Park-and-Ride Access Station Choice Model for Cross-Regional Commuter Trips in the Greater Toronto and Hamilton Area (GTHA)

Transcription

1 Park-and-Ride Access Station Choice Model for Cross-Regional Commuter Trips in the Greater Toronto and Hamilton Area (GTHA) Mohamed Salah Mahmoud, M.Sc. PhD Candidate Department of Civil Engineering University of Toronto mohamed.mahmoud@utoronto.ca Khandker Nurul Habib, Ph.D., P.Eng. Assistant Professor Department of Civil Engineering University of Toronto Khandker.nurulhabib@utoronto.ca Amer Shalaby, Ph.D., P.Eng. Professor Department of Civil Engineering University of Toronto amer@ecf.utoronto.ca Number of Words (Maximum,00):, (words) + (Tables)*0 + (Figures)*0 =, TRB 01 Annual Meeting

2 Abstract The paper presents an investigation on park-and-ride (P&R) access station choices of crossregional commuter in the Greater Toronto and Hamilton Area (GTHA). Data from a household travel survey conducted in 00 in the GTHA is used for empirical investigation. The household travel survey data is supplemented by data from transit service operators regarding P&R station locations, parking lot capacities, parking costs, surrounding land use, and station amenities. Three groups of park-and-ride users are defined herein: individuals who have only local transit (TTC Subway) stations within a reasonable reach, individuals with only regional transit (GO Rail) stations within reach, and individuals who have both GO Rail and TTC Subway stations within reach. Different model structures and specifications are tested and three discrete choice models are estimated. Empirical models reveal that access distance and the relative station direction (toward the work place) are the primary factors affecting transit station choice for parkand-ride options. However, between station distance and relative station direction, commuters are more sensitive to changes in station access distance than to changes in the relative station direction from their households. In addition, the empirical models reveal that local transit parkand-ride users are less sensitive to access distance than regional transit park-and-ride users. The results of this investigation can be useful in future transit station design projects in order to attract more commuters to use park-and-ride. TRB 01 Annual Meeting

3 Introduction With the rapid growth of large cities, the continuous sprawling and extension of suburban neighborhoods around cities central areas, multimodal cross-regional trips shares have increased substantially (1; ). A typical cross-regional trip may involve the use of multiple transit services or the interaction between two different modes. For inter-modal trips, transit access mode is an important component in defining individuals travel choices. Individuals may have options for access modes such as transit feeder services, active modes (walk or bike) or automobiles. However, unlike intra-regional commuting travel (trips originating from and destined to the same region), cross-regional commuters have a limited set of possible travel modes since non-motorized modes are mostly infeasible. Further, cross-regional trips often originate in suburban areas where transit accessibility is usually inadequate. Transit modal integration, including intra-modal integration (e.g. local transit with regional transit) and intermodal integration (e.g. transit with automobiles or active modes) are promising strategies to address this challenge. Transit modal integration refers to facilities that provide combinations of transit services and other motorized or non-motorized modes. The use of private cars for transit access (i.e., parkand-ride) is the most flexible form of modal integration (). Park-and-ride (P&R) facilities have become an important component of urban transportation systems as they facilitate inter-modal integration and provide access to transit for users who may not otherwise consider transit as a travel alternative (-). Typically, park-and-ride facilities are located at major transit hubs which are characterised by wide spacing among facilities making station access choice more complex. Access station choice, a key component of park-and-ride for commuting, is not an isolated choice from the choice of park-and-ride as a travel mode. Commuters choose the access station in accordance with both main mode (transit) and access mode (automobile) choices. In other words, Which station to choose to access the transit system? is an internal/endogenous question of the transit users who choose automobile as their access mode. Many factors may affect this choice including trip maker characteristics, station attributes, station orientation relative to individual s home and work locations, surrounding land use, and the quality of the contiguous transportation network. Previous studies showed that the higher the transit service frequencies along with better station amenities/facilities and network connectivity, the more the station becomes attractive as an access point (). Auto access to regional transit services has become popular in large cities. In the Montreal region, % of the commuter rail users access the system by their private automobiles (). Similarly, in the Greater Toronto and Hamilton Area (GTHA), 1% of GO Rail commuters station access is done by automobiles including park-and-ride, kiss and ride (K&R) and carpool (). In the GTHA, more than 0% of cross-regional commuters who access rail transit (GO Rail and TTC Subway) by their cars choose a station that is not the closest station to their home locations (). Therefore, a better understanding of individuals transit access station is needed to enhance travel demand forecasting at the station level. This can be useful to define factors that affect individuals choices of access stations as well as stations catchment areas. From the transit service provider s point of view, this can be beneficial for station development and/or service improvement planning. Previous studies showed that the current state of practise of modelling access mode and access location choices is not adequate for explaining complex travel 1 TRB 01 Annual Meeting

