ABSTRACT. To understand travel behavior and especially on-demand mobility (ODM) usage, it is essential to assess if attitudes towards modes influence behavior and if so, to what extent. Using measuring techniques such as trip diaries, information about attitudes are usually missing. Nevertheless, they might be relevant to explain observed behavior. Positively oriented individuals, which are willing to use other modes than their private car, may not be able to realize a specific behavior by reasons of lacking mobility offers and services. For this purpose, it is important for transport planning as well as forecasting of travel demand to know more about attitudes and general orientations towards modes.

For this study, we analyzed data from a survey conducting the usage of on-demand mobility services on the way to work and attitudes towards modes in general. The study was conducted in Shanghai, China. In Shanghai, services such as carsharing or ride hailing are widely spread and constantly expanding. As working is the main activity for most people, work trips are in the focus of this study.

Literature review In the past decades, several studies analyzed the influence of psychological factors on travelers‘ behavior in mode choice (Bamberg et al., 2003; Dobson et al., 1978; Gärling et al., 1998; Kroesen et al., 2017). As these “soft” attributes can play a major role in the decision-making process, it became popular to extend the classical discrete choice models with constructions of latent variables (Ashok et al., 2002; Ben-Akiva et al., 1999; Walker, 2001). In order to obtain high-quality data of these factors (such as attitudes and perceptions), it is important to choose suitable survey methods for this purpose (Carrasco and Lucas, 2015). Ben-Akiva et al. (1999) presented an advanced and simultaneous estimation approach that incorporates the latent variables directly into the choice process. Usually, latent variables are defined and integrated by a Multiple Indicator Multiple Cause (MIMIC) model or an ordered logit approach. For a more detailed explanation of the underlying concept and recent progresses of HCMs, the reader is referred to Kim et al. (2014). Recent studies already investigated influencing factors for choosing ODM services for traveling in general (Alemi et al., 2017; Dias et al., 2017; Rayle et al., 2016). However, their research focuses less on the influence of attitudes and perceptions, but much more on the socio-demographic context for the use of ODM services. This study tries to close this gap in literature by specifically examining the relationship between attitudes and the use of ODM services by commuters on their way to work.

Data Our analyses base on a largely unique dataset, as it captures comprehensive information about travel-related aspects such as daily and occasional travel behavior as well as psychological factors (e.g. car use motives and attitudes towards modes). This approach collects information about activities and mode choices by using a travel skeleton (von Behren et al., 2018). Additionally, we implemented two psychological item sets into the survey to investigate peoples’ attitudes (Hunecke et al., 2007; Steg, 2005). The survey has been conducted between October 2016 and January 2017. We designed and implemented a computer assisted personal interview. With this survey, data about travel behavior of one week (including ODM services) as well as socio-demographics and attitudes towards mode usage were conducted. The complete sample size contains 600 observations and is representative concerning age, gender and household size for the urban area of Shanghai.

Methodology We used a hybrid choice model applying an integrated choice and latent variable (ICLV) approach to analyze the effects of attitudes towards conventional modes and the usage of ODM services. In order to derive relevant variables for the choice component, we first estimated a multinomial logit model. In a second step, we extended this model by a random component. Only the random component for the alternative specific constant was found to be a relevant improvement for the model. For the determination of the latent variables, we performed a principal axis factor analysis (PAF) on the indicators from the psychological item sets. Three latent variables were identified. However, one was found not to be significant in the ICLV model. Consequently, we only used two latent variables in the final model: one describing the importance of cars and the experience with public transit. In the next step, we used multinomial logit models to derive possible influencing variables of the latent variables. The final determination took place in the ICLV model estimation.

In the final step, we set up the ICLV model (see Figure 1). For the indicators, we used an ordered approach, as a five-point Likert scale was used. For the latent variables, different influencing variables were significant. The latent variable for the importance of cars contains four influencing variables (car availability, income, stability in mode choice, smartphone ownership). The latent variable for the public transit experience contains nine influencing variables (bicycle availability, car availability, income, driver license, education, stability in mode choice, smartphone ownership, parking situation at work). For the estimation of all models, we used the CMC choice modelling code for R (CMC, 2017).

Results In Shanghai, a frequent usage of ODM services on work trips is observable. 4.34% of all reported work trips in our sample were conducted by ODM services. Moreover, about 15% of the participants have used ODM services for commuting at least once within the reported week.

The HCM helps to illustrate how various socio-demographics as wells as attitudes towards modes influence the usage of ODM services on commuting trips. The results of the choice component reveal an influence of income on ODM usage. Persons with a low income and an age over 35 use ODM services less often than persons with a high income and a age under 35 years. The results of the latent component show a influence of the latent variable describing the importance of the car on the usage of ODM services. Persons who recognize the importance of car as a status or a means to an end are more likely to use ODM services than persons with a low interest in cars. This result is conjecturable, because many ODM services are based on cars, such as car sharing or ride hailing. Especially, the car status has a high influence on the latent variable describing the importance of a car. This is partly because, in China, a car can still communicate status and prestige. The latent variable describing public transport experience is also significant. Persons having positive attitudes and having made positive experiences with public transport also are more likely to use ODM services on their commuting trips. Especially the attitudes towards privacy concerns in public transport systems have an influence on the usage of ODM services. Persons having less privacy concerns are more likely to use ODM services. Multimodal behavior of people is another important issue in relation with ODM service usage. Whenever a person has a less stable behavior and uses different modes during a week, the person is more likely to use ODM services than persons with a high mode choice stability. Mode choice stability often correlates with the domination of car usage.

Conclusion Understanding ODM usage and the influence of attitudes of travelers on the way to work requires a fundamental knowledge about attitudes and their influence on mode choice decisions. To capture persons comprehensively, we used a survey design combining revealed travel behavior and underlying psychological factors. To analyze the influence of attitudes on ODM usage, we applied an ICLV approach. Results show a significant influence of income and age on ODM usage. The results of the latent component highlight the public transit experience for the usage of ODM services. Further research involves a comparison with San Francisco, where a survey using the same method was conducted. We want to investigate whether persons in the U.S. use ODM services on their way to work for some other reasons or not.

Acknowledgements This abstract presents analyses based on the Urban Travel Monitor (UTM) funded by the BMW Group, Munich.

ABSTRACT. This study aims to develop a conceptual framework for a mode choice model in a small island developing state (SIDS). The research focused on household interactions and work tours in the mode choice of Trinidadians. Like many small island developing states, Trinidad has a wide variety of public transportation modes available. Also characteristic of SIDS is the use of multiple modes of transportation in one tour. The island has several para-transit modes that have similar characteristics to private cars. Hence, a cross-nested logit model (CNL) was considered to be the most suitable model. The expected outcome of this study is to develop a mode choice model, which would allow planners to simulate the travel behavior, and mode choice of travelers in response to certain policy measures. This is of particular interest in SIDS where the travel behavior can be more complex compared to developed countries.

ABSTRACT. This study reports the results of a randomised controlled trial to encourage commuters of an urban rail network to travel before the morning peak. The field experiment examined the relative effectiveness of four different reward structures, with pay-outs based on an individual's pre-treatment average entry time (PAET) into the system, in changing travel behaviour. We find that a two-step reward structure, with an intermediate reward for intermediate shifts and a larger reward for larger shifts, to be cost-effective in achieving the policy objectives. Furthermore, a reward structure which does not rely too heavily on financial gains performs the best in sustaining shifts in travel patterns after the initial intervention is removed.

ABSTRACT. In bimodal ridesharing, a private on-demand mobility service operator offers to drop off a passenger at a transit station, where the passenger uses the transit network to get to another transit station, and the service operator guarantees picking up the passenger to drop them off at the final destination. Such collaborations with public transport agencies present a huge potential to increase the ridership. However, most existing studies on dynamic dial-a-ride/ridesharing mainly focus on mono-modal cases only. We consider dynamic bimodal ridesharing problems where real-time information is available to anticipate future demand. A new non-myopic vehicle dispatching and routing policy based on queueing-theoretical approach is proposed and integrated with a non-myopic idle vehicle repositioning strategy to solve the problem. Several experiments are conducted to test the effectiveness of this integrated solution method and measure the benefit of bimodal cooperation. A case study of Long Island commuters to New York City (NYC) suggests having the bimodal operating strategy can cut user journey times and operating costs by up to 54% and 60% each for a range of 10 to 30 taxis per zone. The proposed model and solution algorithm provides useful tools for real-time operating policy design of shared mobility.

ABSTRACT. This paper presents the findings of a tour-based model that jointly investigates individuals’ decisions of tour participation, tour time allocation and accompanying travel choice. Tour participation and time allocation decisions for mandatory, maintenance and discretionary activity tour types are modeled jointly using a multiple discrete continuous extreme value (MDCEV) model framework. Model results exhibit considerable effects of socio-demographic attributes on individuals’ tour type choice and time allocation. For example, individuals belonging into 18 to 24 years age group are highly likely to participate in mandatory and discretionary tour. However, this age group exhibits high satiation (i.e. spend less time) while performing their mandatory activity tour compared to the older individuals. This study exclusively examines the influence of individuals’ tour-based accompanying travel choice on their tour engagement decisions. Accompanying travel models are developed for mandatory, maintenance and discretionary activity tours utilizing mixed logit (MXL) modeling approach. Results indicate significant effects of socio-demographic, time-of-day and built environment attributes on individuals’ tour-based accompanying travel choice. Furthermore, while exploring the effects of accompanying travel decisions on individuals’ tour engagement behavior, this paper finds statistically significant positive logsum values, which suggest that individuals’ probability to participate into a tour as their daily travel decision increases with their higher accessibility towards accompanying travel arrangements.

ABSTRACT. With the growing popularity of activity-based models, there is much interest in flexible but tractable models of time use. Bhat’s Multiple Discrete Continuous Extreme Value (MDCEV) model has recently captured much attention, but it has the important disadvantages of not differentiating between multiple events of the same activity. For example, from both a planning and behavioural perspective it is not equivalent to work during eight hours continuously, than working for two blocks of four hours each during a day, yet MDCEV makes no difference between them.
In this study, we compare three different approaches to expand traditional MDCEV models to address the multiple events issue in time use data. The first approach consists in exogenously splitting the day in three parts: morning, afternoon and evening; and estimating a different (but not independent) model for each part. The second approach implies considering an event of a given activity as the basic alternative, and exogenously enforcing these events to be consumed in an orderly fashion (e.g. the first event of work must be consumed before the second event of work within a day). Finally, we present a theoretical development of a model also based on events rather than activities, but where the consumption order is enforced by the very formulation of the model.
Using a large revealed preferences dataset of time use as a case study, we discuss benefits and limitations of each approach.

ABSTRACT. In this paper we present for the first time a joint time-expenditure allocation and discrete mode choice model, which includes all required components (also expenditures), accounts for inter-block correlations and is estimated using maximum likelihood (ML) method. In the estimation we use the Mobility-Activity-Expenditure Diary data-set for Austria, which allows us to use and model the observed consumption expenditure data and not to artificially input it, as it was done before (Munizaga et al. (2008)). For faster and more robust optimization we have analytically derived the log-likelihood derivatives and used them with nonlinear optimization algorithms from the R package Henningsen and Toomet (2011). The newly implemented estimation procedure allowed us to simultaneously estimate the joint model, to valuate and compare travel-time-reduction (value of travel time savings) and quality (value of time assigned to travel) prioritizing indicators for Austria.

ABSTRACT. The rapid development of Information and Communication Technologies such as mobile broadband, smart devices, secure connectivity or cloud-based storage and services has led to the increasing decoupling of work (and other activities) from specific locations (Hislop & Axtell, 2007; Vilhelmson & Thulin, 2001; Watts & Urry, 2008). While conventional wisdom sees flexible work opportunities to fall mainly onto knowledge-based workers, this perception is restrictive and limiting by the concept of occupations. For example, a computer programmer would be seen as largely unrestricted to work in mobile settings while a hairdresser as virtually excluded from that opportunity.However, this view completely ignores the fact that each occupation consists of a number of tasks, i.e. duties and responsibilities. When viewed through that perspective, a programmer may be required to attend an exhibition to showcase the software, while a hairdresser can still manage their diary or accounting in a mobile setting. This discrepancy reveals a fundamental shortcoming in the way that such practices and the associated travel behaviour, use of ICT and productivity implications are modelled. Hence we claim that in order to reveal and model mechanisms making individuals to choose mobile work practices, there is an urgent need to begin exploring the role of ICT in enabling mobile work practices and the impact on travel behaviour and productivity using the task-based definitions of occupations. Armed with this knowledge, we use the UK Time Use Survey (UK TUS) 2014-15 to empirically investigate the practices of mobile work practices as a function of the tasks associated with occupations. To enable that, we enrich the UK TUS dataset with information regarding typical duties (generalised work tasks) associated with particular occupations contained in the Occupational Information Network (O*NET) database maintained by the US Department of Labor (O*NET, 2017). We match the respondents using the reported UK standard occupational categories (SOC) which we map onto the US SOC allowing the use of O*NET. We subsequently explore the enriched UK TUS dataset for presence of mobile practices often discussed and noted in qualitative studies (Hislop & Axtell, 2007; Vilhelmson & Thulin, 2001; Watts & Urry, 2008). Hence we explore a number of research questions quantitatively, including how a particular work tasks and responsibilities associated with any occupations are correlated with the propensity to conduct work in non-office settings or how the propensity to conduct tasks in mobile setting is related to the duration of work, hourly salary (as a proxy to productivity), level of enjoyment and work-life balance. In doing so, we also identify more complex patterns and implications of work practices within the wider daily time use patterns by relying on time use theory (Jara-Díaz & Rosales-Salas, 2017).

ABSTRACT. Governments urgently require decision tools to deal with what has become a pandemic, road traffic accidents, resulting in millions of deaths around the world every year. Evidence shows that the human factor is one of the major causes of road accidents, and there is much interest in identifying the variables with higher impact on drivers’ perception of risk. To this aim, we designed a series of hypothetical driving scenarios considering some attributes which have been strongly associated with increased accident risks: (i) driving speed, (ii) driving against the traffic in a one-way street, (iii) overtaking on a curve, and (iv) driving under the influence of alcohol or drugs. With this stated choice data, we estimated a hybrid discrete choice model incorporating two latent variables, Driver Concentration and Safe Driving, that depend on socioeconomic characteristics of drivers (i.e., sex, age, occupation, vision/hearing impairment licence restrictions and previous accident experience). Our results may contribute to the design of public policies geared to prevent accidents by encouraging safe driving behaviour. This is particularly relevant in Latin America where over 95% of fatal road accidents have a human factor cause.

ABSTRACT. Context Modellers of travel behaviour are increasingly incorporating travellers’ responses to uncertain conditions by including a representation of attitudes towards risk. The approach is particularly relevant to commuting behaviour where travellers face travel time variability due to congestion. The transport literature adopts non-expected utility from economic research with the Arrow-Pratt constant relative risk aversion (Pratt 1964) to indicate commuters attitudes towards travel time variation (Hensher, Greene & Li 2011). However, the interpretation of a risk attitude parameter α in the stochastic utility model is not straightforward and an arbitrary interpretation of more/less stochastic risk aversion, relying solely on the magnitude of risk parameter, might lead to an incorrect inference (Wilcox 2011). Moreover, this conclusion is specifically based on the analysis of a single monetary attribute with identical expected values (mean-preserving-spread pair) between two risky prospects. Unlike simple monetary games usually tested in economics, in transport stated preference surveys simultaneously present multiple attributes to examine respondents’ trade-offs and choices among alternatives (Louviere, Hensher & Swait 2000). The equivalent expected values of a risky attribute such as travel time variability may lead to dominant profiles and fail to mimic the reality in daily commuting (e.g. travel time distributions in the context of tolled and non-tolled roads). Following the lead from recent research into stochastic utilities of risk, this paper demonstrates that interpreting risk in a mode choice setting is highly dependent on the attribute values included in the choice task, which leads to a number of suggestions in designing a stated preference survey where one attribute is presented as a risky factor. In terms of the mean-preserving-spread pair across risky prospects, risk attitudes are determined by the sign of risk parameter although the interpretation of α as more stochastically risk seeking or averse with the magnitude value is not always clear. However, when the expected values differ across alternatives, the definition of which is the safe alternative and which is the risky alternative is not always clear (Rothschild & Stiglitz 1970). This paper focusses on identifying second-order stochastic dominance and compares the inference on α to an assumption that the alternative with the greater variance presents more risk to the traveller. Theory The paper examines the random utility equivalent to stochastic risk attitude choice axiom whereby the risk attitude is uncovered by choice observation through the estimated differences between utilities – and, therefore, the choice probabilities – between the safe alternative and the risky alternative

Stochastic risk attitude choice axiom: All else being equal, a stochastically more risk averse (SMRA) agent is more likely to choose a safe alternative among two lotteries than a stochastically less risk averse agent

Labelling a stochastic risk attitude in a discrete choice setting: All else being equal, a risk attitude is inferred be the difference in systematic utilities, between a safe and risky alternative as judged by agent A compared to the corresponding difference as judged by an abstract agent N, who holds a risk neutral position. The comparison is represented by:

R_difference=[u(x_Air )-u(x_Ajr )]-[u(x_Nir )-u(x_Njr )] Eq1

where u(x_Air )-u(x_Ajr ) is the difference in systematic utilities for the safe alternative i and the risky alternative j given that at least one attribute, x_r , is random. In a random utility setting the definition of the attitudes ‘risk averse’ is such that R_difference<0 and ‘risk seeking’ is R_difference>0

Transitivity (stochastically more risk averse): an agent is said to be stochastically more risk averse (SMRA) if the difference between systematic utilities for a safe and a risky alternative is smaller than the difference for a stochastically less risk averse (SLRA) agent, such that:

u(x_(A_1 ir) )-u(x_(A_1 jr) )

where smaller indices imply SMRA.

Transitivity implies the R_difference inequalities hold for any two risk positions.

... the remainder of the paper includes embedded figures. please see the attachment.

