Bibliography and Technical Appendices to 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.

Technical Appendix: The Scenario Analysis Tool 97 TECHNICAL APPENDIX: MODEL APPLICATION FOR SCENARIO ANALYSIS This technical appendix to NCRRP Report 4 now presents the full description of the scenario testing tool, which takes the form of a Workbook of spreadsheets in Microsoft® Excel format. Chapter Six presented a shortened, edited version of this full description of the scenario testing tool. 1. Introduction The mode choice models with latent attitudinal variables, described in Technical Appendix: Documentation for the Structural Equation Models, have been implemented in an Excel workbook in order to provide results for user-defined scenario changes. The implementation uses the sample enumeration method. Under this method, the appropriate model parameters are applied for each member of a sample of individual travelers, and the predicted mode shares are expanded and added across the sample to arrive at a total predicted number of trips made using each mode alternative for each travel segment (trip purpose). The resulting predictions can then be compared against a base scenario to determine the change in mode usage that is expected due to the change in the scenario inputs. 2. Excel workbook contents The Excel workbook will be accessible to the public on the TRB website in June 2016 (by searching for “NCRRP Web-Only Document 2”). It contains the following worksheets: Scenario: This is the only sheet that most users interact with. It has cells to define scenario changes, and tables to view the scenario results. This sheet is described in much more detail later in this Appendix. TripData: This sheet contains the reported data for the 5,626 survey respondents for this project who completed the mode choice SP exercise. The data fields in the worksheet for each respondent are: - Hhsize: number of persons in the respondent’s household - Hhadults: number of adults age 18+ in the household - Hhkids: number of children under age 18 in the household - Hhworkers: number of employed persons in the household - Hhlic: number of licensed drivers in the household - Hhveh: number of vehicles owned by the household - Income: household income category - Age: respondent age group - Gender: respondent gender - Education: highest education level completed - Employ: current employment status - ODpair: origin-destination city pair for respondents actual reference trip - Purpose: main destination purpose for actual reference trip - Triplength: nights away from home category for actual reference trip

Technical Appendix: The Scenario Analysis Tool 98 - Psize: travel party size for actual reference trip, including the respondent - Mode: the mode used for the actual reference trip (car, bus, rail or air) All other variables on the sheet are variables used directly in the models that are derived from the variables above. (For example, a “female” 0/1 variable derived from the Gender variable) Los_data_base: For each of the 5,626 respondents, this page has the average mode time and cost variable levels across the SP scenarios seen in the survey. The data fields for each respondent are: - Carparkandrent: Car parking and rental cost, if applicable, in dollars - cargascost: Car fuel cost, in dollars - carpartycost: Car total cost for the travel party, in dollars - busaccdur: Bus access time to the station, in minutes - busdur: Bus in-vehicle travel time, in minutes - busegrdur: Bus egress time to the destination, in minutes - busppcost: Bus fare per person, in dollars - railaccdur: Rail access time to the station, in minutes - raildur: Rail in-vehicle travel time, in minutes - railegrdur: Rail egress time to the destination, in minutes - railppcost: Rail fare per person, in dollars - airaccdur: Air access time to the airport, in minutes - airdur: Air in-vehicle travel time, in minutes - airegrdur: Air egress time to the destination, in minutes - airppcost: Air fare per person, in dollars - airfreqserv: Air origin-destination frequency of service, in departures per day - railfreqserv: Rail origin-destination frequency of service, in departures per day - busfreqserv: Bus origin-destination frequency of service, in departures per day - cbc_carmiles: The distance by car from the origin to the destination, in miles Because the SP experiment levels were customized based on the respondents’ reported trip origins and destinations, they provide a realistic distribution near current service levels. In the future it would be possible to replace the OD-specific levels for fares, in-vehicle times, and service frequencies with values representing specific future operating scenarios. (The access and egress times would be more difficult to specify precisely, as they depend on the exact trip end locations.) However, the focus of the scenarios is on indicative results rather than precise forecasts, so using exact times and costs in the scenarios is not seen as a high priority. Randuniform: This worksheet has 10 randomly generated uniformly distributed numbers in the range 0.0 to 1.0 for each person in the sample. The numbers were generated using Excel functions. Rand_normal: This worksheet has 10 randomly generated normally distributed numbers for each person in the sample, generated with mean 0 and standard deviation of 1.0. The numbers were generated using Excel functions

Technical Appendix: The Scenario Analysis Tool 99 Coeffs: This worksheet has all model coefficients for each person in the sample. These were copied from the “original coefficient table” sheet, using the coefficients corresponding to each person’s reference trip purpose. Since the coefficients do not change across scenarios, they are pre-copied rather than specified using look-up functions in order to speed up the scenario calculations. Los_data_scenario: This sheet calculates the scenario-specific time and cost values, based on the input data in the Los_base_data sheet and any scenario user inputs that change travel times or costs. Attitudes: This sheet contains the calculations that determine the 4 latent attitudinal variables for each person in the sample, based on the input data sheets and the user’s scenario input changes. Mode choice: This is the most calculation-intensive sheet in the workbook. It contains the calculation of the utility functions for the 4 travel modes, and the calculation of the resulting logit model probabilities. It then converts those into a numbers of trips for each mode. It also uses scenario inputs for demographic changes to calculate adjusted expansion factors.

