much larger decrease in O-D volumes resulted in a 30% decrease in the passenger volumes on flights between Boston and the New York airports, as illustrated in Figure 5-4. Figure 5-5 illustrates the net effect of the 50% decline of air passengers traveling between Boston and New York on the number of flights operated. The number of flights between Boston and New York declined by only 16% between 1999 and 2010 (Figure 5-5). The fact that a segment volume decrease of 30% led to only a 16% decrease in flight volume is a result of shifts in plane size. N.B. European data points are represented as diamonds, with a solid trend line: American data points are represented as squares, with a dotted trend line. The two shares for Mardrid-Barc. present shares before and after HSR. Sources: Amtrak 2009, SDG 2006. 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0 60 120 180 240 300 360 420 480 Ra il as Sh ar e of Ra il Pl us Ai rM ar ke t In vehicle Rail Travel Times, in Minutes Albany NYC 97% New York Phil 95% Providence NYC 90% Phil –Washington 89% NYC Washington 63% Boston NYC 55% Boston Phil 17% Boston Washington 7% FRA Cologne 99% Madrid Seville 92% Lon Paris 90% Lon Brussels 90% Paris Marseille 82% London-Manchester 80% Madrid-Barc (new) 52% Milan Rome 43% London-Edinburgh 25% Madrid Barc (old) 10% Figure 5-2. Relationship between rail travel times and rail market share for American and European data for origin-destination markets only. 200,000 400,000 600,000 800,000 1,000,000 1,200,000 1,400,000 1993 1999 2007 2010 Bo st on to N YC O D Ai rP as se ng er s BOS > LGA BOS > JFK BOS > EWR Figure 5-3. Change in segment volumes Boston to NYC. Source: BTS, DB1B data.

67 200,000 400,000 600,000 800,000 1,000,000 1,200,000 1,400,000 1,600,000 1,800,000 1993 1999 2007 2010 Bo st on to N YC Se gm en tA ir Pa ss en ge rs BOS to NYC Connecng BOS to NYC O D Figure 5-4. Change in air passenger volume for connecting and O-D markets. Source: BTS T-100 and DB1B data files. 5,000 10,000 15,000 20,000 25,000 30,000 1993 1999 2007 2010 Bo st on to N YC Fl ig ht s BOS > LGA BOS > JFK BOS > EWR Figure 5-5. Change in number of flights between Boston and NYC. Source: BTS T-100 Files. 0 20 40 60 80 100 120 140 160 180 1993 1999 2007 2010 Av er ag e Pl an e Si ze (p ax ) BOS > EWR BOS > JFK BOS > LGA Total BOS to NYC Figure 5-6. Change in size of plane by airport pair and by total corridor. Source: BTS T-100 Files.

68 The average number of seats per aircraft has declined in two markets and increased in one market. The decrease in segment volumes partially explains these changes. However, the average passengers per flight decreased in these markets; overall, there are additional incentives for airlines to choose to fly smaller aircraft, as discussed herein. Airlines may have a cost incentive to down-gauge. A linear jet operating cost model was used to evaluate changes in cost (Ryerson 2010). The model takes fuel price, seats per opera- tion, and distance as inputs and determines the operating cost (in 2006 dollars). Using this model, a cost per operation and an average cost per seat for four years in 2006 dollars were estimated. A constant fuel price of $2.00/gallon is used so that the change in cost is not overwhelmed by the large fuel price fluctuations between 1993 and 2010. The average cost per seat increased by about 16% (Fig- ure 5-7) for the two markets where seats per operation dropped average. However, when the cost is considered on a per opera- tion basis, cost per operation for these two markets dropped by 18% (BOS → EWR) and 25% (BOS → LGA), as shown in Figure 5-8. Considered on a cost per operation basis, there is an incentive to down-gauge. Additionally, six airlines served the BOS → EWR segment in 2010, according to BTS data. Competition is high, which also keeps aircraft sizes low. Airlines compete in the marketplace on a few key variables: fare, frequency, and service components (like frequent flier programs, first and business class cabins, etc.). However, evidence has shown that airlines can increase their market share by increasing frequency rather than alter- ing the other service variables (Wei and Hansen 2005). Beyond volumes do not translate into proportional reductions in seg- ment volumes. Further, carrier response to passenger volume reductions can cause reductions in plane size, which means less proportional reduction in flight volumes. In the case of the Boston to New York City market, a roughly 50% increase in rail volumes resulted in a similar 50% reduction in Boston to New York O-D air passenger volume. This reduction, in turn, lowered the total air passenger volume between those cities—including connecting passengers—by 30%. Finally, changes in plane sizes meant that the number of flights between Boston and New York decreased by only 16%; the net effect on airport runway loads was only one-third of the effect in O-D air passenger volumes. To the present, mod- els have been built to capture mode shifts between air and intercity rail, but there are not companion models that simi- larly estimate the net effects on airport service and capacity following carriers’ responses to market shifts. These effects are equally important in understanding how HSR will affect airport service and capacity. Response of the Aviation Industry to Shift in Demand Aircraft Size Changes, Boston to New York The Research Team has evaluated the change in costs seen by the operators of aircraft between Boston and New York air- ports: LGA, EWR, and JFK. Significant trends in aircraft size changes were observed in operations between Boston and the New York airports, as discussed in the following paragraphs. Average Aircra Size 1993 1999 2007 2010 BOS --> EWR 118 141 99 101 BOS --> JFK 92 65 79 86 BOS --> LGA 153 158 109 98 Operaons BOS --> EWR 9,511 5,379 4,394 3,978 BOS --> JFK 3,729 8,266 8,089 6,809 BOS --> LGA 11,741 11,959 11,478 10,632 Total Segment Volumes BOS --> EWR 564,070 500,843 289,290 270,680 BOS --> JFK 176,229 249,127 471,292 429,683 BOS --> LGA 790,269 963,948 664,383 498,703 Average Passengers/Flight BOS --> EWR 59 93 66 68 BOS --> JFK 47 30 58 63 BOS --> LGA 67 81 58 47 Total BOS to NYC 61 67 59 56 Source: BTS T-100. Table 5-2. Summary statistics, Boston to NYC.

