1 Introduction

With the most extensive public transit system in North America, New York City (NYC) has one of the highest transit ridership in the world. However, in recent years, it has faced challenges of improving bus service to meet steadily increased demand. One solution developed by policy makers and city planners is bus rapid transit (BRT). NYC’s BRT system is called select bus service (SBS). The SBS system had gradually introduced BRT elements on its routes. Unlike other full-featured BRT systems around the world, NYC’s SBS employs basic BRT elements with relative flexibility. For example, not all SBS routes have off-curb exclusive bus lanes. On the M15 SBS route, both off-curb bus-only and curbside bus-only lanes are implemented. From South Ferry to Houston Street on the M15 SBS route, there is no bus-only lane for SBS buses. According to previous studies (Hensher and Golob 2008), NYC’s SBS can be classified as a light BRT rather than a full BRT system.

For the SBS system, like any other transit service, achieving the goals of increasing mobility and transporting a large number of passengers essentially depends on attracting and retaining ridership. The important role of service quality of public transit in retaining and attracting riders has been well demonstrated by researchers (Hensher and Golob 2008; TRB 2013). Rider perceptions of service quality and performance attributes provide information for determining which attributes will ultimately contribute to ridership growth (Redman et al. 2013; TRB 2013). Therefore, surveys and analysis of rider perceptions of transit attribute-specific performance are commonly used by researchers. Rider perceptions are usually measured by their satisfaction levels. Meanwhile, determining the influence of rider perceptions of attributes on their overall judgment of service quality can help to identify the most important attributes that positively contribute to riders’ higher overall satisfaction.

Transit service is characterized by many diverse attributes and rider satisfaction with some of these attributes can be measured directly, while this might be difficult for other attributes. Among the methods that have been widely used in different fields, structural equation modeling (SEM) is the most appropriate method for capturing the linkages between variables that can be measured directly and those that cannot be measured. Moreover, because attribute-specific satisfaction levels are potentially correlated to each other, SEM is a good approach to account for the correlations between them.

Based on a city-wide survey, this study proposes and tests conceptual constructs using SEM to understand relationships between perceived performance of SBS attributes, overall satisfaction, and customers’ socio-demographic information. As more BRT corridors are to be implemented (NYCDOT 2013), lessons learned from users of current SBS routes can provide operators and authorities with useful information for future service improvement and planning. This study built three separated conceptual constructs considering heterogeneity in and out of Manhattan and the difference made by awareness and unawareness of travel information. From a standpoint of literature reference, results of this study could be useful for cities in similar situations with limited resources to plan or justify transit service taking advantages of some BRT features.

2 Literature review

2.1 A brief introduction to applications of SEM in public transit passenger satisfaction studies

Structural equation modeling has been widely applied in social science, psychology, marketing, management, economics research, natural science, etc. since the 1970s (Joreskog 1973; Wiley 1973). It is a method for establishing, estimating, and testing causality. SEM uses a series of multivariate techniques which include measurement modeling, factor analysis, regression, and path analysis (Golob 2003; Hou et al. 2014). A SEM can contain both observable variables and unobserved variables (latent variables). As a multivariable analysis method, there is no limitation to the number of endogenous and exogenous variables in an SEM. Some of SEM’s advantages make it an appropriate tool for this study. First, SEM allows measurement error in endogenous variables and exogenous variables (Golob 2003). This is beneficial, because measurements of attitude, perception, and behavior using a single indicator often contain errors. SEM estimates a measurement model and a structural model simultaneously. Specifically, the former describes how latent variables are represented by observed variables using confirmatory factor analysis (CFA). At the same time, the latter analyzes the relationship between latent variables by multiple regressions (Qureshi and Kang 2015). Hence, a latent variable could be a dependent variable of some factors and an independent variable in another relationship (Hou et al. 2014). Moreover, unlike traditional regression analysis, SEM can compare and evaluate different theoretical models. Once the conceptual framework is built, using the overall goodness of fit, we can modify and determine the model that is more consistent with the relationship presented by the data.

