Canadian Journal of Civil Engineering

Transcription

1 Effects of Accessibility to the Transit Stations on Intercity Mode Choices in Contexts of High Speed Rail (HRS) in the Windsor-Quebec Corridor in Canada Journal: Manuscript ID cjce r2 Manuscript Type: Article Date Submitted by the Author: 23-Jun-2015 Complete List of Authors: Wong, Billy; University of Toronto, Civil Engineering Nurul Habib, Khandker; University of Toronto, Keyword: Intercity travel, transit accessibility, high speed rail, nested logit model

2 Page 1 of 39 Effects of Accessibility to the Transit Stations on Intercity Travel Mode Choices in Contexts of High Speed Rail (HRS) in the Windsor-Quebec Corridor in Canada Billy Wong, M.A.Sc Department of Civil Engineering University of Toronto Billy.Wong@stantec.com Professor Khandker Nurul Habib Department of Civil Engineering University of Toronto khandker.nurulhabib@utoronto.ca

3 Page 2 of 39 1 Abstract Main objective of this paper is investigating the role of transit station accessibility on intercity travel mode choices in contexts of a proposed High Speed Rail. The study area is the Quebec- Windsor Corridor, which is the most important corridor in Canada and one of the most important corridors in North America. A web-based joint Revealed Preference (RP)-Stated Preference (SP) survey is used to collect data for empirical investigation. To contribute further to travel survey methods, an innovative social media based data collection approach is taken. As opposed to explicit sample frame-based sample selection approach, it applies a reverse procedure of open sample frame-based data collection. The web-based survey is spreaded through social media groups (that are open in sense that information of all individuals are not known explicitly) and the collected responses are screened to match with population distributions. Results prove the potential of such data collection approach in extracting representative samples of the population of concern. The collected dataset, which has close representation of the population, is used to estimate discrete mode choice model (Nested Logit model) of intercity mode choices. Empirical model reveals that intercity travellers are more concerned about access to and egress from transit stations than the main in-vehicle travel while selecting intercity travel modes. The result of this investigate imply that transit station accessibility should be given careful consideration for the success of any innovative travel mode, e.g. High Speed Rail. Key Words: Intercity travel, transit accessibility, high speed rail, nested logit model

4 Page 3 of 39 2 INTRODUCTION The continued economic growth and development of urban areas within Canada have created corridors of population, economic activity, and movement of individuals. Quebec-Windsor corridor is one of such corridor. This corridor serves as the passage of trades and comers between Canada and US. This is the busiest corridor in Canada and one of the busiest corridors in North America. Movement of individuals within such corridor has become the subject of numerous studies from economics, social science, and transportation research groups. Within federal, provincial, and municipal transportation development sectors, considerable efforts are dedicated to assessing the different modes of passenger and freight movement within these corridors (IBI 2002). With increased passenger volumes travelling within these corridors, alternative transportation modes have been proposed and researched in the past without implementation. One such is a high speed rail system (HSR), which has been studied since the early 90 s (Langan 2011) However, increasing oil price in recent years has generated renewed interest in HSR in Quebec-Windsor corridor (Miller 2004). This also generated renewed need for improved understanding on our intercity travel bahaviour. In assessing intercity travel demand, there are issues with data availability on intracity travel. Current publicly available datasets on Canadian intercity travel lack local access and egress information, which is predicted to be a significant factor in intercity mode choice. When travel data is aggregated to the metropolitan level, interpolation of local travel activity is required, which may negatively influence the validity of resulting demand models (Wilson et al. 1990). With the lack of local accessibility aspects, it is necessary to create a new survey framework to collect travel data for demand modeling. This paper contributes to two aspects of intercity travel research: travel

5 Page 4 of 39 3 survey to collect data on intercity travel and evaluating the influences of transit station accessibility on intercity passenger travel mode choice. Contribution to travel survey for intercity travel includes designing a web-based hybrid survey and application of social media based survey data collection. The hybrid survey combines revealed preference (RP) questions on intercity travel experience and stated preference (SP) experiment on mode choices in context of HSR in the Quebec-Windsor corridor. In terms of data collection, we considered an innovative approach of recruitment process. Recognizing the difficulty of defining the sample frame for a large corridor that crosses multiple provinces, we investigated non-incentivized online recruitment. We investigate the potential of using social media based recruitment process for web-based travel survey. Unlike conventional approach of defined sample frame based recruitment, the social media based approach is an open form approach. The open form refers to the uncontrolled or unknown sample frame, which also resembles a snowball sampling approach. Post data collection, the dataset is matched with overall population characteristics and, if necessary, re-sampled from the collected dataset to match the population characteristics. This approach of data collection for intercity travel is validated in the field for the Quebec-Windsor corridor and it is proven to be effective in collecting representative dataset within a short period of time. The collected dataset is then used to develop discrete choice model of intercity passenger travel mode choice. The estimated model reveals that people put higher value on access to and egress from transit stations than the main mode s travel time/cost while choosing intercity mode choice. It becomes clear that competitiveness of various transit options (including HSR) depends largely on accessibility to local station locations.

