Understanding Changes in Demographics, Preferences, and Markets for Public Transportation (2018)

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.

59 Previous chapters explored the role of demographics, location, and preferences separately as influences on the transit markets. This chapter analyzes all of these factors together by using two market research–based travel demand models developed specifically in this study. The chapter is presented in two sections: 1. New scenario-building model. The first section of the chapter presents the results of the application of a new travel demand model that includes both hard factors (e.g., travel times and costs) and soft factors (preferences and attitudes) in one unified analytical format. 2. New attitude-based model. The second section of the chapter presents the results of a new analytical model that focuses primarily on preferences to examine the relationships between key long-term values and shorter-term attitudes and the propensity to use transit. A New Integrated Scenario Forecasting Model for Transit What Can Be Learned from the Creation and Testing of Alternative Scenarios This section presents the results of the new scenario development and testing exercise, which applies a set of advanced hybrid choice models—also often referred to as “integrated choice latent variable” (ICLV) models—to analyze the data from the 2016 TCRP survey, which was a stated choice survey. As described in Technical Appendix 6, these models account for the differences across an individual respondent’s preferences with regard to how the respondent reacts to the level of service variables, such as time and cost, and to the respondent’s baseline preferences for given modes of transport. The explanation of variation in the work used three separate components: 1. The research team sought to explain a substantial share of the variation by linking it to observed respondent characteristics, such as age and education, and trip characteristics, such as purpose. 2. The research team allowed for additional random variation (i.e., differences across an individual respondent’s preferences that cannot easily be linked to observed characteristics of the respondents). This and the first type of variation were allowed for in both the mar- ginal sensitivities to level of service variables (e.g., time and cost) and the baseline mode preferences. 3. The research team allowed for further variation in these modal constants that is linked to attitudinal constructs. These latent attitudes varied both deterministically (e.g., as a func- tion of age) and randomly (i.e., on the basis of unobserved factors) across individuals. At the same time as these attitudes explained a share of the variation in modal preferences across C H A P T E R 6 Understanding How the Factors Fit Together: Integrated Modeling

60 Understanding Changes in Demographics, Preferences, and Markets for Public Transportation respondents, they also explained answers that these same respondents gave to a set of attitudinal questions. A complete description of the development of these ICLV models is presented in Technical Appendix 6. This section presents results of applying the ICLV models described above in various scenarios concerning the future of public transportation in the United States. The scenarios that were tested varied both the level-of-service variables (travel times and costs) of the various modes and potential shifts in attitudes. Some of the results are also broken out by different market segmen- tations: age group, trip purpose, region of the country, and residence neighborhood type. First, brief summaries of the policy-oriented results of the exercise are presented. Then, a more detailed description of how the scenarios were created from a base model and several refinements of that model is presented. Summary of the Results of the Scenario Testing The research team ran several scenarios by changing all the time and cost attributes for specific modes simultaneously. These scenarios provide an idea of the range of response to more pro- nounced overall changes in service levels. The results are presented in Table 18 in the order of their overall effect on transit mode share. (The “total transit” column is for bus and rail combined.) Changes to Service Variables (%) Scenario Bus Train Private TNC Shared TNC Car Total Transita Increase in Transit Share All bus and train better, both TNC worse 36 34 –44 –53 –19 35 All bus and train better 31 30 –27 –32 –24 30 All bus and train better, both TNC better 22 22 2 6 –31 22 All train better –37 78 –17 –21 –15 19 All bus better 78 –40 –19 –23 –16 21 All car worse 9 8 14 15 –18 8 All private and shared TNC worse 6 5 –24 –31 7 6 All shared TNC worse 3 3 10 –38 4 3 All private TNC worse 3 2 –33 9 3 2 Decline in Transit Share All private TNC better –5 –4 55 –12 –6 –5 All shared TNC better –7 –6 –15 73 –7 –6 All bus worse –50 34 8 10 7 –9 All train worse 33 –50 6 8 6 –8 All private and shared TNC better –11 –9 36 54 –11 –10 All car better –12 –11 –14 –16 23 –11 All bus and train worse, both TNC worse –16 –16 –8 –13 26 –16 All bus and train worse –22 –21 20 26 17 –22 All bus and train worse, both TNC better –33 –31 60 91 2 –32 aBus and rail combined. Table 18. Scenario results for overall shifts in mode time and cost attributes, 2016.

