Characteristics of Premium Transit Services that Affect Choice of Mode (2014)

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.

F-1 A p p e n d i x f Awareness and Consideration Models Contents F-2 Joint Bivariate Binary Probit Model Formulations F-4 Transit Awareness and Consideration Analysis F-4 Transit Awareness and Consideration Models F-16 Summary of Awareness and Consideration Models Awareness and consideration of different modes of transportation are key determinants of the choice set generation process. In general, it may be assumed that an alternative is included in the feasible mode choice set only if an individual is aware of the travel mode and actually considers it as a possible option for undertaking a trip. Even if a person is aware of a mode, he or she may choose to not consider it, thus eliminating it from the choice set. If a person is not aware of a travel mode to begin with, then it is automatically not considered and not part of the feasible choice set. As described earlier, the project team formulated a two-step model system where the choice set is formed first by combining a modal awareness model with a modal consideration model. One novel aspect of the work in this project is the development of a choice set generation model for awareness and consideration combined with a discrete choice model for mode choice (as opposed to a pure discrete choice model that does not account for choice set generation). There is a rich body of literature devoted to this topic, virtually all of which has shown and stated that the combination of a choice set generation model and discrete choice model is superior to a pure discrete choice model that assumes constant choice set for all observations (Castro et al. 2011). In much of the research to date, a latent choice set generation approach was necessary because data on awareness and consideration of alternatives were not available. However, in this project, explicit data about awareness and consideration were collected, thus eliminating the need to adopt a latent choice set generation model and allowing for the development of an observed choice set generation model. Findings in the literature, and in this project, show that a model system that combines a choice set generation model with a mode choice model provides greater insights and forecasting accuracy than a model that ignores choice set formation. A model that accounts for mode choice awareness and consideration provides key information to the analyst about the context, person attributes, household attributes, trip attributes, and modal attributes that contribute to a person being aware of a certain mode and then actually considering it for the trip in question. The choice awareness and consideration model provides information on potential policies, strategies, and modal service attributes that can enhance the probability that an individual would become aware of the mode and consider it. By accounting for heterogeneity in

F-2 Characteristics of premium Transit Services that Affect Choice of Mode Awareness and consideration models rely on variables that are also used in mode choice models, so a joint model of awareness, consideration and mode would be ideal. This would provide an opportunity to test on variables in the models. There are, however, practical constraints to jointly estimating models of awareness, consideration and mode that are addressed in Appendix H. Joint Bivariate Binary Probit Model Formulations Awareness and consideration are handled using choice set models as part of a two-step decision process. An individual who has a car available to make the trip is assumed to be aware of the option and always considers it in the choice set. Consequently, the car option enters the choice set in a deterministic way. The complete choice set for each individual q is formed as a result of awareness and consideration of the transit options (bus and rail). The following utility expression for an individual’s awareness is considered: },{, RailBusiwY qiqiiqi In this equation, qiw represents all the factors that affect the individual’s awareness of the transit alternative i. Without loss of generality, it can be assumed that the person is aware of the transit alternative if the utility is greater than zero. The parameter to be estimated, i , includes a constant. Due to common unobserved factors that could affect the awareness of different transit options, the error term qi is allowed to be correlated across transit alternatives. arailqbusqCorr ),( ,, 1,0~ Nqi If it is assumed that the error term qi is standard normally distributed, a joint bivariate binary probit model is obtained. The only additional issue to be incorporated is that the choice set for an individual must not be a null set, so individuals without a car must be aware of at least one of the two transit alternatives. The following terms are introduced to aid in the writing of the likelihood expression: Let 0 if 0 0 if 1 qi qi qi Y Y a and availablenot Car if 0 availableCar if 1 qd With these notations, the following expression serves as the likelihood of the observed awareness of all alternatives: arailqrailbusqbusq arailbusrailqrailrailbusqbusbus aq wwd aawawa L ,, 11 1212,12,12 ,,2 ,,2 choice set composition that is prevalent in the population, the model system is better able to replicate mode awareness, consideration, and choice processes and provides greater levels of accuracy in prediction. The approach adopted in this study is motivated by the evidence in the literature that clearly points to the superior performance of choice model systems that account for awareness and consideration of choice alternatives (choice set generation and composition processes).