Technical Appendix: The Scenario Analysis Tool 100 3. Sample expansion and initial model calibration Because we are using only a partial sample of trips in the Northeast and Cascade Corridors, it was useful to first check the numbers of trips by mode and corridor against similar numbers from other corridors. Two sources were used: the Northeast Corridor Intercity Travel Summary Report (RSG 2015a), which has estimates of annual trips by mode and city pair in the NE Corridor; and the output from the rJourney national long-distance passenger travel demand model (RSG 2015b), which has synthetic estimates of annual trips by mode, trip purpose and city pair in both the NE and Cascade Corridors. Because these two sources use different definitions of geographic areas, and because the SP survey used fairly vague city area definitions, the numbers are not exactly comparable between any of the estimates. The purpose of the scenarios is not to provide precise forecasts, but to begin with a reasonable representation of the current traveling population, so an approximate overall sample expansion was deemed acceptable. Using an expansion factor of 6,000 annual trips per survey respondent gave a reasonable match against “observed” trips in total and by trip purpose. Compared to the observed data, the initial mode share was skewed a bit toward bus and away from rail, so slight changes were made to the mode constants in the models to better match the data. The resulting initial expanded trips for the “base scenario” (no user-supplied changes) are shown in Table 1. Note that the overall rail and bus mode shares are near 14%, which is higher than for the NE Corridor in total because our sample and the resulting models focus on trips with both ends near the major cities (Boston, New York, Philadelphia, Washington), while the entire corridor also contains areas that are relatively distant from those city centers and thus tend to be more car-oriented. Table 1: Base Scenario Total Trips by Purpose and Mode RESULTS BASE SCENARIO Predicted trips (1000/year) Business Vacation Visit Other Total Car 6,114 19,891 19,765 7,641 53,411 Bus 1,812 6,130 2,900 697 11,539 Rail 3,290 4,260 3,551 1,189 12,290 Air 1,887 2,341 2,213 517 6,958 Total 13,104 32,622 28,428 10,044 84,198 Mode share Business Vacation Visit Other Total Car 46.7% 61.0% 69.5% 76.1% 63.4% Bus 13.8% 18.8% 10.2% 6.9% 13.7% Rail 25.1% 13.1% 12.5% 11.8% 14.6% Air 14.4% 7.2% 7.8% 5.2% 8.3%

Technical Appendix: The Scenario Analysis Tool 101 4. Changing the demographic distributions One way that the user can define future scenarios is to adjust the population distribution by gender, age group, education level, employment status and/or income level. Table 2 shows the base scenario distributions that result from the expanded sample of respondents. The user can change the percentages in any of the green shaded rows, and the percentage in the “base” category (white unshaded row) automatically adjusts to maintain the total at 100%. In the model calculations, the expansion factor for each respondent are adjusted using the new scenario percentage divided by the base scenario percentage along for the relevant category for each demographic variable. (For example, if the gender balance for Business trips were adjusted to be 50%/50%, then the expansion factors for Business trips made by females would be multiplied by 0.500/0.411, while the expansion factors for Business trips made by males would be multiplied by 0.500/0.589.) Table 3 shows the resulting changes in rail trips for five different scenario changing the demographic distribution of the traveling population: (1) a shift towards more females, (2) a shift towards more senior citizens, (3) a shift towards fewer non-college graduates, (4) a shift toward fewer non-employed adults, and (5) a shift towards the extremes of the income range and away from the center. The shifts are all fairly modest, at 10% of the initial population share, and the resulting changes in rail trips in Table 3 are also modest, all below 1.5% change in total trips. (The scenarios assume no changes in attitudes within each demographic category.) Table 2: Base Scenario Demographic Distributions Demographics BASE SCENARIO GENDER Business Vacation Visit Other Female 41.1% 57.1% 55.4% 56.4% Male (base, do not edit) 58.9% 42.9% 44.6% 43.6% Total 100.0% 100.0% 100.0% 100.0% AGE Business Vacation Visit Other Under 35 20.7% 24.1% 21.3% 12.6% 35-44 (base, do not edit) 22.0% 19.9% 16.4% 12.9% 45-54 24.0% 18.5% 17.2% 24.7% 55-64 21.7% 18.8% 21.7% 20.8% 65 and older 11.6% 18.7% 23.4% 29.0% Total 100.0% 100.0% 100.0% 100.0% EDUCATION Business Vacation Visit Other Not a college graduate 22.8% 33.9% 28.1% 35.8% College graduate (base, do not edit) 77.2% 66.1% 71.9% 64.2% Total 100.0% 100.0% 100.0% 100.0% EMPLOYMENT Business Vacation Visit Other Not employed 14.5% 32.0% 34.2% 41.9% Employed (base, do not edit) 85.5% 68.0% 65.8% 58.1% Total 100.0% 100.0% 100.0% 100.0% INCOME Business Vacation Visit Other