SEM was applied in other areas of transportation research, e.g., travel demand, organizational behavior, driver behavior, before it began to be applied to model customer satisfaction with public transit. But it has considerable potential for being very useful in this area. While there is a small number of SEM applications modeling public transit rider satisfaction, few of these applications studied BRT rider satisfaction. Table 1 summarizes the service, country, sample size, and latent variables of related studies. Attributes having major effects on customer satisfaction are marked with asterisks. Most of these studies analyzed riders of one mode as a homogeneous group without considering the possible heterogeneity between routes or riders.

Table 1 Studies of customer satisfaction of public transit applying SEM
Full size table

2.2 The impact of travel information and locations of bus routes on public transit

The impact of travel information on public transit has attracted researchers’ growing interest. For various transportation systems, researchers investigated the effects of travel information on ridership (Tang and Thakuriah 2012), route choice (Ben-Elia et al. 2013), and travel behavior (Wang et al. 2015; Zheng et al. 2015), and the situations in which travel information has significant value (Soriguera 2014). Existing studies identified the benefits brought about by implementing travel information in public transit. Real-time transit information was found to significantly increase bus passengers’ level of satisfaction in Tampa, Florida (Brakewood et al. 2014). The real-time bus information systems in Chicago and New York City (NYC) increased bus ridership (Tang and Thakuriah 2012; Brakewood et al. 2015). Therefore, travel information provision as a policy for increasing passenger satisfaction and ridership has been promoted in more and more public transit systems.

All research on the effects of travel information on public transit is based on passengers’ awareness and use of it. However, in reality, there are passengers who are unaware of travel information whether it is schedule information provided at bus stops or real-time information on the Internet and who thus do not use this information. Some surveys and studies have pointed out that the use of travel information is limited to certain passengers with awareness of the information and who have access to it. Tang and Thakuriah (2012) recognized that their research was limited by the possibility that many travelers were unaware of real-time information of Chicago’s bus system. A survey conducted in Staten Island, NYC, by Metropolitan Transportation Authority in 2012, showed that 69 % of all riders surveyed were aware of Bus Time, the real-time information platform in NYC, and only 44 % had used it (Brakewood et al. 2015).

Previous research investigating user satisfaction with existing and potential public transit shows that the degree to which performance attributes are valued varies according to the type of user and the location of the bus corridor (dell’Olio et al. 2011; Pandit and Das 2013). The findings of a market segmentation study of travel information use in Seoul, Korea, indicate that socioeconomic and contextual variables are important for understanding of information acquisition behavior (Pandit and Das 2013). The study of travelers’ responses to advanced travel information in the city of Calgary, Alberta, Canada, found that several factors have significant effects on their decisions (e.g. driving experience, trip purpose, and desire for information) (Kattan et al. 2013). Hence, the awareness and access of travel information and potential heterogeneity between routes in and out of Manhattan are considered for categorizing riders.

3 Study scope, conceptual framework, and hypotheses

3.1 Study scope

Four SBS routes along the initial six corridors of Phase I were chosen for this study. They are M15 SBS and M34/M34A SBS in Manhattan, Bx41 SBS in Bronx, and B44 SBS in Brooklyn. These routes are selected because more BRT elements have been implemented on them than on other SBS routes. They are shown in Fig. 1.

Fig. 1
figure 1

Target SBS routes

Full size image

Based on the results of the pilot survey, surveys conducted by MTA New York City Transit (NYCT) in 2010 and 2011 (MTA NYCT 2011), and an extensive literature review, thirteen attributes were selected. The thirteen selected attributes are: How often buses arrive (frequency), speed of a passenger’s ride on the survey day (speed), adherence to the scheduled arrival and departure times (on-time performance), exclusive right of way (bus-only lanes), signs on off-board ticket machine telling riders to purchase a ticket before boarding (signs on ticket machine), single or double shelter/s at bus stations (shelter), buses with three doors to facilitate faster boarding and unloading of passengers (three-door buses), ease of using off-board ticket machine (ticket machine use), overall comfort of riding and regularity of cleaning and maintenance (comfort and cleanliness), availability of route and schedule information at bus stops (route and schedule information), distance that riders traveled to and from the origin and destination bus stops to finish their SBS trips (proximity of bus stops), utilization of real-time information provided via web and app (real-time information), and longer distance between stops compared to local and limited bus service.