Thank you, Brett

ABSTRACT. Extended abstract 1. Introduction Travel behavior plays two roles in contemporary economic evaluation of transport projects. Firstly, travel behavior plays a positive role through predicting how travelers alter their behavior in response to a transport project. Secondly, travel behavior plays a normative role because economic evaluations such as Cost-Benefit Analysis (CBA) postulate that the utility individuals derive from impacts of a government project (e.g. travel time reductions and safety improvements) can be inferred from observing individuals’ (hypothetical) travel decisions. For instance, the dominant empirical approach to infer Value of Time (VoT) and the Value of Statistical Life (VoSL) is based on experiments in which individuals are asked to make a choice between travel options which differ in terms of travel times, safety and travel costs (e.g. Bahamonde-Birke et al., 2015; Batley et al., 2017; Ehreke et al., 2015; Hensher et al., 2009; Kouwenhoven et al., 2014). The normative role of travel behavior in the evaluation of transport projects has been criticized by various scholars (e.g. Ackerman and Heinzerling, 2004; Hauer, 1994; Mackie et al., 2001). Ackerman and Heinzerling (2004) state that respondents participating in experiments for inferring the VOT and the VOSL are asked to trade-off after tax income, travel time and safety in their role as consumer of mobility, whilst preferences of individuals in their role as consumers may be a poor proxy for how the same individuals in their role as citizens believe that Government should trade-off tax money, safety and travel time. Ackerman and Heinzerling (2004) contest the decision of the US Government against banning cellphone use in the car based on calculations that people who are talking while driving are willing to pay a lot to talk on the phone more than many people who face deadly risks are willing to pay to avoid the risk of being killed. In their view, the consumer values for talking while driving cannot legitimize that some US citizens will end up in the morgue because they are hit by other US citizens distracted by their cellphone while driving a car: “using private market behavior as a standard for public policy overlooks the possibility that people will have different preferences when they take on different roles” (Ackerman and Heinzerling, 2004, p. 191). The argument posed by, amongst others, Ackerman and Heinzerling (2004) is also called the ‘consumer-citizen duality’. Mouter et al. (2017) state that the key distinction between citizen preferences and consumer preferences is that they involve individual preferences inferred from choices within different budget constraints. Namely, while consumer preferences involve an individual’s preferences within his/her personal budget constraint (e.g. after tax income and time), citizen preferences involve an individual’s preferences regarding the allocation of the government’s budget. Moreover, Mouter et al. (2017) conducted several experiments in which respondents were asked to choose either as consumer or citizen between two routes which differed in terms of travel time, accident risk and impact on their after tax income. In one consumer experiment respondents were asked to choose as a car driver between routes which differed in travel time, safety and toll costs. In a citizen experiment individuals were informed that the government decided to allocate non-specific taxes to the construction of a new road and considered two routes that differ in terms of travel time and accident risk. Subsequently, the respondents were asked to recommend one of the routes to the government. Mouter et al. (2017) established that the respondents who were asked to provide recommendations as citizen assigned substantially more value to accident risk than travel time when compared to the respondents who were asked to make route choices as a consumer of mobility. This finding indicates that the ‘consumer-citizen duality’ exists and is relevant for transport research. Moreover, this raises the question how this so-called ‘consumer-citizen duality’ can be explained. To the best of our knowledge, no contribution in the literature exists which aspires to explain empirically why individuals trade-off travel time and accident risk differently as consumer and citizen. This paper aims to ameliorate this gap through administrating a survey consisting of a Stated Choice (SC) experiment in which respondents make trade-offs between travel time and accident risk, and a part in which motivations are inferred. 2. Methodology and data collection The SC experiment consists of two parts. One part is a consumer experiment in which respondents are asked to choose between two routes as a car driver; the other part of the SC is a citizen experiment in which respondents are asked to make a recommendation to the government on which route to construct. After the SC experiment respondents are asked to motivate the differences in trade-offs they make when situated in a citizen and in a consumer role. Moreover, in our study we also investigate whether the motivations provided by the respondents are affected by the risk levels they evaluate and the order in which they conduct consumer and citizen choice tasks. Based on the two design objectives discussed above we designed four different experiments. We asked a survey company (Kantar Public) to draw four random samples from the population of Dutch citizens of 18 years and older. 412 of the respondents recruited by the survey company completed the questionnaire. Below, we describe how the respondents are distributed among the four experiments: Exp. 1: 105 respondents first completed car driver choice tasks in the context of a provincial road (relatively high risk); Exp. 2: 109 respondents first completed car driver choice tasks in the context of a motorway (relatively low risk); Exp. 3: 97 respondents first completed citizen choice tasks in the context of a provincial road (relatively high risk); Exp. 4: 101 respondents first completed citizen choice tasks in the context of a motorway (relatively low risk). 3. Results 4.1 Descriptive results Firstly, we analyzed the choices of respondents with multinomial logit models to investigate the extent to which respondents make different choices in the four experiments. Table 1 presents the estimation results.  

TABLE 1: Estimation results multinomial logit models experiments 1-4

B_D = marginal utility of one additional traffic casualty on a road B_TT = marginal utility of one additional minute travel time Table 1 shows that there is a substantial difference between the ratios of the parameters B_D and B_TT in the car driver and the citizen choices. For experiments 1, 2 and 4 the marginal rate of substitution between additional minutes of travel time and additional traffic casualties on a road is higher in the citizen choices than in the car driver choices. A two-sample t-test shows that in experiments 1 and 2 the marginal rates of substitution of car drivers and citizens are significantly different from one another at conventional levels of significance (α = 0.05). In experiments 3 and 4 differences are not statistically significant which is probably caused by the low parameter B_TT for the citizen choices. Next, we analyze which explanations respondents mention for assigning a relatively high value to safety in the citizen choices and a high value to travel time in the consumer choices. From the 412 respondents that completed the questionnaire 152 respondents provided statements that contribute to the explanation of why citizens assign more value to accident risk than travel time when compared to car drivers. In our analysis, we focus on these statements We coded these statements and clustered them in four cognitive explanation categories and five normative categories. The cognitive explanations involve that individuals make different choices as car driver and citizen because they perceive accident risk differently in these two roles. The normative explanations involve that individuals believe that the government should assign more value to safety when compared to individual car drivers. Figure 1 presents the number of times the nine explanations are mentioned in the four experiments and in total (final column).

  FIGURE 1: Overview of differences mentioned between four experiments

The first observation that follows from Figure 1 is that respondents both provided explanations that are prevailing in existing literature (particularly categories 1-3) and explanations that are new to the literature. Category 4 is one of these new categories. These respondents state that they make a different choice as a car driver and a citizen because, in real life, they do not possess any information regarding accident risk when making a route choices as a car driver, and as a result of this unawareness they are unable to take this information into account when making route choices. Hence, as a car driver, they choose for the fastest route. Various respondents argue that the government has better information available concerning accident risk when making a route choice. For this reason they do recommend the government to select the safest route. Moreover, respondents provided various normative explanations for assigning more value to safety in their role as citizen that are not found in the literature (categories 7-9). Respondent clustered in Category 7 recommend the government to choose for the safest route, because they think that it is not necessarily the government’s duty to reduce travel times. These respondents believe that car drivers have a relatively high own responsibility to reduce their own travel time. For instance, car drivers can try to avoid peak hours and they can start their trip earlier to ensure that they arrive on time. Category 8 respondents argue that car drivers will always have a tendency to try to reach their destination as fast as possible. These respondents think that taking care for the safety of the road network is the government’s correct response to this habitual behavior. It is interesting that the statements in this category indicate that respondents explicitly recommend the government against using their consumer behavior as a standard for public policy. These statements clearly support the assertion of scholars we discussed in the introduction that consumer behavior may be a poor proxy for how the same individuals in their role as citizen believe that government should trade-off travel time and safety. The respondents’ car driver preferences and citizen preferences seem to be communicating vessels. Because car drivers choose for speed, citizens recommend the government to focus on safety. Respondents clustered in Category 9 argue that the government should predominantly focus on the safety of the road network because this allows car drivers just to focus on travel time when making a route choice. Hence, when the government takes care of the safety of the road network this eases the life of the car driver.

Conclusion and discussion The purpose of this study is to empirically explain why individuals trade-off travel time and accident risk differently as consumer and citizen. We administrate a Stated Choice experiment in which respondents make choices as consumer and citizen between routes that differ in travel time and safety. Moreover, respondents are asked to provide motivations for their choices. We identified nine explanations that can be clustered in four cognitive and five normative explanations. Various normative explanations are new to the literature. Particularly two normative explanations are interesting because they involve an explicit recommendation to the government against using their consumer behavior as a standard for public policy: (1) government should respond to car drivers’ tendency to choose the fastest trip by building safe routes; (2) when the government ensures safety of the road network car drivers can choose the fastest route without being concerned about the impact of their route choice on accident risk. The explanations for why citizens prefer safety and consumers desire speed might provide useful input for a debate on whether preferences inferred from choices individuals make as a consumer comprise the single relevant input for the appraisal of transport-related government policy options. For instance, it is interesting to contemplate the extent to which it is defensible to ground the appraisal of government projects on consumer behavior now that it is clear that some individuals explicitly recommend the government against using their consumer behavior as a standard for public policy.

References Ackerman, F., Heinzerling, L., 2004. Priceless: on knowing the price of everything and the value of noting. The New Press. New York. Bahamonde-Birke, F.J., Kunert, U., Link, H., 2015. The Value of a Statistical Life in a Road Safety Context — A Review of the Current Literature. Transport Reviews, 35(4), 488-511. Batley, R., Bates, J., Bliemer, M., Börjesson, M., Bourdon, J., Cabral, M.O., Chintakayala, P.K., Daly, A., Dekker, T., Drivyla, E., Fowkes, T., Hess, S., Heywood, C., Johnson, D., Laird, J., Mackie, P., Parkin, J., Sander, S., Sheldon, R., Wardman, M., Worsley, T., 2017. New appraisal values of travel time saving and reliability in Great Britain. Transportation, in press, 1-39. Ehreke, I., Hess, S., Weis, C., Axhausen, K.W. 2015. Reliability in the German value of time study. Transportation Research Record 2525, 14-22. Hauer, E., 1994. Can one estimate the value of life or, is it better to be dead than stuck in traffic? Transportation Research Part A 28 (2), 109–118. Hensher, D. A., Rose, J.M., Ortúzar, J. de. D., Rizzi, L.I., 2009. Estimating the willingness to pay and value of risk reduction for car occupants in the road environment. Transportation Research Part A, 43(7), 692–707. Kouwenhoven, M., de Jong, G.C., Koster, P.R., van den Berg, V.A.C., Verhoef, E.T., Bates, J., Warffemius, P.M.J., 2014. New values of time and reliability in passenger transport in The Netherlands. Research in Transportation Economics 47, 37-49. Mackie, P.J., Jara-Díaz, S.R, Fowkes, A.S., 2001. The value of travel time savings in evaluation. Transportation Research Part E 37, 91-106. Mouter, N., van Cranenburgh, S., van Wee., G.P. 2017a Do individuals have different preferences as consumer and citizen? The trade-off between travel time and safety. Transportation Research Part A 106, 333-349.

ABSTRACT. This paper develops a multivariate model to investigate the determinants of individuals’ safety perceptions and willingness to adopt autonomous vehicle (AV) technology. The analysis is based on data collected by the authors from the Dallas-Fort Worth Metropolitan Area (DFW), Texas, United States. In the model, AV safety perception and current vehicle ownership/usage behaviors are considered endogenous variables that impact the individual’s future vehicle purchase intentions. Underlying latent psychological constructs representing personal core values and technology-savviness are used to capture individual taste heterogeneity and to create jointness among all model outcomes.

ABSTRACT.  

The activity scheduling literature has long advocated the idea that the demand for travel be derived from the demand for activities. In the last decades, the importance of the influence of “others” in activity patters has been recognised and increasingly incorporated in behavioural models. In this work, we make use of a comprehensive dataset including a 2-week smartphone app-collected travel and activity diary, a name generator & name interpreter (to collect detailed social network data) and a detailed socio-demographic questionnaire. The information collected from the 452 people is therefore extremely rich and suitable for investigating the activity patterns as well as the related social dimension. Some of the behavioural questions that we will address are:

-Which activities are performed with which social contacts?

-How do people interact with specific social contacts in their social network?

-Is the order in which people list their contacts in the name generator reflecting the strength of their relationship?

-Is there an underlying factor that links all these different dimensions?

We apply different models to analyse the different components and link them together through a latent construct.

The present work will contribute to the growing literature aiming to assess the links between social interactions by different means of communication and activity participation and time investment with specific network members. The joint estimation of advanced choice models involving both decision mechanisms can help unveil behavioural dynamics and answer some of the compelling questions about substitution and complementarities between ICT and face-to face interaction.

ABSTRACT. Home-based telework (or telecommuting) has been consistently growing and it is considered by a vast number of policymakers as an attractive travel demand management strategy. Since the beginning of the 1990s telework has been the subject of a substantial amount of literature. Contrary to the expectancy of policy makers, the more recent empirical evidence suggests that the potential of telework to reduce total vehicle travel is at best weak (Nelson et al., 2007; Hjorthol and Nossum, 2007; Zhu, 2012; Zhu and Mason, 2014; He and Hu, 2015; Kim et al., 2015). This results in a great part from the effects on non-work related trips, travel mode choice, and household residential location, all of which can diminish the scope for a net reduction in travel (Mokhtarian et al., 1995).

With the exception of a few recent studies (e.g. Melo and de Abreu e Silva, 2017c; Kim et al., 2015; Zhu and Mason, 2014; Zhu, 2013; Sener and Bhat, 2011; Helminen and Ristimaki, 2007), there is little evidence using nation and region wide large-scale data to study the impact of home telework at the intra-household level. Home telework has implications not only for the travel behaviour of the teleworker, but also for the other household members, in particular the teleworker’s partner. One possible intra-household effect of home telework is that it can encourage residential relocation closer to the workplace of the non-teleworker household member, reducing his/her commute and possibly also the household total commute. There can also be effects resulting from changes in the allocation and sharing of activities and the associated travel between household members (e.g. escorting children to school, shopping). It is not obvious a priori whether the overall net effect of home teleworking on other (non-teleworking) household members should be positive or negative, and other factors, e.g. land use characteristics, may play a role here.

Zhu (2013) used data from the 2001 and 2009 US National Household Travel Surveys to investigate the effect of home telework status of one worker on the commute length and duration of his/her partner and found it did not have a significant effect, suggesting that the longer commute length and duration of teleworker households is largely due to the longer commute of the teleworking members. Using the same data, Zhu and Mason (2014) also found that non-work vehicle miles travelled by a household’s non-working member were not affected by the teleworking status of other household members. Kim et al. (2015) used data from the 2006 Household Travel Survey in the Seoul Metropolitan Area to investigate the effect of home telework on different household members and travel purposes and found that although home telework appeared to reduce home-to-work kilometres travelled, teleworker’s non-work related trips, as well as other household members’ non-commuting trips, were longer compared to non-teleworkers and their household members. Melo and de Abreu e Silva (2017c) used data from GB’s National Travel Survey for the period 2002-2012 and found that the home teleworker status of one of the household’s members did not appear to influence the partner’s commuting travel. However, the authors did not study possible intra-households effects for other travel purposes, [particularly non-work travel.

This paper aims to study the effects of teleworking on total weekly kilometres travelled by mode by workers in two-worker households, while at the same time incorporating telework frequency in a path analysis framework where decisions relating land use characteristics of the residential and employment areas and commuting distances of both workers; as well as household car ownership levels are incorporated. Travel modes are grouped into car-based (both as a driver and passenger), public transport (encompassing different types of transit) and active travel (walking and cycling). This path analysis framework is adapted from previous research which studied single worker households (de Abreu e Silva and Melo, 2017a) and aggregated travel from two-worker households (de Abreu e Silva and Melo, 2017b). It advances previous research on two-worker households by explicitly incorporating intra-household interactions in decisions relating to commuting distance, teleworking frequency and kilometres travelled by mode.

This path analysis framework captures both short-term and long-term travel and residential and employment location decisions. In this framework only socioeconomic individual and household characteristics are considered endogenous to the model, thus accounting for self-selection effects either due to specific needs or preferences of individuals and their households. In this model framework effects go from longer term to shorter term decisions, while at the same time allowing for feedback effects, where shorter term decisions affect longer term ones. Since home telework is considered a shorter term travel related decision it is modelled as a function of the longer term travel-related decisions, namely commuting distance. Residential location is a long-term decision with high transaction costs (Giuliano, 1989, Golledge and Garling, 2003), and hence is much more difficult to reverse than the decision to telework. Although this assumption is supported by previous research (de Abreu e Silva and Melo, 2017a), alternative specifications (with reverse and non-recursive relationships) will also be tested.

This paper uses data from the National Travel Survey (NTS) for England and the period between 2005 and 2016. The NTS collects data for household trip diaries for 7-day periods. A sample of two-worker households is used in this study, and excludes households in which workers work from home and do not have the same work location for at least 3 days a week, are self-employed in companies with only one worker, and work in agriculture, fisheries or mining.

This work contributes to the literature in several ways. It develops a modelling approach that explicitly accounts for endogenous relationships in the chain of decisions relating residence and work location land use characteristics, car ownership, home telework and travel patterns. It uses disaggregated household data from a large-scale nationwide survey, with 7-day travel diaries, whereas most previous studies using national or regional large scale travel surveys only used single-day travel diaries. The use of 7-day travel diaries allows to better capture travel off-set between different days of the week. It uses an ordinal measure of home-based telework which takes into account its frequency, in contrast to previous studies which, except for a small number of cases (Sener and Bhat, 2011; Tang et al, 2008; Walls et al., 2007), have typically used binary measures. Finally by explicitly modelling both household workers travel behaviour, in terms of distances and travel modes, it allows studying the existence, or not, of complementarities in job location between both workers and the influence of the decisions related with telecommuting and travel of one worker on the partner’s travel behaviour. The obtained results will be discussed in terms of its implications on sustainability, the incorporation of distances travelled by mode allows for a rough estimation of emissions, and in terms of the insights the developed model will provide about intra-household interactions.