Understanding How the Factors Fit Together: Integrated Modeling 61 Best for All Transit The effects of changing transit level of service on projected demand also depend on what changes occur in the ride-hailing market. The best scenario for transit decreases all the bus and train time and cost attributes by 25% (“bus and train better”), while increasing all the private TNC and shared TNC time and cost attributes by 25% (“both TNC worse”). Projected demand for both bus and train increases by about 35% (as a percentage of their base scenario mode shares). If bus and train are improved, but with no change to private TNC or shared TNC service (“All bus and train better”), projected demand for both bus and train increases by about 30%. If bus and train are improved, but private and shared TNC also improve (“All bus and train better, both TNC better”), demand for both bus and train increases by about 22%. Best for a Given Mode When just one of the transit modes is improved, but not the other, the model shows a large increase in projected demand for that mode (78%), but a reduction in demand for the other transit mode by about 40%, so that the overall predicted increase in transit demand is about 20% in each case. If all car costs and times increase by 25% (“all car worse”), then the result is an 8% increase in transit demand (but a 15% increase in private and shared TNC demand). If all the times and costs for private and shared TNC increase, transit demand only increases in the range of 2% to 6%. This is because private and shared TNC have relatively small mode shares in the base scenario, so the cross-elasticities for transit are not that large. The second half of Table 18 shows scenarios that are increasingly bad for transit demand, with either the transit times and costs increased, or the competing mode times and costs decreased. The results show a near mirror image of the first half of the table, with the predicted drops in total demand ranging from 5% for “all private TNC better” to 32% for “all bus and train worse, both TNC better.” One difference is that the model predicts that improving both private and shared TNC has a larger influence on transit demand than making both worse because there are more current transit riders to lose when ride-hailing improves relative to the current number of ride-hailing users to gain if ride-hailing service worsens. Variation, by Attitude Another set of scenarios was run leaving the mode travel time and cost attributes constant and assuming future shifts in the attitudes of the travelers in specific ways, as compared with the attitudes simulated in the base scenario. The attitudinal effects in the model are related to the age and education level of the traveler and the traveler’s attitudes toward five attitudinal constructs: (1) willingness to share services, (2) safety in traveling, (3) use of technology while traveling, (4) pro-transit attitudes, and (5) concern for the environment. Although the research team tested more than 100 different scenarios of attitudinal shifts, most showed minor changes in demand. Table 19 shows the results only for those that showed at least a 3% shift in total transit demand. The range of changes in total transit demand in the table is from a 13% increase to an 8% decrease, which is only about one-third of the range of changes shown in Table 18 (35% increase to 32% decrease). The implication is that changes in future attitudes toward specific aspects of transit can influence demand but not nearly as much as changes in the quality and quantity of service that is offered. The largest predicted increase in transit demand (13%) is if all future travelers were to adopt the same attitudes toward all the attitudinal constructs as those in the survey who are under age 30 and highly educated. Conversely, the largest predicted decrease in transit demand (8%) is if all future travelers were to adopt the same attitudes as those who are over age 65 and those with no college education. Other results indicate that of the 13% increase in the best sce- nario, about 8% appears to be related to current differences in attitude that are correlated with