Technical Appendix: The Scenario Analysis Tool 102 Under $25,000 5.6% 6.3% 6.9% 8.9% $25,000 to $49,999 10.4% 15.3% 14.0% 12.9% $50,000 to $74,999 12.3% 16.9% 16.2% 15.3% $75,000 to $99,999 19.0% 20.5% 20.5% 21.3% $100,000 to $149,999 (base, no ed) 23.7% 24.3% 22.3% 21.6% $150,000 to $199,999 12.7% 9.0% 9.3% 10.3% $200,000 and up 16.3% 7.7% 10.8% 9.7% Total 100.0% 100.0% 100.0% 100.0% Table 3: Change in rail trips by purpose for selected changes in demographic distributions Business Vacation Visit Other Total Female share up 10%, male share down to compensate -0.6% -1.2% -0.5% -2.2% -0.9% Age over 65 share up 10%, under 35% down 5% -0.1% 1.1% -0.1% -0.9% 0.1% Not college grad share down 10%, college grad share up to compensate 1.4% 1.0% 1.5% 1.9% 1.4% Not employed share down 10%, employed share up to compensate 0.4% -0.6% 0.5% 1.5% 0.3% Income shares below 50k up 10%, income shares above 150k up 10%, incomes 50-100 k down 10% -0.4% -0.5% 0.4% 0.7% -0.1%

Technical Appendix: The Scenario Analysis Tool 103 5. Changing the mode travel times and costs Table 4 shows how the user can adjust the travel times or costs for specific scenarios. The changes are quite general and are applied in the same percentage to each person in the sample (with the exception of cost changes, which can be specified separately for business trips and non-business trips to represent changes in business or economy fares separately). The base scenario values are given an index of 100, so entering a value of 110 would increase the value by 10% for every person, while entering a value of 90 would decrease the value by 10% for every person. Tests were done increasing each of the values in Table 2 by 10% (entering an index of 110, one cell at a time). The largest influence was seen when increasing the rail non-business fare by 10% with a change in number of trips of -8.2% (a direct fare elasticity of -0.82). The second largest change is when increasing rail in-vehicle time, with a direct travel time elasticity of -0.64. The cross-mode effect for a 10% increase in car in-vehicle time is the third largest with an increase of 5.5% in rail trips (a cross-elasticity of 0.55). In general, the cross-elasticities for changes in car times and costs are larger than for the other modes because of the larger base shares for car. Table 4: User input for changes in travel times or costs Time and Cost Indices (Base=100) Car Bus Rail Air Main mode travel time 100 100 100 100 Access travel time 100 100 100 Egress travel time 100 100 100 Service frequency 100 100 100 Business trip cost 100 100 100 100 Non-business trip cost 100 100 100 100 Table 5: Change in total rail trips when increasing each time and cost variable by 10% Car Bus Rail Air Main mode travel time 5.5% 2.7% -6.4% 0.3% Access travel time 0.5% -2.8% 0.8% Egress travel time 0.3% -1.4% 0.4% Service frequency -0.4% 0.0% -0.2% Business trip cost 0.7% 0.2% -2.0% 0.4% Non-business trip cost 3.1% 0.8% -8.2% 1.7%