Awareness and Consideration Models F-3 Consideration comes in the second stage of the choice set formation step. Even if a person is aware of a transit alternative, it may or may not be considered an option for a specific trip. Consideration is modeled using a utility expression similar to that used for awareness: },{, RailBusiuZ qiqiiqi Here qiu represents variables that affect the consideration decision of the individual q about transit alternative i. Again it is assumed that the person considers the transit alternative if the utility qiZ is greater than zero. The parameter to be estimated i includes a constant. It is assumed that the error terms are independent and identically distributed (i.i.d.) standard normally distributed and the formulation allows the error terms to be correlated across alternatives: crailqbusqCorr ),( ,, 1,0~ Nqi The consideration indicator is defined based on the utility equation (4) as follows: 0 if 0 0 if 1 qi qi qi Z Z c With these notations, the following serves as the likelihood of the observed consideration set of all the alternatives: crailqrailbusqbusq crailbusrailqrailrailbusqbusbus cq uud ccucuc L ,, 11 1212,12,12 ,,2 ,,2 The above likelihood expressions for awareness and consideration are quite straight forward and the parameters can be estimated by the maximum likelihood method. It must be noted that not all observations can be used in the estimation because some observations provide no information on awareness and consideration. An individual can be aware of a transit option only if that particular option is available (or feasible). Similarly, an alternative can be considered only if the individual is aware of that particular alternative. This implies that the following two categories of observations will not provide any information in the estimation of the awareness model: (1) both bus and train are not available (2) car is not available and only one of the transit options is available. It should be noted that there may be instances where the car is not considered by a trip maker even when it is available. Although this is a distinct possibility, there is insufficient information in the survey data set to help establish criteria or develop a model of auto mode consideration. As such, within this study, a deterministic approach was taken with respect to auto consideration. If the auto mode is available, then it is assumed that the trip maker is aware of it and considers it (thus it is included in the awareness set and the consideration set). As the focus of this study is on understanding awareness and consideration of bus and rail modes, this simplification for the determination of auto mode consideration was done. In the above equation, )(2 is the bivariate normal cumulative density function and the expression in the denominator is to ensure that the probabilities of all valid non-null awareness sets add up to 1 for individuals not having a car.

F-4 Characteristics of premium Transit Services that Affect Choice of Mode Transit Awareness and Consideration Analysis The awareness and consideration analysis conducted in Salt Lake City was designed primarily to support development of awareness and consideration models for Chicago and Charlotte. This provided key information to support the need for these models and the design of surveys to collect awareness and consideration of transit models. There were key findings from the analysis of the revealed preference survey for awareness as follows: Travelers appear to be aware of the existence of some form of transit within walking distance of their homes. This understanding is highest among travelers from 0-car households who presumably need to know more about the transit system than travelers from households with cars. Trip-makers are less aware of the full range of all transit sub-mode options within walking distance that are available in their neighborhood. This phenomenon is most pronounced for low income travelers who are less aware of the range of travel options than the population as a whole. Opportunities to drive to transit are less-well understood by survey respondents than their walk access options. This may be a result of the fact that drive access distances are much greater than walk access distances and therefore park-and-ride facilities are located outside of traveler’s normal sphere of activity. There were also key findings for availability of modal alternatives from this analysis: Survey respondents report many fewer modal options as being available for a commonly- made trip than does the travel forecasting model. Transit customers, are particularly likely to report no option or only one option suggesting that they are more likely to be either transit captives or that alternative transit sub-modes are not considered to be viable options. The high proportion of travelers stating that no option exists to their current mode describes a situation in which many travelers perceive themselves to be dependent on their current mode. This is most true for transit users but also exists for automobile users. A • • • • • • typical approach for representing transit dependency is membership in a 0-car household. Survey responses suggest that transit dependency may be more nuanced. For instance a large number of transit travelers from households with 2 or more automobiles stated that they had no transit options while other travelers from 0-car households made trips by automobile. These results provided insights to design questions for the awareness and consideration portion of the survey to support model estimation. Transit Awareness and Consideration Models There is very limited work, in the literature on understanding factors that affect awareness and consideration of alternative modes of transportation. In this study, an attempt was made to test the significance of a number of different attributes with respect to their influence on the awareness and consideration of alternative transit modes. Besides the usual socio-economic and demographic attributes that are known to influence mode usage patterns (age, income, employment status), the study considered a few additional categories of attributes as possibly explanatory entities.