62 Understanding Changes in Demographics, Preferences, and Markets for Public Transportation education level, while 5% appears to be related to current differences in attitude that are correlated with age. Similarly, of the 8% decrease in the worst scenario, about 5% appears to be related to education level, and 3% appears to be related to age. Other scenarios tested less-complete shifts in attitudes. Instead of everyone shifting to the extreme category of age or education, everyone’s attitudes shifted to those of one age group younger or older and one education level higher or lower. If everyone’s attitudes shift one age group lower and one education level higher, the increase in transit demand is about 4%. If every- one’s attitudes shift in the opposite direction—one age group higher and one education level lower—the decrease in transit demand is about 5%. (The former change seems more likely, as age cohorts grow older and generally have higher education levels than previous age cohorts.) The one single attitudinal effect that appears to have the highest effect on transit demand is the relationship between age and the attitude toward shared services. If all age groups adopted the attitudes of those under age 30 for this construct, the predicted increase in transit demand would be about 4%, whereas if all adopted the attitudes of those over age 65, transit demand would decrease by 4%. Another outcome of note in Table 19 is that the changes for the private TNC and shared TNC modes are typically in the same direction as the changes for bus and train; they are usually even larger than the changes in transit demand—particularly the age-related effects. This result indicates that the attitudinal shifts that most strongly favor transit will also strongly favor ride-hailing services; thus, much of the potential increase in transit demand may be attracted away by Uber, Lyft, and similar services, particularly if those services become more convenient or less expensive through automation technology. Table 20 shows the variation in the base mode shares of reference point scenarios by national region. The mode shares for the reference point scenarios are also similar across regions. However, transit use is much higher in the northeast and northwest regions than in the other regions. The model results indicate that the variations in mode shares are more an outcome of different Results (%) Scenario Bus Train Private TNC Shared TNC Car Total Transit All adopt under age 30 and graduate degree attitudes 11 15 13 19 –26 13 All adopt grad degree attitudes 6 10 –3 7 –11 8 All adopt under age 30 attitudes 6 5 18 12 –15 5 All shift one category younger and one education level higher 4 5 9 8 –10 4 All adopt under age 30 attitude toward service sharing only 5 3 14 6 –10 4 All adopt over age 65 attitudes –5 –1 –23 –11 13 –3 All adopt over age 65 attitudes toward service sharing only –5 –3 –15 –6 11 –4 All shift one category older and lower education –5 –5 –11 –8 11 –5 All adopt no college attitudes –4 –6 0 –5 8 –5 All adopt over age 65 and no college attitudes –9 –7 –23 –16 21 –8 Table 19. Scenario results for shifts in attitudes.

Understanding How the Factors Fit Together: Integrated Modeling 63 levels of transit service offered in the regions than they are of any major differences in people’s underlying mode preferences or attitudes across the regions. Variation, by Neighborhood Type The scenario results were also segmented by four neighborhood types, as self-reported by the respondents: (1) urban or (2) suburban/small town areas and (3) mixed-use or (4) residential neighborhoods. As shown in Table 21, transit mode shares are rank-ordered in the direction that one would expect, rising from 42% in suburban/residential areas to 52% in urban/mixed use areas. Mode share for private TNC plus shared TNC also increases from 16% to 22%. This segmentation by neighborhood type was not included in the model explanatory variables, so any differences in predicted mode share arose from differences in the demographics of the people who live in the different types of neighborhoods. In the actual current situation, transit and ride-hailing levels of service also tend to vary a great deal across these neighborhood types, so the actual variations in mode shares are much more pronounced than in the base scenario. This data analysis suggests that it is not some inherent characteristics of residents of different regions of the United States that leads to transit ridership, but rather the actual times and costs of the services they are exposed to. In and of itself, this suggests that the study of migration of populations is less a study in the differences of region (e.g., moving from Providence to Phoenix) but more a study of the difference in the quality of service experiences at the residential location in the two regions. Definition of Elasticities and Cross Elasticities The research team ran several scenarios to gauge the elasticities of the model outcomes to the changes in the mode travel time and cost attributes that were varied in the stated choice experiment. Each scenario changed only one of the attributes, increasing or decreasing the Base Mode Share (%) Scenario Bus Train Private TNC Shared TNC Car Northeast 24 24 10 10 33 North Central 26 19 9 9 37 Northwest 23 27 9 9 32 Southeast 21 23 9 9 38 Southwest 24 19 11 9 37 Table 20. Base mode shares of reference points used in the modeling process. Mode Share (%) Scenario Bus Train Shared TNC Car Suburban/town residential 21 21 8 41 Suburban/town mixed use 24 21 9 37 Urban residential 25 23 11 31 Urban mixed use 26 26 Private TNC 8 9 10 11 11 26 Table 21. Base mode shares of reference points, by residential demographic composition.