6 Page 5 of 39 4 The paper is organized into a number of sections. The next section presents a brief literature review on intercity mode choice investigation. This section is followed by the section on survey design, data collection and empirical investigation. The paper concludes with key findings and policy recommendations. LITERATURE REVIEW Intercity travel is defined as a trip that passes through the boundaries of an urban center. Typically an intercity trip would start and end within an urban center. Intercity travel demand is the frequency of trips made between urban centers as well as the modes of travel used. While intercity travel demand research has been published since the 1960s, these models have not been improved as rapidly as the more urban center specific models. Some hypothesized reasons for the slow development of intercity models are due to fewer intercity travel corridors of interest to policy makers in comparison to urban regions, unclear jurisdiction of intercity corridors, larger private sector stake resulting in proprietary information, open-ended definition of study area, limited existing data available for research purposes, and unwilling commitment to invest in long-term research by governments and public agencies (Langan 2011). Application of models such as the logit model to assess intercity travel demand has not been as successful as the urban center counterparts. Aside from the lack of suitable data, there are other fundamental reasons. Compared to daily local journeys (trips to work, school, shopping), intercity trips are made less frequently than urban trips. Due to the lower frequency of intercity travel, there is difficulty modeling a trip maker s decision to visit another city with logit models using existing data. Unless daily intercity trips are made, a longer time period of trip information collection is required (Sonesson 2011).

7 Page 6 of 39 5 An empirical study in using disaggregate choice model of intercity travel demand in the Quebec City Windsor Corridor, published in 1988 used the 1969 Canadian Transport Commission (CTC) survey to estimate mode choice in the Quebec-Windsor Corridor (Ridout and Miller 1989). The major weaknesses in the dataset was the lack of automobile choice, which limited the accuracy of the demand forecasting logit model as automobile comprises a large market share in intercity travel. While the 1980 Canadian Travel Survey (CTS) was available as a dataset alternative, the aggregation of trip origins and destinations to the Census Metropolitan Area (CMA) would not provide relevant trip access and egress information. The resulting models were all calibrated using a standard maximum likelihood logit estimation procedure. Three models were estimated with each model representing different market segmentation (business, pleasure, or personal). For the business market, it was observed that access was a generic attribute while egress was alternative specific. For non-business and non-personal intercity trips, access was estimated as an alternative specific attribute and egress was generic. For personal intercity trips, only local access, as a generic attribute, was estimated to be statistically significant to mode choice. There was difficulty in estimating the access and egress term, which may be attributed to the use of access/egress distance instead of cost and time. Socioeconomic variables were found to have minimal explanatory function in intercity mode choice. Wilson et al. (1990) used the 1985 Canadian Travel Survey (CTS) to estimate MNL models for intercity passenger travel. They developed models for both business and non-business trips in the eastern and western Canadian regions. They proved that the CTS can be effectively used for intercity mode choice modelling when datasets for such investigations are scarce. Forinash and Koppelman (1993) applied and compared a Nested Logit (NL) model against a MNL model using an RP dataset generated from the VIA Rail s (a national carrier

8 Page 7 of 39 6 serving along the Windsor-Quebec corridor) 1989 Passenger Review dataset. They investigated the effect of rail service improvements for weekday business travel in the Toronto-Montreal corridor (which is a part of Windsor-Quebec corridor). They justified the advantage of the NL over the MNL by relaxing the Independent and Irrelevant Alternatives (IIA) property of the MNL to capture correlation among unobserved attributes of similar modes. Bhat (1995) used the same dataset to examine the impact of improved rail service on intercity business travel in the Toronto Montreal corridor. The focus of this paper was mostly to test an advanced formulation of discrete choice models in terms of performance over the conventional logit model. He found that, compared to the MNL, the Heteroskedastic Extreme Value (HEV) model predicts smaller increases in rail shares and smaller decreases in non-rail shares. The HEV model should allow for cross-elasticity among alternatives compared to a nested logit model and require less computational complexity compared to the multinomial probit model. To test the application of the HEV model, the 1989 Rail Passenger Review from VIA Rail was used. The main focus was on paid business travel in the Quebec-Windsor Corridor and confined between the air, rail, and auto travel modes as travel for non-personal business purposes had less than 1% market share. Five different models were estimated; a multinomial logit model, three nested logit models, and the heteroskedastic extreme value model. From model estimation, the nested logit structure was not significantly better than the multinomial logit models. Compared to both MNL and NL, the HEV model was able to predict smaller changes in level-of-service changes, which may point to an improvement of the HEV model over the commonly used MNL and NL model formulations. Mandel et al. (1997) present an intercity mode choice model for Germany. They use a logit model with nonlinear Box-Cox transformations of key level of service variables. Their efforts were mostly focused on capturing non-linear responses with respect to travel time and travel cost