On public transit, all riders are offered the same service. But rider perceptions of service quality tend to differ among different user groups (Andreassen 1995). From the results of pilot survey and a survey done in 2012 by MTA in Staten Island, NY, 44 % of SBS passengers have never used the real-time information (RTI) provided via web and mobile devices (Brakewood et al. 2015). Some passengers neither use nor have awareness of the route and schedule information (RSI) at bus stops. Thus, according to their experience of using travel information, different hypotheses for different rider types are developed. The hypotheses for Model 1 are for riders with experience using travel information and the hypotheses for Model 2 are for riders with no experience using travel information. Secondly, because the characteristics and locations of SBS routes affect rider perceptions of service (dell’Olio et al. 2011; Pandit and Das 2013), with the assumption that riders’ overall satisfaction is driven by different attributes of perceived performance in specific areas, for Model 1 the hypotheses are tested for SBS routes according to their locations in Manhattan (MModel 1) or outside Manhattan (OModel 1). Riders included in each model are shown in Fig. 2. Moreover, three latent variables for MModel 1 and OModel 1 and two latent variables for Model 2 are defined. Table 2 presents the latent variables and observed variables. Hypotheses for each model are explained in Table 3. Specific explanations of each hypothesis are presented in Sects. 3.2 and 3.3.

Fig. 2
figure 2

Rider classification for models

Full size image
Table 2 Latent variables and observed variables
Full size table
Table 3 Model hypotheses
Full size table

3.2 Hypotheses for riders with experience using travel information (MModel 1 and OModel 1)

The hypotheses made are based on the literature review, the empirical findings of a related survey of NYC SBS, results of pilot survey, and the service characteristics of SBS.

As opposed to the Null Hypotheses (i.e. no relationships between variables), six alternative hypotheses are proposed for MModel 1 and OModel 1. For each test, considering the sign of the coefficient, the Null Hypothesis will be rejected and the alternative hypothesis will be accepted when the p value is less than 0.5.

Hypothesis 1 (H1 1 ): Satisfaction with service quality is positively related to SBS riders’ overall satisfaction.

Several studies of bus service and BRT systems (Currie and Wallis 2008; Kain and Liu 1999; Taylor et al. 2009; Hensher and Golob 2008) indicate that improvement in reliability, increased service levels, and reduced fares have the greatest association with increased satisfaction levels and ridership. Since SBS, local buses, and subways all have the same fare rate in NYC, fare is not considered in this study.

Hypothesis 2 (H1 2 ): Satisfaction with bus stop amenities has a positive influence on riders’ overall satisfaction.

Bus stop features such as off-board ticket machines and safe waiting environments have been found to significantly increase the efficiency of BRT routes and user satisfaction, respectively (McDonnell and Zellner 2011; Iseki and Taylor 2010). Meanwhile, RSI provided at stops was assumed to increase riders’ satisfaction of the overall service. Hence, to measure this latent variable, four observed variables are considered based on the literature review and local conditions. Specifically, they are signs on off-board ticket machines, shelters, ease of using off-board ticket machines, and RSI at bus stops.

Hypothesis 3 (H1 3 ): There is a positive relationship between satisfaction with the other SBS attributes and overall satisfaction.