REFERENCES de Abreu e Silva, J. and Melo P. C. (2017a), Home telework, travel behavior, and land use patterns: A path analysis of British single-worker households, presented at the World Symposium on Transport and Land Use Research 2017, 2-6 July, Brisbane, Australia. de Abreu e Silva, J. and Melo P. C. (2017b), The effects of home-based telework on household total travel: A path analysis approach of British households, presented at the 20th EURO Working Group on Transportation Meeting, EWGT 2017, 4-6 September, Budapest, Hungary Giuliano, G., (1989), Research policy and review 27. New directions for understanding transportation and land use. Environment and Planning A, 21(2), 145–159. Golledge, R. and Garling, T. (2003), Spatial Behavior in Transportation Modeling and Planning in Transportation Systems Planning, Methods and Applications, K. G. Goulias (ed.), CRC Press. He, S.Y. and Hu, L. (2015). Telecommuting, income, and out-of-home activities. Travel Behaviour and Society, 2(3), 131–147. Hjorthol, R. and Nossum Å. (2007), Teleworking - Possible Interaction with Travel Patterns. In: TRANSPORT, A. F. E., ed. European Transport Conference, 2007 Leiden, Netherlands. Kim, S.-N., Choo, S., Mokhtarian, P.L., 2015. Home-based telecommuting and intra-household interactions in work and non-work travel: a seemingly unrelated censored regression approach. Transport. Res. Part A: Policy and Practice. 80, 197–214. Melo, P.C. and de Abreu e Silva, J. 2017. The impact of home telework on household and intra-household commuting in Great Britain. Transportation Research Part A: Policy and Practice Vol. 103, pp. 1-24 Mokhtarian, P. L., Handy, S. L. and Salomon, I. (1995), Methodological issues in the estimation of the travel, energy, and air quality impacts of telecommuting. Transportation Research Part A: Policy and Practice, 29, 283-302. Nelson, P., Safirova, E. and Walls, M. (2007), Telecommuting and environmental policy: Lessons from the e-commute program. Transportation Research Part D: Transport and Environment, 12(3), pp.195–207. Sener, I. N., & Bhat, C. R. (2011). A copula-based sample selection model of telecommuting choice and frequency. Environment and Planning A, 43(1), 126–145. Tang, Wei; Mokhtarian, Patricia L; & Handy, Susan L. (2008). The Role of Neighborhood Characteristics in the Adoption and Frequency of Working at Home: Empirical Evidence from Northern California. Institute of Transportation Studies. UC Davis: Institute of Transportation Studies (UCD). Retrieved from: http://escholarship.org/uc/item/13x2q3rb Walls, M., Safirova, E., & Jiang, Y. (2007). What drives telecommuting? Relative impact of worker demographics, employer characteristics, and job types. Transportation Research Record, (2010), 113–122. Zhu, P., 2013. Telecommuting, household commute and location choice. Urban Stud. 50, 2441–2459. Zhu, P., Mason, S.G., 2014. The impact of telecommuting on personal vehicle usage and environmental sustainability. Int. J. Environ. Sci. Technol. 11, 2185–2200.

ABSTRACT. Whilst there has been a significant number of study exploring the time use of individual across different group of travelers and across different days of the week, most of the previous studies neglected the role of household members in influencing the individual’s daily time allocation. This study aims to contribute to this research gap. Using a three week of time use and activity diary from Bandung Metropolitan Area, Indonesia, this study aims to explore the plausible impacts of altruism between adult members of the household to the day-to-day variability of time use allocation of the household members. The use of data from developing countries also would reveal the impact of unique local culture in allocating responsibility, trip and time allowance within household to the time allocation of each household member. A joint estimation model of time use allocation between 2-adults’ household case, on six typical shared activities (i.e. grocery shopping, household chores, babysitting, picking up children, relaxing, and social activities) was developed.The results show that: The weights of altruism differ substantially between husband and wife and between different activity types. Income and the presence of children polarise husband’s altruism behaviours (weights). Accessibilities to wider crowd and amenities matter in stimulating the altruism behaviours within the household. Unlike the results from previous studies which based on developed countries’ data, the results of the preliminary analysis show a very strong tendency of delegation of activities/responsibilities within the household. Nobody always in charge, but definitely not an equal distribution of roles at all activities as what has been widely promoted in developed western society.

Keywords: Time use, day-to-day variability, intra household interaction, Bandung, Indonesia

ABSTRACT. 1 The link between leisure trips and interaction modes

Social activities are responsible for a substantial portion of trips (Axhausen, 2005). Related trips, intending to allow social activities, are influenced by new ways of social interaction, oc-curring since the rise of new information and communication technologies. Urban and transport planners need to take these changing communication patterns into account as they impose new demands on urban environments and transportation services. Therefore, it is im-portant to study face-to-face interactions, implying physical movements, as well as ICT-mediated communication patterns for social purposes, that do not rely on travel. From a social network perspective, a certain amount of contact between two persons is nec-essary to maintain their social tie. This contact can include face-to-face meetings as well as contacts mediated by different ICT tools. In recent years the possibilities for ICT-mediated contacts have increased tremendously. This will have an effect on face-to-face interactions and therefore also on travel. Regarding the effects of ICT on activity travel patterns generally four options are possible: substitution, complementarity, neutrality and modification (Salo-mon, 1986; Mokhtarian, 1990; Graham and Marvin, 1996). Previous studies have shown that, for leisure or social activities, the effect of ICT is generally complementary (e.g. Mokhtarian and Meenakshisundaram, 1999; Senbil and Kitamura, 2003; Mokhtarian et al., 2006; Wang and Law, 2007; Frei and Axhausen, 2008; Mosa et al., 2010; van den Berg et al., 2012; Car-rasco, 2011). Lately, the relationship between ICT and travel patterns has received a substantial amount of attention in the literature on travel behaviour. However, still little is known regarding the effect of ICT on travel for social activities. Thus far, the relationships between communication patterns by different ICT-mediated modes, face-to-face interactions for leisure purpose and travel have been largely neglected in the travel behaviour literature. Recently, data collection efforts have been developed that can bridge this gap, namely in Toronto, Canada by Hogan et al. (2007), in Zurich, Switzerland by Frei and Axhausen (2007), in Eindhoven, the Netherlands by Van den Berg et al. (2009), in Switzerland by Kowald and Axhausen (2012), in Concepción by Carrasco and Cid-Aguayo (2012) and again in Zurich, Switzerland, by Guidon and Axhausen (2017).

2 Social networks and transport research

The social network approach has been developed and used for decades in sociology (e.g. Wasserman and Faust, 1994). It regards social networks as a composition of nodes (people) and links (the ties between people). Within the social network approach social networks can be studied as whole networks in a given setting (e.g. a school, a company, or a city), which requires knowledge of all existing relationships among people in that setting. Another way to study social networks is the egocentric approach. In this approach the personal network mem-bers (alters) are elicited for the given individuals (egos). To elicit the alters name generating questions are used (Degenne and Forsé, 1999; Marsden, 2005), with the choice of name gen-erators depending on the purpose of the study. To collect more information on each alter (or ego-alter relationship) additional questions (name interpreters) are used. These questions may for instance include age and gender of the alter, to assess the influence of homophily (the tendency to interact with people that are simi-lar to you) (McPherson et al., 2001), tie strength and duration, and role relationship (e.g. fami-ly vs. friend). In addition, social network data collections in the field of transportation have also included geographical distance between the homes of ego and alter and contact frequen-cy by different communication modes (face-to-face, telephone, SMS, email), as these are im-portant aspects of social activity-travel behaviour. The study of social networks in transportation research is relatively new, starting with the work of Axhausen (2005), Dugundji and Walker (2005) and Páez and Scott (2007). The above mentioned six studies from four countries all employed the personal network approach and aim to explain the influence of social network characteristics on the generation of (social) activities. Analyses from each of these studies have indicated that incorporating social net-work characteristics is crucial in studying social activity-travel behaviour. An in-depth com-parison of the distance patterns between social network members in the first five datasets is reported in Kowald et al. (2013). However, a comparative investigation of the factors that in-fluence social interaction frequency among social network members with different communi-cation modes has not been conducted yet and is the objective of the present study.

3 Influences on interaction modes and frequencies

The frequencies of communication and the choice of communication mode differ per indi-vidual and per ego-alter-relationship. Previous research has indicated that contact frequencies are influenced by personal characteristics such as gender, age, time constraints (e.g. work hours), household composition, and the availability (and costs) of mobility and communica-tion tools. For example, communication frequencies with different ICT-mediated modes not only depend on the availability of these modes, but also on familiarity with these modes. The first ones to adopt new ICT tools in general are young, highly educated males. Some studies have indicated that these groups still have higher contact frequencies with ICT-mediated modes such as e-mail and SMS (e.g. Frei and Axhausen, 2008; van den Berg et al., 2012b). Regarding time constraints previous studies have indicated that full-time work and the presence of children in the household can be a constraint for longer out-of-home face-to-face social activities (e.g. Carrasco and Miller, 2006). With respect to access to mobility tools Ban-ister and Bowling (2004) found that elderly people with access to a vehicle (and people with access to good local transport) were likely to undertake more (face-to-face) social activities, which seems an intuitive finding. On the other hand though, Farber and Páez (2009) found that people who are more automobile-reliant tend to participate in fewer social activities. In addition to personal and household characteristics, social network characteristics have been found to influence communication frequencies. People with a large social network have been found to have more social interactions compared to people with a small social network (e.g. Carrasco and Miller, 2006; Silvis et al., 2006; van den Berg et al., 2012b). Boase et al. (2006) on the other hand, found the number of face-to-face and mediated interactions to be lower for people with a large social network. This may suggest that communication frequen-cies are reduced in order to maintain larger social networks (Dijst, 2009). On the level of ego-alter relationships, age and gender homophily might play a role in the sense that people tend to interact more frequently with others that are similar to them. How-ever, in his study on face-to-face, telephone and Internet frequency of interaction in Chile, Carrasco (2011) only found a significant effect of gender homophily in face-to-face interac-tion frequency. Sharmeen et al. (2014) analyzing the frequency of face-to-face social interac-tion, found a positive effect of homophily in gender and educational level. They found how-ever a negative effect of age homophily. Previous studies have found emotional closeness and relational role (e.g. family or friend) to affect communication frequencies. For instance, Boase et al. (2006), Tillema et al. (2010) and van den Berg et al. (2012) found that interaction frequencies with all modes were higher for very close ties compared to less close ties. However, Rivière and Licoppe (2005) found some differences between different cultural contexts regarding text messaging in Japan and France. Their study indicates that in France text messages are mainly sent to close ties, where-as in Japan, text messages are sent to all contacts, independent of emotional closeness. On the other hand, telephone calls, which are relatively expensive in Japan, were mainly used for con-tacting emotionally very close people. Regarding relational role Frei and Axhausen (2008) found lower face-to-face and telephone contact frequencies between work mates and higher telephone contact frequencies between relatives. Van den Berg et al. (2012) also found higher telephone contact frequencies between relatives, and lower face-to-face and email frequencies compared to friends. A final variable that is important in explaining communication frequencies with different modes is the geographical distance between ego and alter. The geographical distance between social network members is increasing (Axhausen 2002; Urry 2003; McPherson et al. 2006). This is probably (partly) due to the fact that opportunities to maintain long-distance contacts are increasing with the development of new ICT tools and decreasing prices for telecommuni-cation. However, according to Mok et al. (2010) geographical distance still is an impediment for interaction in the post-Internet era. Especially regarding face-to-face interaction frequen-cy, distance has been found to have a negative effect (Boase et al., 2006; Larsen et al., 2006; Frei and Axhausen, 2008; Carrasco and Miller, 2009; van den Berg et al., 2012; Sharmeen et al., 2013). These studies also found a negative effect of distance on telephone contact fre-quencies, except for Boase et al. (2006) who found no relationship. On the other hand, email contact frequencies have been found to increase with geographical distance by all studies ex-cept for Frei and Axhausen (2008) who found no significant relationship. In conclusion, the literature review shows that that communication patterns differ among countries and cultures. However, an in-depth comparison of the factors that influence the fre-quency of interaction through various communication tools is not available yet.

4 Contribution of the paper

The paper includes a discussion on the existing literature on social networks and activity-travel behaviour, the relationship between ICT-use and social travel behaviour and the factors influencing communication frequencies with different communication modes. Furthermore, it provides a comparative overview of the six datasets in terms of survey method. This is neces-sary as both, sampling strategy and survey instrument differ between the survey-studies and as information on social networks are sensitive to small changes in these issues. Therefore, a methodological comparison between the six datasets is an obligatory starting point for any comparison of statistical figures. In terms of descriptive statistics, the paper includes detailed information on the egos´ char-acteristics, the alters´ characteristics and the ego-alter relationships, especially in terms of in-teraction modes and frequencies. This descriptive analysis is accompanied by national statis-tics (e.g. wage level and transport costs) that can help to understand differences between the datasets. Besides descriptive statistics, the datasets are analysed and compared by employing hierar-chical regression models. Such multilevel models allow to include two or more levels of ex-planatory variables and quantify their influence on mode specific interaction frequencies. In terms of the personal network data there are two levels of explanatory variables: First, at the ego-network level the respondents’ socio-demographic characteristics and characteristics of the personal network can explain communication frequencies with alters. Second, at the ego-alter level explanatory variables are alters’ socio-demographics and geographical distance be-tween ego and alters’ homes and other characteristics of the ego-alter relationship. Findings on both levels are important from a transportation research perspective, as they allow a recon-struction of the generation of social activities and the relationship between ICT-mediated communication and travel demand for social purposes. The analysis and comparison between the six datasets offers theoretical and practical in-sights into social activity-travel behaviour and cultural differences in maintaining social rela-tionships. Knowledge on the factors that influence the frequency of interaction through vari-ous communication tools in different cultural context increases the understanding of the maintenance of social networks and related social travel demand.

ABSTRACT. Extensive and continuous datasets of public transit transactions collected by Automated Fare Collection (AFC) systems can be processed to understand transit usage and travel behavior. This paper expands on the analysis of AFC data to understand individual travel patterns and detect trip purpose for smartcard transit riders. In this paper we compare data mining algorithms to infer trip purpose, and propose a methodology to consider individual spatial and temporal travel patterns to estimate missing alighting locations and infer trip purpose.

ABSTRACT. - Abstract The travel mode choice model is an important component of the travel demand module in the transportation planning workflow where the decision-making process of a person’s choice of travel mode for trips are simulated. This paper presents two artificial intelligences (AI) created using a deep machine learning algorithm to predict the travel mode choice. The two models are similar in that they both have aggregate household-level attributes (number of cars, adults, and children) and person-level attributes (age, sex, occupation, and driver’s license) in addition to the travel episodes for the person’s day. The difference is in the household-level analysis in which the information of the rest of the household members, including travel episodes are included as ancillary inputs to the AI models. Both AI models are constructed using a similar deep neural network topology with 7 hidden layers and 1024 neurons each layer. The neural network’s topology is constructed through the trial and error method maximizing the accuracy of the models. The traditional logistic regression method is also used to estimate the mode choice model with training similar input data model at both household and person levels and compared with the other AI models. In general, the final results indicate that the neural network approaches predict travel mode choices with considerably higher accuracy than the logistic regression approaches.

- Mode Choice Model Literature Mode choice modelling has been done historically through variations on three techniques: probit, logit, and machine learning. Use of artificial intelligence (AI) techniques for mode choice modelling has been growing, and this trend is expected to continue in the future for border transport as well. In recent studies, it has been shown that neural networks have been achieving better accuracy than more the more conventional logit model and warrant further exploration (Ratrout, et. al., 2014). Within the area of machine learning Artificial Neural Networks (ANN), Support Vector Machine (SVM), and logistic regression model (LRM) have been explored in different studies for the prediction of travel mode choice. A challenge for machine learned algorithms is that care needs to be taken to ensure that the models are not being overfit during the learning phase. In (Omrani, 2015) this was addressed by separating the dataset into a training set and a validation set. Omrani 2015, presented four machine learning methods namely artificial neural net-MLP, artificial neural net-RBF, multinomial logistic regression, and support vector machines, for predicting travel mode of individuals in city of Luxembourg. The results show that the artificial neural networks perform better compared to other alternatives. Sarada et. al. (2013) used Fuzzy-logic approach to predict mode choices of trip makers in Port Blair city, India and then compared the results with traditional logistic regression model. They concluded that the fuzzy logic models were better able to capture and incorporate the human knowledge and reasoning into mode choice behavior. 

- Data Preprocessing The observed data used to train and build different models in this study are gathered from 5% of households living in the GTHA by Transportation Tomorrow Survey (TTS) agency in City of Toronto. Various input features are used to build the mode choice models. The input variables of the models are produced from household’s and person’s features. In this study, two types of mode choice models are developed based on (Artificial Intelligence) AI topology and Logistic Regression. In this regard, the models are trained with two sets of input features distinguished at person and household levels. Both types of models have aggregated household attributes (including number of adults, cars, and children) and person attributes (including age, sex, occupation, and driver’s license) in addition to the travel episodes for the person’s day. The difference is in the household-level analysis in which the information of the rest of the household members (maximum 5 members), including travel episodes are included as ancillary inputs to the AI and logistic regression models. In order to standardize the input data for building the model at the household-level and to overcome intensive runtime happened in the training process, maximum five household members are considered in all models. Each record of input data has at least one person (in a household size of one) and the rest of input variables related to the rest of household members set to zero in this case. In addition, each record has just one active person who makes the trip besides the attributes of the rest of household members. It should be noted that before constructing the models, the categorical input data are reshaped based on one-hot-encoding method and the numerical input data are rescaled and normalized based on Min-Max rescaling method. In total, 1593 and 319 input features have been generated for the household- and the person-level respectively. The preprocessing of the input data is necessary to avoid biases may be happened in training the model raising from different input data structure and scales. The response/target variable of the models contain four travel modes including Auto, Transit, Active (Cycling and Walking), and Passenger. For training the input dataset in the AI approach, each mode of the response variable is labeled numerically and then is reshaped and vectorized based on one-hot-encoding method as below: Auto [1, 0, 0, 0] Transit [0, 1, 0, 0] Active [0, 0, 1, 0] Passenger [0, 0, 0, 1] The one-hot-encoded classes of the response variable enables the AI model to estimate the probability of choosing all possible travel modes by each person at any specific trip similar to the logit model.