Awareness and Consideration Models F-5 The survey asked respondents to provide information on awareness and consideration of transit modes for a reference trip. The characteristics of the trip itself may influence how people view alternative transit modes. A variety of hypotheses may be constructed in this context. An individual may be more aware of transit options on weekdays than on weekends. An individual may be more likely to consider transit options for commute travel than for non-commute travel. The regularity of the commute trip, the likelihood that transit service is competitive during the peak commute hours, and auto costs (e.g., parking costs) may motivate an individual to consider transit for commute travel more than for non-commute travel. Party size (number of people traveling together on the trip) is another trip attribute that may influence modal consideration. The survey results reported in Appendix C showed that party size tends to be smaller for transit trips, particularly in Charlotte. The difference in party size distribution is less pronounced in the Chicago data set. In Charlotte, it may be the case that travelers are less likely to consider transit alternatives when undertaking joint travel. This study attempts to explicitly account for attitudes, perceptions, and values in modeling transit awareness and consideration. The results of the factor analysis described in the previous section demonstrate the strong correlation between attitudinal factors and mode choice. If attitudinal factors are strongly correlated with mode choice, then one would expect the factors to be also strongly correlated with modal awareness and consideration (because choice is inextricably linked to awareness and consideration). The models in this study consider attitudinal factors as possible explanatory variables to account for these attitudes, perceptions, and values that are traditionally unmeasured, unobserved, and relegated to being absorbed in the random error term. A key question that merits consideration is the extent to which modal level of service variables should enter the awareness and consideration model specifications. It may be hypothesized that people are more aware of and would give greater consideration to transit modes when transit level of service is greater, more competitive with the auto, and of high quality. In the current study, transit awareness and consideration is modeled whenever transit is available (as determined by the presence of a valid skim value). While it is certainly possible to include transit level of service attributes and measures of competitiveness in the model specification for transit awareness and consideration, such variables are not included in the model specifications presented in this report. The inclusion of such variables in models of awareness and consideration may be fruitful directions for future research, with due consideration given to the ramifications of including such level of service attributes in both models of awareness and consideration and mode choice. As noted in the survey results (Appendix C), it is found that a large percent of individuals who did not consider transit options in the revealed preference portion of the survey actually chose a transit option in the stated preference portion of the survey. This is not unexpected in any way as one would expect individuals to choose transit options when information about their service attributes is presented in a clear way vis-à-vis service attributes of the auto mode. In a stated preference survey design, transit service attributes are designed such that the transit alternatives are competitive against the auto mode thus facilitating the determination of trade-offs in attributes during choice processes. While this suggests that transit competitiveness is important in the “choice” process, it does not necessarily imply that transit service competitiveness is important in the “awareness and consideration” process. Again, the research team believes that this notion needs to be tested in

F-6 Characteristics of premium Transit Services that Affect Choice of Mode which future research by explicitly including transit competitiveness measures and service attributes in models of awareness and consideration. In this study, the team has taken a more simplified approach where the awareness and consideration models include the extent to which a person is “informed” about transit as an explanatory variable. In the stated preference survey, respondents are being “informed” about the presence of transit service alternatives and the levels of attributes associated with service variables. In essence, the formulation in this study assumes that individuals may become aware and consider a mode of transport when they have information about the modal alternative. As a first attempt at developing a modal awareness and consideration model, the study includes extent the to a person is informed about transit as an explanatory variable. In FIGURE C-9 of Appendix C, it is readily apparent that this is an important explanatory variable influencing transit usage. If an individual is very well informed about a modal option, then he or she is likely to be aware of it and consider it. Given the objective of this TCRP project, it is possible to make the case that primary interest lays in traveler awareness and consideration of, and therefore choice between, premium and non-premium modes of transport – as opposed to the traditional modal designations of “bus” and “rail”. The study team recognizes that there is considerable interest in understanding the value that travelers place on premium mode attributes and how this translates into awareness, consideration, and choice. Although it is theoretically possible to replace the term “bus” with “non-premium mode” and “rail” with “premium mode” in this study, it is not practically feasible to do so. The entire survey instrument that was presented to survey respondents labeled the modal options as bus and rail, and it is therefore important to label the choice set elements in the same way that the options were presented to respondents. It is likely that there is a strong correlation between modal labels and the premium label, but travelers are generally more familiar with thinking of modes using the traditional labels of bus and rail. As such, the choice modeling effort of this study treats modal options as bus and rail, with the idea that the value placed on premium attributes can be inferred – to a strong degree – from the awareness, consideration, and choice models. Awareness Models for Chicago and Charlotte TABLE F-1 presents estimation results for the joint bivariate probit awareness models for both Chicago and Charlotte. Utility functions are estimated for both transit modes – bus and rail – in the two geographical contexts, and the error terms of the modal utility equations are allowed to be correlated with another. Thus, if there are unobserved factors that affect the awareness of both bus and rail in a particular city, then the model is able to account for the presence of such common unobserved factors. As noted previously, the inclusion of auto mode choice in the choice set is based on a deterministic rule and hence the awareness models apply only to transit modes. Some interpretations of the results is provided below but should not be interpreted as conclusions.