69 entire day—a June day at LGA in 2010—are then obtained from the Aviation System Performance Metrics database. Delay can then be estimated in both scenarios. In scenario two, with decreased operations, there is no observed delay. For scenario one, there are two time regions of observed delay: from 9:00 a.m. to 1:00 p.m. and from 4:00 p.m. to 7:00 p.m. The total delay is 11 aircraft hours, or about 1.2 min- utes per flight for flights scheduled to arrive during the delayed period. There is almost no incentive for airlines operating at LGA to up-gauge their aircraft as the delay savings are negligible. The negative externalities are high even though this is a low level of delay. If all the delay occurs in the air, the fuel con- sumption waste could be up to 33,000 lbs. of fuel (Ryerson and Hansen 2011). Cross-Sectional Analysis of Flight Service from Logan Airport In this section, the flight schedule from Logan airport is analyzed to determine whether competition from intercity increased frequency, airlines can increase their competitiveness through targeted frequency, most notably by scheduling flights at times very close to those of their competitors as put forward by Borenstein and Netz 1999. This is an example of airlines act- ing in their own self-interest; however, it is also a classic game theory problem—once one carrier schedules an operation at a particular time, the other carriers must follow, depleting the breadth of available flights for everyone. Congestion Changes in airport congestion due to aircraft down-gauging were also considered. Two scenarios of demand were compared at LGA. The first scenario, the baseline, is the 2010 operational level. The second scenario is the 2010 operation level if all flights from BOS to LGA were on the maximum aircraft size observed across the 4 years shown in Figure 5-8. The number of opera- tions between BOS–LGA on this maximum aircraft size was found; operations in the 2010 schedule were reduced to reflect the decreased level of operations. The operational demand and the airport acceptance rate (AAR) over time for the 1 2 3 4 5 $/ O pe ra o n 0 000 000 000 000 000 1993 1999 Year 2007 2010 BO BO BO S > EWR S > JFK S > LGA Figure 5-8. Cost per operation. 0 5 10 15 20 25 30 35 40 45 $/ Se at 1993 1999 2007 Year 2010 0 20 40 60 80 100 120 140 160 180 Av er ag e Se at sp er O pe ra o n BOS > E BOS > J BOS > L "BOS > Operao BOS > J BOS > L WR FK GA EWR Average n" FK Average Se GA Average S Seats per ats per Operat eats per Operaon ion Figure 5-7. Average cost per seat and seats per operation.

70 craft size so that airlines can maintain frequency in order to compete with the rail service. Results confirmed this expec- tation: aircraft sizes for segments from the New York airports are, in aggregate, 18% less than those predicted. For other segments originating from BOS, observed aircraft size was 7% less than predicted. BOS is primarily a low-gauge air- port, but the effect is pronounced on the segments with rail competition. In summary, an early cross-sectional analysis provides ini- tial evidence that rail competition affects air service from BOS by encouraging down-gauging, but it does not show evidence that overall seat capacity is affected. Further analysis would be required to verify and refine these results in later research studies. Other American Diversions from Air to Rail The corridor between New York City and Boston was selected for this research project for several reasons, includ- ing the clear-cut improvement in travel times associated with the implementation of electrification between New Haven and Boston and the simultaneous commencement of the Acela service line. By comparison, incremental improve- ments on the service between New York City and Washing- ton, D.C., have occurred over several decades, beginning with the development of Metroliner services in the 1970s. While a full analysis of the impact of such changes on air ridership requires the kind of airline data presented in this Chapter, some observations of scale can be made about the scale of diversion in the NYC/D.C. corridor. As shown in Figure 5-9, Amtrak reports that between 2004 and 2013 the mode share of O-D passengers grew from 50% to 75%. Given that the section of the NEC between New York City and Washington has the highest ridership, it would be desirable for further research in this area to undertake an analysis similar to that presented here for the relationship of rail service characteris- tics to change in air passenger volumes. The research has not identified any other corridor in the United States where significant diversion to rail has occurred. rail has a discernible impact on airline service. Specifically, cross-sectional regression analyses are performed on airline seats (expressed as log seats per month in July 2007) provided and average aircraft size between BOS and U.S. domestic des- tinations. Observed values are compared with model predic- tions for two classes for destinations: those with significant competition from Amtrak and those without significant rail competition. In practice, the rail competitive class contains only the New York area airports of EWR, LGA, and JFK. In 2008, rail accounted for 49% of the total travel by rail and air between Boston and New York in 2008 (Amtrak 2014). The only other market from Boston with meaningful competition was Phil- adelphia, where