9 Page 8 of 39 7 for high-speed rail. Hensher et al. (1999) applied the HEV model to estimate an intercity travel mode choice model using a combination of RP and SP datasets with the objective of identifying the market for a proposed high-speed rail service in the Sydney Canberra corridor. They found that much more uncertainty exists in the evaluation of non-car modes than of the car mode for that corridor. It is often hypothesized that an individual s responsiveness to level-of-service variables affects that individual s mode choice (Bhat 1998). In this paper, Bhat accommodates variations in this responsiveness within a multinomial logit based model. Monte Carlo simulation techniques were also incorporated to approximate the choice probabilities, which are a technique that has been used in empirical applications in the economics field and relatively new to transportation researchers. The 1989 Rail Passenger Review dataset from VIA Rail is used once again to develop the travel demand models. Weekday business-based market segment was selected and recorded trips by bus were omitted from the dataset due to a small percentage of market shares. Three models were estimated in this paper; a multinomial logit model, a fixedcoefficient logit model, and a random coefficients logit model. Level-of-service variables included service frequency, total travel cost, in-vehicle travel time and out-of-vehicle travel time along with socioeconomic variables; income, sex, travel group size, and large city indicator. From model estimation, both the fixed-coefficient logit and random coefficient logit models showed differences in sensitivity of level-of-service variables based on the socioeconomic characteristics of the trip maker, rejecting the response homogeneity assumption of the MNL model. Bhat (1998) shows that not accounting for variations in responsiveness across individuals may lead to inappropriate estimation of mode choice, which may affect policy decisions.

10 Page 9 of 39 8 Wen and Koppelman (2001) proposed a general model structure which they refer to as the general nested logit model (GNL), and also applied it to the same 1989 VIA Rail dataset to analyze the effect of rail service improvements. However, the study s main focus was the general nature of the GNL model formulation and the derivation of the other GEV model structures as special restrictive cases of the GNL model or as approximations to restricted versions of the GNL model. The GNL model is conceptually appealing because of its very general structure and flexibility, while maintaining closed form expressions. However, in practice, informed restrictions that are customized to the application context must be imposed on the GNL model formulation, and thus the flexibility of the GNL model can be realized only if a large number of dissimilarity and allocation parameters are estimated. Grisolía and Ortúzar (2010) recently applied mixed logit approach to investigate the role of travel time perception of on inter-island mode choice model for a limited number of modes (plane and ferries). However, majority of recent studied on intercity passenger travel related studies focus mostly specific aspects of specific intercity mode choices. Among many examples of such investigations, Ahsan et al. (2003) investigated market segmentations and role of socio-economic variables on preferences variations among the passengers of intercity bus travellers. Rojo et al. (2008) investigated satisfactions of passengers of intercity bus traveller and Rojo et al. (2011) investigated how gender plays role in defining passenger satisfactions of intercity bus services. Debrezion et al. (2009) applied discrete choice model to investigate relationship between access mode and station choice for intercity trains. Hsio and Hensen (2011) presented models for forecasting demands of intercity air transportation. Rojo et al. (2012) investigated the role of service qualities in influencing choice of bus transit for intercity bus only. They did not conduct any systematic modelling exercise of intercity mode choice modelling.

11 Page 10 of 39 9 The typical barrier to developing comprehensive understanding on intercity mode choice behaviour is the lack of sufficient data for estimating empirical models. Since intercity trips are not often part of household travel surveys, there are no systematic data collection programs normally available for intercity trips. National travel survey data used for intercity mode choice investigations lack significant details of intercity trip components. Datasets used for majority of research investigations (as reviewed in this section) are collected as parts of specific projects by different government or private organization. Such dataset are mostly designed to collect information of specific aspects of intercity trips than to comprehensively understand intercity travel behavior. To fill this gap in intercity travel research, we designed a comprehensive data collection program considering the Quebec City to Windsor Corridor (QWC) of Canada as the study area. SURVEY DESIGN The first task is to create a survey that can collect the required travel information while addressing existing research and dataset gaps. The survey used for this paper collected information on a respondent s previous experiences of intercity trip, travel behaviour related to non-intercity trips, socioeconomic characteristics in addition to a stated preference (SP) choice experiment. The following sections detail each aspect of the survey design. Revealed Preference (RP) Travel Information of Intercity Trips The respondent is asked to provide travel details of his or her last intercity trip within the QWC including basic travel information such as; the destination city, travel mode, travel cost, and trip purpose. Aside from correlating the collected revealed intercity travel data with existing national