Bus-only lanes give BRT vehicles an exclusive right-of-way with the goals of reducing delay and travel time for BRT vehicles compared to vehicles in general traffic lanes. Buses with three doors can improve their boarding efficiency. Both characteristics are highlighted as important elements of BRT for increasing service levels (Currie and Delbosc 2011). The comfort and cleanliness of vehicles is an attribute discussed in investigations of several public transit modes. A study of the public transit services of Chicago Transit Authority shows that improvement of comfort significantly increases customer satisfaction (Foote 2004). The attractiveness of bus stops tends to be positively related to their proximity (Shaaban and Khalil 2013). However, while limited/express bus stops can improve SBS operations by reducing travel time, they can increase walking distances for some riders (McDonnell and Zellner 2011). Because of the existence of local bus service along SBS corridors, limited bus stops are supposed to have a positive influence on riders’ overall satisfaction. Real-time information which helps riders to estimate bus arrival times has the potential to positively contribute to customer satisfaction (Eboli and Mazzulla 2007). As shown in Table 2, six attributes are determined for this latent variable.

Hypothesis 4 (H1 4 ): Satisfaction with service quality is positively related to satisfaction with stop amenities.

Hypothesis 5 (H1 5 ): Satisfaction with service quality is positively related to satisfaction with the other SBS attributes.

Hypothesis 6 (H1 6 ): Satisfaction with stop amenities is positively associated with satisfaction with the other SBS attributes.

When BRT elements are implemented together, they reinforce each other (McDonnell and Zellner 2011). Since most of the observed attributes of stop amenities and other SBS features are intended to increase travel convenience and reduce trip time, a positive relationship was expected between them. The literature (Stuart et al. 2000; Karlaftis et al. 2001) supports the positive correlation between satisfaction levels with attributes of public transit service.

Figure 3 presents the proposed conceptual structure of Model 1 and Model 2.

Fig. 3
figure 3

Conceptual structure

Full size image

3.3 Hypotheses for riders with no experience using travel information (Model 2)

Hypothesis 1 (H2 1 ): Overall satisfaction is positively influenced by satisfaction with service quality and the ticket system.

For riders who have not used travel information, we expect that their overall satisfaction is also positively affected by service quality. Moreover, from the regression analysis results of the final report of the project related to this study (Wan 2015), these riders’ overall satisfaction is positively affected by the signs for off-board ticket machines and the ease of using them. So the first hypothesis for Model 2 is as follows.

Hypothesis 2 (H2 2 ): Satisfaction with the other SBS attributes positively influences overall satisfaction.

Even though we expect that the other SBS attributes may have only a minor influence on satisfaction, these attributes should not have a negative impact on rider satisfaction levels. As explained in Sect. 3.1 above, several studies have shown that these other variables have positive effects on satisfaction.

Hypothesis 3 (H2 3 ): A positive correlation exists between satisfaction with service quality and the ticket system and satisfaction with the other SBS attributes.

Similar to Model 1, a positive correlation is hypothesized between satisfaction with service quality and the ticket system and satisfaction with the other SBS attributes.

This model’s conceptual structure is also presented in Fig. 3.

4 Sample survey and analysis

4.1 Data collection

A survey was designed to collect SBS rider socio-demographic characteristics and satisfaction levels of SBS’s overall service and attribute-specific performance. Respondents are requested to rank their overall satisfaction from 1 to 10, with 1 representing the lowest and 10 representing the highest. A Likert scale (1–5) is used to measure the riders’ attribute-specific satisfaction as follows: 1 = Very Unsatisfied; 2 = Unsatisfied; 3 = Neutral; 4 = Satisfied; and 5 = Very Satisfied.

1700 SBS riders were randomly interviewed at bus stops and on the buses during both peak and off-peak hours of weekdays and weekends. The number of records for each route is as follow: 435 for M15 SBS, 413 for M34/M34A SBS, 450 for Bx41 SBS, and 406 for B44 SBS. The overall response rate was 90.6 %. Considering the average annual ridership of these four SBS routes as our study population, our sample sizes are significant at a confidence level of 95 % for each route (Tyrinopoulos and Antoniou 2008; Johnson and Wichern 2007).