Awareness and Consideration Models F-7 TABLE F-1. Awareness models. Explanatory Variable Chicago Charloe Bus Rail Bus Rail Alternative specific constant 0.124 (1.1) 0.415 (2.1) 3.745 (5.6) 0.202 (1.4) Individual Demographics Full-time student -0.501 (-2.2) Full-time employed 0.298 (2.3) 0.379 (2.8) Retired -0.535 (-2.0) Female -0.365 (-2.8) Longtime resident (> 5 years) -0.339 (-2.7) -0.241 (-1.7) 0.226 (2.0) Has mobility problem 0.448 (2.1) Household Demographics Family income (Income in $ per year). -0.351 (-5.7) More drivers than vehicles 0.757 (1.5) Kids present in household 0.197 (1.3) Trip characteristics Weekend trip 0.332 (1.6) -0.534 (-3.3) 1.045 (4.2) Traveler Attitudes and Latent Variables Pro-transit attitude 0.456 (5.3) 0.785 (8.1) 0.180 (2.3) Consciousness 0.189 (2.4) 0.220 (3.1) Pro-car attitude -0.221 (-2.5) -0.226 (-3.1)

F-8 Characteristics of premium Transit Services that Affect Choice of Mode Explanatory Variable Chicago Charloe Bus Rail Bus Rail Low transit comfort level -0.276 (-3.1) -0.486 (-3.9) Willing to walk not more than 2 minutes -0.317 (-1.3) -0.739 (-2.6) Willing to walk 10 or more minutes 0.674 (4.6) 0.319 (1.8) Very informed about transit 0.367 (2.3) 0.665 (3.9) 0.374 (1.7) Error Correlation rho 0.288 (2.7) 0.178 (1.3) Model Fit Statistics Log-likelihood (final) -543.6 -518.3 Log-likelihood (constants) -683.9 -630.1 Pseudo rho-squared 0.205 0.177 Number of observations 801 748 TABLE F-1. (Continued).

Awareness and Consideration Models F-9 There are interesting differences in the awareness model estimation results between Chicago and Charlotte. The alternative specific constant for bus mode is statistically insignificant in the awareness model of Chicago, but statistically significant in the corresponding model of Charlotte. On the other hand, for the rail mode, it is found that the alternative specific constant in Chicago’s model is statistically significant but that in the Charlotte model is statistically insignificant. This suggests that there is an overall propensity for respondents in Chicago to be aware of the rail mode (relative to bus) while there is an overall propensity for respondents in Charlotte to be aware of the bus mode (relative to rail). This is presumably due to the more extensive rail network in the Chicago area, and hence respondents are likely to be more aware of the rail mode. On the other hand, Charlotte has very limited rail service, and as a consequence, travelers are more likely to be aware of bus service that has greater coverage and presence in the area. This broad finding is consistent with the finding that 70 percent of Charlotte survey respondents had no rail service available, while the corresponding percentage for Chicago is just over 50 percent. In Charlotte, less than 50 percent of respondents had no bus service available (as evidenced by the absence of a skim value) – which is considerably less than the percent that had no rail service available. In Chicago, about 50 percent of respondents have no bus service available – which is almost identical to the percent of respondents who have no rail service available. With respect to individual demographics, it is found that several variables influence the awareness of bus and rail modes. Full-time students, perhaps due to their transient nature, are less likely to be aware of bus services in Chicago (relative to rail service, which tends to be more visible). This finding is not seen in Charlotte, presumably because rail service is quite limited. On the other hand, full-time employed individuals are more likely to be aware of rail service in Chicago; in Charlotte, full-time employed individuals are more likely to be aware of bus services. In other words, regular commuters are likely to be aware of the more prevalent transit services in their respective geographic areas – that would be rail in Chicago and bus in Charlotte. In Chicago, females are less likely to be aware of rail service possibly because there is a lower prevalence of full-time commuters among them and because females continue to be more auto-dependent as they carry a greater share of household obligations and serve-passenger/child trips. A longtime resident (more than 5 years) is less likely to be aware of both bus and rail services in the Chicago sample; however, between the two transit modes, it is found that longtime residents are less aware of bus relative to rail (as evidenced by the larger negative coefficient in bus awareness model), suggesting that rail is more visible in a rail rich market such as Chicago. In Charlotte, on the other hand, longtime residents are more likely to be familiar with the bus service as evidenced by the positive coefficient on that variable in the bus awareness model. Essentially, longtime residents are more aware of rail in Chicago (relative to bus) and more aware of bus (relative to rail) in Charlotte, once again pointing to greater awareness of the more visible transit mode in the respective contexts. Individuals with a mobility problem are more likely to be aware of rail service in Chicago, presumably because the rail service in Chicago accommodates the needs of mobility-challenged persons and provides greater geographical coverage. With respect to household demographics, persons in households with higher income are less likely to be aware of bus services in Charlotte. Such income-based differences are not