4 options such as park-and-ride. This stresses the importance of studying the access component of the trip in order to fill this gap in knowledge by developing high-fidelity operational demand models which are essential for planning activities. In this paper, access station choice of park-and-ride cross-regional commuters in the GTHA is examined. Three multinomial logit models were developed for regional commuter rail (GO Rail) park-and-ride access station choice, local transit (Toronto Transit Commission TTC Subway) park-and-ride access station choice, and a model for those commuters with possible access to both TTC Subway and GO Rail. Literature Review Park-and-ride is a complex modal option that has been subject to a wide interest in the literature (-1). Park-and-ride, as a mode choice option, is often treated under the umbrella of access to transit. Previous studies showed that the park-and-ride mode is a complicated modal option that requires focused investigation (1). Li et al. (00) applied a network equilibrium formulation to model park-and-ride in a multimodal network context considering three modes: auto, transit with walk access and park-and-ride (1). The study showed that traditional demand models often misestimate park-and-ride demand. The study concluded that introducing park-and-ride facilities may affect the overall system, positively or negatively, depending on several factors including parking charges and offered spaces. Vijayakumar et al. (0) conducted a study with an objective of understanding the variables that affect driving distance to suburban rail stations and their demand in the Montreal region (). A multivariate regression model was developed to measure how individuals socioeconomic as well as station characteristics affect driving access to stations. Results showed that additional parking spots at park-and-ride locations and higher train frequencies attract more users to drive further to rail stations. In addition, better street connectivity to a rail station, as a measure of station accessibility, contributes to the total demand served by this station. However, previous studies have not investigated travel choice trade-offs within a multimodal network with focus on auto access to transit. A limited number of explicit park-and-ride access station choice models are found in the literature. Kastrenakes (1) studied rail station choice for New Jersey transit (1). The developed multinomial logit model incorporated variables including station access time, frequency of service at the boarding station, station location relative to home location, and generalized trip cost from access station to the final destinations. However, other variables such as parking availability and parking fees showed counterintuitive signs and were dropped from the model. With more focus on parking attractiveness for station choice, Wardman and Whelan (1) studied rail station choice for inter-urban trips in London (1). The study concluded that parking availability and better passenger related facilities are important features that make stations more appealing to trip makers. Station access location is often modelled conditional on access mode choice in a nested model structure with access mode choice in the upper-level and station choice in the lower-level. Mukundan (11) developed a nested logit model to study access mode and station choice of Metro rail trips in Washington, D.C. (1). The access station choice set was defined, based on TRB 01 Annual Meeting