Technical Appendix: The Scenario Analysis Tool 104 6. Changing the attitudinal variables The spreadsheet also allows the user to simulate predefined shifts in the four latent attitudinal variables, where one or more groups defined by age, gender, education and/or employment status take on the attitudes of another group along that demographic dimension. Such a shift is simulated by entering a value greater than 0 in the cell for any of the four attitudes, as shown in Table 6. Table 6: User input for specific pre-defined shifts in attitudes Car- oriented Technology Urbanism Privacy Shift all ages to under 35 attitude 0 0 0 0 Shift under 35 to 35-44 attitude 0 0 0 0 Shift all ages to over 65 attitude 0 0 0 0 Shift all ages one group younger 0 0 0 0 Shift female to male attitude 0 0 0 0 Shift male to female attitude 0 0 0 0 Shift no college to college attitude 0 0 0 0 Shift college to no college attitude 0 0 0 0 Shift no job to employed attitude 0 0 0 0 Shift employed to no job attitude 0 0 0 0 Table 7: Change in total rail trips when shifting attitudes, on at a time and all at once Car- oriented Technology Urbanism Privacy All at once Shift all ages to under 35 attitude 17.9% 2.5% 0.0% -3.4% 16.4% Shift under 35 to 35-44 attitude -1.7% 0.0% 0.0% 0.0% -1.7% Shift all ages to over 65 attitude -11.9% -3.4% 0.0% 10.4% -5.7% Shift all ages one group younger 6.1% 1.4% 0.0% -2.5% 4.9% Shift female to male attitude 2.3% -0.4% -0.3% -0.4% 1.2% Shift male to female attitude -1.8% 0.3% 0.2% 0.4% -1.0% Shift no college to college attitude 1.2% 0.1% 0.1% 2.7% 4.2% Shift college to no college attitude -3.6% -0.3% -0.2% -7.5% -11.4% Shift no job to employed attitude -0.6% 0.2% 0.0% 1.3% 0.9% Shift employed to no job attitude 1.2% -0.4% 0.0% -2.5% -1.7% The results of the attitudinal change for each the cells above, in terms of total rail trips, are shown in Table 7. The final column also shows the result if all four attitudinal variables are shifted at once. The age variables tend to have the largest effect. If all age groups were to adopt the current attitudes of the “Millennials” (under 35), then rai trips would increase by 16%, with the largest effect from a shift away from the “car-oriented” attitude, plus a positive effect form the technology aspect of being able to use devices in the train, but an offsetting negative effect from the relative lack of privacy in the train. On the other hand, if all age groups were to adopt the attitudinal tendencies of the current age 65+ group, rail

Technical Appendix: The Scenario Analysis Tool 105 trips would decrease by almost 6%, due to opposite offsetting effects as described for the shift to “younger” attitudes. The most realistic future trend in attitudes by age may be the “shift all ages one group younger”. This would be the situation 10-15 years from now if attitudes for any given person stay the same and go together with the age cohorts, rather than changing with age. The result for that test is a 5% increase in total rail trips. Other than age, the largest effect would be if all persons adopted the “no college education” attitude, regardless of education level. This would result in an 11% decrease in rail trips, but seems like a very unlikely scenario. 7. Four illustrative scenarios Finally, combinations of the attitudinal shifts above were run to represent four possible future scenarios. Only the trends with age are varied – the attitudinal differences related to gender, education and employment are assumed not to change in these scenarios. The most optimistic scenario for rail is that people will keep their current attitudes toward auto orientation and technology as they age (“Go with cohort” below) and that the next generation will have the same attitudes as current Millennials, but that all age groups will adopt the current attitudes toward privacy of the current age 65+ group. The most pessimistic scenario for rail is that each age cohort will adopt the attitudes toward auto orientation and technology that the previous cohort had at that age (“Current trend with age” below) that the next Generation Z will not reflect the current Millennials but will reflect the current post-Millennial 35-44 age group, and that the cohorts will keep their same attitudes toward privacy as they age. The “Mixed A” and “Mixed B” have different combinations of those assumptions. Table 8: Definitions of four scenarios, relative to the Base scenario Name Change in Values 1 Pessimistic for Rail 2 Mixed A 3 Mixed B 4 Optimistic for Rail Age effects on auto orientation Current trends with age Gen Z same as current 35-44 Current trends with age Gen Z same as current 35-44 Go with cohort Gen Z same as Millennials Go with cohort Gen Z same as Millennials Age effects on technology orientation Current trend with age Go with cohort Current trend with age Go with cohort Age effects on privacy attitude Go with cohort All adopt Over 65 Go with cohort All adopt Over 65 Gender, employment, education effects on all attitudes Current trends Current trends Current trends Current trends

Technical Appendix: The Scenario Analysis Tool 106 Table 9: Change in rail trips under the four scenarios, relative to the Base scenario Name Change in Rail Trips 1 Pessimistic for Rail 2 Mixed A 3 Mixed B 4 Optimistic for Rail Business -5% 13% 3% 22% Vacation -4% 8% 2% 14% Visit Friends/relatives -2% 9% 4% 15% Other -11% 14% 9% 35% Total -4% 10% 4% 18% The results are that the pessimistic scenario results in a 4% drop in total rail trips, while the optimistic scenario results in an 18% increase. The mixed scenarios result in a 10% and 4% increase in rail trips respectively. The “Other” trip segment, which is the smallest segment, is the most volatile, while the other trip purposes all show fairly similar trends across the scenarios. References RSG (2015a) Northeast Corridor Intercity Travel Summary Report. Report prepared for the Northeast Corridor Infrastructure and Operations Advisory Commission. April 2015. RSG (2015b) Foundational Knowledge to Support a Long-Distance Passenger Travel Demand Modeling Framework Implementation Report, prepared for the Federal Highway Administration. Washington. http://www.fhwa.dot.gov/policy/modelframework/model_framework.pdf