12 Page 11 of travel datasets, the collection of this data may reveal any mode switching tendencies given the introduction of a new travel mode in the QWC. Revealed Preference Information on Non-Intercity Travel A set of local travel revealed preference questions was also included in the survey. The collection of this data may be used to correlate daily travel patterns with intercity modal choice. However, the main purpose of the local revealed preference data is to cross reference the demographics of the respondents from this survey with demographics from other travel datasets; such as the ones from Transportation Tomorrow Survey (TTS) from University of Toronto s Data Management Group. Socioeconomics Information of the Respondent This section of the survey inquires about the respondent s individual and household socioeconomic information. The range of questions included is historically shown to influence transportation-related choices including; age, gender, marital status, employment, household size, household auto ownership, and household income. Stated Preference Survey of Intercity Mode Choice The section most useful to modeling mode choice is the intercity mode choice stated preference section. While revealed preference information indicates current travel patterns, the potential modal shift due to the introduction of a currently non-existent high speed rail (HSR) mode cannot be modeled using RP data. The basis of the stated preference survey is presenting the

13 Page 12 of respondent with a hypothetical intercity travel scenario and inquiring about the respondent s mode choice. For this paper, a hypothetical intercity trip between an individual s specified home location in the Greater Toronto Area (GTA) to a location in Montreal is tested. Montreal was chosen as the destination location because existing travel surveys indicate that the greatest percentage of intercity trips within the Quebec City/Windsor corridor and originating from the GTA has a destination in Montreal (Statistics Canada 2010). Addressing for previous research gaps, this paper requires greater geographical disaggregation (than previous research) to specifically focus on the effects of local accessibility on intercity mode choice. Unlike existing travel surveys that record origin and destination locations at the cities level, a higher degree of geographical disaggregation is required to obtain local accessibility characteristics. In this paper, the respondent s origin location is recorded at Forward Sortation Area (FSA) levels, which are the first three digits of a postal code. Utilizing FSA s is the optimal compromise between stated and perceived local travel times, where further geographical disaggregation may not yield significantly improved estimation results. Similarly, the destination location in Montreal is aggregated to its 19 boroughs. A higher degree of aggregation was chosen for the destination to reduce the necessity to search out a FSA code. Attributes and levels in the SP Experiment: A set of attributes was generated to provide the respondent relevant information regarding each travel mode. An initial list of attributes was based on a TSRB study on a high speed rail line in London, UK (Burge et al. 2011) and modified to better relate to the Canadian region and existing

14 Page 13 of Canadian travel mode alternatives. The attribute levels are the various values that can be associated for each attribute. Levels can either be continuous values associated with the attribute (such as a dollar amount with travel cost) or categorical values that associated with a description (such as 1 for booked seating and 2 for representing arranged seating). Table 1 presents the list of the attributes used in the stated preference survey. With the list of attributes and its respective levels established, an orthogonal stated preference design was completed. A current lack of empirical models that also incorporate all the proposed attributes hindered the use of an efficient SP design. Figure 1 below is one of six scenarios shown to the respondent in the web-based survey. The values populating the SP table is based on the respondent s stated origin and destination locations and outputs the appropriate travel time and costs while accounting for the different attribute levels. All elements shown in Figure 1 are automatically generated and stored prior to deployment of the survey. Time and distance based elements (access/egress time, travel time, and travel costs) are aggregated at the FSA level and associated times and costs were used during the survey design stage via the use of mapping tools such as Google Maps. Travel time and costs were determined using publically available tools such as Google Maps. As FSA level geographical aggregation was used, the same base-values would be presented to all respondents in the same FSA. This data was collected and stored as static backend elements that were tied to a specific web URL that referenced the FSA. When the survey asks for the respondents FSA origin, the appropriate subset of survey scenarios would be pulled from the survey backend. Total travel time and total travel costs values were also displayed at the bottom of the table to provide assistance to the respondent and reduce survey fatigue.

15 Page 14 of Access/egress time for the SP experiment: Access time is defined as the time required traveling from a respondent s home location (aggregated to the centroid of the home location s FSA address) to the specified intercity mode departure station, using the corresponding access mode. The purpose of the access time attribute is to assess if local accessibility has any influence on intercity modal choice. For example, it is relatively easy to travel to all three departure stations from a Toronto downtown location; however, a similar trip may take longer if travelling from an Ajax home. The baseline access times were determined by using Google Maps to query directions between a FSA centroid location and the three intercity departure locations (Toronto Coach Terminal, Union Station, and Lester B. Pearson Airport). Google Maps was used as the service provided travel times by automobile, transit, as well as non-motorized transport. Access/Egress Mode: The access mode attribute is defined as the local mode choice option that the respondent would use to travel to the departure station of an intercity travel mode alternative. This access trip should originate from the respondent s home location (aggregated to the geographical center of the respondent s home FSA address) and end at the location of the intercity mode s departure station. In conjunction with access time the inclusion of the access mode is to assess whether or not the local accessibility of transportation alternative has an effect on a respondent s intercity mode choice. The possible access mode alternatives are typical travel methods currently used to access the designated intercity departure stations. The cost of utilizing various access modes is also presented to the respondent and is a stratified price based on the access mode. The decision to include a generalized range of travel cost was done to simplify the survey design. In addition,