the rail share was 17%. Philadelphia will not be considered rail competitive for this analysis since the rail share is significantly less than in the Boston-New York market. Using the Official Airline Guide flight schedule of July 2007, seats per month were regressed from BOS to 68 domes- tic destinations. Two main explanatory variables were used: total airport seats per month at the destination airport and distance to the destination airport. A log-log model was used and introduced second-order terms (such as the square of airport seats per month) that were found to be statistically significant. The resulting model, based on 68 flight seg- ments, has an adjusted R2 of 0.74 (Table 5-3). If rail com- petition had an impact on airline capacity supply then the model would over-predict supply in the rail competitive markets. This did not turn out to be the case. In two of the three segments—JFK and LGA—the model under-predicted seat supply; only in EWR did it over-predict. Overall, the model slightly under-predicted seat capacity from BOS to the New York airports. The effect of rail competition on aircraft size was then analyzed. A previously developed aircraft size model (Wei and Hanson 2005) was used on flight segments with at least one end at a major U.S. airport. The prediction errors of this model for segments originating from BOS were compared. One would expect rail competition to result in reduced air- Dependent Variable: log(seats per month) Parameter Estimates p eulaV t dradnatS retemaraP FD elbairaV etamitsE Intercept 1 -3.20261 3.96205 -0.81 0.4219 lstm log(distance in stat. mile 2030.0 22.2 6951.1 1175.2 1 )s lap_spm2 log(seats per month at destination airport)^2 log(seats per month at destination airport)*log(distance in stat. miles) 1 0.08512 0.02248 3.79 0.0003 R2=.7382 Table 5-3. Model results for cross-sectional analysis.

71 RPA in New York City for the PANYNJ represents a major breakthrough in the application of both multimodal and multi- jurisdictional methods. The RPA study evaluated NEC HSR improvements as one of six categories of potential investments to reduce delays at New York region airports. The other cat- egories included improved air traffic control systems, build- ing a new airport, encouraging the use of outlying airports in the region, expanding runway capacity at the three major airports, and managing demand to reduce peak period flights (Regional Plan Association 2011). With regard to the impact of true “California-style” (i.e., with line haul speeds significantly higher than that of pres- ently used rail technology) HSR on New York Region airport use, the report states that, on an average day, almost 160,000 people left the New York region by either air or rail in 2008. Of these, 145,200 flew, including connecting passengers, and 13,200 used intercity rail. However, for the five destinations on the spine line of the NEC with air service (Baltimore, Boston, Philadelphia, Providence, and Washington), the number of daily departing rail passengers is reported as 9,200, with a rail share of over 50% (air passengers were 8,400). However, 9,200 rail passengers is less than 6% of the total air passen- gers using the three airports on a daily basis. True “California- style” HSR in the NEC is estimated to divert 2,049, or 24.4%, of the 8,400 total departing passengers (as of 2008) to the five NEC cities. This is because connecting passengers make up more than half of departing air passengers, and they were not included in the pool of passengers who might divert to rail. The percentage of passengers connecting from other flights to Boston are 67%, 60%, and 22% from JFK, EWR, and LGA, respectively. To closer cities like Philadelphia, the Part Two: Data Needs and Modeling Capabilities Applied in the NEC Recent Data Analysis Activities in the NEC This section of Chapter 5 now turns to the review of analyses, by others, concerning how air and rail interact together in the Northeast Corridor of the United States, with an emphasis on the data needs and modeling capabilities employed. At present, the Federal Railroad Administration is undertaking a Tier 1 Environmental Impact Statement examining all aspects of the future of rail operations in the Northeast, notably merging the analysis of long-distance rail with that of metropolitan com- muter rail operations. That project includes the analysis of the relationship of rail operations and airport activities in the Cor- ridor. It is particularly relevant that the FRA project empha- sizes the integrated analysis of commuter rail operations with those of long-distance operations, as two of the region’s largest airports, John F. Kennedy International Airport and Philadel- phia International Airport are currently served by rail opera- tors providing commuter rail services. This section of Chapter 5 reports on relevant multimodal analyses done at Amtrak, the FRA and the Office of the Inspector General. The section commences with a review of an important intermodal study undertaken by the New York’s Regional Plan Association, on behalf of the Port of New York and New Jersey. The 2011 RPA Study of the New York Region’s Airports The January 2011 report, “Upgrading to World Class: the Future of the New York Region’s Airports,” prepared by the Pe rc en t R ai l Rail Share of Air+Rail Market Figure 5-9. Market share for two NEC markets. Source: Amtrak, 2014.