F-10 Characteristics of premium Transit Services that Affect Choice of Mode observed in Chicago; this is not unexpected given the transit-rich market that is Chicago. In a transit market such as Charlotte, higher income individuals are not likely to use – and therefore familiarize themselves with – bus service (which likely caters to more transit-captive lower income riders). Respondents in vehicle-deficit households (where the number of drivers exceeds the number of vehicles) show a greater level of bus awareness in the Chicago sample. This is presumably because these individuals are more transit-captive and more dependent on bus services. This variable does not differentially affect awareness of bus and rail in Charlotte. This is not to say that the variable has no impact on awareness of bus and rail; the absence of the variable in the utility equations simply indicates that the variable has no differential effect on awareness of bus versus rail in Charlotte. It appears that respondents in vehicle-deficient households in Charlotte are likely to be equally aware of bus and rail services. The only trip characteristic that is found to significantly impact modal awareness is the indicator corresponding to a weekend trip. For weekend trips, it appears that Chicago respondents are more aware of rail service. In the Charlotte area too, respondents exhibit a greater level of awareness of rail for weekend trips and a lower level of awareness for bus service. Attitudinal variables and factors are found to be quite important in shaping the awareness of transit modes among individuals. Those with a higher pro-transit attitude are more likely to be aware of bus and rail services in Chicago and more likely to be aware of bus service in Charlotte; this finding is consistent with the exploratory analysis of the relationship between factor scores and mode choice presented in the previous section of the report. It is likely that the rail mode awareness is not significantly affected by this attitudinal factor in the Charlotte model because of the limited rail service in the region. Those with a higher level of consciousness demonstrate a greater proclivity to be aware of bus services in both Chicago and Charlotte (over rail services), presumably due to their predisposition towards being aware of transit alternatives that are environmentally friendly and allow them to be productive while traveling. It is true that the same argument applies to rail mode as well, but there appears to be a differential in the awareness of bus versus rail depending on level of consciousness. Pro-transit people are likely to be more aware of both bus and rail services (relative to non pro-transit people); being “conscious” adds an additional awareness level in the context of the bus mode – a mode that traditionally does not necessarily garner the same attention as rail. Those with a pro-car attitude are expected to be less aware of transit services. The estimation results are consistent with this expectation, although there is a differential impact on the awareness of bus versus rail in each city. Those with a pro-car attitude are less likely to be aware of rail services in Chicago, and less aware of bus services in Charlotte. In other words, it appears that the pro-car factor is more negatively associated with awareness of the more visible or prevalent transit service in each region. It is likely that those with such a pro-car attitude in Chicago live in the outlying suburbs not served well by rail. Such a residential self-selection pattern is not likely to be prevalent in Charlotte where pro-car folks are likely to be more uniformly spread throughout the region. Isolating these residential self-selection effects is a challenge that should be addressed in future research endeavors. As expected, the transit aversion factor is not significant in any of the models and this finding is consistent with the rather weak trends seen in the exploratory descriptive statistics and charts (presented in the previous section) for this particular variable. Those who consider it difficult to access transit and have a generally lower level of comfort with transit usage do not show any specific tendencies in the Chicago model, but show a clear tendency to be less aware of both bus and rail services in