16 Page 15 of it was assumed that travel cost in each access mode alternative were mutually exclusive. For example, the cost of transit would range between $3 and $6 whereas a trip by taxi with the same origin and destination would be over $10. It is assumed that a respondent would only require one access mode to get from their home location to the intercity mode departure station. However, incorporating multiple access modes would introduce more complexity into the SP design SURVEY IMPLEMENTATION AND DATA The survey is designed as a web-based survey that utilizes online social media sources as the primary method of data collection. With a heavy focus on online social media, a data collection strategy was designed to assess the benefits of web-based data collection. Data Collection Method The survey invitations were distributed among different social media sources. Each social media source contains different number of members and has different online activity patterns. So, sample recruitment through these multiple social media sources can be analogized to a pyramid, where each increasing level up the pyramid adds additional survey responses to the total amount; however, the number of expected respondents at each increasing level is decreased due to a reduced social network reach. Under this pyramid analogy, the first step would be to distribute the web-based survey using a social network platform that would produce a large number of respondents knowing that a large percentage of these respondents would fit a narrow socioeconomic demographic. Existing social platforms fitting this would be Facebook and Twitter. The next step would be to distribute the survey onto other networks that may expand to different socioeconomic and demographic

17 Page 16 of groups. For example, university mailing lists may include people outside of the author s personal Facebook social networks; however, there are fewer people on the listserv and the lack of recognition to the author may lower probability that potential respondents would complete the distributed survey. This sampling technique has an inherent bias of oversampling some respondents, resulting in a biased socioeconomic profile, while under-sampling responses, with a more representative socioeconomic profile, from limited social network reach. Given a large set of collected data, it is possible to draw a subset from the oversampled demographics; where multiple draws from the same larger subset can be used to validate the model. Data Collection Results Overall, the data collection procedure was satisfactory in obtaining complete survey responses. While the number of total responses did not meet initial expectations, the costs to obtain the responses were kept at a minimal. Table 2 below is a summary for each collector source. There are several metrics that can be used to gauge the relative success of each collector source. Each metric has its own validity and has other explanatory variables that are hard to explicitly measure, such as an individual s social reach. Some possible metrics are; ratio of completed surveys to collector size, ratio of completed surveys to surveys started, ratio of surveys started to approximate size, completed surveys accounting for social reach and collector size. It is evident that the relationship between the approximate size of the collector source and the number of completed surveys is not linear. This observation is in line with initial expectations that posting survey requests in social media networks outside of an individual s own social network yields a lower number or responses due to the reduced familiarity between members in the social network

18 Page 17 of with the researcher and/or research project. Especially with a lack of monetary or prize incentive, there is less willingness for individuals to dedicate time on something with little immediate payout. To validate the geographical distribution of collected responses, population and dwelling counts from Canadian Census data was used as comparison baselines. While there are other potential metrics to validate the collected responses, the emphasis on local accessibility is the main motivator to use household population and dwelling counts. One of the main concerns during the data collection process was a possibility where outlying cities in the Greater Toronto Area are underrepresented in comparison with the City of Toronto. With the relatively small number of collected responses compared to the Census population and dwelling counts, the count of responses, population, and dwellings were calculated as percentages relative to the specified list of cities in the survey design. By changing counts into percentages, it is possible to compare the relative distribution of values across the GTA. Figure 2 below compares the distribution of origin locations between survey respondents and Statistic Canada s 2011 Census figures (Statistics Canada 2013) of population and household totals. Looking at the distribution of respondents, the overall trend of respondent distribution does follow the distribution of household of 2011 Census; however, the major areas of variance are the outlying areas away from downtown Toronto. This variance is most apparent at the FSA level but is less pronounced when aggregated to the cities level. It is also clear that the spatial coverage of the sample is smaller than the population coverage as represented in census in the study area. However, it should be noted that Toronto is over-represented than other parts in the collected dataset.