72 benefit to air travelers making use of the available slots for higher value, longer distance travel. In 2010, about 104 million people accessed the three major New York airports. The report expects “the demand for pas- senger volumes would reach 150 million, if the capacity is available, as early as 2030.” With this overall demand, the report projects 19 peak hour slots freed up from “California- style” HSR in the NEC distributed as follows: 3.5 slots at JFK, 3.2 slots at EWR, and 12.3 slots at LGA. The much higher impact on LGA is due to its larger percentage of traffic to nearby destinations and a smaller share of passengers con- necting to other flights. The report concludes that: “a successful expansion or reconfiguration at Kennedy and Newark, along with NextGen (ATC), can meet the twin goals of capacity and delay reduction in the 2030s and beyond. . . . The inability of the combined impacts of NextGen, outlying airports and faster intercity rail to stem the need for eventual airport capacity expansion should not be viewed as a reason to deemphasize these actions. To the contrary, they are each of great value. . . . Faster rail travel, particularly in the Northeast Corri- dor, will divert travelers from the highways and knit together the economies of the Northeast” (Regional Plan Association 2011). In summary, this RPA study demonstrates that building a HSR system, even in the NEC, is not a panacea for solving the airport congestion problem. The contribution to reduc- ing delays at airports or increasing airport capacity for higher value long-distance routes is “only one of the environmental, economic and social benefits of having a high-quality passen- ger rail service and only one of the many reasons that justify major investments in our rail network.” Data and Analysis Needs Identified in the RPA Report for Integrating Air Rail Planning The RPA report, like most United States studies of the sub- stitution of rail trips in place of air trips, suffered from severe limitations of available data. The published report describes these limitations: “Both the rail and air data are station-to-station (or airport- to-airport), and do not provide information of the specific ori- gin or destination within the metropolitan areas for each end of the trip. More refined trip data would have made it possible to create a more nuanced demand model. However, these data either do not exist or are not available from the carriers. Reli- able intercity automobile travel data is unavailable. If it were, the interplay among the three modes and their shares would have been of great interest. The lack of auto data has long been a handicap to intercity travel modelers, and its continued absence prevents credible estimates from being made of how well speedier rail service can attract auto travelers” (Regional Plan Association 2011). percentages are much higher, at 94%, 89%, and 90%, respec- tively. The diversion of air passengers to Boston, including connecting passengers, was also 24%, while the diversion to Washington was higher at 36%; it was much lower to closer cities like Philadelphia (9%) since so much of the air travel is connecting passengers. Even “California-style” HSR would divert approximately 2,050 passengers—or 1.4% of the total 145,200 daily air passengers—departing the three major New York airports in 2008. This study is unusual in relating the potential diversion of air passengers to the total air passengers using the airport(s), espe- cially considering the difficulties faced by the report authors with the input data discussed herein. The study dealt with both connecting and true O-D passengers. In addition, the reported results are for diverted air passengers between specific cities rather than for all passengers using specific airports. Interestingly, the RPA study estimates available airport capacity from the diversion of air passengers to a specific HSR proposal. This additional capacity could be used for flights to other destinations not served by rail and/or to reduce delays at the airport(s). Either way, the additional capacity provides benefits. Calculating the additional capacity at New York’s three major airports involves making important assumptions about the supply response of the airlines to the reduced number of passengers, in the context of important institutional con- straints imposed by operations at these airports. Quoting the study, “JFK and EWR serve as major hubs for travel to long-distance and international markets. The airlines rely on shorter flights to feed these routes. Rather than eliminating flights or reducing the frequency of service that could create longer connections, airlines may instead shift to smaller aircraft and keep the same number of flights. . . . Further, the size of the flights or individual markets may not make it practical for the airlines to reduce the number of flights. Their reluctance to drop flights may also stem from their interest in retaining peak hour slots where capacity is capped by the FAA” (Regional Plan Association 2011). The report makes the important observation that “desti- nations with large markets such as Boston or DCA, are the most likely candidates for fewer flights where the airlines can accommodate the loss of air passengers to rail more easily” because of the higher frequency of flights (Regional Plan Association 2011). Given the uncertainties in making these estimates of available slots, the estimates in the report “are likely to be the maximum possible values for capacity freed up, rather than probable impacts” (Regional Plan Association 2011). Thus, introducing HSR in the most heavily traveled intercity corridors maximizes the benefits to potential HSR users and non-users who remain on air between the same cities, particularly connecting passengers, and maximizes the

73 and Washington to serve the fast-growing intercity rail market, provisionally known as the NEC Next Generation High-speed Rail or “NextGen HSR” system. . . .This Vision is currently being further evaluated and refined so that it may serve as one of the bases for the more significant planning efforts in FY 2012 and beyond (Amtrak 2012). Amtrak’s $117 billion “Vision for High-Speed Rail in the Northeast Corridor” proposal would cut the travel time between New York and Boston to 1 hour 24 minutes and 1 hour 36 min- utes between New York and Washington for its “Nextgen High- Speed Express” service. Amtrak projects that NEC Nextgen HSR ridership in 2050 would be five times Acela levels; overall NEC rail ridership would be three to four times the current level of 11.8 million riders. Interestingly, Amtrak’s 2050 projection predicts only 23% of the overall NEC ridership increase to be diverted from air, with the rest diverted from auto (47%) and induced new riders (30%) as shown in Figure 5-10. However, between Boston and New York, Amtrak envisions a larger shift to rail from air and auto “where the most dramatic improve- ment in rail travel times is predicted.” In 2050, their diversion from air would eliminate “true O-D” air travel between New York and Boston. Early docu- mentation does not address the persistence of feeder flights. However, the aforementioned studies do not investigate the response of the airports and airlines in scheduling flights to achieve the congestion reduction benefits at specific airports. This is especially relevant to this study of the New York–Boston corridor since the delays in minutes per air passenger using the three major New York region airports [Newark (EWR), LaGuardia (LGA), and Kennedy (JFK)] are the highest in the country, with