5 predetermined modal impedance functions, as the two best access stations for the walk mode and the six best access stations for all the other modes. Fan et al. (1) studied access mode and access station choices of rail/subway morning peak period work trips in the Greater Toronto Area (GTA) (1). A multinomial logit model was developed to model subway automobile access station choice from a choice set which consists of the closest five stations. However, for commuter rail users, the two closest stations on the two closest lines defined commuter rail users choice set. Then, a nested logit model with access mode choice at the upper level and access rail station choice at the lower level was developed. Although this study has provided a solid foundation for future research in station choice modelling, it suffered from some limitations. Both models did not incorporate station-specific attributes such as parking prices or integration with other transit services. Further, the models were developed using service level attributes which are obtained from an aggregate traffic assignment tool. Debrezion et al. (00) developed a multimodal logit model to study the Dutch railway passengers choice of departure railway station (). The model showed that the effect of service frequency is relatively small compared to the effect of distance between home and station locations. However, no accessibility measures or station-specific features were introduced within the model. In addition, the study was limited to aggregate choices made by trip makers at household post code area level without considering the trip purpose or the trip destination. In a later study, Debrezion et al. (00) developed a joint model of access mode and railway station choice (1). They estimated a nested logit model that was capable of jointly capturing both access mode and station location choice preferences. Inherent assumption of their model is that commuters decide to take transit first and then choose the access mode as well as departure station. In their empirical model, the provision of parking spaces and bicycle racks showed a significant positive effect on drive and bike access mode choices, respectively. However, both studies did not primarily focus on examining station choice for auto access (P&R) commuters. Further, the sample frame used for the two studies incorporated a significant percent of intercity trips which are different in characteristics than long-distance or cross-regional commuting trips. Recently, Chakour and Eluru (01) studied commuter rail users mode and station choice behaviour in Montreal region using data from an on-board survey (). The study focused on relaxing the hierarchical nature of the joint choice using a latent segmentation approach, in which access mode and station choice are tested to determine individuals choice sequence. Socioeconomic, trip and service level attributes as well as built environment factors were used to model commuters behaviour. The results showed that the developed latent segmentation model has higher explanatory power than the traditional nested logit models. However, the same inherent assumption of Debrezion et al. (00) is also valid for this study. Studies on commuting mode choice in large regions reveal that regional transit with local transit access, regional transit with walk access, local transit park-and-ride and regional transit park-and-ride are often independent modal options. In the GTHA, it is proven that these four modal options are Independent and Irrelevant Alternatives (IIA) (0-). In this paper, we are interested in studying park-and-ride cross-regional commuter trips in the GTHA. Hence, we focus on access station choice to further improve our understanding on parkand-ride mode choices. TRB 01 Annual Meeting

6 Study Area and Data Description Situated directly to the north west of Lake Ontario in the province of Ontario, the Greater Toronto Hamilton Area (GTHA) forms Canada s largest urban region. The GTHA s current population is over million, with a projected growth to approximately. million by 01 (). The GTHA consists of the City of Toronto and five other regional municipalities. It has nine local transit systems and a regional transit service operating under the administration of Metrolinx, an agency of the Government of Ontario, which was created to improve the coordination and integration of all modes of transportation. That is, the GTHA provides a generic case-study of a multimodal integrated transportation network. This study focuses on cross-regional inter-modal commuting trips. The trip, demographic, and socioeconomic characteristics of the study were extracted from the 00 Transportation Tomorrow Survey (TTS) dataset. The TTS is a trip-based household survey conducted every five years in the Greater Toronto and Hamilton Area (GTHA) among % of its population. Crossregional commuter trips (i.e., trips with trip ends in different regions) represent around % of total daily commuting trips, which has grown by % since 1 (1; ).The majority of crossregional commuter trips (around 0%) are made by private automobiles. The dataset provides a detailed disaggregate individual trip records with exact locations of households, employment and access station. Data of cross-regional commuting trips in the morning peak period (:00 :00 Am) on regular weekdays were extracted from the 00 TTS dataset. In order to study individuals access station choice, trip records with transit as the chosen main mode and automobile as access mode were used in this analysis. In the GTHA, park-and-ride users choose between GO Rail park-and-ride locations which serve trips across the region, and/or TTC Subway park-and-ride locations which cater to trips within the City of Toronto. The total number of complete trip/passenger records used in this study is 1,. Figure 1-(a) shows the distribution of household locations (with at least one cross-regional commuting trip) across the GTHA and Figure 1-(b) shows the distribution of their employment locations. It can be clearly seen that most of the cross-regional morning commuting trips originate from the outer suburbs and end in the City of Toronto. 1-(a) Distribution of Household Locations 1-(b) Distribution of Employment Locations Figure-1 Distribution of Household and Employment Locations of Cross-Regional Commuters TRB 01 Annual Meeting