- Defining Model Topology and Strategy The python programing language is use in this study. The python data science library namely “Keras” is used on top of the Google “TensorFlow” API for constructing deep neural networks and machine learning processes. TensorFlow is an open source software library for numerical computation using data flow graphs. Models in Keras are defined as a sequence of layers. The neural network’s topology is constructed through the trial and error method maximizing the accuracy of the models. To do this, a sequential model is constructed in Keras and layers with different neurons are added one at a time until a high accuracy is achieved for a network topology. The multi-layer perceptron (MLP) neural network with fully-connected network structure with 1 input layer, 7 hidden layers, and 1 output layer is defined. A deep neural network with 1024 neurons at each hidden layer is developed. The rectifier (ReLU) activation function is used on the hidden layers. The softmax function is utilized on the output layer to ensure our network output is between 0 and 1 and easy to map to either a probability of mode choice classes or snap to a hard classification of either class with an argmax function. The softmax function is a generalization of the logistic function, through which the output of the model represents the categorical or probability distribution over different possible mode choices. In order to prevent the neural network from overfitting, the 20% dropout is defined between layers during the process of training the model. Dropout is a technique where randomly selected neurons are ignored during training process. In addition to the neural network models, the logistic regression models are used at both household- and person-level analyses. The topology of the logistic regression model is similar to the neural network which is simplified in one neuron with softmax activation function for the multiclass output variables. Similarly, the softmax function in the multinomial logistic regression estimates the categorical or probability distribution over different possible mode choices. Now that the model is defined, and the model needs to be compiled and trained. Compiling the model uses the efficient numerical library-TensorFlow under the covers (the so-called backend). The backend automatically chooses the best way to represent the network for training and making predictions to run on the hardware. Training a network means finding the best set of weights to make predictions for this problem. The multiclass logarithmic loss function (which is categorical cross-entropy in Keras) is specified to evaluate a set of weights of the neural network and the ADAM optimizer is used to search through different weights for the network. ADAM optimizer is the efficient gradient descent algorithm tuning the learning rate during the training process.

- Study Results and Discussions Each model is trained on 70% of data and the accuracy of the constructed model is then assessed. The performance of the model is then evaluated on the testing dataset (30% of data). To do that, the confusion matrix of each model is constructed and three assessment metrics including Overall Accuracy, Precision, and Recall are calculated. Below are the definitions of the assessment metrics used for evaluating comparing the performance of different models. Overall Accuracy: is the most intuitive performance measure and it is simply a ratio of correctly predicted observation to the total observations. Precision: is the ratio of correctly predicted positive observations to the total predicted positive observations. Recall: is the ratio of correctly predicted positive observations to the all observations in actual class. The results demonstrate that neural network model at household-level predict different travel mode choices slightly with higher accuracy (around 2%) compared with the person-level model. However, predicting Auto mode are very similar at both levels. Evaluating the overall accuracy of both neural network models indicates that the accuracy of the model tested on the validating dataset (testing data) is around 10% lower than the accuracy of the trained model. In addition, Recall and Precision accuracy assessment illustrates that both models predict Active and Passenger modes on the testing dataset around 20% lower than what the model is trained based upon (training dataset). The results indicate that the logistic regression model has a better performance at the person-level compared to the household-level as the logistic estimator is not very good at hyperparameter scale. It can be concluded that the accuracy of the deep neural network is significantly higher than the logistic regression model. The reason is that the neural network model is more advanced with more complex architecture than the logistic regression model. The logistic regression model consists of one processing layer of decision-making with one neuron through which all the model decision processes are aggregated in that single neuron with one utility function. In contrast, in the neural network model there are different layers of analysis with more neurons which allows simulating the process of decision making significantly better with multiple utility functions developed at different layers. The neural network is able to better mimic the human brain in which each layer of neural network represents layer of human decision-making, in a way that a human creates different factors out of aggregation of observed inputs in a decision-making process and depends on the complexity of the problem continues this process to create different layers of aggregated factors to reach a reasonable decision.

- References Ratrout, N.T., Gazder, U. and Al-Madani, H.M.N. (2014) “A review of mode choice modelling techniques for intra-city and border transport”, World Review of Intermodal Transportation Research, Vol. 5, No. 1, pp. 39–58. Omrani H. (2015). “Predicting Travel Mode of Individuals by Machine Learning”, 18th Euro Working Group on Transportation, EWGT 2015. Sarada P., Ashutosh A., and Madhu E., (2013). “Use of Artificial Intelligence for Mode Choice Analysis and Comparison with Traditional Multinomial Logit Model, In Procedia”, Social and Behavioral Sciences, Vol. 104, pp. 583-592.

ABSTRACT. See the attached pdf.

ABSTRACT. In a context of car use reduction, particularly in urban areas this paper set out to estimate the factors that determine non-use of the private car among young adults for home-work and home-school trips. Data from Household Travel Surveys conducted in the Lyon conurbation in France in 1995 and 2006 show that if socioeconomic factors have a significant impact, characteristics of individual’s areas of residence and activity have also a strong explanatory power for young adults’ modal choice. The influence of the urban structure via population/employment density, nearby facilities and public transport accessibility was even greater in 2006 than in 1995. The objective of this paper is to estimate the factors that determine non-use of the private car among young adults by focusing on home-work and home-school trips and to measure the specific effects of accessibility and urban form of both the area of residence and work or study.

ABSTRACT. Although classic household-travel surveys have been a powerful tool in describing and analyzing mobility behavior, they still leave a number of significant gaps that haven’t been extensively explored in urban research. Access to urban amenities cannot be summarized as daily mechanical activity, because it involves instead a series of subjective components, such as individuals schedules and personal preferences. Thus, while the use of quantitative data is essential when estimating accessibility, qualitative methods offer an alternative approach, that complement  quantitative techniques and allow a better / deeper understanding of the variegated forms and complexity of travel behavior. This paper seeks to grasp travel behavior and access to urban amenities in the suburban areas of the Metropolitan Areas of Lille (France) and Belo Horizonte (Brazil), by reporting the key results of a set of qualitative interviews conducted in the suburban areas of the referred cases of study. The results underlines that qualitative methods can help fill the gaps left by the classic household-travel surveys, by identifying the diversity of decision-making processes behind travel behaviors and mobility strategies.

ABSTRACT. The presentation focuses a new psychometric measurement for perceived accessibility in daily travel (3 studies/versions). The measures capture the traveler perspective of accessibility, a dimension that has previously been overlooked in empirical research. Alas, contrary to objective accessibility, perceived accessibility is not about setting up a priori assumptions of the (most) important indicators of accessibility, as these may vary between individuals, groups, cultures and contexts. Instead, perceived accessibility consist of perceptions of the level of ease to access and use the built environment and transport system, the ability to live the life one wants, and access to activities of choice. The measure has been validated in three versions – for single mode assessment (n= 750), daily travel assessment (n= 2711), and sustainable travel assessment (n= 1926) – consisting of four items each, that can be assessed as an index for perceived accessibility (PAC) or separately. This contribution is important, as the choices individuals face when deciding how to travel (such as transport mode), when they are able to, or prefer to travel (e.g. peak-hours or low service), and the routes they actually do travel are not fully captured with conventional accessibility-measures

Insights into gender differences and differences between (main) daily travel modes, and residential areas are also discussed.

ABSTRACT. Abstract goes here. Writing anything here.

ABSTRACT. In this study, we utilize Wi-Fi communications obtained from smartphones to predict mode of transportation, i.e. walking, biking and driving. In the preliminary stage of the study, WiFi data have been collected from 4 participants in Toronto and three decision tree based algorithm have been developed. Promising results obtained in this stage lead us to further develop the algorithms and data collection process to assess the applicability of our method in real life scenarios.

ABSTRACT. This study aims at finding ways to initiate the use of big data collected by various public agencies and private companies for travel demand modeling especially with the cases of newly developed cities. Centralized development has been the strategy of South Korea for faster development for over fifty years, but about a decade ago, with the purpose of providing better opportunity for less developed regions, the South Korean government made the decision to build 15 new cities and spatially redistribute its government functions.  Over the course of the development, each new city gained population and employment in various industry, which resulted in increase of short and regional trips in each city’s own pattern.  The biggest challenge in analyzing travel demand of these cities is that it is almost impossible to grasp the changes in travel demand and the factors causing the changes only by the traditional household travel survey which is conducted every five years in South Korea.  Therefore, an alternative method to analyze the new cities with more responsive data collected with shorter time interval is needed.  This study includes examples of travel demand analysis using the data collected from the toll collection system, which is installed throughout the over-300 tollgates in the expressway system, and the mobile phone user distribution data collected and processed by one of the major mobile network companies of Korea.  The results show the potential of using big data for travel demand analysis, and we make recommendations for major policy directions to promote big data analytics for travel demand analysis

ABSTRACT. In order to establish more appropriate tourism strategies, the data on tourists' traveler behavior (e.g. tourist site destination choices, trip-chain making and tour choices, and dwelling time at each spot) would be quite informative. Traditional paper-based questionnaire survey or GPS-based survey methods may not be suitable for large spatial scale analysis of tourists' travel choices because of its high costs and inefficiency. On the other hand, there have been recent several applications of Wi-Fi based approach for human monitoring or travel survey. This study seeks applicability of Wi-Fi based continuous monitoring technology for collecting tourists’ travel behavior through the data analysis from two large-scale field experiments in Japan. Monitoring equipment has been located into some popular spots and transport nodes in two popular sightseeing districts (Central Hokkaido and Main Island of Okinawa) during tourist season. After appropriately cleansing the full records at each site to obtain the dataset of probably tourists, we explore (i) day-to-day/within-day variation of tourists’ concentration across different places; (ii) staying duration of tourists at each site; and (iii) spatio-temporal patterns of tourist movements partially checking the validity of the data collected from Wi-Fi with actual observations of tourist count. The future possibility of the model development of tour patterns with machine learning techniques and discrete choice approaches are also discussed.

ABSTRACT. The concept of Mobility as a Service (MaaS) aims to break the determining role of mobility tools in favour of a pay-per-use approach for all modes. Translating fixed (sunk) costs back into marginal costs allows for a more time- and cost-efficient travel. Despite first insights into design and usage patterns of MaaS services, a number of questions still remain. For example, it is yet unclear, how use of MaaS services depends on the spatial structure, on the time of day or on seasonal effects; whether it leads to more multi-modal trip chains and how it influences activity locations and time use. This study tries to address such questions by analysing a rich set of travel diaries from an MaaS experiment in Switzerland, where respondents were provided with a 1^stclass national season ticket for public transportation, a BMW i3 electric car (including recharging), free parking at a train station as well as credit for the use of car-sharing and bike-sharing schemes across the country. In return, participants paid 12’200 CHF (approx. 9’900 USD at PPP) and agreed to keep an electronic travel diary for the whole duration of the field test (one year) and respond to an entry survey. As data collection has just been concluded, the authors will report on the pre-processing of one-year GPS travel diaries as well as preliminary analyses on how such a “mobility flatrate” impacts its users’ aggregate travel behaviour, their mode choice decisions, and their activity spaces.

ABSTRACT. See attached file.

ABSTRACT. Advances in technologies have made the idea of Autonomous Vehicles (AV) very close to become a reality in near future. While over the past few decades most researchers in related fields focused on developing tools and techniques that made the AVs available, because of the novelty of related technologies, it is not clear how they would affect our cities and daily life. Particularly since these vehicles are capable of picking up passengers without a driver, it is widely speculated that they may end up disrupting our life, if not regulated. In the absence of knowledge about AVs' effects, and people's behavior in their presence, simulation is one way of developing a better understanding of such environment. Although simulations are based on many assumptions and not necessarily accurate, analyzing corner case scenarios could help to estimate the magnitudes of impacts. One of the speculated scenarios for future transport system is the large scale adoption of Level 4 AVs, which are assumed to be driven without human presence and/or interaction. In this paper, we develop a mixed integer optimization model to estimate the optimized number of level 4 AVs to serve a household travel needs. Then we will apply the model to the entire population of the Chicago region to estimate the total number AVs that is needed to serve the region’s population. It should be mentioned that as a corner case study, this paper assumes that the region is served only by shared private AVs. That is, each household owns its AVs, and shares it with its members. However, members could choose to ride a taxi if it helps them to lower the total cost of household travels.

ABSTRACT. Automated Vehicles (AVs) offer their users a possibility to perform new non-driving activities while being on the way. The effects of this opportunity on travel choices and travel demand have mostly been conceptualised and modelled via a reduced penalty associated with (in-vehicle) travel time. This approach invariably leads to a prediction of more car-travel. However, we argue that reductions in the size of the travel time penalty are only a crude proxy for the variety of changes in time-use and travel patterns that are likely to occur at the advent of AVs. For example, performing activities in an AV can save time and in this way enable the execution of other activities within a day. Activities in an AV may also eliminate or generate a need for some other activities and travel. This may lead to an increase, or decrease in travel time, depending on the traveller’s preferences, schedule, and local accessibility. We present an optimisation model which rigorously captures the time-use effects of travellers’ ability to perform on-board activities. The model builds on the core ideas of the classical time-allocation frameworks (Becker 1965). The main interactions captured in our time-use model are as follows. An individual maximises utility by selecting activities to be performed, activity locations (stationary and/or on board a travel mode), and travel mode(s) to reach the activity. In these choices, the individual is constrained by two main time constraints. First, the time spent in stationary activities and travelling should not exceed the total time (this is the same constraint as is used in classical time-allocation frameworks). Second, the time spent in on-board activities during a trip should not exceed the trip duration (this is a new constraint necessary to model on-board activities). We formulate the problem as a mixed-integer linear model.

Reference: Becker, Gary S. (1965): A theory of the allocation of time. In The Economic Journal 75 (299), pp. 493–517.

ABSTRACT. Introduction Japan is facing a super-aging society. It is required to provide an environment where all the people including mobility handicapped persons such as the elderly, disabled people and baby stroller users, travel safely, securely and comfortably. Generally, travel is considered a demand derived from the desire to engage in activities at destinations and travel time is also considered a wasteful time. Then people sometimes prefer virtual mobility using ICTs (the Internet and smartphones) to physical mobility, in order to save travel time and engage in activities in cyber space (e.g., telecommuting, e-shopping and e-conferences). However, travel has an element not only of a disutility but also of a “positive utility” (Mokhtarian and Salomon, 2001). For example, Redmond and Mokhtarian (2001) found that an average ideal commute time was not zero but 16min. Mokhtarian and Salomon (2001) discuss three elements of the positive utility of travel: (i) the activities conducted at the destination, (ii) the activities that can be conducted while traveling, and (iii) the activity of traveling itself. The rapid spread of ICTs has provided people with much activity opportunity while traveling. Existing studies revealed what activities travelers engage in while traveling and the influence of activities conducted while traveling on travelers’ subjective evaluation of travel, e.g., travel liking (Ory and Mokhtarian, 2005), scale of travel satisfaction (Ettema et al., 2011), travel happiness (St-Louis et al., 2014) and irritation level (Ohmori et al., 2004). Also Mokhtarian et al. (2015) discuss extrinsic and intrinsic motivations of travel by reviewing a number of theories and typologies of motivations in psychology. On the other hand, in the near future the new technology of autonomous driving might change travel time for driving a car into time for engaging in activities. This study examines a positive utility of travel for various types of people including mobility handicapped persons. Methods are focus groups and questionnaire surveys to ask preferences for saving travel time by the “teleportation door” and for changing travel time into activity time by the “autonomous vehicles.”

Survey Firstly, we conducted focus group interviews in Utsunomiya city (a local city in Japan) during July 2015 and January 2016. Respondents were wheelchair users (2 groups, 12 individuals), visually impaired persons (1 group, 5 individuals), hearing impaired persons (2 groups, 8 individuals), women rearing children (1 group, 4 individuals), foreign hostesses (1 group, 3 individuals), and university students (2 groups, 11 individuals). We asked their socio-economic attributes (age, sex, living place, etc.), travel behavior in their daily lives (trip purpose, travel mode, frequency, etc.), and especially asked the following questions: “What travel difficulties do you face in your daily lives?”, “What types of travel are enjoyable for you in your daily lives?” and “For what types of travel would you want to use the teleportation door in your daily lives, if it were hypothetically available?” Following the focus groups, in January 2016, questionnaire surveys were conducted for a total of 204 students of Utsunomiya University and 27 wheelchair users to collect information on the current and ideal travel (frequency, travel mode and travel time), and preferences and willingness to pay for saving travel by the teleportation door. In November 2017, we are conducting a similar web-based questionnaire survey to collect data from a total of 1,000 respondents living in Tokyo, where the target segment will be persons rearing children and car drivers. The survey will focus on preferences for the teleportation door and three types of autonomous vehicles (fully-automated driving, partly-automated driving and driverless taxis).

Results From the data obtained by focus groups, we found the differences in travel characteristics among people with different mobility constraints for their enjoyable travels and the types of travels they wanted to save by the teleportation door if possible. Also we found that the common elements which all the groups wanted to save travel by the teleportation door were time constraints, travel cost, physical strength and weather. In the questionnaire survey for university students and wheelchair users, we asked the current and ideal travel modes (including the teleportation door) and travel times for different trip purposes. The highest share of the current travel mode for the university students was bicycle for all trip purposes. Whereas, for the ideal travel mode, the teleportation door is the highest share for commuting (39%) and shopping trips (37%), being less share for bicycle. Regarding wheelchair users, car driving is the highest for the current travel mode for all trip purposes. However, the teleportation door is the highest for the ideal travel mode for commuting (33%) and shopping trips (37%). Analyses of the ideal travel time revealed that the average ideal travel times were less than the current ones but not zero (6.3min. for students’ commuting, 5.3min. for students’ grocery shopping, 24.9min. for wheelchair users’ commuting, and 8.1min. for wheelchair users’ grocery shopping). Respondents were asked the reasons why they were choosing the current travel mode. As for commuting and shopping, the top reason was “to save travel time” for both university students and wheelchair users, whereas it was “for refreshment” for walking/driving. The result suggests that even commuting and shopping travel which is considered as a derived demand have some elements of a positive utility such as “exercise” and “refreshment.” We also asked their preferences for using the teleportation door when commuting and their willingness to pay for using it in different travel situations. It was found that only 31% for students and 56% for wheelchair users preferred to use the teleportation door on fine days, but when facing more difficult travel situations such as on rainy days, when in hurry, with heavy luggage, and in bad health, more respondents preferred to save the commute travel and their willingness to pay were larger than that on fine days, especially for university students. The amount of willingness to pay for wheelchair users is three to ten times larger than that for university students, which reflect their travel difficulties.