Awareness and Consideration Models F-11 the Charlotte sample. It appears that those with this attitude are not necessarily less aware of the transit services in Chicago (transit services in Chicago are quite noticeable and one would have to be aware of them to develop a comfort or discomfort level and make a judgment regarding ease of access), but rather less likely to just use the mode of transport. Charlotte is a less transit- rich market and therefore lower levels of awareness are likely to be associated with an attitudinal factor that represents low transit comfort level. Those with a low tolerance to walking are less likely to be aware of rail services, while those with a high tolerance for walking time are more likely to be aware of rail services. It is likely that accessing rail stations generally tends to be more demanding than accessing bus stops, and as a result, walking time sensitivity more significantly impacts awareness of rail service as opposed to awareness of bus service. Those who are willing to walk not more than two minutes to access transit are less likely to be aware of rail services that generally demand a greater level of walking. On the other hand, those willing to walk more than 10 minutes are more likely to be aware of rail services as they seek out and are willing to endure more effort in availing of a premium service mode. Finally, those very informed about transit services are more likely to be aware of bus services relative to rail services. This finding is consistent with expectations. In general, as rail services tend to be more visible, it is not necessary for individuals to be “very informed about transit” to be aware of rail services. However, there is a difference when it comes to bus service. In the case of bus services (that are not so visible), if a person is not “very informed about transit”, then the likelihood of being aware of bus services is lower than that for rail services. When a person is very informed about transit services, the difference is likely to manifest itself in the bus arena. The error correlation between the transit utility functions is statistically significant for the Chicago area suggesting that there are correlated unobserved factors affecting awareness of these two modes in Chicago. On the other hand, the measure is statistically insignificant in the Charlotte context, possibly due to the lower level of complexity (and hence unobserved factors) associated with explaining modal awareness in a context where rail service is dwarfed by the bus system. Consideration Model Results for Chicago and Charlotte TABLE F-2 presents estimation results for the consideration models for both Chicago and Charlotte. Utility functions are estimated for both transit modes – bus and rail – in the two geographical contexts, and the error terms of the modal utility equations are allowed to be correlated with one another. Thus, if there are unobserved factors that affect the consideration of both bus and rail in a particular city, then the model is able to account for the presence of such common unobserved factors. As noted previously, the inclusion of auto mode choice in the choice set is based on a deterministic rule and hence the consideration models apply only to transit modes. Some interpretations of the results are provided below but should not be interpreted as conclusions. The set of consideration models for the two city samples also provides plausible behavioral indications. Consideration is generally a step that follows the awareness stage. Once an individual is aware of a certain mode of transport, the question is whether the individual will actually consider using the mode for the particular trip in question. The alternative specific

F-12 Characteristics of premium Transit Services that Affect Choice of Mode constants show a clear differential between rail and bus; the alternative specific constants (in the consideration utility equations) are higher for rail than for bus. This finding suggests that, all else being equal, individuals are more likely to consider rail over bus. This finding is consistent with the statistics reported in Appendix C where it is found that the ratio of “the number of individuals who consider rail to the number of individuals who do not consider rail” is consistently greater than the ratio of “the number of individuals who consider bus to the number of individuals who do not consider bus”. Among individual demographic attribute effects, homemakers are less likely to consider bus than rail (in Chicago), presumably because bus is not a convenient mode for taking care of household obligations and serve child trips. In Charlotte, there is no differential consideration effect between bus and rail – presumably homemakers give (or do not give) equal levels of consideration to bus and rail. Long-time residents are less likely to consider bus (relative to rail), presumably because the bus mode is not viewed in the same vein as the (more premium) rail mode (this is also seen in Appendix C). This finding is consistent across both data sets. Those with mobility problems are more likely to consider bus mode of travel, presumably because the bus mode is easier to access. Accessing rail service may entail longer access distances and times which may be inconvenient for those with mobility challenges. As such, even though there may be instances where mobility-challenged persons are more likely to be aware of rail service (as in Chicago), the fact is that they are more likely to consider bus services due to the reality of their mobility-challenged situation. Those aged less than 35 years are more likely to consider rail in Chicago (relative to bus); no such differential is seen in Charlotte suggesting that the greater prevalence of rail in Chicago makes a difference in its consideration level for this demographic.