7 Data on park-and-ride station locations, parking lot capacities, parking costs, surrounding land use, and station amenities was obtained from Metrolinx and the Toronto Transit Commission. Figure presents park-and-ride locations of GO Rail and TTC Subway across the region. Figure -(a) shows parking lot capacities and parking cost while Figure -(b) shows morning peak parking demand of cross-regional trips (the number of trip records in the dataset without applying the expansion factors) at each location. It should be noted that all GO Rail park-andride facilities provide free parking to their customers on a first-come-first-serve basis with options to purchase annual/monthly reserved parking spots. On the other hand, park-and-ride facilities operated by TTC charge their users a daily parking fare ranging from $-$, however no additional parking charges are levied to the TTC Subway (Metro) monthly pass holders. Figure shows park-and-ride stations catchment areas for cross-regional commuting trips. This defines an approximate area that covers households access station choices based on the observed trip records. As shown in Figure, a substantial overlap between stations catchment areas exists which indicates that individuals whose home locations are approximately situated in the same area make different access station choices. The shape of these catchment areas shows the travel direction pattern of cross-regional morning commuting trips from elsewhere towards the City of Toronto. Legend Legend Lot Capacity Cost - 1 Free 1 - $0 - $ - $ - $ - 1 $ - $ 1-11 $ - $ 11 - Legend (a) Park-and-ride Lots Capacities (Size) -(b) Parking Demand (number of trips) at Parkand-ride Locations and Parking Cost (Color) Figure- Park-and-ride Lots Capacities, Parking Cost and Parking Demand TRB 01 Annual Meeting

8 Legend GO Rail Station Catchment Area < 1. Sq Km Sq Km Sq Km Sq Km Sq Km Figure- Park-and-ride Catchment Areas Econometric Model To ensure the availability of all variables of concern, a subset of the collected data was selected. Three datasets were prepared for modelling and empirical investigations. The first dataset included all park-and-ride stations (including GO Rail and TTC Subway) and their corresponding trip records. The second dataset included regional transit (GO Rail) trip records and finally the third dataset included local transit (TTC Subway) trip records. Household locations were used to define the choice set for each individual. The park-and-ride GO Rail stations are widely dispersed across the region; however, the 1 park-and-ride TTC Subway stations are concentrated in the City of Toronto, mainly near the terminal stations of the three subway lines. Therefore, it was assumed that the five closest stations defined the access station choice set for GO Rail users. On the other hand, the three closest stations defined TTC Subway users access station choice set. Respectively, % and 0% of cross-regional park-and-ride GO Rail and TTC Subway users observed access station choices fell within the pre-defined choice sets. We considered that individuals gain a certain level of utility by choosing one station from their pre-defined choice set. The utility function (U) for each station is composed of systematic (V) and random (ε) components. The systematic component explains the deterministic utility of choosing the corresponding alternative station as a function of linear-in-parameters of the observed variables and their corresponding coefficients (βx).the random utility component explains the unobserved random variations in choice (). It was assumed that trip makers are rational in selecting an access station among a set of feasible alternatives and to choose the alternative with the highest utility value. U s V ( x) [1] s s s s Where the subscript s indicates one of the stations in the choice set. TRB 01 Annual Meeting

9 Model variables may affect the utility function by different levels. For instance, previous studies showed that the access distance (i. e., from the trip origin to the station) is an important variable in access station choice models (; ; 1). In this study, we introduced station features such as the direction of the station relative to home and work locations, parking lot capacity, parking cost and other station amenities which may potentially affect individuals station choice. Previous studies often included service frequency as a variable in the utility functions. However, in this study, subway services run at high frequency levels (around two-minute headways) during the morning peak period, resulting in indifferent service frequencies among stations in concern. For regional transit services, we assumed that individuals check the publicly available train schedules and they plan their trips accordingly. Therefore, service frequency was not included as a variable in the station choice utility function. In order to derive the probability function for access station choice, a distributional function of the random error component needs to be assumed. We assumed that the error term follows the Independent and Irrelevant Distribution (IID) of Type I Extreme Value distribution. Such assumption results in the Multinomial Logit (MNL) model () of the form: exp( V ) s Pr( s) S [] exp( V / ) s / s 1 Where the subscript s indicates the chosen station and S indicates the maximum number of stations under consideration. In this paper, the empirical models were estimated using the mlogit package in the statistical software R and using the MAXLIK component for maximum likelihood estimation (; ). Empirical Models We divided the park-and-ride users into three groups: individuals who have only TTC Subway park-and-ride stations within a reasonable reach, individuals with only GO Rail stations within the reach, and individuals who have both TTC Subway and GO Rail stations within the reach. Three multinomial logit models were estimated accordingly. Hence, the three models represent three market segments. Having such segments, we developed three models for better capturing choice behaviour. Table 1 presents definitions of variables that are used in this study. TRB 01 Annual Meeting