delays at these airports rippling through the entire national aviation system. However, the Next Generation rail analysis is complemented by the results of the PANYNJ/ RPA study described in the paragraphs herein. This lack of trip end intercity travel data by all modes handicaps any attempt to estimate the relative advantage of city center rail service over air service since air airports tend not be located in city centers per se. A series of additional data shortcomings were cited by the RPA study authors: • A depository of intercity travel data by all modes is needed; such a depository could be accessed by analysts performing studies like the RPA study. • The intercity travel data for each mode should include trav- elers by trip purpose, group size, household income, and other socioeconomic characteristics that influence travel mode choice and trip making. • The authors of the RPA study desired better access to a wide variety of data collected, or funded, by Amtrak, including station-to-station flows and general data about hypothe- sized highway vehicle flows, a major weakness in American multi-state data. • The authors desired data in which both the rail destination and ultimate destination were known, together with similar data for air trips; but no one believed data were collected in this manner. In general, the RPA study authors wanted better data. The authors also commented on models for forecasting rail rider- ship and revenue, especially the interaction of rail improve- ments with air. • They argued that there should not be a single model required by all federal agencies or other governmental body. This is because conditions between studies vary, both the study context and the study goals, to say nothing about time and budget restrictions; • They suggested that a peer review process is needed to review the models and provide reasonableness tests; and • They took the position that the federal authorities should provide modeling guidance and standard setting, but they should not impose models. Amtrak’s Next Generation High-speed Rail Proposal In September 2010, Amtrak issued “A Vision for High-Speed Rail in the Northeast Corridor.” As described in Amtrak’s 2012 Business Plan, The “Vision” outlined a conceptual framework and provided an initial review of the feasibility of improving the existing NEC alignment to handle growth in regional, commuter and freight services, while simultaneously planning and building a new, dedicated, two-track, high-speed rail alignment between Boston Figure 5-10. Sources of new ridership in Amtrak’s next generation plan. Source: Amtrak.

74 In this approach, this type of sensitivity analysis would occur late in the analysis process and could be applied to a wide variety of responses by the airlines, with “each of the scenarios modeled to determine their ridership and revenue impact.” Mode Choice Modeling to Forecast the Diversion of Intercity Trips from Air to HSR Much has been written about mode choice models over the last 50 years, and this brief review is certainly not intended to duplicate the many volumes and articles on the subject. For example, the classic text book on the subject is Discrete Choice Analysis (Ben-Akiva and Lerman 1985). Until the release of HSIPR Best Practices: Ridership and Revenue Forecasting there had been no current text available that comprehensively covers forecasting demand for HSR, much less remedies the need to be identified herein for some guidance on this subject. This book has complete chapters on Binary and Multinomial Choice models, which are the two main modeling types briefly referred to here. A thorough international review of methods for forecasting the competition between air and rail was under- taken at the University of Leeds for the European Organization for the Safety of Air Navigation, EUROCONTROL (Wardman et al. 2002). Travel choice models predict the travel behavioral response of people to the transportation choices confronting them. The best mode choice models aspire to be “behavioral models,” meaning they include all the important variables that affect modal choice. These variables include all the important characteristics of the transportation system affecting modal choices, and all the important influencing socioeconomic char- acteristics of the trip makers themselves. Consistent with the analysis presented in the OIG Best Practices reports, three forms of mode choice models are reviewed here: diversion choice, multinomial choice, and nested choice. Binary Diversion and Multinomial Choice Models. Two types of models are commonly used to forecast HSR ridership and revenue. Binary or diversion models simply forecast the diversion of travelers from one mode (e.g., air) to another (e.g., HSR). These diversion models have a binary form, as shown in Figure 5-11. The Analyses of the Office of the Inspector General The OIG Report on the Completion of the NEC The NEC has been analyzed using a variety of methods and tools. Substantial work to evaluate the impact of NEC rail travel time improvements on rail and air travel in the NEC was undertaken by the Office of The Inspector General (OIG) of DOT with the assistance of Charles River Associates (OIG 2008). Presently the Boston–New York travel time is 3 hours 20 minutes at best, with the current schedule showing 3 hours 26 minutes. The current Acela schedule shows New York– Washington running times of 2 hours 45 minutes or longer. The OIG report focused on two scenarios for HSR. The first scenario achieves travel times initially envisioned in the 1976 enabling legislation (i.e., 3-hour service between Boston and New York and 2.5-hour service between New York and Washington). Scenario two estimated the impact of achiev- ing travel times that are .5 hours shorter on both ends. The results were 10.6 % and 20.3% reductions in NEC air travel for scenarios one and two, respectively. Air travel between Boston and New York was reduced by 11% and 21% for the two scenarios, respectively (OIG 2008). The OIG Best Practices Report One response to the desire for more modeling guidance noted by practitioners has been the release in 2012 of the document High-Speed Intercity Passenger Rail Best Prac- tices: Ridership and Revenue Forecasting, prepared by Steer Davies Gleave for the OIG at U.S. DOT. The document is one of five separate reports prepared for OIG on “best practice” relevant to the examination of High-Speed Intercity Passen- ger Rail (HSIPR) projects in the United States. According to the OIG, the report was positively received by the Federal Railroad Administration (FRA) and will influence their man- agement of the analysis process for candidate investments. Their report is largely consistent with the assessment of the status quo reported earlier in this Chapter: The OIG report addresses the problem of dealing with airlines’ service altera- tion uncertainty in reaction to change in competition. The OIG study took the position that: [T]he likely competitive response of common carrier service providers to the introduction of HSR service is impossible to predict with any certainty in advance. Absent information to the contrary, it is generally assumed that future common carrier LOS characteristics will mirror base year conditions, so the precision of future year level of service characteristics is of second order importance. Accordingly, the impacts of potential changes in future year service frequencies, fare levels or other LOS variable may be most appropriately examined in the context of a sensitivity analysis (SDG 2012). Figure 5-11. Diversion choice model example.