4.2 Characteristics of respondents

After data selection, the sample sizes of MModel 1, OModel 1 and Model 2 are 493, 495, and 454, respectively. Based on studies of SEM sample size (Boomsma 1982; Comrey and Lee 1992; Wolf et al. 2013), it is concluded that our sample size is considered to be good for SEM. The socio-demographic information for the respondents included in these three models is illustrated in Fig. 4. Table 4 presents descriptive statistics for riders’ satisfaction levels.

Fig. 4
figure 4

Socio-demographic information of respondents

Full size image
Table 4 Descriptive statistics for satisfaction levels
Full size table

It is noteworthy that:

  1. 1.

    Model 2 is built for SBS riders who have not used travel information. Compared to riders who have used this service, the rider composition of Model 2 has less young riders under 30 (24.4 %), male riders (40.1 %), and riders traveling for work or school (51.3 %), but more riders taking SBS less often (26.62 %). The percentages for each of these characteristics for the riders used in MModel 1 are 32.7, 49.1, 63.1, and 21.2 %, respectively. For OModel 1, they are 39.8, 42.8, 66.9, and 11.2 %, respectively. A comparison of socio-demographics of riders between model groups reveals that young riders (18–29 years old) are less likely to be included in the Model 2, indicating that young drivers are more likely to use travel information while making a transit trip. Male riders and riders traveling for work and school tend to use travel information more than female riders and riders traveling for other purposes. Riders who had used the bus service that was replaced by SBS are more likely to use travel information.

  2. 2.

    Riders on SBS routes outside of Manhattan are younger (39.8 %), more likely to have used the bus service that was replaced by SBS (16.3 %) and use SBS more frequently (71.8 % with a riding frequency of at least 4 days per week) than the rider groups for the other two models.

  3. 3.

    Riders on SBS routes in Manhattan are more likely to be male (49.1 %), more likely to have not used the previous bus service (37.6 %), and more likely to be traveling to work (55.2 %) than the riders for the other models.

4.3 Testing for reliability

Cronbach’s Alpha was calculated for each latent variable in the three models to assess their strength and adequacy. A Cronbach’s Alpha value between 0.70 and 0.90 indicates an acceptable level of internal consistency (Al-Refaie 2015; Nunnally and Bernstein 1994). The results of the reliability test for the three models are given in Table 5.

Table 5 Cronbach’s alpha values
Full size table

As shown on Table 5, the Cronbach’s Alpha values are all within the acceptable range. Hence, the three models have sufficient reliability to warrant further analysis.

4.4 Empirical results and model refinement

Each structural model consists of two parts. The first part is the measurement model to define relationships between observed and latent variables. Specifically, it measures the relationships between SQ, SA, OA and their observed variables for MModel1 and OModel 1. As for Model 2, it is the evaluation of relationships between ST, O and their observed variables. The second part is the structural model. It evaluates the relationships among latent variables and their associations with overall satisfaction.

The first trials for MModel 1, OModel 1 and Model 2 were analyzed using IBM SPSS Amos 23. A probability level less than 0.001 indicates that the model has a good fit. In the literature, some studies recommended the use of the goodness of fit (GOF) measures to determine the model’s fitness. The first measure is the Normed Chi square (NC), which is the ratio of Chi square over the degree of freedom (DF). Bollen (1989) proposed that a value between 2.0 and 3.0 indicates an acceptable fit. Then, a root mean square error approximation (RMSEA) below 0.05 indicates an excellent model fit, while a value between 0.05 and 0.08 suggests a reasonable fit (Brown and Cudech 1993). The other two measures are the comparative fit index (CFI) and the goodness of fit index (GFI). The values of these indices range between 0 and 1. The closer a value is to 1, the better a model fit is. When the values are greater than 0.9, the models have a good fit with the data (Hu and Bentler 1999). Finally, the parsimonious goodness of fit index (PGFI) and the parsimonious normal-fit index (PNFI) above 0.5 indicate an acceptable fit between the measurement model and the data (Qureshi and Kang 2015).

