Intercity Passenger Rail in the Context of Dynamic Travel Markets (2016)

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.

96 9.1 Introduction This chapter gives a brief summary of new research done by NCRRP to better understand how the competition from air may be different from the competition from other modes. The chapter explores a newly documented interaction between improvement of rail travel time on the one hand and variation in air price on the other hand. The new implications of this interaction are summarized for the policy maker. Passenger rail systems interact with aviation systems in several ways. This chapter will help the transportation community to understand the manner in which rail makes a contribution to the intermodal system by diverting traffic from congested airports. In this diversion process, the full system may become more efficient as airports become more focused on critical long-distance trip making; rail can efficiently transport people in shorter-distance contexts. For this exploration of rail in a competitive mode, the data, tools, and methods need to be in place to support the analysis of multimodal and intermodal systems and strategies—with particular attention to the day-to- day competition with air services. The NCRRP 03-02 research team examined the market-based performance of these competitive services, commenced the examination of the adequacy of the analytical tools available, and developed new tools in response to the gaps revealed. Several models exist concerning forecast travel behavior involving mode choice decisions between rail and air. In some cases, those models are considered proprietary, such as those used by Amtrak for detailed market research in its competitive environment. On the West Coast, an elaborate model of rail demand has been developed in order to meet the very exacting require- ments for the California High-Speed Rail Authority’s project development and environmental documentation. While that model is extremely thorough in its approach, it was never designed to serve as an analysis tool for quick and cost-effective analyses of public policy options. Thus, a gap has existed in the ability of researchers and practitioners to model the interaction of air and rail. The NCRRP 03-02 research team continued the development of the ACRP’s air/rail diversion model to fill this gap. The continued work on this air/rail diversion model (Figure 51) was envi- sioned to produce a strategically designed analysis tool to address specific questions about consumer preferences and tradeoffs in response to conventional rail/high-speed rail (HSR) and air service modifications. The model was designed as an efficient, quick-response tool useful for realistic planning-level scenario analysis and it includes only tradeoffs between air service and both conven- tional rail and HSR, omitting tradeoffs with both auto travel and intercity bus. This NCRRP project has developed refinements to the model to help improve understanding of the relationship among several key factors in the explanation of the choice of mode between air and rail, which, when seen in a scenario of improved rail services, is associated with the diversion from air to rail. The model is best used as a learning tool at an early concept stage of planning to determine basic impacts C H A P T E R 9 Competition Between Rail and Air

Competition Between Rail and Air 97 of a series of actions/investments so that a decision can be made on whether to engage in more detailed analysis. 9.2 Air/Rail Diversion Model The air/rail diversion model allows policy analysts to study possible air and rail diversion for two study years (2008 and 2040) by altering seven global (system-wide) variables for rail/HSR and air service levels. These variables are rail in-vehicle time (IVT), air IVT, auto IVT for auto access to airport or rail station, rail fare, air fare, amount of rail service (frequency), and amount of air service (frequency). The model users can alter the scales or levels of service for each of these variables to analyze implications under different scenarios. The model has been applied in two North American study areas—the NEC (“East Coast”) and California (“West Coast”)—where there is considerable availability of both air and rail modes, meaning that many long-distance travelers have a reasonable choice between the modes. There are largely five categories of data input for the model: socioeconomic data, rail station access data, airport access data, rail service description data, and air service description data. The model takes the input variable parameters and implements an eight-step model procedure to produce the output of rail and air trips by origin and destination. 9.2.1 Air/Rail Diversion Model Sensitivity Analysis In order to understand how the global variables affect the outcome (number of rail trips) in the model, the research team conducted a sensitivity analysis. The basic idea behind sensitivity analysis is to measure how variations in the output can be attributed to variations in the input. The goal of sensitivity analysis in the context of the air/rail diversion model is to measure relative impact or Figure 51. Model of air/rail diversion, expanded for NCRRP.