Conclusions This study examined, by focus groups and questionnaire surveys, the differences in enjoyable travel and travel which is desired to save by the teleportation door among various types of people including mobility handicapped persons. It was found that only 30% to 40% of university students and wheelchair users preferred to use the teleportation door as an ideal travel mode for commuting and grocery shopping trips, and their average ideal travel times were less than the current ones but not zero. Also we found that university students and wheelchair users preferred to save commute trips using the teleportation door when facing more constraints and their willingness to pay varied across different travel situations. The coming data will provide interesting results of understanding what types of travel people want to save by the teleportation door and in what situations people enjoy the positive utility of travel by the autonomous driving.

References 1) Mokhtarian, P. and Salomon, I., 2001. How derived is the demand for travel? Some conceptual and measurement considerations. Transportation Research A. 35(8): 695-719. 2) Redmond, L. and Mokhtarian, P., 2001. The positive utility of the commute: modeling ideal commute time and relative desired commute amount. Transportation. 28(2):179-205. 3) Ory, D. and Mokhtarian, P., 2005. When is getting there half the fun? Modelling the liking for travel. Transportation Research A. 39(2-3):97-123. 4) Ettema, D., Gärling, T., Eriksson, L., Friman, M., Olsson, L. and Fujii, S., 2011. Satisfaction with travel and subjective well-being: development and tests of a measurement tool. Transportation Research F. 14(3):167-175. 5) St-Louis, E., Manaugh, K., van Lierop, D. and El-Geneidy, A., 2014. The happy commuter: a comparison of commuter satisfaction across modes. Transportation Research F. 26:160-170. 6) Ohmori, N., Hirano, T., Harata, N. and Ohta, K., 2004. Passengers’ waiting behavior at bus stops. Proceedings of ICTTS 2004. 157-164. 7) Mokhtarian, P., Salomon, I. and Singer, M., 2015. What moves us? an interdisciplinary exploration of reasons for traveling. Transport Reviews. 35(3):250-274.

ABSTRACT. Understanding travel behaviour and being able to forecast flows on the public transport network is essential for urban planning. Traditional public transport route choice models assume rational travellers maximizing their utility level (or alternatively minimizing their cost) associated to the different travel alternatives. These models propose a fully-compensatory behaviour, where travellers evaluate the attributes of all alternatives before making their choice. In the context of route choice, where the set of potential alternatives can be extremely large (compared to other choice contexts, such as travel mode), such behaviour might not be reasonable. As travellers might not be able to process and evaluate all possible alternatives, there is a need to explore alternative choice frameworks from a behavioural point of view. The purpose of this study is to enhance the behavioural framework for modelling travel decisions within public transport systems, by appliyng heterogeneous discrete choice mechanisms to model public transport behaviour.

ABSTRACT. INTRODUCTION

Ridership on the San Francisco Bay Area Rapid Transit (BART) system has increased rapidly in recent years due to population and employment growth. Between 2004 and 2016, ridership grew by about 40 percent overall and 75 percent in the Transbay corridor connecting San Francisco’s financial district and East Bay cities via the underwater Transbay Tube. Generally, increased demand can be accommodated through expanded physical capacity or through more efficient use of existing capacity. More efficient use of existing capacity can be an effective short-term solution for alleviating capacity constraints while infrastructure improvements, which often take longer periods of time to bring online. BART Perks was a six-month test program that provided incentives for travelling during the shoulder hours of the peak period, in order to reduce peak period, peak direction crowding and improve person throughput in the Transbay tube that connects San Francisco with the cities of the East Bay. The first program of its kind in North America, BART Perks was modeled after a similar program deployed in Singapore. Perks was a joint project of BART and the San Francisco County Transportation Authority (SFCTA), the congestion management agency for San Francisco, and was funded by the Federal Highway Administration, BART, and the SFCTA.

ENROLLMENT

Participants enrolled in the program through a mobile-friendly website. At signup they accepted a user agreement and were prompted to enter their transit smart card ID number. Linking each user with their smart card was necessary to award them points based on the frequency, timing, and length of their trips, and to evaluate program effectiveness. Participants were recruited four ways: • Direct outreach • Earned media coverage / word-of-mouth • Employer partnership program • Friend referral

PROGRAM DESIGN

In the initial four months of the program, participants earned one point per mile for all travel on BART and between three and six points per mile for travel initiated during morning peak shoulder periods, depending on their status in the program. Program status was determined by the number of shoulder-peak trips made in a given time period. For the last two months of the program, the design was changed so that participants earned 17 points per trip (equivalent to the average points earned per trip during the first part of the program), rather than one point per mile traveled. The intent was to compare the effectiveness of the two approaches, especially for certain travel markets, such as longer distance origin-destination pairs where participants would receive relatively more incentives under the initial scheme, and relatively fewer under the later scheme.

In addition to earning points per mile or per trip, randomly selected participants also received occasional special offers (“bonus boxes”) of additional points during the program. The purpose of these bonus boxes was to test the sensitivity of participants to different types and levels of targeted incentives, and to encourage additional shift among the target market.

PARTICIPATION

Active enrollment peaked at about 17,800 participants. Most enrollment was obtained through the direct outreach and earned media coverage in the first few days of the program. A minority of participants (6 percent) signed up for the program through a friend referral or through their employer. The project team targeted marketing efforts at the Embarcadero and Montgomery stations during the morning peak period in order to capture peak period Transbay travelers. Sweepstakes law, which applied to BART Perks because of the Spin-to-Win component, precluded more direct methods of recruitment such as qualifying participants based on their travel history. However, peak period Transbay travelers were ultimately a small share of program participants. Although 50% of program participants were regular AM commuters, regular AM peak hour Transbay commuters comprised only 13% of all program participants.

POINTS REDEMTPION

Participants had three options for exchanging points for cash rewards: • Autoplay: For those enrolled in autoplay, all points earned in the previous week were automatically entered into a game (effectively, a random rewards generator) once per week. Depending on the outcome, the participant could receive nothing, or a reward ranging between $1 - $100. • Spin-to-win game: Participants could turn off autoplay and use points to play an online chutes-and-ladders game to win more points or cash rewards. Like autoplay, this game was essentially a random rewards generator, but included an interactive element. • Cash buyout: Participants could exchange points at a rate of $1 per 1,000 points. Note that 1,000 points was equivalent to 1,000 miles travelled on BART - a typical commuter might travel this distance in about a month and a half.

Points accrued automatically and most users retained the default auto-play setting, meaning that their points were entered into a random rewards generator every week. About 86% of program cash rewards were paid out this way, compared to 13% paid out through manual online play of the Spin-to-win game and less than 1 percent through the cash buyout. The average participant made about one trip per weekday, and earned a total of $13, or a little over $2/month, for the duration of the program. However, the distribution of rewards varied considerably by participant

BEHAVIORAL EFFECTS

Perks’ effect on participant travel behavior was evaluated by comparing behavior before and during the program, estimating the number of trips shifted, and examining whether shifts persisted after the program ended. In addition, a longitudinal, disaggregate model of the shift in participants’ departure time choice was also estimated, and will be documented in the full paper. In order to estimate the number of trips that were shifted by the program, the shares of daily participant trips that entered the system in the peak hour before Perks and during Perks were calculated using data participants’ smart card data. These shares were then applied to the average daily trips made by participants. It is estimated that systemwide during the six-month trial period about 10% of all peak riders, or an average of 250 participants (roughly two full train cars of passengers), shifted their ride either before or after the peak morning rush hour each weekday. Inbound Transbay peak hour trips were reduced by 11% or 180 trips.

Upon termination of the program, it was possible to observe whether the behavioral shifts from the peak hour persisted. Regular commuters reduced their share of peak hour trip-making from 29.9% before the program to 26.2% during the program (slightly more change than the reduction from 26.6% to 23.7% of all Transbay inbound trips). However, after the program ended, this share then rose back to 28.6%, suggesting that, for the target market of AM peak hour inbound Transbay commuters, the program effects partially, but not completely, persisted after the program ended.

PROGRAM LIMITATIONS

The program design had several limitations, most notably with respect to the use of a static, universal definition for the shoulder peak period (6:30-7:30 AM & 8:30-9:30 AM) and peak (7:30-8:30 AM) travel. In fact, the peak travel period on BART varies significantly by line, day of week, season, and in response to erratic delays, therefore a single overall peak period may or may not apply on any given day. Additionally, a key goal of the program was to reduce crowding in the Transbay Tube in the peak hour in the peak direction. However, what constitutes a ‘peak departure’ varies based on the individual’s origin station and distance from the Transbay tube.

CONCLUSIONS

The Perks program demonstrated that incentives can be successfully used to shift the departure times of peak period travelers. Among all program participants there was a reduction in peak hour inbound Transbay travel of -10.9%, and a reduction in peak hour overall system travel of -9.6%. While promising, this shift was on a small base of peak period trips and therefore had little impact on crowding. The 17,800 active participants only generated about 1700 Transbay A.M. peak hour, peak direction trips on a typical weekday, and about 10.9% of these trips shifted during the program, representing a shift of approximately 170 trips.

Recommendations for future programs include:

• Obtain sufficient peak travelers. To significantly increase peak shift, future programs will need to recruit larger number of peak riders in the targeted travel market. This can be achieved through multiple means such as scaling up enrollment to ensure that sufficient number of peak riders are participating, providing higher levels of rewards to entice more peak riders to sign up, or qualifying participants based on their travel history.

• Focus rewards on behavior change. Future programs should ideally be structured to reward behavior change rather than pre-existing behavior, such as by establishing a behavior baseline (e.g. frequent peak travel) and rewarding change from that baseline.

• Consider more efficient methods of participant recruitment and retention. BART Perks rewarded all BART riders to some degree for traveling on BART, and this likely was important factor in explaining the appeal of the program and rapid enrollment of nearly 18,000 participants within a few weeks of launch. However, the broad appeal of the program resulted in high level of participation among individuals who were not frequent BART commuters.

ABSTRACT. In this paper, an activity-based model is proposed for scheduling individuals’ daily activity-travel patterns (DATPs) in multi-modal transit networks with the customized bus (CB) service. As an innovative mode of public transit, the CB attracts increasing attention and has been in operation in many cities in China. This study investigates the change of individuals’ activity and travel choice behavior after CB service is introduced in multi-modal transit networks. A super-network platform is adopted to simultaneously consider individuals’ activity and travel choices before and after CBs are operated. To describe the CB subscription process considering capacity constraint, a day-to-day learning and adjustment process is incorporated in the proposed model. A solution algorithm without prior DATP enumeration is proposed to solve the DATP scheduling problem on the super-network. Numerical examples are conducted to verify the proposed model and the solution algorithm. 

ABSTRACT. Passenger transport information is available via different sources, such as maps and timetables, bus stop displays, mobile applications, travel websites etc. Among the sources of traveller information, internet-based information has increased in popularity over the last few years also due to the diffusion of mobile devices in everyday life.

Since its introduction, transport planners and researchers emphasise the potential benefits of Ubiquitous Real-Time Passenger Information (URTPI) in terms of route choice optimisation, triggered by a more accurate knowledge of network and service conditions. Some studies have considered information as a potential tool to influence traveller behaviour but the impact of URTPI on passenger choices is still inadequately explored.

Our study addresses this gap by means of an intercept survey among bus passengers in Edinburgh, UK. Edinburgh has a well-developed and largely used Public Transport and about 21% of the population use buses as their main mode of travel. The questionnaire included 17 questions regarding the trip the respondents were making at the time of the survey. A total of 1645 passengers took part in the study.

We find that passengers’ choices after consulting URTPI influence PT demand distribution in 39% of revealed cases. The demand distribution across bus runs and bus lines is affected in 17% and 15% cases respectively. These changes may have important consequences on on-board crowdedness, above all in peak times.

The spatial dimensions of route choice are strongly influenced by the contents of consulted URTPI. In particular, the decision of changing bus line is influenced by passenger demographics and contents of information. Therefore passenger demographic profile and disseminated contents of information could be useful to anticipate the change in demand for bus services.

ABSTRACT. This study provides a new specification for a context-dependent reference point: we adopt the average value of an attribute within the choice set as its reference point. We also adopt regret function in the random regret minimization model to generate asymmetric preference in the loss and gain domain which characterizes loss aversion. We show the model’s capability to replicate various context effects including compromise and decoy effects. We also test the performance of this specification with an empirical dataset of shopping choice: our model not only performs better than the random utility model (RUM) and the classical random regret model (RRM), but also outperforms the muRRM which is a generalized version of RRM with more parameters. The results demonstrate the potential of our model.

ABSTRACT. The random utility theory has been widely criticised for its inability to explain and predict travel behaviour accurately. Previous latent and hybrid choice models tried addressing the limitations of the rational choice models by incorporating attitudinal variables to explain the random error of the aforementioned models. However, recent studies in behavioural science and consumer behaviour suggest that a person’s level of ego has a direct influence on consumer behaviour. However, this study focuses on the impact that narcissism (egotism) and narcissistic traits have on travel mode choices. A sample size of 2,000 randomly selected households across the city of Edinburgh, Scotland was analysed. Preliminary findings based on the initial 24 responses are presented in this paper demonstrate that narcissism (egotism) and high scorers on NPI component of Authority are more likely to travel by car. Understanding the relationship between narcissism and travel mode choice behaviour could provide a useful guide to policy-makers in the development of public transport policy to nudge Narcissist and influence their travel mode choice behaviour.

ABSTRACT. Congestion, a term usually associated with road networks and private vehicle use, is becoming increasingly prominent within the Public Transport space. Increased patronage on public transport by large metropolitan workforces accessing Global Economic Activity Centres like the Sydney CBD has resulted in a rail network operating at near ceiling capacity and excessive dwell times due to passenger congestion at key stations.

The emergence of passenger management strategies, such as Responsive Passenger Information Systems leverage new technologies to address congestion, increase operational capacity and maintain a minimum level of customer service.

This extended abstract will explore passenger congestion in the context of a global activity centres, introduce the capabilities of the RPI System technologies and the strategies proposed to manage passenger behaviours within the Micro Transport Environment. It will also discuss the implication for current practices and articulate the need for theory development to support this new area of research.

ABSTRACT. 1. INTRODUCTION The increasing urbanization emphasizes the importance of sustainable transport modes. Therefore, a shift toward active transport modes, like cycling, in the urban environment has been advocated by several cities worldwide. To realize this shift, safety and flow efficiency of cyclists need to be guaranteed. In order to assess these, a model that captures bicycle traffic behavior is needed for their evaluation. The development of such a model requires understanding the behavior of cyclists, i.e. how they interact with other traffic participants and with the infrastructure. This microscopic operational behavioral level encompasses decisions of individuals while cycling, such as changing speed, overtaking and crossing or stopping. These decisions are affected by several attributes, which may reflect the characteristics of the rider, the bicycle or the built environment, as well as environmental and traffic conditions. The most relevant attributes that affect cyclists’ decisions at the operational level need to be known so as to be included in the model. Results from literature are, however, limited. The study by Yuan et al. (2017) examined the effect of gender on the steering behavior of bi-directional cyclists on collision course using data collected in a controlled experiment. To the best of our knowledge, with the exception of this study, the effect of different attributes on cyclists’ decisions at the operational level has not been researched. However, a number of studies have looked into desired speed and acceleration profiles under different circumstances (Shepherd 1994, Vansteenkiste et al. 2013, Ma and Luo 2016, Schleinitz 2016, Twaddle and Grigoropoulos 2016). This leaves numerous opportunities for research into the determinants of cyclists’ operational decision making. This research aims to formulate a conceptual model linking different attributes to cyclists’ decisions based on the results of a survey we conducted in the Netherlands. The conceptual model, which is the main contribution of this study, is useful for the design of a controlled experiment, whose data will guide the development of a mathematical model. In this paper, we present and analyze the survey data and formulate the conceptual model based on the results. Moreover, we draw conclusions and future research directions.

2. SURVEY DESIGN The aim of the survey is to investigate which attributes influence decisions that cyclists make when confronted with a situation on the road where they interact with each other, or with the infrastructure. In the survey, the specificities of the situation are outlined. The situation always involves cycling during daytime on road infrastructure designated for cyclists and separated from other traffic, referred to as bike path. The respondents are asked to name the attributes that influence their decision making in each situation. More specifically, three situations are considered: (i) overtaking or staying behind a single or a small group of cyclists; (ii) stopping or continuing at a red traffic signal, and; (iii) going ahead or stopping at a crossing to allow cyclists with priority to merge or cross. The justification for choosing to focus on interaction decisions rather than individual preferences of speed and positioning on the bike path, is the reliability of data that can be obtained from a survey. According to Rasmussen (1983), riding a bicycle can be classified as skill-based, indicating the lack of conscious control. This, however refers to automated actions, such as steering and pedaling, or, potentially, stating the decision they would make. As the respondents were not asked about these but rather about the attributes they take into account when faced with a particular situation, the obtained data can be considered reliable. With respect to the attributes, a list is provided to the respondents for each situation based on behavioral hypotheses regarding the most influential ones. Each list contains ten attributes that are randomly ordered, along with three empty fields for respondents who wish to enter attributes not included in the list. A selection of at least three and at most ten attributes is requested per situation, ranked according to their subjective importance. Additionally, general information about the respondents (age, gender, nationality) is collected, along with their years of cycling experience and the frequency of using the bicycle per trip purpose. The survey was distributed digitally and remained online between June and August 2017. The target group was cyclists within the Netherlands, and, therefore, the survey was posted on bicycle Dutch fora (wereldfietser.nl and fietsersbond.nl), as well as to university students and parents (ouders.nl). In total 444 complete responses have been obtained, which are analyzed following the methodology of the next section.