The Modification Index (MI) indicates the covariance of error (Al-Refaie 2015). Reviewing the MI values, improvements were made to the first measurement models of SEMs to achieve a better model fit. Covariance between error pairs which have large MI values was included. The final trials of the three SEMs with overall satisfaction are illustrated in Fig. 5. The direct effects of latent variables on the overall satisfaction are represented by unidirectional arrows. Double-headed arrows indicate the covariation between latent variables, which test the association relationships between them. The GOF indices are shown in Table 6. All the GOF indices are within the acceptable ranges, which indicates an acceptable fit of the three SEMs and their data. Thus the SEM results can be used to test the hypotheses.

Fig. 5
figure 5

Final trials of the three SEMs

Full size image
Table 6 Overall fit indices for the three structural models
Full size table

5 SEM results and discussion

5.1 SEM results

Figure 6 depicts the final structure models with estimates of standardized regression weights. The coefficient values and significance indicate the relationships between variables. For illustration, in MModel 1, when the satisfaction level with service quality increases by 1, overall satisfaction will increase by 0.615 at a confidence level of 0.001.

Fig. 6
figure 6

Standardized structural models

Full size image

Table 7 presents the final results for the measurement model parts of the three SEMs. All the observed factors have significant loading on their latent variables at the confidence level of 0.001. A greater value of the estimate indicates a more important role for the observed variable in its latent variable. For instance, in MModel 1, when the satisfaction level with service quality (SQ) goes up by 1, the satisfaction level with frequency (SQ1) goes up by 0.775, while the satisfaction level with speed (SQ2) goes up by 0.678 (Al-Refaie 2015). This indicates that satisfaction with frequency is relatively more important for its latent variable.

Table 7 Measurement model estimates
Full size table

5.2 Main findings

The results show that some of the proposed hypotheses for the three types of SBS riders in NYC are supported by the models which were developed using the survey data. The results are shown in Table 8. A critical ratio (C.R.) value more than 1.96 indicates a significant relationship. Details of the findings are discussed in the following sections.

Table 8 Results of SEM hypotheses
Full size table

Satisfaction levels with service quality are positively related to overall satisfaction at confidence level of 0.001. The acceptance of H11 is not surprising. It indicates that frequency, speed, and on-time performance are the significant factors that positively affect riders’ overall satisfaction in urban areas, which is consistent with findings of previous studies on bus or public transport (Currie and Wallis 2008; Currie and Delbosc 2011; de Oña et al. 2013; Del Castillo and Benitez 2013; Eboli and Mazzulla 2011; Mouwen and Rietveld 2013). The larger coefficient of SQ in MModel 1 than in OModel 1 indicates that service quality is more important for SBS assessment in Manhattan. Considering the uniqueness of riders’ socio-demographics, SBS features, and land use of each model, some possible explanation could be given for the difference between models for Manhattan and outside Manhattan. As demonstrated in Fig. 4, SBS routes in Manhattan have more riders with no riding experience of previous replaced limited bus lines and more riders using SBS less frequently (at most 3 days per week). These riders have less familiarity with SBS features and pay more attention to service quality related attributes. Not like BRT systems in other cities, the SBS system in NYC gradually launches BRT features. The target SBS routes are not fully featured with all BRT attributes. SBS routes in Manhattan have relatively more BRT attributes than routes outside Manhattan. This makes less difference among SBS stations in Manhattan. What is more, the land use of areas served by routes in Manhattan are mainly commercial while outside Manhattan are mostly residential (New York City Planning). Combining with riders’ trip purposes, it could indicate that SBS riders in Manhattan are more time-sensitive focusing on attributes that can increase travel efficiency directly. Moreover, frequency and on-time performance are more important than speed. Hence, to increase riders’ overall satisfaction in Manhattan, transportation planners and operators should consider adopting enhancements in service quality attributes in light of how important these attributes are. Ultimately, service quality enhancements should result in ridership growth (TRB 2013).