98 Intercity Passenger Rail in the Context of Dynamic Travel Markets importance of each of the seven global variables, i.e., rail IVT, air IVT, auto IVT, rail fare, air fare, rail service (frequency), and air service (frequency). To complete the sensitivity analysis for the NCRRP 03-02 project, the research team chose to use partial inclination coefficients (PICs) because they allow any number of simulations to be run. PICs estimate a model response in relation with all model input variables (Chalom and de Prado 2015). It is important to keep in mind that the resulting coefficients only reveal relative impact among input variables rather than their absolute impact on the outcome. In other words, these coefficients cannot be interpreted the same way, for example, the coefficients of linear regression models would be interpreted. In addition, there is no measure of goodness of fit for PICs that linear regression models otherwise produce to indicate how close to the “truth” the models are. Instead, one can infer from PICs which of the input variables affects the outcome of the model the most relative to one another. For example, model inputs A, B, and C with PICs of 2, -10, and 5, respec- tively, would indicate the following: • The input variable B impacts the outcome more strongly than A or C and has a negative rela- tionship with the outcome variable. • The input variable C has the strongest positive impact on the outcome, yet this impact is, in an absolute value sense, 50% as strong as the impact of B. To implement PIC analysis, the research team first ran 200 simulations of East Coast air/rail diversion using random samples for each of the seven input variables. The input variables are in the units of percentage scales with their default values set to 1, indicating 100% of the existing values. For example, the default scale of the input variable air fare (scale = 1) means existing aver- age air fares. As the scale moves up to 1.2, this indicates a 20% increase from the existing values. If the current average air fare for New York to Boston were $237, an input variable scale of 1.2 would mean a 20% fare increase, or $47 ($237 × 0.02, rounded), for a total of $284. Likewise, the default scale for rail IVT (scale = 1) means the current average in-vehicle times for rail. When a user sets the scale to 0.7, this indicates 30% reduction in IVTs or inversely a 30% increase in rail travel speeds. For the purpose of running simulations, all seven input variable scales were varied simultaneously by using values from random samples. The PICs are shown in Table 33 in the descending order of relative magnitude. The cor- responding confidence intervals and standard errors were estimated by using 100 bootstrap replicates. Bootstrap produces measures of accuracy to sample estimates by performing ran- dom sampling with replacement within the existing sample. The bootstrapped 95% confidence intervals indicate that a PIC value will fall within the confidence interval in 95% of simulations. For example, variable air fare has a confidence interval of 3.74 and 4.25 and its PIC falls squarely in the interval. Therefore, its PIC of 3.94 can be trusted with a high level of confidence. On the other hand, variable auto IVT has a confidence interval of -0.19 and 0.27. A confidence interval Input Variable PIC (std. error) Confidence Interval (CI) Lower Bound Upper Bound Air Fare 3.94 (0.12) 3.74 4.25 Rail IVT 3.26 (0.11) 3.44 2.98 Rail Fare 1.77 (0.10) 1.94 1.55 Air IVT 1.03 (0.11) 0.81 1.23 Rail Service 0.94 (0.11) 0.70 1.18 Air Service 0.91 (0.12) 1.11 0.64 Auto IVT 0.03 (0.12) 0.19 0.27 Table 33. Partial inclination coefficients.

Competition Between Rail and Air 99 that contains 0 brings into question the validity of the PIC value because the PIC could eas- ily be a negative number as well as a positive. Since the confidence interval of -0.19 and 0.27 means that the true value lies within the range, the direction of impact is indeterminable. As mentioned before, the PIC values should be used only to analyze the relative impact of the input variables. 9.2.2 Sensitivity Analysis Results Based on the results shown in Table 33, the following factors are found to be positively or negatively correlated with the number of rail trips. Factors That Are Positively Related to the Number of Rail Trips • Air fare seems to have the most impact on the number of rail trips. The impact of air fare on rail trips is positive, such that an increase in air fare will increase the number of rail trips by diverting passengers from air to rail. Air fare is explored more in Section 9.2.3. • Air IVT has the next biggest positive impact on the number of rail trips. Its positive PIC indi- cates that as the IVT increases for air travel, more passengers will divert from air to rail. How- ever, among the top four variables, which consists primarily of fare and travel time, air IVT has the least impact on the outcome. Specifically, rail IVT has far bigger impact on the number of rail trips than air IVT. In other words, improving rail speeds will divert plenty more passengers from air to rail than faster air speeds will divert passengers from rail to air. Considering that air travel is considerably faster than rail at the current level, this finding may imply that the speed improvements after a certain threshold have diminishing returns. • Rail service also has a positive PIC with the least relative impact on the outcome among the positively related variables. Increased rail service, which means more frequent rail service, will divert some passengers from air to rail but with minimal impact among these positively related variables. Unlike intracity transit in which frequency plays a big role in determining ridership, intercity travel via air and rail seems to compete more on fare and travel time rather than fre- quency. Within their own modes, different service providers do compete on frequency. For example, it is well acknowledged in the literature that airlines compete with one another on frequency of service to attract passengers. However, frequency seems to be less of a factor in diverting passengers between air and rail. Factors That Are Negatively Related to the Number of Rail Trips • Rail IVT is the variable with the most negative impact on the number of rail trips. Rail IVT inversely indicates rail travel speed. In other words, rail travel speed is highly positively related to the number of rail trips. As rail speeds improve, the rail IVT becomes smaller. • Rail fare has the third most overall impact on the outcome of the model after air fare and rail IVT and the second most negative impact after rail IVT. Including the fourth variable with most impact, air IVT, the top four input variables with the highest impact on the outcome of the number of rail trips are fare and travel time. • Air service has the least negative impact on the number of rail trips. Along with rail service, these service frequency variables have the least impact on the competition between air and rail. Factors That Have No Statistical Correlation with the Number of Rail Trips The input variable with the least impact overall on the number of rail trips is auto IVT. This seems intuitive because the auto access/egress time is relatively small compared to the long travel time for intercity travel. A small benefit gained from faster auto access time to rail stations, for example, is likely to be overshadowed by slow travel speeds on rail for much longer distances.