75 Figure 5-12. Multinomial choice model example. Figure 5-13. Nested choice model example. also by the intercity mode they currently use. This approach has the great advantage of simplicity and transparency. For example, a research study (Brand 1996)(Table 5-4) shows the values of travel time savings by HSR for intercity air travel- ers for trips under 500 miles in the United States estimated in studies using these models in several major corridors in the United States, reflecting the conditions operant in those corridors. Modal Level of Service (LOS) characteristics Influencing Air Rail Mode Choice In these models, the most important LOS characteristics influencing the diversion of air trips to a potential HSR service include: • Line haul travel times; • Access/egress times and costs; Market Segment Corridor Incremental Value of Time Mean Gross Annual HH Income Percent (%) of “Wage Rate” Air Business Texas-Southwest Airlines $35.65 $83,502 85 Texas-Other Airlines $55.29 $86,370 128 California $39.07 $77,438 101 Pooled Corridors $50.77 $77,000 132 Air Texas-Southwest Airlines $24.38 $74,644 65 Nonbusiness Texas-Other Airlines $27.10 $62,191 87 Pooled Corridors $27.29 $68,500 80 Auto Business Texas $28.95 $52,825 110 California $23.38 $59,304 79 Pooled Corridors $26.19 $55,000 95 Auto Texas $13.59 $42,632 64 Nonbusiness California $12.45 $54,278 46 Source: Brand 1996. Table 5-4. Implied values of time savings for air and private vehicle business and nonbusiness travelers derived from HSR mode choice models in several intercity corridors (1992). The second type of mode choice model is the multinomial model. This model includes the choice between all available modes in a single mathematical formulation. The general form of a multinomial model is shown in Figure 5-12. The third type of mode is the nested choice model. This model restricts the number of choices in any one nest to a sub- set of the available alternatives, as shown in the Figure 5-13. Different modeling approaches have been applied in the studies described in this Chapter. Both work done directly for Amtrak and work now being undertaken for the FRA’s NEC Future project utilize the nested choice model approach. Work done for the OIG has used the diversion choice approach. Because this Chapter is most concerned with the potential for substituting rail trips with air trips, it begins with a discussion of a simple binary or diversion choice model approach. This approach involves developing separate single mode choice models of the attractiveness of HSR with air. Separate models would be developed for separate trip purposes (e.g., business and nonbusiness). Consequently, intercity travelers’ prefer- ences for a new mode can vary not only by trip purpose but

76 gage handling, safety, and security. Some of these attributes are included, at least in part, in the more straightforward variables defined herein (e.g., reliability in on-time perfor- mance and security delays can be measured and included in access times). However, the most common way of measuring their influ- ence is in the modal constant. The modal constant, like all constants in such statistical models, measures the effect of the “unobserved variables” in the model or equation. In this case, the unobserved variables are simply the LOS attributes not included as separate variables in the mode choice model (i.e., in the list of LOS variables given herein). Modal con- stants in binary diversion models measure the preference for one mode over the other ceteris paribus, everything else being equal (i.e., all the other variables in the model, including the time and cost variables having the exact same values). The constants are usefully described as the equivalent fare in dol- lars. This results in the traveler of a certain type and traveling for a certain trip purpose indifferent between the two modes. An example would be a value of $5 in favor of HSR, mean- ing that an airfare reduction of $5 would make the traveler indifferent between air and HSR if the values of all the LOS variables explicitly included in the model were the same. In fact, $5 in favor of HSR is a common default value of the air modal constant in NEC studies, but it must be stressed that the value is completely dependent on the LOS variables explicitly measured and included in the diversion model. For example, airline security delays are explicitly included in access time in these models. Attempts to “unbundle the modal constant” using very sophisticated stated preference surveys have been made in the United States. A more subjective way of quantifying the modal constant is included in the SDG Report on competi- tion and complementarity. In this case, the model developers assigned “factor scores” by city pair, separately to rail and air, for four “other factors” (Table 5-5). These other factors were airport links (for connecting passengers), price vari- ability (availability of a low-priced air carrier), reliability, and service quality. Air was superior on the first two factors, while rail did better on the last two. These factor scores were included in the model in place of a statistically estimated modal constant. Data Collection and Modeling in the Northeast Corridor by Amtrak and FRA The Tier 1 Programmatic Environmental Impact State- ment process for the future of rail in the Northeast has been begun by the FRA. That process, known as the NEC Future project, will include a significant refinement of the mod- els previously used by Amtrak in its Next Generation rail studies. While Amtrak has historically considered its ongo- • Fares; • Frequency; • On-time performance; and • Number of transfers. Most of these variables are simple to define and measure. The costs include all the out-of-pocket costs of travel, includ- ing operating and parking costs per passenger for auto access/ egress to airports and