19 Page 18 of To validate the socioeconomic distribution of collected responses, age and income attributes were examined based on empirical results. Table 3 presents the descriptive statistics that summarizes the distribution of respondent age and income and compares the median collected values with both TTS and Statistics Canada data. The mean age from survey respondents was considerably lower than that of the 2011 TTS data; however, this variance may be explained by the larger bias of younger respondents from the data collection program. When looking at mean income, the values between the survey respondents and the 2012 Statistics Canada were similar. The higher mean income of survey respondents may be due to the sampled group. EMPIRICAL INVESTIGATION Validating the data as representative of GTA residents, a discrete choice model of intercity mode choice is developed by using the collected dataset. It is assumed that the utility of a choice alternative i in the choice set t ( ) is based on the combination of a systematic component ( ) and a random component ( ). Considering the Independent and Irrelevant Distributed Type I Extreme Value distribution of the random element, the multinomial logit model (MNL) is developed, which is often used for intercity mode choice analysis (Ashiabor et al. 2007). In this paper, we are clear that no all alternatives are independent. They may be number of alternatives having common features that may have influences on mode choices. So, we used nested logit model instead. The nested logit model the limitations of the multinomial logit model, considering that there are multiple levels of conditional choices included in the model in the form of nests (Ben-Akiva and Lerman 1985). In the nested logit model, the probability of choosing an

20 Page 19 of alternative i will equal to the probability of choosing the alternative i conditional to choosing some subset k. The full equation to determine the probability of choosing alternative i is: P alternative i choice set = µ Σ µ µ Σ µ µ µ Σ µ µ µ µ Σ µ (1) Where, µ k, µ p defines the scale parameters of nest k and p µ defines the root scale parameter The model is estimated by STATA/IC 13.0 with intercity mode choice as the dependent variable. The nested logit structure was chosen to separate the automobile mode with commoncarrier modes. The model is estimated by using full information maximum-likelihood estimation technique. Based on the possible explanatory attributes collected from the stated preference survey, the common-carrier modes were grouped by the relative location of the primary departure stations in the GTA. This nesting structure was based on the assumption that travel time is a generic utility while the choice of travel modes is alternative specific and based on the departure station location. In this paper, bus was placed in its own group (Toronto Coach Terminal), rail and high speed rail were placed together (Union Station), and airplane was placed in its own group (Pearson International Airport). Figure 3 presents the diagram of the double nested structure used in this paper and. The NL model is structured to account for the potential effect of local accessibility on intercity mode choice. These nests are organized to separate the automobile mode away from the transit modes.

21 Page 20 of The resulting NL model was estimated using 344 of the total 430 respondent entries. The remaining 86 entries were used as model validation. The entries used for model estimation where randomly chosen and represents all data collector sources. Table 3 is a summary of the estimated model parameters. In the NL model, the primary explanatory variables (intercity travel time, intercity travel cost, access time, and egress time) were all statistically significant with the expected signs. The variable specifications in the model were developed based in understandings from similar studies in similar contexts. For example, investigations of Ridout and Miller (1989) provided guidance for specifying transit service related attributes, e.g. access, egress etc. In the final model the sign and magnitude of intercity travel time and cost coefficients are also found consistent with Bhat s (1995) HEV and MMNL (1998) models of travel demand within the same corridor. This proves the justification of nested logit model as opposed to multinomial logit or mixed logit model. Multinomial logit is clearly not suitable as some alternative modes are found to have correlated random utility components and hence the nested logit model is estimated. Similarly, unlike other studies, a mixed logit model was not needed in our context. Possible explanation is that the nested logit formulation is sufficient enough to capture correlated random utilities of and data evidence does not support any further significant overlap between the alternatives as well as any random variations of attribute effects across the population. The magnitude of local access and egress time coefficients were higher than that of intercity travel time, indicating the importance of local accessibility and transit station locations on intercity travel demand. While these results are inconsistent with prior studies, the local travel time information was shown to the respondent and not implied post-data collection and thus should be more accurate than existing datasets used in prior studies.

22 Page 21 of Aside from local travel time, local access and egress mode parameters were also statistically significant and had intuitive signs. The NL model revealed a preference for being dropped off at Union Station, which is considered a disutility when dropped off at Toronto Coach Terminal or Pearson International Airport. There is also a preference for taking the taxi to the airport over the bus or rail stations, which may be due to the higher travel cost associated with air travel compared to the other intercity travel alternatives. When arriving at Montreal, respondents prefer local transit for an egress mode for all alternatives. Only air travelers consider taxi as a beneficial egress mode choice, which may be correlated with intercity travel costs. The magnitude of these local travel mode coefficients are relatively high when compared that of intercity travel time and cost, which another indication to the importance of local accessibility on intercity travel demand. Income and age socioeconomic attributes were found as statistically significant attributes to this intercity mode choice model. Household income did not appear to have a direct relationship on intercity mode choice as individuals from different household incomes still preferred automobile over transit-based intercity travel alternatives. On the contrary, an increase in age increases the utility of rail, air, and high speed rail travel while decreasing the utility of bus travel. These findings imply that the age has a predictable influence on intercity mode choice but may also be affected by the non-linear effect of household income. It is apparent that local accessibility is a significant contributing factor to intercity mode choice. The effect of local accessibility is most apparent when long local access or egress times are required for a given intercity travel mode. The use of a SP survey design is instrumental to understanding how changes in local accessibility level-of-service influence intercity modal