100 Intercity Passenger Rail in the Context of Dynamic Travel Markets Likewise, any loss of time due to traffic congestion on the way to airports is likely to be more than compensated by faster travel time on airplanes. Additionally, auto IVT is the only variable whose bootstrap confidence interval includes 0, which indicates unclear direction of impact. 9.2.3 Interpretation of the Variable Interactions The top four input variables in the descending order of impact are air fare, rail IVT, rail fare, and air IVT. These variables indicate that air and rail for intercity travel compete largely on these two dimensions of fare and travel time. Notably, the analysis implies that air and rail compete mainly on air fare and rail IVT; their PICs are relatively high at around 3 in absolute terms while the PICs for the other two variables, rail fare and air IVT, are around 1. This finding is intuitive. In relative terms, air travel is typically faster and more expensive than rail trips. Given the existing superiority of speed for air, the factor that causes the most diversion from air to rail is the comparative price of air service. At the current rail speeds, rail cannot compete with air directly on speed. Passengers instead make tradeoffs between speed and price and the analysis indicates that passengers respond to air fares the most in deciding to divert to rail from air. Rail speed improvements can change this dynamic. Rail speed improvements will shorten rail IVT and make rail travel highly competitive with air on speed. However, rail IVT interacts with air fare in complex ways. Figure 52 shows the relationship between air fare and number of rail trips categorized by rail IVT levels. Positive slopes of the graphs indicate the positive relationship between air fare and number of rail trips. Each of the four curves is for each category of rail IVT. The top graph shows the highest rail speed improvement (lowest rail IVT category) and the bottom graph shows the lowest level of rail speed (highest rail IVT category) where rail speeds have dropped below the current level. In general, the graphs show that, as rail IVT becomes smaller (i.e., rail speed improves), the number of rail trips increases for any given air fare. However, as air fare becomes too small or too large, the incremental gain in the number of rail trips for rail IVT categories becomes smaller. The tail ends of the four curves cluster closer together than the middle of the curves. The converging tails of the graphs indicate that the ability to divert passengers from air to rail by rail speed improvements depends largely on the level of competing air fare. In other words, the effect of air fare seems to become so significant at these extreme levels that the rate of diversion from air to rail due to rail IVT improvements becomes relatively smaller. This points to Figure 52. Interactions between air fare and rail in-vehicle time.