HSR stations. For business trips, peak period access/egress times from the local MPO coded net- works are normally used; for nonbusiness trips, off-peak times are used. For frequency, the average headways during peak and off-peak periods are used. On-time performance can be more complicated. U.S. DOT compiles statistics on the percent- age of flights arriving within 15 minutes of their scheduled arrival time, and the percentage of flights cancelled. Amtrak is the only source of on-time performance data for rail in the United States, and these data can be hard to obtain. The main challenge in specifying these service variables, however, is likely to be the on-time performance of the proposed HSR improvement. Presuming that the excellent on-time perfor- mance records of HSR in Europe and Japan translate easily to applications in the United States may be too simple and incorrect. Due diligence is required. Determining the influence of each of these variables on diverting air trips to HSR is, of course, the key element in modeling HSR ridership and revenue. The Brand (1996) article provides the values for the first four of these variables for a series of forecasting studies in the United States. Similar infor- mation for a series of studies in Europe is found in “Air and Rail Competition and Complementarity” (SDG 2006), particularly in Chapter 4, “The Market Share Model.” The SDG model created for their European Union study of air/rail competition had four kinds of input. They describe the four inputs as, • Schedule related factors: These include journey time, check-in time, time required to leave the rail station or air- port, and frequency. • Price: We have estimated average one-way ticket price for each mode on each route, based on a large sample of fares collected from the operators. • Access time and cost: The average time required to access the terminals and the cost of journeys to/from the terminals. • Other factors: We have also taken into account other fac- tors such as reliability, service quality and connections to airports (SDG 2006). Other LOS attributes of air and HSR influence the diver- sion of air trips to HSR. These so-called “non-traditional” LOS variables are much harder to define and measure. They include descriptors like comfort, convenience, reliability, bag-

77 tion included in available descriptions of the AECOM model, which will be very helpful in future studies of the substitution of HSR for air trips between city pairs in the NEC. In 2014 the FRA’s NEC Future Project will complete a new survey to be used to develop a new model, whose struc- tures are being developed as diagrammed in Figure 5-14. The ing modeling process to be partially proprietary in nature, some details have appeared in the professional literature (ConnDot 2010). As described by AECOM, the 2007 sur- veying effort included new travel surveys in Maryland, New Jersey, and Massachusetts; these were supplemented by a telephone survey of Amtrak customers and a telephone sur- vey of random NEC travelers (AECOM 2007). The highway surveys generated over 4,600 completed responses, with 5,000 phone surveys from Amtrak users and 10,000 from the random travelers. They report that the highway survey was “used to adjust data from the random traveler telephone survey to account for under-reporting of auto trips.” The survey of Amtrak users was used to increase the understand- ing of trip purpose (not available from ticket sampling) and to model “stated travel intentions.” These factors are used to predict overall rail demand rather than diversion from air. • Level of Service; • Travel time (line haul and access); • Departure frequency and time slot; • On-time performance (OTP); and • Travel cost/income. The document is also interesting in that it places predicted rail volumes in the context of highway vehicle volumes, which are difficult to forecast; it also places predicted rail volumes in the context of bus ridership forecasts, which are nearly impossible to forecast given the lack of agreement about base case volumes. In any event, there is a great deal of informa- Source: Reproduced from SDG 2006. N.B., Original table notes that “Ten is Best” Table 5.5. “Other Factors” influencing choice of mode. Figure 5-14. Structures being developed in the FRA NEC Futures model. Source: AECOM 2014.

78 Recap: Responses to Key Questions Chapter 5 has reviewed several aspects of the way in which air and rail compete in the Northeast Corridor of the United States. In interviews with leaders in this area, three common questions were explored; those questions, with their evident answers, are presented herein: 1. Did local planners and managers have access to data that would allow a quick summary analysis of descriptions of performance for the long-distance trips in the corridor? Interviews with key players suggest the answer is no. The NEC is the most well-studied intercity rail corridor in the United States, as noted at the beginning of this Chapter. Even project is designed to produce between 10,000 and 15,000 completed household surveys in the study area, defined in Figure 5-15. This will build directly upon an existing data base from 10,000 non-customers, and 5,000 existing Amtrak customers (NEC Future 2012). The study managers believe this will provide a solid data base to describe current travel by trip purpose, by mode and geography and trip length, pro- viding an adequate number of survey records for each com- bination of mode and purpose, including intercity bus. At the same time, working in cooperation with the Northeast Corridor Commission, a highway intercept study was under- taken in 2013 to better understand travel flows in the region, using a highly innovative program to survey drivers who use the region’s E-ZPass Program. Figure 5-15. Data collection area for FRA’s NEC Future modeling process, 2014. Source: FRA.