3. METHODOLOGY The objective of the analysis of the survey data is to formulate a conceptual model, linking the main attributes to the decisions that cyclists make. In order to find these main attributes, a principal component analysis (PCA) is performed (Abdi and Williams 2010). The variables used for the PCA are all the attributes given for the three situations, which have a binary character: they were either mentioned or not. The alternative would be to consider the attributes along with their rank. However, since several respondents stated that they had difficulties ranking the attributes based on their importance, the rank is not considered a trustworthy indicator. The number of components is decided upon based on a trade-off between four criteria: (a) the resulting scree plot. It shows the relative importance of each factor depending on its eigenvalue and it is a reliable criterion since the sample is bigger than 200 participants (Stevens 2012). According to Kaiser (1960), factors with eigenvalues greater than 1 should be retained, while Cattell (1966) argued that the cut-off point for selecting factors should be at the inflexion points of the curve. These alternatives are compared on the remaining criteria. (b) the Kaiser-Meyer-Olkin (KMO) test. It is a measure indicating how suitable the data are for factor analysis, with values above 0.8 being considered as adequate (Hutcheson and Sofroniou 1999). (c) the percentage of explained variance. It is desired to be as high as possible while keeping as few factors as possible. Therefore, the starting point is the amount of factors with eigenvalues greater than 1, whose performance is compared to the performance of the inflexion points. It is obvious that the more factors that are used, the higher the percentage of explained variance would be, and thus the interpretability of the components becomes crucial. (d) the interpretability of the extracted components. Each variable is associated with the extracted components through a loading coefficient. The higher this coefficient, the more related the variable is to the corresponding component. In order to divide the variables among distinct components, varimax rotation is applied (Kaiser 1958). It ensures that the components remain orthogonal (i.e. independent) while the variable clusters are intersected by the factor to which they relate most. Moreover, small coefficients are suppressed, since only those with factor loadings greater than 0.4 are recommended for interpretation (Stevens 2012). After suppression, the factors with smaller loadings are removed and the extraction is repeated until all variables have a factor loading greater than 0.4. Additionally, it is interesting to study whether the answers differ among subgroups of the respondents. The distribution between genders is 58% male – 42% female, while the nationalities are strictly dominated by Europeans (72% Dutch and 15% rest of Europe). In order to have a sensible sample for the nationality subgroups, the Dutch are compared with the non-Dutch. The indicator on which they are compared is the frequency of an attribute being mentioned per situation. Since the measurement level of the variables is not an interval, the Mann-Whitney U test is performed to examine significant differences in the distribution of each variable across the subgroups. The results of these analyses and the resulting conceptual model are presented in the next section.

4. RESULTS The PCA extracted 9 components with eigenvalues greater than 1, while the inflexion points of the scree plot are at 3, 5 and 7 components. These are four alternatives for the number of components. The KMO test for all of them returned values higher than 0.8, hence rendering the sample adequate for the factor analysis. By comparing their performance, the 5 components alternative is selected. These 5 components explain 43.12% of the total variance. Table 1 shows the rotated component matrix, i.e. the factor loadings per component and variable, sorted by size. The first column contains the names of the variables which represent the attribute and the situation they refer to. More specifically, “Q1”, “Q2” and “Q3” are used for overtaking, stopping at red light and giving priority, respectively. The other columns display the loadings greater than 0.4. Based on these loadings, a name is assigned to each component for ease of reference. The first component is named “Effort” because it contains variables that represent the amount of effort, either physical (slope or weather) or mental (time pressure), that needs to be exerted for each maneuver and which can be compensated for based on the bicycle type being ridden. The second component, “Visibility”, captures the effect of visibility, especially in the decision to give priority, because the variables within the factor are all observable characteristics of the situation. “Rules and infrastructure usage” is the third component, since the variables describe the attitude towards rules as well as the way in which the bike path is being used (i.e. amount of directions and intention to turn or continue straight). The fourth component is referred to as “(Social) interaction”, as its variables relate to the characteristics of surrounding cyclists. Finally, the component “Familiarity and traffic” is mostly relevant for the decision to stop at a red light, in which case the familiarity is important but also the traffic situation on conflicting streams and on the bike path. According to the Mann-Whitney U test, there is a statistically significant difference between males and females in the importance they put on time pressure when deciding to overtake and in the slope and direction of approaching cyclists when deciding to give priority. Females mention both attributes more frequently. The two nationality subgroups use different attributes to decide whether to stop at a red traffic signal. The non-Dutch mention more frequently the obedience to traffic rules, while the Dutch focus on their familiarity with the intersection and the amount of conflicting traffic. Also, when giving priority, the Dutch pay more attention to the direction where they approaching cyclists are headed. Based on these findings we formulate the conceptual model displayed in Figure 1. The effect of gender and nationality on each factor is indicated and the factors are linked to the decisions weighed by their contribution to each decision. The weight is higher as the thickness of the lines increases. Therefore, “Effort” is the key determinant when deciding to overtake, while “Visibility” is the most important factor in the decision to give priority. With respect to the decision to stop at a red light, the “Familiarity and traffic” is the main determinant, followed by “Effort”. The actual influence of these factors on each decision can be described by a probability curve of a “yes” or “no” decision. By varying the values of the attributes within the factors, the shifting point of each curve can be investigated in future research.   Table 1: Rotated Component Matrix. Component 1 2 3 4 5 Q3: Weather conditions 0.716 Q1: Weather conditions 0.666 Q1: Bicycle type 0.599 Q2: Weather conditions 0.597 Q1: Time pressure 0.578 Q2: Bicycle type 0.535 Q1: Slope of bike path 0.531 Q3: Slope of bike path 0.488 Q3: Number of approaching cyclists 0.645 Q3: Visibility (obstacles) 0.590 Q3: Width of bike path 0.453 Q3: Cycling alone or in group 0.443 Q2: Visibility (obstacles) 0.437 Q3: Obeying traffic rules 0.675 Q2: Obeying traffic rules 0.655 Q1: Going straight or turning 0.530 Q1: One-way or two-way bike path 0.455 Q2: Cycling alone or in group 0.556 Q3: Speed of approaching cyclists 0.503 Q3: Direction of approaching cyclists 0.486 Q1: Number of other cyclists 0.481 Q1: Cycling alone or in a group 0.465 Q2: Familiarity with intersection 0.721 Q3: Familiarity with crossing 0.615 Q2: Amount of conflicting traffic 0.475 Q2: Number of queuing cyclists 0.412 Rotation converged in 7 iterations.

Figure 1: Conceptual model of factors influencing the operational decision making of cyclists and the effect of gender and nationality on these factors. Each decision is assigned a color, while the width of the relations lines corresponds to the relative importance of each factor per decision.

5. CONCLUSIONS This study presented a novel dataset from a survey conducted within the Netherlands, investigating the key determinants for decisions that cyclists make when interacting with each other and with the infrastructure. Principal component analysis was performed on the responses and revealed five factors that influence the decisions. Moreover, the effect of gender and nationality was investigated. Based on the findings, a conceptual model linking the attributes to the decisions has been formulated, which shows that the decision to overtake is driven by “Effort”, while “Visibility” greatly affects the decision to give priority and “Familiarity and traffic” is the main determinant of the decision to stop at a red light. By varying the values of the attributes within the factors, the shifting point of each probability curve can be investigated in future research. More details on the results and the conceptual model will be presented in the full paper, while future research will focus on the design of a controlled experiment to investigate the shape of the probability curves for different factor settings.   ACKNOWLEDGEMENTS This research was supported by the ALLEGRO project (no. 669792), which is financed by the European Research Council and Amsterdam Institute for Advanced Metropolitan Solutions. Acknowledgements are also due to Jenneke Bijpost and Larissa Eggers, who helped with the digitalization of the survey, and Maria Salomons who assisted with its distribution.

REFERENCES Abdi, H. and L. J. Williams (2010). "Principal component analysis." Wiley interdisciplinary reviews: computational statistics 2(4): 433-459. Cattell, R. B. (1966). "The scree test for the number of factors." Multivariate behavioral research 1(2): 245-276. Hutcheson, G. D. and N. Sofroniou (1999). The multivariate social scientist: Introductory statistics using generalized linear models, Sage. Kaiser, H. F. (1958). "The varimax criterion for analytic rotation in factor analysis." Psychometrika 23(3): 187-200. Kaiser, H. F. (1960). "The application of electronic computers to factor analysis." Educational and psychological measurement 20(1): 141-151. Ma, X. and D. Luo (2016). "Modeling cyclist acceleration process for bicycle traffic simulation using naturalistic data." Transportation research part F: traffic psychology and behaviour 40: 130-144. Rasmussen, J. (1983). "Skills, rules, and knowledge; signals, signs, and symbols, and other distinctions in human performance models." IEEE transactions on systems, man, and cybernetics(3): 257-266. Schleinitz, K. (2016). "Cyclists’ road safety-Do bicycle type, age and infrastructure characteristics matter?" Dissertation, Technische Universität Chemnitz. Shepherd, R. (1994). "Road and path quality for cyclists." 17th ARRB Conference, Gold Coast, Queensland. Stevens, J. P. (2012). Applied multivariate statistics for the social sciences, Routledge. Twaddle, H. and G. Grigoropoulos (2016). "Modeling Speed, Acceleration, and Deceleration of Bicyclists for Microscopic Traffic Simulation." Transportation Research Board 95th Annual Meeting. Vansteenkiste, P., G. Cardon, E. D’Hondt, R. Philippaerts and M. Lenoir (2013). "The visual control of bicycle steering: The effects of speed and path width." Accident Analysis & Prevention 51: 222-227. Yuan, Y., W. Daamen, B. Goni-Ros and S. P. Hoogendoorn (2017). Investigating cyclist interaction behavior through a controlled laboratory experiment. Velo-city. Nijmegen, The Netherlands.

ABSTRACT. One element that is often overlooked in most researches is that a not negligible proportion of individuals in many countries lacking a cycling culture indeed use their bikes, but only as a form of recreation or as a sporting activity, without considering them as a true travel mode options for their trips. The goal of the present paper is to better understand the differences between "utilitarian bikers" (those who cycle at least sometimes for their trips), "hedonic bikers" (those who use their bikes only for leisure, sport, etc.) and "not bikers".

ABSTRACT. There are several methods in existing literature for evaluating the Level of Service provided by bicycle infrastructure. Models to explain or predict quality of service usually rely on physical variables, but the experience, the context, the environment and other sensorial influences appear to be relevant to satisfaction for active modes of transportation. This research explores the potential of perceptions to explain the perceived satisfaction of biking in Bogotá, Colombia. We conducted quantitative research methods, which involve a survey to a representative sample of cyclists in Bogota (n=434) administered in May 2017. Through linear regression models, we compared the effect of perceptional variables in the users’ Level of Service. Results are promising and have direct implications for street design theory. We found that the most influential variables were: quality of the pavement-surface, quality of signaling and traffic lights, presence of physical separation from pedestrians, traffic safety (accidents), and presence of trees and aesthetics. Identifying these variables allows to include new parameters in urban and transportation planning in the city to provide a better quality of service to users of the infrastructure and to attract more citizens to use the bicycle as a growing mode of transportation. The use of bicycles in Bogota is in constant increase and the need of new bike infrastructure with an adequate design keeps growing.

ABSTRACT. 1. Introduction Cycling is widely regarded as environmentally sustainable and healthy (e.g. Morris and Hardman, 1997). In addition, it has been associated with better mood during and after travel as compared to motorized travel modes (Olsson et al., 2013; Mao et al., 2017; St-Louis et al, 2014; De Vos et al., 2015, Friman et al., 2017a). Despite its positive effects, cycling is often not an option, due to its limited distance range and physical limitations that apply to part of the population (Popovich et al., 2014, MacArthur et al., 2014). E-cycling has been introduced over the past decade as a way to overcome limitations of distance and physical limitations. Over the past decade, e-bikes have obtained a non-trivial market share in various places in the world (Fishman and Cherry, 2016). This raises questions about the health and well-being effects of a shift from other travel modes to e-biking. Notably, travel satisfaction using e-bikes has not yet been addressed in transportation literature. Also, those taking up e-biking often use their e-bike to make trips previously made by other modes, potentially resulting in a change in travel satisfaction (Abou-Zeid et al., 2012). Studying such a shift is important in order to learn about the well-being effects of a shift to e-bike, but also since mood or satisfaction associated with new behaviours may be an important predictor of adherence to the new behaviour (Standage et al., 2012). An additional relevant issue in this respect is how travel satisfaction develops over a longer period, as this may have implications for adherence on the longer term. To address the effects of a shift to e-biking on travel satisfaction, this study uses unique longitudinal data of satisfaction with the commute trip among participants in an e-cycling stimulation programme (B-riders). Conclusions are drawn regarding the direct and longer-term effects of the shift to e-bike on travel satisfaction, and on the personal and contextual factors influencing these effects. The outcomes will be interpreted in the context of psychological theories about the dynamics of satisfaction and well-being, such as the hedonic treadmill theory, and theories about biases in individuals’ expectations.

2. Materials and Methods Study design With the e-cycling incentive program (B-Riders), the province of North-Brabant in the Netherlands aimed to stimulate a switch from car commuting towards the use of the e-bike. E-bike use was stimulated by giving participants financial compensation depending on their e-bike use. On average, it would take up to a year to reach the maximum financial incentive. E-bike use was monitored using a smartphone app that tracked participants’ travel behaviour using GPS. In order to measure behavioural change and satisfaction with travel, three questionnaires were conducted. The baseline questionnaire (T0) recorded the travel modes used for commuting in a regular week before starting to commute by e-bike. In addition, respondents reported their experienced satisfaction with current car-travel to work, their expected satisfaction with e-cycling to work and a set of personal and household characteristics. The second questionnaire was held a month after the start of participation in the program (T1). It included questions about frequency of travel modes used for commuting (including the e-bike), and experienced satisfaction with the e-bike commute. The third questionnaire (T2) again recorded the frequency of travel modes used for commuting (including the e-bike) and experienced satisfaction with the e-bike commute and change in physical health since entering the program. Satisfaction with the commute was measured using the 9-item Satisfaction with Travel Scale (STS), which includes cognitive and affective items.

3. Results

3.1. Descriptive analyses Table 1 shows the average STS scores at T0, T1 and T2 per STS sub-scale and the overall STS score per participant. Both the expected satisfaction with e-cycling (T0) as well as the experienced satisfaction with e-cycling (T1 and T2) are significantly higher than satisfaction with the car commute for all participants. This holds for the aggregate nine-item STS score as well as the three subscales. This increase is found both for commuters only using the car at T0 and those occasionally cycling to work at T0. Apparently, the e-bike is an attractive alternative for the car commute, irrespective of prior cycling experience. Experienced STS of e-cycling at T1 is somewhat lower than the expectation at T0. Hence, participants seem to slightly overestimate their travel satisfaction. Nevertheless, the increase in experienced STS as compared to commuting by car is much larger than the small difference with the expectation. Experienced STS at T2 shows an increase compared to T1, bringing the experienced STS back to the initial expectations on most STS subscales. Apparently, over a 5-month period, multimodal commuters learn to appreciate e-biking more.

Table 1: Satisfaction with car-commuting (T0), expected e-cycling (T0) and experienced e-cycling T1 and T2 Variable Car experience T0 Expected E-bike T0 E-bike experience T1 E-bike experience T2 All participants (N=547) Positive deactivation - negative activation 0,87 2,18 2,05 2,15 Positive activation - negative deactivation 0,27 2,06 1,85 1,92 Cognitive evaluation 0,55 1,83 1,66 1,78 Satisfaction with travel 0,56 2,02 1,85 1,95 Only car-commuters (N=172) Positive deactivation - negative activation 1,05 2,15 2,02 2,09 Positive activation - negative deactivation 0,35 2,02 1,84 1,88 Cognitive evaluation 0,69 1,81 1,65 1,77 Satisfaction with travel 0,70 1,99 1,83 1,91 Multi-modal car-commuters (N=375) Positive deactivation - negative activation 0,79 2,19 2,06 2,17 Positive activation - negative deactivation 0,23 2,07 1,85 1,94 Cognitive evaluation 0,48 1,84 1,67 1,78 Satisfaction with travel 0,50 2,03 1,86 1,96

3.2. Regression models of satisfaction with e-cycling The overall STS was regressed on personal, household, commute, route and spatial context characteristics for different times and travel modes. At T0 the experienced satisfaction with car commuting was used as dependent variable as well as the expected satisfaction with e-cycling. Next the experienced satisfaction with commuting by e-bike at T1 and T2 were considered as dependent variables. Each model included personal and household characteristics, work place related circumstances, and route characteristics as explanatory variables (Table 2).

With respect to the expected satisfaction with e-cycling at T0, the estimation results indicate that the youngest group (25 – 40 years old) expect to be less satisfied with e-cycling compared to the oldest group of participants (50 – 65 years old). It is found that relative to participants with an excellent condition, the other groups expect the e-bike commute to be less satisfying. Irrespective of physical condition, however, the more strenuous the commute is perceived the less satisfied participants expect to be. Route characteristics influencing expected satisfaction include difficulty in wayfinding, the route being boring or going through urbanized area. As compared to the expected STS with e-cycling at T0, the experienced STS with e-cycling at T1 is partly determined by other determinants. First, the STS with e-cycling is positively influenced by the share of e-cycling in all commuting trips in a week. This may be due to a selection process, in which people who enjoy e-cycling more use the e-bike more frequently for commuting. At T1, not only younger commuters but also male participants value e-cycling less. The results further indicate that participants with a worse physical condition (relative to excellent) indeed have a lower experienced STS at T1. The degree to which the commuting trip by e-bike is regarded as strenuous affects the STS negatively. Being single has a negative impact on STS compared to couples with children, where couples without children at home have a higher STS. Results of the model for T1 further indicate that perceived safety, ease of wayfinding, degree of green, liveliness and cosiness all influence STS positively. As compared to the expectation of e-cycling at T0, route safety and greenness influence the experience at T1, but not the expectation at T0, and distraction by billboards influenced the expectation at T0 but not the experience at T1. To explore the medium-term effects of switching to e-cycling, we regressed satisfaction with e-cycling after six months (T2) on the same set of explanatory variables. Compared to the models for T0 and T1, we find that the effect of physical condition has diminished, and only those with a poor physical condition now have a lower STS. This may be due to the fact that participants’ condition improves as a result of e-cycling. In support of this, we find that an improvement in physical condition scores between T2 and T0 shows affects the STS positively. Similar to the results at T1 the environment has a significant impact on the STS, where in addition to easy wayfinding, greenness and liveliness also lack of height differences and slopes is found to be associated with a higher satisfaction with e-cycling. Finally, we note that satisfaction with e-cycling at T2 is not affected by unsafety as at T1, which may be attributed to habituation with the route.