The rejection of H12 and H13 in MModel 1 suggests that, for SBS riders in Manhattan, satisfaction levels with station amenities and other attributes do not directly contribute to overall satisfaction. Thus, improvements in these attributes will not significantly affect riders’ overall satisfaction. But for SBS riders outside Manhattan, acceptance of H13 suggests that the satisfaction level with the other attributes positively influences riders’ overall satisfaction. As mentioned in the last paragraph, riders in Manhattan might be more time-sensitive and their perceptions are not significantly affected by attributes that affect travel efficiency indirectly. Meanwhile, outside Manhattan, SBS riders experienced relatively more differences among stations and there are less SBS dependent riders who use SBS at least four days per week. Hence, for riders outside Manhattan, besides the service quality related attributes, improving or developing other attributes could also positively affect riders’ overall satisfaction. Among them, proximity of bus stops, buses with three doors and limited stops are more important.

The covariance between latent variables is significant at a 99 % confidence level. The acceptance of H14, H15, and H16 suggests that BRT elements reinforce each other.

The respondents for MModel 1 are from the two SBS routes serving the central business district of NYC. So the results may also reflect patterns found in similar routes. Generally speaking, for SBS riders in Manhattan, the satisfaction level with service quality is the only direct determinant of overall satisfaction. Thus, as for increasing riders’ satisfaction levels with SBS, improving the performance of service quality attributes should be the priority. Among the service quality attributes, increased frequency and bus reliability (on-time performance) are the major factors most highly rated by riders. Meanwhile, the improvement of performance of the other attributes could reinforce the positive effects of improving service quality. This is also supported by other studies of BRT, which indicate that the aggregate implementation of enhancement measures would result in greater efficiency and attractiveness than individual implementation (McDonnell and Zellner 2011).

The significant positive standardized regression weight of ST provides evidence for the acceptance of H21. Satisfaction levels with service quality have larger (positive) coefficients than attributes pertaining to off-board ticket systems. The most consistent finding from the complete models is that frequency and on-time performance have more important influences on perceptions of overall service than other attributes. Satisfaction with speed was the next most important attribute having a major effect on overall satisfaction. Signs on ticket machines requesting that customers buy a ticket before boarding have a more important role in increasing overall satisfaction than the ease of using the machines. Improvement in ticket machine-related attributes is believed to be beneficial (McDonnell and Zellner 2011). Specifically, accepting credit cards, selling and refilling metro cards, eye catching signs on machines asking passengers to buy a ticket before boarding, frequent checks of paper supply, and lighting around off-board ticket machines could be priorities. Some of these improvements were also pointed out in a previous survey and phase II plan (Metropolitan Transportation Authority 2011; NYCDOT and MTANYCT).

H22 was rejected. When service quality and ticket systems are kept unchanged, efforts to improve other attributes might result in lower overall satisfaction. A possible explanation of the negative coefficient is that, with less frequent riders and fewer riders under 30 years old, a certain portion of riders for Model 2 may not have enough knowledge and familiarity about the effects of these attributes. In the instance of bus-only lanes, because they are not physically separated from other lanes, some riders do not fully understand their effects on reducing travel time and increasing schedule reliability. Riders who used the SBS more frequently paid more attention to the enforcement of bus-only lanes especially in Manhattan.

The H23 was accepted. Although the covariance between these two latent variables was relatively small, higher satisfaction with other attributes is positively related to higher satisfaction with service quality and the ticket system.

5.3 Limitations

In this study, applying SEM to investigate relationships between unobserved variables and SBS riders’ overall satisfaction has several limitations. First, the observed variables in this study are limited to those that could be easily perceived by riders. And, considering the distance between stops and the extent of respondents’ patience, the questionnaire was designed to be finished in 2–3 min on average. Hence there could be some unmeasured variables that have potential significant effects on riders’ overall satisfaction. Second, the use of satisfaction levels with perceived service performance as the measurement for satisfaction could be biased. Riders’ perception could be affected by many personal factors, such as previous critical experience (Friman et al. 2001), psychological conditions (responding emotionally rather than rationally), and knowledge. To a certain extent, the application of SEM can manage these issues. Finally, as a confirmatory approach to verify the logical or conceptual structure, SEM is only able to test hypotheses. The revealed influences of variables are not unquestionable (Qureshi and Kang 2015). The results cannot be used for prediction, but only for understanding the underlying mechanisms.