Competition Between Rail and Air 101 the dynamic relationship between fare and travel time since it shows that once speeds for air and rail have almost equalized, rail and air will compete on fare. After air fare and rail IVT, rail fare has the third largest impact on the number of rail trips. As mentioned previously, the relative magnitude of rail fare’s PIC is considerably smaller than air fare and rail IVT. Cheaper rail fares make rail more competitive as it offsets slower speeds of rail. However, the analysis indicates that air fares command more influence than rail fares probably because air speed benefits are significantly higher than what cheaper rail fares can overcome. Air IVT also has a relatively small PIC. The subsequent analysis of elasticity in the following sub- section shows that a 50% increase in air IVT, i.e., 50% reduction in air travel speed, only increases number of rail trips by 8% while holding all the other input variables at their default settings (scale = 1). Likewise, a 50% reduction in air IVT or 50% increase in air travel speed, with all other variables staying the same, reduces the number of rail trips (or increases air trips) by 8%. This may indicate that the current air travel speeds are so highly competitive that further air travel speed improvements or speed reductions will not divert passengers significantly. 9.2.4 Elasticity Another way to think about these input variables relative to one another is to investigate their elasticities. Elasticity measures how responsive a variable is to a change in another variable. Applied to the model, elasticity for each variable measures how the outcome of the number of rail trips responds to changes in each of the input variables while holding the rest of the input variables at their default setting (scale = 1). Since it is unreasonable to assume a linear relationship between input variables and the outcome, elasticity is estimated at two different change points: 50% reduc- tion (scale = 0.5) and 50% increase (scale = 1.5) of input variables. First, the model is run with one of the input variable scales changed to 0.5 while everything else is held at 1. This is then repeated for the same variable changed to 1.5. This is done for all of the seven input variables. The resulting outcome of numbers of rail trips is summarized in Table 34. Second, the percentage changes of the resulting outcomes from their default settings are calcu- lated. At the default setting where every input variable is set to their current existing level (scale = 1), the model estimates 9,584,950 rail trips. Elasticity is calculated by dividing the number of rail trips for 0.5 scale and 1.5 scale, respectively, for each variable by the default number of rail trips. Table 35 summarizes elasticity for each of the variables in the descending order of input elasticity. The order of input variables according to elasticity corresponds to the ordering of variables by their PICs. Air fare has the highest elasticity; 50% reduction in air fares while holding everything else constant reduces the number of rail trips by 30% and 50% increase in air fares likewise increases the number of rail trips by 20%. Rail IVT has the second highest elasticity in the opposite direction; Variable Number of Rail Trips Scale = 0.5 Scale = 1 Scale = 1.5 Air Fare 6,735,100 9,584,950 11,487,350 Rail IVT 11,899,600 9,584,950 7,795,250 Rail Fare 10,804,800 9,584,950 8,369,900 Air IVT 8,789,600 9,584,950 10,326,100 Rail Service 9,118,500 9,584,950 9,776,050 Air Service 10,032,100 9,584,950 9,429,300 Auto IVT 9,580,400 9,584,950 9,587,950 Table 34. Number of rail trips for elasticity estimation.

102 Intercity Passenger Rail in the Context of Dynamic Travel Markets 50% reduction in rail IVT increases the number of rail trips by 24% while 50% increase reduces it by 19%, everything else held constant. They are followed by rail fare with 13% and -13% elastic- ity for 0.5 scale and 1.5 scale, respectively. Elasticity for the remaining four input variables quickly drops below 10%. Auto IVT has the smallest elasticity of -0.05% and 0.03%; 50% reduction in auto in-vehicle time to access airports/rail stations results in 0.05% increase in the number of rail trips. Since auto IVT includes access time to both airports and rail stations, the direction of change is ambiguous. However, the negligibly small elasticity for auto IVT indicates auto access constitutes a very small part of the competitive decision between air and rail. 9.3 Case Study Example: New York City–Boston For a case study, the research team simulated the New York City–Boston intercity corridor to illustrate the results of the sensitivity analysis. In Figure 53, the dotted line represents the level of rail trips between New York City and Boston when only air fare is changed while all the other variables are held at the existing level (scale = 1). Point A is therefore the default level of rail trips when all the variables are held at the existing level including air fare. Using the average air fare from US DOT ticket data for 2013 (around $237), the model predicts about 423,000 rail trips between New York City and Boston. Assume that in the near future Amtrak improves its average Variable Change from Exisng Values (Scale = 1) Scale = 0.5 Scale = 1 Scale = 1.5 Air Fare 30.00% 0% 20.00% Rail IVT 24.00% 0% 19.00% Rail Fare 13.00% 0% 13.00% Air IVT 8.00% 0% 8.00% Rail Service 5.00% 0% 2.00% Air Service 5.00% 0% 2.00% Auto IVT 0.05% 0% 0.03% Table 35. Elasticity for input variables. Figure 53. Impact of air fare on rail trips (NYC–BOS).