23 Page 22 of choice. While socioeconomics does factor into the NL model, it is a smaller degree compared to local accessibility attributes. Waiting time is an important factors of transit related modes and air transportation. In case of transit modes for inter-city travel time, it is expected that passengers know about the schedules and arrive at the stations accordingly. So, waiting time becomes irrelevant in such context. This is the reason, waiting time for transit modes become insignificant and are left out of the model. In case of air transportation, waiting time (boarding time) is also systematic and more or less same for any types of trips. In case of such a systematic factor, the effects are constant and hence difficult to capture in the model. However, we believe that waiting time have effects and even though are difficult to capture in the model those are indirectly captured in the modespecific socio-economic variable coefficients. Model Validation To validate the NL model, the model coefficients were applied to the 86 entries not used for estimation. Comparing between the frequency of stated mode choice and predicted mode choice of both sets of respondents, the resulting modal shares for all five alternatives did not vary beyond 8%. This validation effort is a confirmation that the NL model is an accurate predictor of intercity mode choice. Figure 4 below is a graphical representation of stated and predicted modal shares. Value of Travel Time Savings The coefficient estimates from nested logit can be used in other forms of econometric analysis. One such form is the calculation of value of time travel savings (VTTS), which is a cost-benefit analysis that assesses the trade-offs between travel cost and travel time. The concept of VTTS

24 Page 23 of arises from the idea that the necessity of travel derives from demand for activities and the idea that travel time has a negative demand. Depending on the importance of the activity, individuals place a certain importance to reduce the travel time required to reach the activity and may be willing to pay a higher travel cost [15]. Some reasons why individuals may want to reduce his or her travel time are to use the time saved to yield a monetary benefit, spend the saved time in recreation or other activities, and to reduce any undesired attribute of travel such as discomfort or fatigue. VTTS is calculated by taking the ratio of the coefficient of travel time to coefficient of travel cost. Alternatively, the access and egress VTTS values measure the additional intercity travel cost that an individual would pay for a departure station to be located closer to his or her origin location. VTTS values obtained from the model estimation coefficients are listed below. Main intercity trip - $24.67/hour Local access - $40.54/hour Local egress - $45.01/hour The VTTS for the main intercity trip is similar to the values of existing literature (EcoTrain 2011). For access and egress VTTS, the respective local travel times were divided by the intercity travel cost rather than the cost of local travel. The difference between the intercity VTTS and access/egress VTTS values may be attributed by the relatively lower local travel times. While there are inherent weaknesses to this approach, as the intercity travel cost is independent of local travel time, the survey design would have been too complex to include the extra dimension of local travel cost.

25 Page 24 of CONCLUSIONS There were two primary contributions of this research: assessing the effectiveness of collecting responses using web-based methods (including social network based sampling) and evaluating effects of transit station accessibility on intercity mode choice. With the design of a web-based survey, the procedure to sample from multiple online social networks was used. It became clear that the use of a single individual s immediate social network was not enough to capture a representative sample of the target population. The process of sampling from multiple online social networks expanded the overall socioeconomic and geographic profiles of respondents at the cost of low response rates when sampling from larger social networks. Another alternative was to utilize the immediate social network of a number of individuals, which yielded improved response rates. The main limitation encountered during data collection was the low ratio between completed surveys and potential audience. The online social networks that have the largest potential audience were typically the ones where one individual has the least impact and audience. This relationship was expected and was supplemented by utilizing an online survey panel for additional responses. Overall, online social networks were observed to be a viable survey recruitment source when appropriate sampling techniques and communication tools are used to generate interest. The survey used in this paper compiled trip information and travel preference datasets with greater geographical disaggregation compared to currently available public data. From the nested logit model, it is evident that individuals have greater sensitivity to local access and egress travel time compared to intercity travel time on a per unit time level. Additionally, the NL model also found a significant relationship between access/egress modes and intercity stations. These findings validate initial prediction that local accessibility has a relatively large influence on