Awareness and Consideration Models F-13 TABLE F-2. Consideration models. Explanatory Variable Chicago Charlotte Bus Rail Bus Rail Constant 0.018 (0.3) 0.767 (5.9) 0.779 (4.2) 1.021 (5.0) Individual Demographics Homemaker -0.774 (-2.0) Female 0.296 (1.9) Longtime resident (> 5 years) -0.483 (-2.8) -0.493 (-3.2) Has mobility problem 0.405 (1.4) 0.748 (1.8) Age less than 35 years 0.428 (2.1) Trip Characteristics Group travel 0.737 (5.8) Weekend trip 0.988 (2.2) Makes stop for groceries -0.407 (-2.7) Makes stop for other reasons 0.349 (2.0) -0.407 (-2.7) Non-commute trip 0.374 (1.8) -0.668 (-3.7) Traveler Attitudes and Latent Variables Pro-transit attitude 0.499 (4.1) 0.503 (4.4) 0.567 (5.2) Consciousness savings 0.434 (4.0) 0.229 (2.3) 0.154 (1.6) Pro-car attitude -0.312 (-2.5) -0.242 (-2.2) -0.155 (-1.0) Transit averse

F-14 Characteristics of premium Transit Services that Affect Choice of Mode Explanatory Variable Chicago Charlotte Bus Rail Bus Rail Low transit comfort level -0.316 (-2.8) -0.318 (-1.7) Willing to walk not more than 2 minutes -0.679 (-1.7) Willing to walk 10 or more minutes 0.609 (3.5) 0.706 (5.1) 0.356 (2.1) 0.443 (1.6) Very informed about transit 0.301 (1.3) Error Correlation Rho 0.798 (12.1) 0.737 (5.8) Model Fit Statistics Log-likelihood (final) -265.6 -236.8 Log-likelihood (constants) -341.0 -287.0 Pseudo rho-squared 0.221 0.175 Number of observations 584 550 It is interesting to note that there are no household demographics that significantly impact the consideration of transit alternatives. It appears that individuals are largely indifferent to household influences when it comes to considering different modes of transportation. Rather, it is their own demographic and attitudinal characteristics, and the characteristics of the trip they are undertaking, that influence consideration of transit alternatives. Group travel is positively associated with consideration of rail in Charlotte (relative to bus), while no such differential impacts are seen in Chicago. It appears that individuals in Chicago consider bus and rail as equally conducive (or not conducive) to group travel, while those in Charlotte view rail as being more conducive to accommodating group travel. Charlotte respondents are more likely to consider rail for weekend trips (relative to bus); further research is warranted to fully explore the context of weekend travel that may explain this differential in consideration between the transit modes. Making a stop for groceries or other reasons generally reduces consideration of rail alternatives in the Chicago context, possibly because of the difficulty associated with accomplishing multi-stop trip chains using rail. Making a stop for other reasons positively impacts consideration of bus in the Chicago context. Similarly, if the trip is a non-commute trip, then there is a greater likelihood of consideration of the bus alternative in the Chicago context. It appears that bus is viewed as a fairly flexible and accessible mode of transport in the Chicago context and hence the greater level of consideration of this mode for multi-stop journeys or non- commute trips. On the other hand, Charlotte respondents are less likely to consider the bus mode for non-commute trips, a finding consistent with expectations in light of the lower levels of transit service (both bus and rail) in the Charlotte area when compared with Chicago. TABLE F-2. (Continued).