10 Table-1 Definitions of Variables Variable Name Description DIST Airline access distance, in meters, from household location to park-and-ride station location alpha (α) Relative station direction, in degrees, between a straight line from home location to regular work location and a straight line from home location to park-and-ride station location as shown in Figure LotCapacity Park-and-ride lot capacity ParkCost Parking cost at morning peak period, in CAD, at park-and-ride location RefresKiosk =1 if station has a refreshment kiosk; =0 otherwise Washroom =1 if station has a washroom facility; =0 otherwise CarpoolResPark =1 if station has reserved parking option for carpool; =0 otherwise ResPark =1 if station has reserved parking option; =0 otherwise Regional =1 if station is a regional transit (GO Rail) station; =0 otherwise ConnectLT =1 if a GO Rail station connects to local transit services; =0 otherwise ConnectRT =1 if a TTC Subway station connects to a regional service (GO Rail); =0 otherwise TTCPass =1 if the individual possess a TTC Subway (Metro) pass; =0 otherwise GOPass =1 if the individual possess a GO Transit pass; =0 otherwise Figure- Relative station direction (alpha α i ) of park-and-ride Stations (S i ) from Home Location (H) to Regular Work Place (W) As shown in Figure, the relative station direction (alpha) indicates whether the station in consideration is on the way to work (S1) or not (S). Stations with higher value of alpha (i.e., out of the way to work) are expected to be less attractive to trip makers. The dataset included other variables that were omitted from the final model specifications as they were found not to be useful in explaining individuals choice of access stations. Different model structures and specifications were tested and the final model specifications are reported in Table. Mixed logit models for regional commuter rail and local subway park-and- TRB 01 Annual Meeting

11 ride access station choice assuming access distance and relative station direction as random variables (drawn from a normal distribution) were examined; however, the standard deviation of the random variables were statistically insignificant. Therefore, a multinomial logit model structure was used. Parameter estimation results of GO Rail and TTC Subway park-and-ride access station choice multinomial logit model are reported in Table -(a). A total of parameters were estimated using a subset sample of currently park-and-ride users who have at least one GO Rail and one TTC Subway park-and-ride location within their choice set which are composed of the five closest park-and-ride stations. All the reported parameters in the final model specification are statistically significant with t-statistics greater than 1.. The reported adjusted Rho-Square value (also referred to as likelihood ratio test) (), as a measure of goodness-of-fit, is 0.. This is relatively a high value considering the small sample size used in this analysis. In general all variable coefficients have the expected correct signs. Two variables are used to define stations proximity to home locations considering the usual work place, namely, access distance and relative station direction. The increase of both access distance and relative travel angle has a negative impact on access station choice; however, it can be seen that the access distance and the relative station direction have different impacts on individuals choices. The majority of park-and-ride locations in the GTHA are running at capacity during morning peak periods, and as such more parking spaces increase individuals chances to find a parking spot. Interestingly, if a subway and a regional transit station happened to be in the same choice set of a cross-regional trip maker, the regional transit station is less likely to be chosen. In the GTHA, a typical park-and-ride cross-regional trip destined to the City of Toronto can be as follows: driving to a nearby GO Rail station, taking the GO train to Union station and then transferring to TTC Subway to the final destination. Therefore, if a subway station is feasibly accessible from individuals home location, it will be more attractive than a regional transit station. This will save the trip maker an unnecessary mode transfer, thus affording the commuter travel time and cost savings. One of the other factors that may affect individuals access station choice is whether the trip makers possess a transit pass or not. Two types of transit passes were considered, namely, TTC Subway (Metro) monthly pass which gives individuals unlimited TTC subway, bus and streetcar rides in addition to free parking and park-and-ride locations, and GO Transit pass as a discounted pre-paid set of tickets which was recently replaced by the PRESTO smart card. In order to account for transit pass type, a dummy variable is introduced to the model in association with access distance. Results show that TTC pass holders are more likely to drive longer to TTC Subway park-and-ride stations than GO pass holders drive to GO Rail park-andride stations. Similarly, Tables -(b) and -(c) show parameter estimation results for GO Rail and TTC Subway park-and-ride access station choice multinomial logit models, respectively. It should be noted that a mixed logit model of regional transit park-and-ride access station choice was estimated. The model was developed using similar model specifications considering the relative access station direction as a normally distributed random parameter. However, the reported MNL model outperformed the estimated mixed logit model in terms of goodness of fit measures. TRB 01 Annual Meeting