79 Analysis Framework, which occurred after the interviews for this project, has been undertaken to deal with these histori- cally recurring issues and concerns. Bibliography AECOM. 2007 (May 25). “Summary of New Amtrak Northeast Corridor (NEC) Travel Demand Forecasting Model Data.” AECOM. 2014, Personal Communication from AECOM. Amtrak. 2009 and 2014, (Personal communication from Amtrak). Amtrak. 2011 (December).High Speed Rail and the Future of the North- east Corridor, Presentation by Drew Galloway to the Governor’s Transportation Conference, Norfolk, VA. Amtrak. 2012 (January). Fiscal Year 2012 Budget and Comprehensive Business Plan. Ben-Akiva, M. and S. Lerman. 1985. Discrete Choice Analysis. MIT Press, Cambridge, MA. Borenstein, S. and J. Netz. 1999. “Why Do All the Flights Leave at 8 am?: Competition and Departure-time Differentiation in Airline Mar- kets.” International Journal of Industrial Organization, Vol. 17, No. 5, pp. 611–640. Brand, D. (Charles River Associates). 1996 (June 1). “The Values of Time Savings for Intercity Air and Auto Travelers for Trips under 500 Miles in the US.” Prepared for the US DOT, Office of the Secretary. ConnDot. 2010. Appendix H, “Summary of Amtrak Travel Demand Forecasting Models.” HSR Grant Application. This was located at http://www.nhhsrail.com/GrantApplication.aspx. NEC Future. 2012. “Household Survey Design” Presented to: Ridership & Revenue TWG, October 10, 2012. Office of the Inspector General, U.S. D.O.T., for the Federal Railroad Administration. 2008 (June 26). “Analysis of the Benefits of High- Speed Rail on the Northeast Corridor.” Report CC-2008-091. Regional Plan Association. 2011 (January). “Upgrading to World Class; the Future of the New York Region’s Airports.” Ryerson, M. 2010. “Optimal Intercity Transportation Services with Heterogeneous Demand and Variable Fuel Price.” University of California Dissertation Series. Available at: http://www. uctc.net/resdearch/UCTC-DISS-2010-07.pdf. Ryerson, M. and M. Hansen. 2011 (June). “Fuel Consumption and Oper- ational Performance.” In Proceedings of the Ninth Annual FAA/ EUROCONTROL Air Traffic Management Research and Develop- ment Seminar, Berlin, Germany. Steer Davies Gleave. 2006 (August). “Air and Rail Competition and Com- plementarity” Prepared for the European Commission DG TREN. London, U.K. Steer Davies Gleave. 2012 (April). “HSIPR Best Practices: Ridership and Revenue Forecasting.” Prepared for the Office of Inspector General, US DOT, under subcontract to Charles River Associates. Virgin Trains—The Media Room. 2011 (March 28). “Former Fliers Get on Track.” Press release available at http://www.mediaroom. virgintrains.co.uk. Wardman, M., A. Bristow, J. Toner, and G. Tweddle. 2002 (May). “Review of Research Relevant to Rail Competition for Short Haul Air Routes.” European Organisation for the Safety of Air Navigation EUROCONTROL, Brussels, Belgium. Wei, W. and M. Hansen. 2005. “Impact of Aircraft Size and Seat Avail- ability on Airlines’ Demand and Market Share in Duopoly Markets.” Transportation Research Part E, Vol. 41, No. 4, pp. 315–327. the comprehensive RPA study of the substitution of HSR for rail trips lacked access to the detailed trip end (geographic) distribution data for air and rail. The local practitioners felt that Amtrak’s (then) traditional approach to the proprietary nature of their demand data added to the problems in under- taking a “quick summary analysis” of multimodal strategy options. The contributions of Amtrak to FHWA’s Travel Analy- sis Framework (which occurred after the Research Team’s project interviews) will go a long way in improving the quality of data available to groups such as the RPA. In interviews with the RPA and the Port Authority, par- ticipants are concerned with the poor availability of regional intercity data and the models used to help interpret it. RPA representatives were particularly concerned with the lack of good data on intercity automobile flows. 2. Did the airport strategists feel they have adequate informa- tion about the possible use of rail in various ways to help with the major capacity issues looming over major airports, such as New York City area airports? No. The RPA planners made a heroic effort to model the diversion of air trips to HSR, but they had the serious data problems detailed herein. The airport managers interviewed had, in fact, just financed the RPA study, so they were in a position to understand the strengths and weaknesses of the data and tools available. The Research Team’s interview with NJ Transit revealed a concern about the quality of the modeling done in some cases. Concern was raised that many participants in the planning process do not understand the need to base the stated pref- erence modeling process (used in analyzing access to air- ports and diversions from air) on the experience with these particular modal circumstances, and not simply lifted from the modeling process used in the metropolitan areas. Like colleagues at New Jersey Transit (NJ Transit), the PANYNJ staff was directly aware of possible misuses of the modeling process when traditional metropolitan planning models are applied to such subtle issues as rail-as-feeder or sub-mode of access to airports in general. 3. Would they benefit from the creation of an accepted set of longer distance travel descriptions, including data organized at a sketch planning level for all modes? Again, those interviewed felt a certain amount of frustra- tion with the lack of the most basic multi-state travel informa- tion, such as automobile trip tables, etc. There was a feeling that the government might be able to help, along with a healthy skepticism of what might happen with too much fed- eral intervention. The development by FHWA of the Travel