64 Understanding Changes in Demographics, Preferences, and Markets for Public Transportation specific time or cost level by 25% of the reference level for each trip in the survey sample (Table 22). The values shown in the first column are direct elasticities—the percentage change in demand for each percentage change in a service attribute for the same mode. The other cells are the cross elasticities: the percentage change in demand for each percentage change in a service attribute for a competing mode. The shaded cells are the largest cross elasticities for each variable: blank cells represent cross elasticities so low as to be deemed not meaningful for this discussion. By looking at the patterns of unshaded and shaded cells, one can see that bus and train compete more closely with each other than they do with the nontransit modes. The same is true for the private TNC and shared modes, which also compete most closely with each other. It also appears that private TNC and shared are the closest substitutes for the car mode, although the cross elasticities for private TNC and shared for changes in car attributes are not that much higher than the cross elasticities for bus and train. Finally, for changes in bus and train attributes, the cross elasticities for private TNC and shared are typically somewhat higher than for the car, meaning that ride hailing appears to be a somewhat closer substitute for transit. Note: A column for “car” would show no cross elasticities large enough for inclusion. aDirect elasticity = (percentage change in demand + percentage change in own mode attribute). Variable Elasticitya Bus Train Private TNC Shared TNC Table 22. Elasticities to mode travel cost and time attributes.

Understanding How the Factors Fit Together: Integrated Modeling 65 A New Model for the Impact of Values and Attitudes on Transit Ridership A separate model was built to focus on attitudes and values, with little emphasis on supply characteristics. The study of preferences in the explanation of variation in public transportation ridership involves multiple attitudinal factors, some of which involve longer-term decisions (e.g., where to live and how many cars to own), and some of which involve shorter-term judgments (“I think transit is stressful or nonstressful”). This research project applied the method called structural equation modeling to examine not only the direct relationship between a given preference and the propensity to use transit, but also the indirect impact, where an independent variable may influence a second variable, which in turn influences the outcome variable. An example of this might be long-term values about urbanism, which influence the density of one’s location, which in turn influences the amount of transit consumed. A powerful tool within structural equation modeling is the calculation of total effect, the sum of the direct impact of the factor plus the sum of the indirect factors. The total effect allows the quick observation of the total impact of a given independent factor on the outcome factor. The method employs the concept of latent factors, which help examine multiple observed vari- ables together, as documented in Technical Appendix 6. Elements of the Attitudinal Model The attitudinal model of the 2016 TCRP survey (Figure 24) has four major component parts, as follows: • Longer-term values. On the left-hand side of the diagram, four longer-term values are defined that are hypothesized to cast influence on the next three component parts, both directly and through intervening factors. Transit Trip Making Values Information Technology Values Urbanism Values Privacy In Location Values Auto Orientation Transit Trip Enjoyable Transit Trip Unsafe Transit Trip Green Transit Trip Difficult Social Support for Transit Trip Residential Density Transit Accessibility Car Availability Longer-term values that influence... ...the residential setting, which influences... ...short-term attitudes that influence... ... the outcome Figure 24. Conceptual diagram of attitude-based model.