12 Table- Parameter Estimation Results Table- (a) Regional and Local Transit park-and-ride Access Station Choice MNL Model Number of Observations Log-Likelihood (Full Model) -11. Log-Likelihood (Null Model) -1. Rho-Square Value 0. Adjusted Rho-Square Value 0. Systematic Utility Function: Variables Parameter t-statistics DIST * alpha (α) * LotCapacity * Regional * DIST * TTCPass * DIST * GOPass * Table- (b) Regional Transit (GO Rail) park-and-ride Access Station Choice MNL Model Number of Observations 1 Log-Likelihood (Full Model) -0 Log-Likelihood (Null Model) -11. Rho-Square Value 0. Adjusted Rho-Square Value 0. Systematic Utility Function: Variables Parameter t-statistics DIST * alpha * LotCapacity * CarpoolResPark+ResPark 0.0.* RefresKiosk+Washrooms 0.00.* ConnectLT * Table- (c) Local Transit (TTC Subway) park-and-ride Access Station Choice MNL Model Number of Observations 1 Log-Likelihood (Full Model) -.01 Log-Likelihood (Null Model) -. Rho-Square Value 0. Adjusted Rho-Square Value 0. Systematic Utility Function: Variables Parameter t-statistics DIST * alpha * ParkCost RefresKiosk+Washrooms ConnectRT * * Significant at the % level of confidence TRB 01 Annual Meeting

13 The reported parameters are highly significant with the expected correct signs except for the Parking Cost variable s parameter which was retained as it provides behavioural insights into the model. As expected, park-and-ride locations with higher parking costs are less likely to be chosen over stations that provide parking at a lower cost or free. Similarly, stations that provide better local/regional service integration and better station amenities/facilities are preferred more. Both regional commuter rail and local subway park-and-ride access station choice models showed acceptable adjusted rho-square values of 0. and 0., respectively. A forecast of access station choice for park-and-ride cross-regional trip maker was conducted and compared to actual observations. The percent correctly predicted analysis, as another goodness-of-fit statistics (), showed that the three models have relatively high prediction accuracy. The combined model performs at.% prediction accuracy, while the regional commuter rail and local subway park-and-ride access station choice models attain.% and.1%, respectively. In order to investigate the sensitivity of park-and-ride users to station access distance and relative station direction, we estimated access distance and relative station direction elasticity for each individual in the three datasets. Average elasticity was calculated for the number of stations in each individual s choice set and kernel densities were plotted as shown in Figure. Equations for direct elasticity calculations are available in (). Figure shows that the access distance has a bi-modal distribution; however, the relative station direction has a uni-modal distribution. The bi-modal distribution suggests two groups of users with different mean access distances. Such a distribution, perhaps, results from different land use patterns near each station. Further investigation showed that different regions (planning districts) have different access distance distributions, which can be explained by the variation of stations spacing in each region. Users from regions with higher station densities have more flexibility in driving to a further station within their acceptable access distance. However, as the station density decreases, users become more inelastic (captive) to drive longer distances to access a nearby station (e.g. Hamilton Region). This provides an explanation for the existence of two groups of users; those more flexible (elastic) to access distance and those whom are more constrained in terms of access distance. The sample average elasticity of the station access distance for the combined model is -.1 with a standard deviation of 0. and the sample average elasticity of the relative station direction is -0. with a standard deviation of That is, the access distance is elastic and relative station direction is inelastic. Individuals are more sensitive to changes in station access distance than to changes in the station relative direction. Similarly, the regional commuter rail and local subway park-and-ride access station choice models reported a sample average elasticity of station access distance of -.0 and -.1 with a standard deviation of.1 and 1., respectively. On the other hand, the models reported a sample average elasticity of the relative station direction of and -0.1 with standard deviation of 0. and 0., respectively. This indicates that TTC Subway park-and-ride users are less sensitive to access distance than regional transit users. Hence, GO Rail users choices are more elastic with respect to the station location than TTC Subway users. This can be explained by the spatial wide dispersion of park-and-ride GO Rail stations across the GTHA compared to fewer options of TRB 01 Annual Meeting