Awareness and Consideration Models F-15 It is found that attitudinal factors play a key role in shaping consideration of transit modes. Those with a pro-transit attitude are more likely to consider transit alternatives and this finding is consistent across both geographical contexts. In Chicago, the magnitudes of coefficients in the consideration model associated with this factor are virtually identical suggesting that there is no differential impact of this factor in consideration of bus or rail. In Charlotte, however, pro-transit attitudes are associated with greater consideration of bus travel mode. This is consistent with expectations; given the limited rail service, it is likely that pro- transit attitudes make a differential impact on bus mode consideration. Those who are conscious of the environment and value productivity are found to give greater consideration to the bus mode relative to the rail mode (in both Chicago and Charlotte). Once again, it appears that this attitudinal factor has a net additional impact on bus consideration (similar to awareness) over and above the pro-transit attitude. Those with a pro-car attitude are less likely to consider transit alternatives, particularly in Chicago. It is found that the coefficient associated with rail is less negative than that for bus, suggesting that pro-car individuals are less likely to consider bus relative to rail. In Charlotte, there is no appreciable difference in the consideration of bus versus rail modes as a function of this particular factor. As shown in the previous chapter, the respondents in Charlotte are more auto-centric in general (in their mode usage patterns); this implies that Charlotte respondents are less likely to consider transit alternatives in general, and being pro-car does not have a differential impact between the two transit modes. While the transit aversion factor is not found to affect transit mode consideration, the factor representing a low level of comfort with accessing and using transit is found to significantly impact transit mode consideration in the Charlotte case study. In Charlotte, it is found that a low level of comfort with transit negatively impacts consideration of both transit modes (to a similar degree). However, no such statistically significant indications are found in the Chicago case study suggesting that this factor does not have an additional net impact over and above other attitudinal factors (for Chicago). Consistent with the sparse rail service in Charlotte, those not willing to walk more than two minutes are less likely to consider rail service in that city (relative to bus). On the other hand, those willing to walk more than 10 minutes are more likely to consider bus and rail for their trip in both cities, with the rail consideration consistently higher than the bus consideration. This finding is consistent with expectations as those willing to walk further distances are more likely to consider rail alternatives (due to the desire to access premium mode). Those who are very informed of transit are not necessarily likely to show a different transit mode consideration pattern than those who are not very informed of transit. Perhaps transit information campaigns affect awareness, but do little to affect consideration. Rather it is the individual characteristics, attitudinal factors, and trip characteristics that are more likely to determine consideration of a transit mode alternative. There is a weak positive impact (not statistically significant at the 95 percent confidence level) of being informed about transit on rail consideration in the Chicago context. The error correlation between the two transit utility functions in the consideration model is statistically significant in both cities. This suggests that there are significant common unobserved factors that affect the consideration of both rail and bus. Further research is needed to explore what these common unobserved factors may be and how best they can be accounted for in model specifications that purport to predict consideration choice sets for individuals. An

F-16 Characteristics of premium Transit Services that Affect Choice of Mode example of a common unobserved factor may, for example, be fuel price. Spikes in fuel costs may suddenly spur a die-hard single-occupant vehicle mode user to potentially consider transit alternatives. The individual may look into the possibility of using either bus or rail services, thus potentially influencing the presence of both transit alternatives in the consideration set. In other words, a common unobserved factor (fuel price/cost, which is not included as an explanatory variable in the consideration model) may affect the consideration of both bus and rail alternatives. A rise in fuel price would presumably lead to greater consideration of both bus and rail alternatives. As a result of this common unobserved factor having the same directional influence (positive) on both bus and rail mode consideration, the consideration model error correlation is positive. This is but one example of a common unobserved factor and there may be several other common unobserved factors that influence consideration of transit modes. In the future, it should be possible to enhance model specifications to explicitly include transit competitiveness, auto operating costs, transit service attributes, and other supply measures in the model specifications to better account for such attributes on awareness and consideration. The testing of such enhanced specifications remains a future research exercise. Summary of Awareness and Consideration Models One primary question for the awareness and consideration models is whether the mode choice with and without these constraints on awareness and consideration are substantially better than with these constraints. During the development of these models, a test was performed to estimate mode choice models with and without the awareness and consideration models. These tests showed that the model estimation statistics were improved when awareness and consideration models constrained the choice sets for mode choice: In Chicago, the rho-squared was 0.712 and 0.782 for commute trips and non-commute trips, respectively, with awareness and consideration models to constrain the choice set and was 0.556 and 0.752 without these constraints. In • • Charlotte, the rho-squared was 1.408 and 1.027 for commute trips and non-commute trips, respectively, with awareness and consideration models to constrain the choice set and was 0.686 and 1.241 without these constraints. For rho-squared, all but the non-commuters in Charlotte were improved with the awareness and consideration models. The log-likelihood statistics improved across all segments for both cities. These statistics support the hypothesis that the awareness and consideration models contribute to improving mode choice models.