6 Conclusions

Empirical evidence shows the positive relationship between passenger satisfaction on public transit and its ridership growth. When the success of a public transit system highly depends on the number of passengers it transports, enhancing passenger satisfaction is crucial for transit planners and operators who want to encourage travelers to shift from other modes and to retain current riders. In order to help prioritize the most important attributes positively related to customer satisfaction with SBS, a “light” BRT service, this study applies SEM to data collected by a survey of NYC SBS.

A series of three conceptual structures was proposed based on a literature review and a pilot survey. The three structures are nested with two levels considering two important factors: riders’ awareness and access to travel information and locations of SBS routes. These structural models aim to investigate relationships between unobserved variables and their effects on overall satisfaction. Hypotheses were developed to explore variables that could contribute to higher overall satisfaction. Models were tested using the survey results. The findings provide useful information for identifying the specific attributes which are most concerned by customers when they evaluate the overall BRT service.

The results derived from modeling show different patterns for different types of SBS riders. For all riders, offering quality BRT service is essential to increasing customer satisfaction. This is to increase riders’ overall satisfaction from the supply side. Specifically, introducing more SBS buses could increase the frequency of bus arrival. Installation of cameras enforcing the right of way for exclusive SBS-only lanes could improve buses’ adherence to their schedules and maintain their high speeds. Strict enforcement by the police department could also help with this. When capital funds and the required space are available, off-curb and physically separated bus-only lanes could have a significant impact on satisfaction levels and ridership (dell’Olio et al. 2011). Given that, in the three structural models, frequency and on-time performance consistently have larger contributions to satisfaction with service quality than speed, increasing frequency and improving on-time performance would be more efficient for increasing passenger satisfaction. However, these service quality improvements are usually costly. And there are delicate relationships between different enhancement measures. Infrastructures increasing speed and frequency need relatively higher initial cost but could reduce operating cost as less driving time would be needed. In the meantime, it is possible that the cost would be higher if only bus frequency has been increased with no punctuality improvement.

The relationship between higher overall satisfaction and other significant attributes helps to prioritize the important attributes for enhancement with limited capital resources. For routes outside of Manhattan, proximity of bus stops, real-time information and limited stops are relatively more important than bus-only lanes, buses with three doors, and bus comfort and cleanliness. In urban areas, when walking environments have favorable features for riders, they will walk farther to BRT stations (Jiang et al. 2012). Hence it is important to consider urban design when determining where to locate stations. Currently, real-time information panels are under design for NYC SBS stations. They will increase rider satisfaction and achieve further development of the BRT market. Besides service quality, bigger, clearer and eye catching signs on off-board ticket machines are important for riders who have not used travel information. Meanwhile ticket machines accepting cash and selling metro cards are desired by riders. In addition a regular check of ticket machines can avoid the inconvenience caused by running out of paper.

The ultimate goal of all the attributes enhancement measures is to increase customer satisfaction and increase ridership. From a point of demand, it is noteworthy that this study focuses on current SBS riders and has no information about the potential riders’ mode choice and travel behavior changes. Ridership increases might mainly result from attribute improvements encouraging more frequent SBS use of transit dependent customers rather than a shift from private cars.

To conclude, based on the study findings for riders on different routes in and out of Manhattan and those with and without awareness of travel information, diverse tactical strategies for improving important attributes could be employed to increase satisfaction and to generate more ridership. Meanwhile, the study results show that the good performance of the SBS attributes has positive effects on each other. As an application of SEM to study customer satisfaction with a light BRT system, this study shows the feasibility of this modeling approach in other large cities deploying similar BRT systems.