14 park-and-ride TTC Subway stations within the City of Toronto (i.e., regional transit users have more choices in different relative directions). (a) Access Distance Average Elasticity (b) Relative station direction (alpha) Average Elasticity (i) Regional (GO Rail) and Local (TTC Subway) Transit park-and-ride Access Station Choice Model (c) Access Distance Average Elasticity (d) Relative station direction (alpha) Average Elasticity (ii) Regional Transit (GO Rail) park-and-ride Access Station Choice Model (e) Access Distance Average Elasticity (f) Relative station direction (alpha) Average Elasticity (iii) Local Transit (TTC Subway) park-and-ride Access Station Choice Model Figure- Marginal Effects of Access Distance and Relative Station Direction for the Three Models 1 1 Conclusions and Recommendation for Future Study The paper presents a park-and-ride access station choice model for cross-regional commuter trips in the GTHA. Two sources of data were used; cross-regional commuting trips in the morning peak period on regular weekdays were extracted from the household travel survey, and park-andride station locations, parking lot capacities, parking costs, surrounding land use, and station amenities were obtained from transit service operators. The five and three closest stations defined access station choice set for GO Rail and TTC Subway users, respectively. Three datasets were prepared for modelling and empirical investigations. The park-and-ride users were divided into three groups: individuals who have only TTC Subway stations within a reasonable reach, individuals with only GO Rail stations within the reach, and individuals who have both TTC Subway and GO Rail stations within the reach. Different model structures and 1 TRB 01 Annual Meeting

15 specifications were tested and three multinomial logit models were estimated accordingly for better capturing choice behaviour of each market segment. In general, the estimated variable coefficients were found to be statistically significant with the expected correct signs. Access distance and the relative station direction were the primary factors that affect individuals choices. The increase of both access distance and relative travel angle has a negative impact on access station choice. The estimated model parameters were used to investigate park-and-ride users sensitivity to station access distance and relative station direction. Results showed that individuals are more sensitive to changes in station access distance than to changes in the station relative direction. In addition, TTC Subway park-and-ride users are less sensitive to access distance than regional transit users. Hence, GO Rail users choices are more elastic with respect to the station location than TTC Subway users. Kernel density charts of the average access distance (across all individuals in each of the three datasets) showed that the access distance has a bi-modal distribution which refers to two groups of users with different mean access distances. Such distribution, perhaps, results from different land use patterns near each station. The results of this study are potentially useful for transit station design in terms of facilitating modal integration for cross regional trips, which are growing increasingly common in large regions. The empirical models developed in this study allow for a quantitative investigation of the important factors that should be considered in identifying the transit station characteristics that maximize park-and-ride type of trips. In addition, the developed empirical models are considered to be a part of a regional scale transit service planning tool that will combine access station choice into a comprehensive cross-regional commuting mode choice model. Such a tool will be essential to devise policies for increasing competitiveness of transit to private automobiles. Therefore, the next step of this research includes testing advanced model structures to capture different levels of joint/sequential decisions of main mode, access mode and access station choices. This requires disaggregate service level attributes for mode in consideration. Consequently, improved network models that explicitly capture modal interaction/integration are needed. Acknowledgements The authors acknowledge the Data Management Group (DMG) of the Department of Civil Engineering at the University of Toronto, Metrolinx, and the Toronto Transit Commission (TTC) for sharing travel survey and stations service level attributes data for research purpose. 1 TRB 01 Annual Meeting