Competition Between Rail and Air 103 train speed by about 25%. In other words, now the rail IVT would be about 25% less than the existing level (scale = 0.75). The model indicates that such reduction in rail IVT would move the number of rail trips from point A to B, about 550,000 rail trips including diversion from air. In this scenario, airlines can regain the diverted trips from rail by reducing air fare. By reducing air fares, the number of rail trips would move down along the graph to point C. The quick downward tail of the graph also indicates that as airlines keep reducing fares, the rate of diversion from rail to air increases. On the other end of the graph, there seems to be a saturation effect where increases in rail speeds result in smaller gains for each rail IVT scale category. In other words, as air fares increase to almost double their existing levels (scale = 2), the rail trips seem to hit a ceiling and flatten out around 800,000, indicating there is less and less to be gained from rail speed improvements when air fares become increasingly high to near prohibitive levels. One implication is that, in the event that future air fares increase to pre-deregulation levels (maybe due to fuel shortage), most passengers will have already diverted to rail and rail speed improvements will only add incrementally to the diversion. Likewise, if air fares substantially decline from the current levels (maybe due to new fuel technology), passengers will divert from rail to air, but rail can retain some of the potential diverted passengers by improving rail speeds to be more competitive. In this case, rail improvements will potentially bear future benefits if air fares subsequently increase diverting passengers from air to rail. In all, air fare and rail speeds seem to be major factors that influence the dynamic relationship between air and rail passengers. Rail IVT has a negative relationship with rail trips where increase in rail IVT (or decrease in rail speed) will decrease the number of rail trips, diverting passengers from rail to air. Both Figures 53 and 54 also show that as air fares become cheaper (air fare scale decreases), the overall number of rail trips decreases, as shown by the descending curves for each air fare scale category. The effect of decreased air fares can be counteracted by faster rail speeds. However, when air fares hit extremely low levels (see air fare category “Less than 0.5”), rail speed improvements will only recover a small portion of the diverted passengers. For the other three categories of air fare scale (0.5 to 2), rail speed improvements can divert a significant number of passengers from air to rail. Figure 54 sug- gests that, after a certain point in rail speed improvements (around rail IVT = 0.5), there appears to be a ceiling for number of rail trips, as is shown by the converging tails on the left end of the curves. Figure 54. Impact of rail in-vehicle time on rail trips (NYC–BOS).

104 Intercity Passenger Rail in the Context of Dynamic Travel Markets The only curve that does not converge is air fare scale category “Less than 0.5,” again highlighting the drastic effect that extremely low air fares have on passenger diversion. 9.4 Conclusion In this analysis, the research team employed the air/rail diversion model to study the inter- action between air and rail/HSR service levels and how these service-level variables affect rail demand. The seven service variables included were rail IVT, air IVT, auto IVT for auto access to airport or rail station, rail fare, air fare, amount of rail service (frequency), and amount of air service (frequency). The variables affecting rail demand the most significantly were found to be air fare and rail IVT. The impact of air fare on rail trips is positive (such that higher air fares encourage air travelers to divert to rail), while the impact of rail IVT is negative. Compared with air fare and rail IVT, the remaining variables either had a small or nonstatistically significant effect on rail demand. The implication of the sensitivity analysis result is that, when considering rail demand, there is significant interaction between rail speeds (or IVT) and air fare. A reduction in air fare can divert passengers from rail to air; however, the effect of decreased air fares can be counteracted by faster rail speeds. Yet the results of the analysis suggest that there are diminishing returns to rail speed improvements. Rail speed improvements in general divert passengers from air to rail as rail speeds become more competitive with air speeds. Yet competing air fare can significantly alter the level of diverted passengers. Specifically, low air fares, around 50% of existing average air fares, would significantly reduce the number of diverted passengers from air to rail. In addition, as rail speeds improve, there seems to be a ceiling where no further speed improvements will divert a substantial number of passengers from air to rail. Although the exact point of diminishing returns will be different depending on the routes, the result suggests that improving rail speed has a small effect on rail ridership when air fare is relatively low. When considering rail speeds, there may be a point after which further investment in rail speed improvements will start to divert fewer and fewer passengers from air to rail for each dollar spent. In short, air fares, for which relative rail demand is highly responsive, may be more impor- tant to rail ridership than rail travel time. Overall, the results indicate that any initiative focused on increasing rail ridership must consider that the responses from competing modes of travel, especially air, can change the policy outcome. The findings underscore the criticality of analytical modeling tools through which the tradeoffs between modal service variables can be analyzed and for these modeling tools to be cross-modal. As the largest impact on rail ridership is a service-level variable pertaining to air service, it is clear that intercity planning must take place in a cross-modal context with high-fidelity analysis tools that provide actionable and policy-relevant information.