66 Understanding Changes in Demographics, Preferences, and Markets for Public Transportation • Residential setting. Next, the residential setting of the participant is reported in terms of density, design, accessibility, and car ownership. These indexes of residential setting are hypothesized to influence transit use directly and also indirectly through shorter-term attitudes, which in turn influence propensity to use transit. • Shorter-term attitudes. Five latent variables represent shorter-term attitudes with direct impact, including four concerning the evaluation of the transit trip and one representing perceived normative influence and peer influence. • Outcome/ridership. On the right-hand side, a latent variable represents the outcome factor (transit ridership). Eleven Factors Help to Explain Variation in Transit Use The model is not designed to predict behavior but rather to contribute to understanding of the relationship between and among factors, given the relationships hypothesized in Figure 24. Structural equation models can reveal the combination of direct and indirect effects of one fac- tor on another, called the standardized total effect (STE). For example, to explain the meaning of the total effect, the AMOS software program states: The standardized total (direct and indirect) effect of ‘Values Auto Orientation’ on Transit Use is –.264. That is, due to both direct (unmediated) and indirect (mediated) effects of Values Auto Orientation on Transit Trip making, when Values Auto Orientation goes up by 1 standard deviation, Transit Trip making goes down by 0.264 standard deviations. Table 23 shows the ranking by level of importance of 12 explanatory factors, of which one was found to be statistically insignificant. The table shows that the strongest factor in the explana- tion of transit use is the normative factor, or the idea that those in one’s social network would Rank Order by Absolute Value of STE STE Factora 1 0.46 Normative (social support) 2 –0.35 Car available 3 0.35 Values urban setting 4 –0.26 Values auto orientation 5 0.22 Transit trip green 6 0.21 Transit trip enjoyable 7 0.21 Design and accessibility 8 0.16 Density 9 0.13 Values productivity, ICT 10 –0.13 Transit trip difficult 11 0.12 Values suburban house 12 ns Transit trip unsafe aBold italic type indicates latent factors for the four basic values, italic type indicates the three items concerning residential setting, and roman type indicates the five short-term attitudes (see Figure 24); ICT = information and communications technology; ns = not significant. Table 23. Rank order of importance of factors in the explanation of transit use.

Understanding How the Factors Fit Together: Integrated Modeling 67 approve of one using transit and that they would use transit themselves. In Figure 24, this factor representing normative pressure is labeled “Social Support for Transit Trip.” The data in Table 23 can be interpreted in several ways. While the STE is usually expressed in the scale of a 100% increase in the independent factor, a more realistic interpretation can be stated in terms of a 10% increase in the independent factor. By way of example, Table 23 shows the following: • A 10% increase in the value of the factor “values urban setting” would be associated with a 3.5% increase in the outcome factor, “transit trip making.” • A 10% increase in the value of the factor “values auto orientation” would be associated with a 2.6% decrease in the outcome factor, “transit trip making.” • A 10% increase in the value of the factor “transit trip enjoyable” would be associated with 2.1% increase in the outcome factor, “transit trip making.” Exploration of the Interactions Between Factors As shown in Table 23, the most important explanatory factor in interpreting transit use is normative (social support), with an STE of +0.46. Phrased differently, the propensity to believe that people in one’s social network either use public transportation or would approve of one’s use of public transportation is a powerful factor in the prediction of transit use. The role of peer influence is a major theme in the social psychology of transportation behavior and in the social psychology of multiple other behaviors.4 Other important factors in the explanation of transit use are as follows: • The factor of car availability has strong (negative) impacts on transit ridership, with an STE of −0.35. • Another powerful explanatory factor reported in the table is that of basic values toward urbanism (STE of +0.35), which plays a key role in the choice of residential setting, which in turn plays a key role in the propensity to choose transit. • The fourth most important explanatory factor in Table 23 concerns one’s values and prefer- ences with regard to the automobile, with an STE of −0.26. Logically enough, the stronger one feels about the importance of owning a car, and the greater the pleasure one derives from the car, the lower will be one’s propensity to use transit. • The factor representing neighborhood design and accessibility has an STE of +0.21. This factor is clearly interrelated with residential density and is a good predictor of transit use. 4 The potential role of social normative pressure was noted in a paper by Popuri et al. (2011) that was based on an early attitudinal survey fielded by RSG in 2010.