16 References [1] DMG. Data Management Group. 1 Transportation tomorrow survey. Joint Program in Transportation, University of Toronto. < 1. [] DMG. Data Management Group. 00 Transportation tomorrow survey. Joint Program in Transportation, University of Toronto. < 00. [] Cairns, M. R. The Development of Park and Ride in Scotland. Journal of Transport Geography, Vol., No., 1, pp. -0. [] Bos, I., R. van der Heijden, E. Molin, and H. J. P. Timmermans. Traveler preference for parkand-ride facilities: Empirical evidence of generalizability. Transportation Research Record: Journal of the Transportation Research Board, Vol. 1, No. -1, 00b, pp [] Chakour, V., and N. Eluru. Analyzing Commuter Train User Behavior: A Decision Framework for Access Mode and Station Choice.In Transportation Research Board nd Annual Meeting, Washington D.C., 01. [] Forsey, D., K. M. N. Habib, E. J. Miller, and A. Shalaby. An Evaluation of the Impacts of Introducing a New Transit System on Commuting Mode Choice and Transit Ridership: A Case Study of the VIVA BRT-Lite System in Toronto.In Transportation Research Board 1st Annual Meeting, Washington D.C., 01. [] Holguín-Veras, J., J. Reilly, F. Aros-Vera, W. Yushimito, and J. Isa. Park-and-Ride Facilities in New York City. Transportation Research Record: Journal of the Transportation Research Board, Vol., No. 1, 01, pp. 1-. [] Debrezion, G., E. Pels, and P. Rietveld. Choice of Departure Station by Railway Users. European Transport, Vol., 00, pp. -. [] Vijayakumar, N., A. M. El-Geneidy, and Z. Patterson. Driving to Suburban Rail Stations. Transportation Research Record: Journal of the Transportation Research Board, Vol. 1, No. 1, 0, pp. -. [] Mertrolinx. GO Transit Rail Passenger Survey Report.In, 0. [] Bos, I. D., R. Van der Heijden, E. J. Molin, and H. J. Timmermans. The Choice of Park and Ride Facilities: an Analysis Using a Context-Dependent Hierarchical Choice Experiment. Environment and Planning A, Vol., No., 00, p. 1. [1] Li, Z.-C., W. H. Lam, S. Wong, D.-L. Zhu, and H.-J. Huang. Modeling park-and-ride services in a multimodal transport network with elastic demand. Transportation Research Record: Journal of the Transportation Research Board, Vol. 1, No. 1, 00, pp. 1-. [1] Tsang, F. W., A. S. Shalaby, and E. J. Miller. Improved modeling of park and ride transfer time: Capturing the within day dynamics. Journal of advanced transportation, Vol., No., 00, pp. -1. [1] Washbrook, K., W. Haider, and M. Jaccard. Estimating commuter mode choice: A discrete choice analysis of the impact of road pricing and parking charges. Transportation, Vol., No., 00, pp. 1-. [1] Fan, K.-S., E. J. Miller, and D. Badoe. Modeling rail access mode and station choice. Transportation Research Record, No. 11, 1. [1] Kastrenakes, C. R. Development of a Rail Station Choice Model for NJ Transit,. Transportation Research Record, Vol., 1, pp [1] Wardman, M., and G. Whelan. Using Geographical information systems to improve rail demand models. Final report to Engineering and Physical science Research council, 1. 1 TRB 01 Annual Meeting

17 [1] Mukundan, S. An Access Mode and Station Choice Model for the Washington DC Metrorail System. 11. [1] Debrezion, G., E. Pels, and P. Rietveld. Modelling the joint access mode and railway station choice. Transportation Research Part E: Logistics and Transportation Review, Vol., No. 1, 00, pp. 0-. [0] Habib, K. M. N. A joint discrete-continuous model considering budget constraint for the continuous part: application in joint mode and departure time choice modelling. Forthcoming in Transportmetrica (): 1-1, 01. [1] Habib, K. M. N., N. Day, and M. E.J. An investigation of commuting trip timing and mode choice in the greater Toronto area: Application of a joint discrete-continuous model.. Transportation Research Part A : -, 00. [] Weiss, A., and K. M. N. Habib. Evolution of modal captivity and mode choice for commuting trips: A longitudinal analysis by using crossectional datasets..in Transportation Research Board nd Annual Meeting, Washington D.C., 01. [] Mertrolinx. The Regional Transportation Plan for the Greater Toronto and Hamilton Area (GTHA): The Big Move. < 00. [] Habib, K. M. N., M. S. Mahmoud, and J. Coleman. The Effect of Parking Charges at Transit Stations on Park and Ride Mode Choice: Lessons Learned from a Stated Preference Survey in Greater Vancouver.In Transportation Research Board nd Annual Meeting, Washington D.C., 01. [] Ben-Akiva, M. E., and S. R. Lerman. Discrete Choice Analysis: Theory and Application to Travel Demand. The MIT Press, 1. [] Croissant, Y. Estimation of multinomial logit models in R: The mlogit Packages.In, R package version 0.-, URL [] Train, K., and Y. Croissant. Kenneth Train's exercises using the mlogit package for R.In, R package version 0.-, URL [] Train, K. Discrete choice methods with simulation. Cambridge university press, TRB 01 Annual Meeting