Table 2: Regression analysis of STS for car-commuting (T0), expected e-cycling (T0) and experienced e-cycling T1 and T2 Car T0 E-bike T0 E-bike T1 E-bike T2 Age 25-39 years old -0,243 -0,301 40-49 years old Gender Male -0,219 Physical condition Phys. cond. bad -0,267 -0,375 -0,286 Phys. cond. neutral -0,225 Phys. cond. good -0,176 -0,181 Change in physical health - - - 0,152 Car ownership 1 car per household Household Income < 3.000 3.000 - < 4.000 Household composition Single -0,354 Single parents Couple 0,343 Urbanization (very) Strong urbanized 0,288 Moderate urbanized 0,255 Less urbanized 0,269 Start/end time Flexible Cycling distance 0-5 km 5<10 km 10<15 km 0,208 15<20 km Commuting days a week 1-3 days a week 0,241 4 days a week Car commuter Habitual cycling Cycle share T0 E-cycling E-cycle share - - 0,541 0,441 Strenuous commute Not at all - Very -0,309 -0,109 -0,115 -0,104 Crowdedness during commute Very quiet - very busy Freedom of speed determination Under control - by other road users -0,067 Annoyed by road users Own control - by other road users Threatened/Unsafe by road users Not at all - very much Route unsafety Very safe - very unsafe -0,065 Wayfinding Easy - difficult - -0,143 -0,133 -0,168 Distraction by billboards Not at all - very much - -0,090 Green Very little - very much 0,090 0,077 0,117 Openness Sheltered - open Aesthetics Pretty - ugly Liveliness Lively - boring -0,091 -0,094 -0,096 Atmosphere Cosy - distant -0,081 -0,083 Height difference Many - little 0,066 Landscape Alternately - monotonous Urbanization Little urban - strong urban 0,065 Goodness-of-fit (R2) 0,447 0,329 0,354 0,388

4. Conclusion Our study finds that travel satisfaction with e-cycling is higher than for car-commuting. With an increase of about 1.4 on a 7-point scale, this suggests that a shift from car to e-bike generates a considerable increase in commute satisfaction, and therefore possibly in overall well-being. Notably, a similar increase was observed for commuters who only used the car before the programme and those who occasionally used the bicycle. We find that the initially experienced satisfaction with e-cycling is slightly lower than the expected satisfaction. This suggest that both the focussing illusion (Schkade and Kahneman, 1998) and forecasting bias (Wilson and Gilbert, 2003) might be present: participants slightly overestimate the positive aspects of e-cycling. However, initial satisfaction with e-cycling is still high. Where some well-being literature indicates a hedonic treadmill effect (Diener et al., 2006), our study finds that travel satisfaction remains high for a period up to six months, and actually slightly increases, which might have positive and long-lasting effects on e-cycling. Interestingly, the negative impact of physical condition diminishes over time, suggesting that improved condition as a result of e-cycling might be a reason for the slightly increased satisfaction. Again, this adds to the attractiveness of e-cycling on the longer term and adherence of participants to e-cycling. In addition, our study finds that the attractiveness of the route influences travel satisfaction. Factors like greenness and liveliness of the environment contribute to a positive travel satisfaction. In this respect, e-cycling resembles conventional cycling.

References Abou-Zeid, M., Witter, R., Bierlaire, M., Kaufmann, V., & Ben-Akiva, M. (2012). Happiness and travel mode switching: Findings from a Swiss public transportation experiment, Transport Policy, 19, 93–104. Diener, E., Lucas, R. E., & Scollon, C. N. (2006). Beyond the hedonic treadmill: revising the adaptation theory of well-being. American psychologist, 61(4), 305. Fishman, E. & Cherry C. (2016). e-bikes in the Mainstream: reviewing a decade of research, transport reviews, 36 (1), 72-91. Friman, M., Olsson, L. E., Ståhl, M., Ettema, D., & Gärling, T. (2017a). Travel and residual emotional well-being. Transportation research part F: traffic psychology and behaviour, 49, 159-176. MacArthur, J., Dill, J., & Person, M. (2014). Electric Bikes in North America: Results from an online survey. Transportation Research Record: Journal of the Transportation Research Board, 2468, 123–130. Mao, Z., Ettema, D., & Dijst, M. (2016). Commuting trip satisfaction in Beijing: Exploring the influence of multimodal behavior and modal flexibility. Transportation Research Part A: Policy and Practice, 94, 592-603. Morris, J. N., & Hardman, A. E. (1997). Walking to health. Sports medicine, 23(5), 306-332. Olsson, L. E., Gärling, T., Ettema, D., Friman, M., & Fujii, S. (2013). Happiness and satisfaction with work commute. Social indicators research, 111(1), 255-263. Popovich, N., Gordon, E., Shao, Z., Xing, Y., Wang, Y., & Handy, S. (2014). Experiences of electric bicycle users in the Sacramento, California area. Travel Behaviour and Society. Schkade, D. A., & Kahneman, D. (1998). Does living in California make people happy? A focusing illusion in judgments of life satisfaction. Psychological Science, 9(5), 340-346. Standage, M., & Ryan, R. M. (2012). Self-determination theory and exercise motivation: Facilitating self-regulatory processes to support and maintain health and well-being. St-Louis, E., Manaugh, K., van Lierop, D., & El-Geneidy, A. (2014). The happy commuter: A comparison of commuter satisfaction across modes. Transportation research part F: traffic psychology and behaviour, 26, 160-170. De Vos, J., Schwanen, T., & Witlox, F. (2017). The road to happiness: from obtained mood during leisure trips and activities to satisfaction with life. In 2017 World Symposium on Transport and Land Use Research (WSTLUR). Wilson, T. D., & Gilbert, D. T. (2003). Affective forecasting. Advances in experimental social psychology, 35, 345-411.

ABSTRACT. The study examines the tour-level vehicle allocation decisions among individuals during mandatory- and discretionary-activity tours based on travel accompanying arrangements, specifically, solo travel (i.e. traveling alone) and joint travel (i.e. traveling with household/non-household members). Latent segmentation-based random parameter logit (LSRPL) models are developed in this study to explore the factors affecting vehicle allocation behavior within households, including travel characteristics, built environment and accessibility measures. For instance, presence of children in joint mandatory- and discretionary-activity tours increases individuals’ probability of getting SUVs from their households’ existing vehicle fleet. Also, tour complexity identified by higher number of activity stops, exhibits positive coefficient value for SUV allocation in case of discretionary-activity tours. One of the unique features of this study includes evaluating the effects of individuals’ attitudes on vehicle allocation decisions at tour-level. Results suggest that due to a positive attitude towards active transportation, individuals are observed to decrease their likelihood of using vehicles during both solo and joint mandatory-activity tours. The vehicle allocation decisions in the households, however, vary across two segments. For example, older-higher income individuals in segment one tend to get SUVs from their household vehicle fleet during a joint discretionary-activity tour while living in higher mixed land-use areas, however, younger-lower income individuals in segment two exhibit negative relationship for SUVs. In addition to the heterogeneity across segments, preference of SUVs during a joint discretionary-activity tour might vary among individuals within the same segment, as indicated by statistically significant standard deviation of ‘land-use index’ variable. The models developed in this study will generate a newer module within the Halifax iTLE model as an extension of the vehicle ownership decision model.

ABSTRACT. The contribution to the community development inside the neighborhood is essential in neighborhood planning. Neighborhood planners often have focused on physical design, and attempted to build a sense of community via specific design elements, for example, by integrating residential and public spaces, careful designs and layouts of infrastructure in the neighborhood. However, Madanipour (2001) noted that neighborhood planning is under severe criticism because it emphasizes the physical rather than the social setting. Meanwhile, such planning may create physical proximity among residents, but it may not create social bonds among them. In addition, environmentally friendly neighborhoods have been emphasized on reducing vehicle travel miles, providing employment and services in a neighborhood, and so forth. Since with rapid motorization and complex modern lifestyles, residents could not entirely conduct their activities inside a neighborhood, especially for new neighborhoods in Asia where the residents typically travel to a central business district nearby for work, the environmental friendly neighborhood with regarding non-working activities within the neighborhood would be more feasible. Therefore, it is necessary to examine social fabric regarding the non-working purposes within a neighborhood, which contribute to nourishing the community development in a neighborhood. Although social network plays an important role in understanding and predicting travel behavior is generally acknowledged, but from the neighborhood perspective is relatively new. There are a variety of terms implying endogenous social effects, depending on the context, being called as “social interactions”, “neighborhood effects”, “peer effects”, “conformity effects” and so on. According to Manski (1993), social interaction effects can be classified into three components (endogenous effect, exogenous effect (contextual effect), and correlated effect) which is difficult to distinguish empirically (known as “reflection problem”). Lee et al. (2014) mention that reflection problem does not really happen when a real social network is used instead of a reference group since endogenous effects are heterogeneous across individuals, and correlated effects can also be controlled by introducing random-effect or fixed-effect terms to take of account group-level unobserved factors. The remaining concerning is to take account into the whole social network needs to be observed. Therefore, handling social network formation as the real one in the neighborhood based on the partial observed information is necessary. Previous studies on social network formation focus on the small size of population as students’ network within a class, school where agents/firms know others within the population size well (Mayer and Puller, 2008). It seems not feasible to collect social network information of the entire population in a neighborhood. A dominant approach to develop a social network formation is exponential random graph model which predict the probability of a given set of agents as the graph based on an exponential function of possible configuration (Robin et al., 2007). Another approach is agent-based model focuses on agents’ behavior and network dynamics, the objective function of agents represents observed tendencies in reality as homophily, geographic distance, and geographic transitivity (Arentze et al., 2012). Because the availability of transitivity can be known after a network has been created, transitivity still is a promising field of research (Arentze et al., 2012; 2013). Since this study focus on the travel behavior inside a neighborhood where majority activities are non-working purposes, the geographic distance among agents within a neighborhood would be ignored. Therefore, this study is to develop a social network formation with an objective function related to homophily in a neighborhood for use in neighborhood planning. Motivated by the above shortcomings of existing studies on neighborhood planning and lack of studies focusing on developing countries, this study attempts to capture the endogenous effects of social network on destination choice a network formation model from partial observation a in the context of Hanoi, Vietnam. A two-day travel diary data were collected in 2015 from 450 respondents with 1,706 trips made for discretionary activities (the trips with purposes are shopping, personal business, meeting with friends/acquaintance, doing exercise, eating out, taking around, leisure activities, and so on) at three new urban areas in Hanoi Metropolitan Area, Vietnam. Based on preliminary analyses we did, around 56% of discretionary activities tend to be done inside the neighborhood. Since new neighborhoods in Hanoi were not planned for providing jobs or quality hospitals/universities at the beginning, it would not be easy to reduce traveling out for mandatory purposes. The average number of acquaintance inside each area is highest in Ecopark (16.96), followed by Van Quan (8.77) and Viet Hung (7.95). The residents in three neighborhood areas are on the way to shape their acquaintance network inside the area. About a half of respondents having 1-10 acquaintances within the areas, and the number of respondents answering more than ten acquaintances is quite large. In addition, for an agent’s network, the list of six people who the agent has had the most frequent regular face to face contact with over the past six months. Concretely, a destination choice behavior for discretionary activities is modeled, where the average choice probability of acquaintances are incorporated in the model as an endogenous variable. Two methodological challenges are addressed in this paper: (1) observing the whole social network in a neighborhood is not feasible, and it is simulated based on the partial information of social network, and (2) a structural estimation method under the control of unobserved neighborhood-level characteristics (which is achieved by introducing a fixed-effect term) is adopted to reduce biases in the estimated social interaction effects. These two contributions are not independent of each other. Indeed, as there will mention later in details, the iterative process of structural estimation allows us to naturally utilize the simulated social network to update the average choice probability of (simulated) acquaintances.

Model specification: The employed model structure is basically analogous with Lee et al. (2014). Consider the destination choice of residents in a neighborhood g in a sample of G neighborhood networks. The population of neighborhood g is Ng and the sample size collected through the survey is ng. The utility Ugit that an individual i (i = 1, 2, …, Ng) who travels on a t-th trip (t = 1, 2, …, T) chooses destination ygit = 1 (1: inside the neighborhood area, 0: outside the neighborhood area). The social network among Ng individuals is represented by a Ng  Ng adjacency matrix , where wgi is the i-th row of Wg (i.e., wgi = (wgi1,…, wgij,…, wgiNg)). The diagonal elements of the matrix (i.e., wgii) are assumed to be zero. In general, since no specific information is available about the importance of acquaintances, wgij is assumed to be equal one if j is an acquaintance and zero otherwise, meaning that individual i’s acquaintances receive equal weight. In this study, a row-normalize the values of the row of the weights matrix is added so that the row sum is 1. The utility Ugit is defined as follows:

(1)

where β, and  are vectors of parameters, xgit is a vector of explanatory variables (including socio-demographic, trip purposes, mobility and accessibility), pgit is a vector of acquaintances’ probability of choosing the destination inside the neighborhood (i.e. destination choice decision will be influenced by his/her connected acquaintances in the neighborhood’s network), ug denotes the neighborhood fixed effects which are unobserved effects of the common factors faced by the same neighborhood members, and εgit is an error term which is Gumbel-distributed. Controlling the correlated effects by introducing the group fixed-effects ug may be appropriate in this study, since the number of groups is only three (Viet Hung, Van Quan and Ecopark). The probability that individual i chooses destination inside a neighborhood given neighborhood’s network g can be written as the following standard logit formulation.

(2)

The nested pseudo maximum likelihood (NPL) estimation proposed by Aguirregabiria and Mira (2004) (Aguirregabiria, 2004). This estimation procedure is similar to Hoz and Miller (1993) but takes further an iterative process as follows. Let p0git be the nonparametric frequency estimator just like the CCP estimator. The K-stage NPL estimator can be defined as follows,

(3)

where the sequence of probability distributions { : K≥1} are constructed recursively as

(4)

The above iterative process will be continued till the parameters θK are converged.

Keywords: Social network formation; Social interaction effect; Destination choice; Neighborhood planning.

  Reference: Aguirregabiria, V. (2004) Pseudo Maximum Likelihood Estimation of Structural Models Involving Fixed-Point Problems. Economics Letters, 84(3), 335-340. Arentze, T., van den Berg, P., Timmermans, H. (2012) Modeling Social Networks in Geographic Space: Approach and Empirical Application. Environment and Planning A, 44(5), 1101-1120. Arentze, T. A., Kowald, M., Axhausen, K. W. (2013) An Agent-Based Random-Utility-Maximization Model to Generate Social Networks with Transitivity in Geographic Space. Social Networks, 35(3), 451-459. Lee, L.-f., Li, J., Lin, X. (2014) Binary Choice Models with Social Network under Heterogeneous Rational Expectations. Review of Economics and Statistics, 96(3), 402-417. Madanipour, A. (2001) How Relevant Is 'Planning by Neighbourhoods' Today? On Jstor. Town Planning Review, 72(2), 171-191. Manski, C. F. (1993) Identification of Endogenous Social Effects: The Reflection Problem. The Review of Economic Studies, 60(3), 531-542. Mayer, A., Puller, S. L. (2008) The Old Boy (and Girl) Network: Social Network Formation on University Campuses. Journal of Public Economics, 92(1), 329-347. Robins, G., Pattison, P., Kalish, Y., Lusher, D. (2007) An Introduction to Exponential Random Graph (P*) Models for Social Networks. Social Networks, 29(2), 173-191.