26 Page 25 of intercity modal choice. Policy changes related to of local accessibility in intercity travel plans may include; the effect of station locations on travel demand, partnerships with local transit or transportation services, and induced intercity travel due to improved local accessibility. Additionally, the lack of local accessibility attributes for the automobile mode may be one of the primary reasons why automobile has a higher modal share compared to existing intercity travel modes. From this model, the introduction of a HSR line in the QWC may result in a large modal shift towards HSR; however, long local access or egress times or a lack of local travel integration may greatly reduce the appeal of HSR. The paper presents an investigation and empirical result on inter-city travel mode choices while collecting data through an innovative and new approach of data collection. Specially, the potential of using social media for travel survey sampling frame is tested. However, this also results in some limitations in the empirical study. Online social network based sampling methods resulted in a skewed sample set towards the socioeconomic characteristics of the individual distributing the survey. The sample collected using social networks did have a bias towards the university population; however additional responses were collected via an online survey panel, which has more socioeconomically representative respondents compared with census data. Overall, the university population bias is somewhat subsided. Such, bias may not be as evident in other forms of data collection like call centers, street intercept, or mail outs; however the efforts of data collection was greatly reduced. In addition the distribution of revealed travel modes were similar between sampling from online social networks versus a more demographically representative survey panel for this project. Further research should be conducted to assess whether there are any links between internet use and travel preferences. However, it becomes clear from this study that there is a potential that social media can be used as a supplementary

27 Page 26 of survey collection method to reduce the necessity to rely on more traditional sampling methods. References Ashiabor, S., H. Baik and Trani, A Logit models for forecasting nationwide intercity travel demand in the United States. Transportation Research Record: Journal of the Transportation Research Board 2007: Ben-Akiva, M. and Lerman, S Discrete choice analysis. The MIT Press, Cambridge Massachusetts. Bhat, C. R A heteroscedastic extreme value model of intercity travel mode choice. Transportation Research Part B: 29(6): Bhat, C.R Accommodating variations in responsiveness to level-of-service measures in travel mode choice modeling. Transportation Research part A: 32(7): Burge, P. C., Kim, W. and Rohr C Modelling demand for long-distance travel in Great Britain. Rand Corporation, Santa Monica, CA. Available from [accessed 01 September 2015] EcoTrain Updated feasibility study of a high speed rail service in the Quebec City - Windsor corridor. Transport Canada. Available from [accessed 01 September 2015] Grisolía, J.M. and Ortúzar J.D Forecasting versus observed outturn: Studying choice in faster inter-island connection. Transportation Research Part A 44: Ahsan, H.M., Rahman, M.M. and Habib, K.M.N Socio-economic conditions and travel

28 Page 27 of behaviour of inter-city bus passengers: Bangladesh perspective. Journal of Civil Engineering, Institution of Engineers of Bangladesh 30(2): Rojo, M., dell Olio, L., Moura, J.L. and Gonzalo, H User satisfaction in interurban bus transit: Modelling relevant variables and their influence. Proceedings of the 10th International Conference on Applications of Advanced Technologies in Transportation, Athens, Greece. Rojo, M., Gonzalo, H., dell Olio, L. and Ibeas, A Modelling gender perception of quality in interurban bus services. Proceedings of the Institution of Civil Engineers, Transport 164 (1), Rojo, M., H. Gonzalo, H., dell Olio, L. and Ibeas, A Relationship between service quality and demand for inter-urban busses. Transportation Research Part A 46: Debrezion, G., Pels, E. and Rietvelt, P Modelling the joint access mode and rail station choice. Transportation Research Part E 45: Hsio, C-Y. and Hansen, M A passenger demand model for air transportation in a huband-spoke network. Transportation Research Part E 47: Mandel, B., Guadry, M. and Rothengatter, W A disaggregate box-cox logit mode choice model of intercity passenger travel in Germany and its implications for high-speed rail demand forecasts. Annals of Regional Science 31: Hensher, D.A., Louviere, J. and Swait, J Combining sources of preference data. Journal of Econometrics 89: IBI Group Making transportation sustainable: A case study of the Quebec City-Windsor corridor. Environment Canada, Ottawa.

29 Page 28 of Langan, P All Canadian HSR studies. Available from [Accessed 01 September 2015S]. Miller, E.J The trouble with intercity travel demand models. Transportation Research Record: Journal of the Transportation Research Board 1895: Ridout, R. and Miller, E. J A disaggregate logit model of intercity common carrier passenger modal choice. 16(4): Sonesson, T Inter-urban travel demand elasticities with emphasis on trip generation and destination substitution. Journal of Transport Economics and Policy 35: Statistics Canada Travel survey of residents of Canada 2006 to Ottawa. Available from [Accessed 01 September 2015] Statistics Canada Census profile - age, sex, marital status, families households, dwellings and language for Canada and forward sortation areas, 2011 census. Ottawa. Available from =0&GK=0&GRP=1&PID=104353&PRID=0&PTYPE=101954&S=0&SHOWALL=0& SUB=0&Temporal=2011&THEME=90&VID=0&VNAMEE=&VNAMEF= [Accessed 01 September 2015] Data Management Group Transportation tomorrow survey 2011 Version 1.0 data expansion & validation. Toronto. Available from [Accessed 01 September