ABSTRACT. Introduction The Value of Travel Time (VOTT) is a prominent element of many infrastructure projects, travel demand models, and social justice analyses. A common way to assess VOTT is to estimate a mode choice model and then compute the ratio of parameter estimates for travel time and travel cost, indicating the tradeoff travelers are willing to make between time and money, and thereby yielding a monetary value of time spent traveling. In addition to providing a monetary value for travel time, such studies have also aimed to determine how VOTT varies among different segments of the population, typically by estimating different mode choice models for each segment. Athira et al. (1), for instance, report that the VOTT associated with work trips tends to be greater in higher-income groups, while distance correlates positively with VOTT. Furthermore, a meta-analysis of travel time in Europe (2) points to income, journey purpose, mode used, mode perception, and mode value as factors affecting VOTT. A number of authors (e.g. 3) have noted that the increasing ability to multitask while traveling is likely to influence the VOTT, and in particular to reduce the disutility of travel time (diminishing the magnitude of its negative coefficient). To our knowledge, however, only one study has operationalized this conjecture, and then only with stated response data. Ettema and Verschuren (3) found that multitasking attitudes and behavior have a significant impact on the VOTT and should not be ignored in policy analyses. In this study, by taking an already estimated mode choice model as a base, we aim to model the coefficients of IVTT and OVTT as a function of attitudinal factors, mode perceptions, multitasking behavior, and socio-economic variables. The result will be a richly-nuanced portrait of taste variations in the impact of travel time on mode choice. In the following sections, we describe the approach of this work-in-progress in more detail, and present some preliminary results. Dataset and Methodology This study uses survey data collected from more than 2,000 commuters in Northern California, in the fall and winter of 2011-2012 (4). The survey collected socio-economic characteristics, attitudes, personality traits, commute mode choice and commuters’ engagement in multitasking. The person with average characteristics for this sample is a female, around 45 years old, college graduate, and lives in a household of 2.7 people owning 2.1 vehicles and earning $75,000 - $99,999 annually. Using this dataset, Malokin et al. (5) estimated a discrete choice (multinomial logit) model for the primary commute mode. They categorized the modes into five groups: bicycling, rail (intercity/commuter), transit (bus, light rail, subway), shared-ride (carpooling, vanpooling, employer shuttle), and driving alone. The model estimates an alternative-specific in-vehicle travel time (IVTT) coefficient for biking, and a generic one for the remaining alternatives. In view of the distinctive nature of bicycling compared to the other modes, and the relatively small number of bicycling commuters in the sample, we focus the present study on the remaining four modes, for which the IVTT coefficient is -0.016 (p < 1%). The out-of-vehicle travel time (OVTT) coefficient is, on the other hand, generic across all alternatives, and equal to -0.048 (p < 1%). Consistent with conventional wisdom (6), it is about three times larger in magnitude than the generic IVTT coefficient, signifying the greater physical and mental discomfort associated with waiting for a vehicle and/or transferring between vehicles. Taking the Malokin et al. model as our starting point (and also referring to its age-segmented counterpart in Malokin et al. (7(, in progress), our aim in this study is to model the coefficient of travel time by introducing interaction terms with IVTT and OVTT into the mode choice model. More formally, we estimate, interpret, and analyze the equation: β_(tt,n)=α_0+α_1 z_(1,n)+α_2 z_(2,n)+⋯+α_M z_(M,n)+η_n , (1) where β_(tt,n) is the coefficient of TT specific to individual n (tt = IVTT, OVTT), the zs are M variables that influence β_(tt,n), the αs are the unknown (and to-be-estimated) coefficients of those variables, and η_n denotes the unobserved residual of β_(tt,n). Some of the z variables will be mode-specific and thus Eq. (1) is mode-specific, but we suppress the mode subscript for simplicity. While the initial results we present here pertain to the linear form of Eq. (1), ongoing research will experiment with non-linear transformations of the z variables. Multiplying both sides of Eq. (1) by the travel time variable, and therefore constructing the interaction terms, yields: β_(tt,n) 〖TT〗_(i,n)=α_0 〖TT〗_(i,n)+α_1 z_(1,n) 〖TT〗_(i,n)+α_2 z_(2,n) 〖TT〗_(i,n)+⋯+α_M z_(M,n) 〖TT〗_(i,n)+η_n 〖TT〗_(i,n) (2) The last term in Eq. (2) will be absorbed into the error term (i.e. the unobserved residual) of the utility function for the associated mode. Although we could have modeled taste heterogeneity using a mixed logit model (estimating the parameters of an assumed distribution for the random variable β_(tt,n)), doing so would not have yielded individual-specific coefficients, and thus would have offered only limited insight into the factors influencing β_(tt,n). We could have parameterized the parameters of the assumed distribution of the β_(tt,n) parameter (e.g., expressing the mean and/or standard deviation of β_(tt,n) as a function of z variables), but that would simply replace the single random variable β_(tt,n) with N random variables – a distinctly-distributed random coefficient for each person – and it is simpler (and, in our view, more insightful) to allow each person to have her own coefficient that is constant, but systematically varying across people. Preliminary results Although our analysis is still in progress, we have so far obtained promising results. First, we assembled a pool of conceptually plausible variables from among those available from the survey. These variables include: attitudinal factors, multitasking attitudes and propensities, mode perceptions, and socio-economic variables. Then, we interacted them with IVTT, singly and a few at a time, and re-estimated the mode choice model. Table 1 shows the coefficients for IVTT, the significant IVTT-related interaction terms, and the associated interacted variables (“main effects”). Table 1: Estimated travel time interaction variables in the mode choice model Variable Coefficient P-value IVTT -0.0688 0.001 Interaction terms IVTT * Age 0.0011 0.027 Mode benefit perception 0.0064 0.008 Propensity to use laptop (Rail) 0.0064 0.008 Per capita income (Drive alone) -0.0002 0.007 Completed graduate degree (Rail) 0.0178 0.013 Other main effect terms Age -0.0051 0.390 Mode benefit perception 0.2335 0.006 Propensity to use laptop 0.6891 0.065 Per capita income -0.0045 0.125 Completed graduate degree -0.2107 0.159

Figure 1 shows how, for each mode, the βIVTT (as defined in Eq. 1) changes with respect to each of the interacted variables, while the others are held at their averages.

As Table 1 and Figure 1 show, age shows a positive correlation with the coefficient of IVTT, indicating that older commuters find a given amount of travel time to be less onerous than do their younger counterparts. We suggest that this points to a habituation effect, i.e. older commuters are less impatient and more used to the pattern of their commute, and/or have been better able to optimize their commute over time. The perceived mode benefits, expressed by a factor score, tend to be higher when a mode is perceived as being good in terms of its effect on the environment, ability to carry things, cost, avoiding congestion, and physical effort. A higher score for a mode, as Table 1 shows, is positively correlated with the coefficient of IVTT, indicating that IVTT can be less of a burden when commuters have more positive perceptions of the benefits of the mode. Among rail commuters, those who have a higher propensity to use a laptop have a less negative IVTT coefficient, indicating that the ability to use a laptop can reduce the disutility of travel. This interaction was only significant for commuter rail, pointing to the superior ability of this mode to support productive activities. The main effect of the laptop propensity variable is not highly significant, but (as in the case of other interaction terms) we needed to keep it in the model so as to be able to correctly estimate the coefficients (8). The main effect, nevertheless, has a positive sign, further indicating that using laptop improves the utility of the associated mode. For the drive alone mode, higher income makes the IVTT coefficient more negative. This again is according to reason and literature, indicating that those who earn more have a higher value of time and a longer IVTT can be more onerous for them compared to those who earn less. The main income effect is not significant at a 0.1 significance level, but having a negative sign indicates that, compared to the drive alone mode (the base alternative for the individual-specific variables), it also is negatively associated with the utility of the corresponding modes. Finally, a dummy variable indicating a high level of education (complete graduate studies) results in a less negative IVTT coefficient for the rail mode. This is plausible, in view of the fact that those who are well-educated may (1) be better equipped to use, and (2) have jobs better suited to using, their travel time more productively than those who are less well-educated. However, it points to the fact that the benefits of travel time productivity are not at all evenly distributed. A careful look at Figure 1 shows that β ̂_(tt,n) will be positive for about 6% of cases in the sample. The existence of positive coefficients of travel time is controversial (9), and we will carefully analyze those cases to better understand them. It is likely that in many cases the positive value arises from the linear extrapolation of what is really a non-linear relationship to the extremes of the observed range (e.g. for age). On the other hand, we can also imagine that the ability to travel multitask may in some cases make a longer trip more preferable than a shorter one, because the longer trip allows for a meaningful block of time to be devoted to productive activities whereas the shorter trip may comprise essentially wasted time. Consistent with that conjecture, it is telling that proportionally more commuters on rail and transit – potentially the two modes most conducive to productive multitasking – seem to have positive βIVTT than for the other two modes. This observation may have implications for the impacts on travel demand of the adoption of autonomous vehicles in the future, where commuters are freed from the burden of driving and can spend their travel time in more relaxing and/or productive pursuits. Further plans The results so far are only for IVTT, and our exploration has not yet yielded any significant interactions with OVTT. Regardless, we will further explore the variables and then report any conclusion. In addition, once we have finalized a model with interaction terms, we will more intensively analyze the resulting individual-specific travel time coefficients. Using the estimated αs from eq. (1), we will compute β ̂_(tt,n) for each person, and plot the distribution of β ̂_(tt,n) across the sample. We will particularly focus on cases whose estimated betas are positive. Further, we will compute the utility part-worths for IVTT as a whole (β ̂_(tt,n)×IVTT) and for each set of interacted variables, and see for how many (and which) of the cases the joint effect on utility is negative or positive. For example, we speculate that β ̂_(tt,n) is more likely to be positive when IVTT is small.

References [1] Athira, I. C., C. P. Muneera, K. Krishnamurthy, and M. V. L. R. Anjaneyulu. Estimation of Value of Travel Time for Work Trips. Transportation Research Procedia, Vol. 17, No. Supplement C, 2016, pp. 116-123. [2] Wardman, M., V. P. K. Chintakayala, and G. de Jong. Values of Travel Time in Europe: Review and Meta-analysis. Transportation Research Part A: Policy and Practice, Vol. 94, No. Supplement C, 2016, pp. 93-111. [3] Ettema, D., and L. Verschuren. Multitasking and Value of Travel Time Savings. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2010, 2007, pp. 19-25. [4] Neufeld, A. J., and P. L. Mokhtarian. A Survey of Multitasking by Northern California Commuters: Description of the Data Collection Process. Institute of Transportation Studies, University of California, Davis, Research Report UCD-ITS-RR-12- 32, 2012. Available at: http://www.its.ucdavis.edu/?page_id=10063&pub_id=1802, accessed October 20, 2017. [5] Malokin, A., G. Circella, and P. L. Mokhtarian. How Do Activities Conducted While Commuting Influence Mode Choice? Testing Public Transportation Advantage and Autonomous Vehicle Scenarios. Presented at Transportation Research Board 94th Annual Meeting, Washington D.C., 2015. (Under review for publication; available from the authors) [6] Wardman, M. A Review of British Evidence on Time and Service Quality Valuations. Transportation Research Part E: Logistics and Transportation Review, Vol. 37, No. 2, 2001, pp. 107-128. [7] Malokin, A., G. Circella, and P.L. Mokhtarian. Do Multitasking Millennials Value Travel Time Differently? A Revealed Preference Study of Northern California Commuters. Presented at Transportation Research Board 97th Annual Meeting, Washington D.C., 2017. [8] Brambor, T., W. R. Clark, and M. Golder. Understanding Interaction Models: Improving Empirical Analyses. Political analysis, Vol. 14, No. 1, 2005, pp. 63-82. [9] Hess, S., M. Bierlaire, and J. W. Polak. Estimation of value of travel-time savings using mixed logit models. Transportation Research Part A: Policy and Practice, Vol. 39, No. 1, 2005, 221-236.

ABSTRACT. See attached file.

ABSTRACT. The use of public transportation based on online application is getting popular in developing countries. The high demand also came from tourism site which has a dramatic impact on the change of traveler behavior nowadays. This research explores the linkage between tourist social networking and online transportation usage in Yogyakarta, Indonesia. By using a face-to-face interview survey data collected at 18 tourism sites in Yogyakarta, we develop a mode choice model which takes into account social interactions through social media. The main assumption we made is that the degree of social interaction effects is proportional to the number of shared social media. The preliminary results show that mode choice decisions are significantly affected by the average probability of choosing online transportation by others who are connected through social media.

ABSTRACT. As a consequence of climate change, an increase in the number of heat waves and very hot summers is expected for European cities. Almost no empirical study so far has investigated to which extent and in which way this may lead to adaptation of heat burdened urban residents. Therefore, this paper aims at (i) exploring the intentions of Viennese residents to adapt to the increasing heat by doing summer retreat trips (‘Sommerfrische’), to (ii) understand the internal (cognitive) determinants for such travel decisions, and to (iii) estimate if rural mountain destinations located close to large agglomerations may benefit from this adaptation. These objectives were pursued by conducting a quantitative online-survey in the source region of Vienna with 877 persons. The survey is based on the Theory of Planned Behavior (TPB) by Ajzen (1991) as theoretical framework extended by the construct of (positively formulated) social norms (Anable 2005).

The results indicate a strong overall interest in visiting ‘Sommerfrische’-destinations. The intention to travel to ‘Sommerfrische’-destinations increases in expectance of a higher number of heat days. Almost half of the sample experienced last heat summers to be burdening and about one fifth stated that they have already changed their travel behaviour in response to the increase of heat The model results show that travel choices might be less a matter of individual attitudes and preferences, but that they are strongly influenced by a person’s social environment including family and friends.

ABSTRACT. INTRODUCTION Travel and tourism industry is undergoing transformation with the flourishing of online sharing economy marketplaces such as Uber (for taxi services), Eatwith (for community restaurants), and Airbnb (for accommodation). The shared housing market place Airbnb with its large inventory and wide reach across the globe is redefining the hospitality sector. Airbnb is unique in its design as it does not own any properties but provides a platform for ordinary people (sellers) to rent their residences (entire house/apartment or a room) to tourists (consumers) (Botsman & Rogers, 2011). Airbnb accommodation system is quite easy to use: a consumer searches for an entire home or private (or shared) room based on their travel dates, destination on the Airbnb website (www.airbnb.com). The user is provided with a list of housing alternatives based on the user preferences. The success and wide adoption of the system is based on available review information and background check procedures for renters and tourists. Airbnb charges a service fee for each transaction. Initiated in 2008, popularity of this sharing hospitality platform has rapidly grown with over 200 million guests having stayed in about 3 million listings in more than 65,000 cities and 191 countries (Airbnb, 2017). In fact, since 2016, over 100 million people have enjoyed the accommodation through Airbnb while over 1 million new listings worldwide have been added to the market place. The growth of Airbnb impacts transportation and urban systems along two major directions. First, Airbnb provides a unique snapshot of the hospitality industry and can serve as a surrogate for the health of tourism industry in the region. The number of available listings on Airbnb can serve as a proxy for tourist interest in the region. Airbnb provides renters with an opportunity to immediately respond to tourist demand by allowing for a simple listing process (without any substantial capital costs). In the event of a drop in tourist demand, renters on the website remove their listing. On the other hand, traditional hospitality industry with hotels respond to tourist demand slowly due to the large capital costs involved in increasing capacity. In addition, the traditional hospitality sector cannot dismantle their infrastructure as easily in response to the reduced tourist demand. Thus, with its ease of adding a listing, the Airbnb listings provide a unique snapshot of the health of tourism industry. Second, an analysis of Airbnb listings will allow transportation and urban regional professionals examine the demand arising from these tourists on transportation and urban infrastructure. Cities such as New York that receive significant expenditures from tourists can provide improved services by enhancing infrastructure in response to emerging tourist locations.

Current Study Given these afore-mentioned implications, the proposed research conducts a comprehensive analysis of Airbnb listing in New York City region drawing on data from January 2015 to September 2017 (http://insideairbnb.com/get-the-data.html). The listings dataset provides information on zip code, longitude and latitude, city and street name, accommodation information such as residence type (full apartment or private/shared room), number of bedrooms and bathrooms, price, amenities information and review of customers. The listings data is aggregated at a census tract level (2166 census tracts) in the New York City region. The analysis examines the evolution of Airbnb listings at a census tract level by listing type – entire home or private/shared room. The dependent variable is defined as the number of listings in the census tract by listing type. Given that each census tract has two dependent variables with multiple repeated observations for each CT, observed and unobserved factors affect these variables. While observed variables can be included in the univariate models, the consideration of the influence of unobserved factors requires a panel multivariate or joint modeling approach. Earlier research efforts on modeling count variables have developed simulation oriented multivariate models that stitches together the various dimensions within a maximum simulated or Bayesian approach (see Yasmin, and Eluru (2017) for an extensive literature review). Alternatively, bivariate copula framework that treats the variable dimensions as a joint distribution have also been developed (see Nashad et al., 2016). The first approach allows for accommodating unobserved attributes affecting the joint distribution as well the individual count components. The copula approach only allows for the influence of unobserved factors on the joint distribution within a closed form framework. In our proposed research, we build on these two model structures to accommodate for repeated measures by developing a unified framework that accommodates for dependency within a joint copula framework while also allowing for random parameters within each count model. To the best of the authors’ knowledge, this is the first attempt to employ such a unified framework for examining count events.

Variables Description The first step in data assembly for analysis is sample formation to generate the dependent variables for the analysis (count of availability of home/room) from disaggregate listing data. The density distribution of full apartment/home and private or shared room for 31 months for each census tract level of NYC is shown in Figure 1. Of the 2166 census tracts, 120 tracts have no Airbnb listings. In terms of the two dependent variables, around 17% of the census tracts have zero full apartment/home listings while the corresponding number for private/shared room is about 10%. Further, the figures indicate that major portion of the Airbnb listings are observed in Manhattan and Brooklyn boroughs. Given that the NYC tourism industry is concentrated in these two boroughs the trend is expected. In addition to the listing database, the explanatory attributes considered in the empirical study will also be generated at the CT level. The selected explanatory variables can be grouped into three broad categories: (1) sociodemographic characteristics, (2) built environment and land use attributes and (3) transportation infrastructure attributes. The built environment and transportation infrastructure attributes are derived from New York City open data (https://www.nycopendata.socrata.com/) while the socio-demographic characteristics are obtained from US 2010 census. Sociodemographic characteristics considered will include population density, job density and establishment density. Land use attributes considered will include proportion of office and industrial land use, number of conference or convention center, distance from points of interests (such as Metropolitan Museum of Art, Statue of Liberty and Times Square). The number of restaurants (including coffee shops and bars), area of recreational parks and shopping centers in the CT level will also be considered. Transportation infrastructure attributes to be included are distance from metro and bus stations, roadway network density and CitiBike availability.

Econometric Methodology see PDF file

Expected Results The research methodology will follow the proposed steps. First, independent negative binomial models by room type will be estimated. The model development will include variables from the following three categories: (1) sociodemographic characteristics, (2) built environment and land use attributes and (3) transportation infrastructure attributes. Second, a copula model that accommodates for common unobserved factors affecting the two count variables is estimated. The model estimation will consider all six copulas for considering the dependency. Third, we develop a panel random parameter within the estimated copula model. Finally, the model results will be employed to conduct a policy analysis based on elasticity estimates.

REFERENCES Airbnb, 2017. About us. Airbnb, Retrieved on March 24, 2016 from https://www.airbnb.ca/about/about-us. Bhat, C. R., & Eluru, N. (2009). A copula-based approach to accommodate residential self-selection effects in travel behavior modeling. Transportation Research Part B: Methodological, 43(7), 749-765. Botsman, R., & Rogers, R. (2011). What's mine is yours: how collaborative consumption is changing the way we live. Cameron, A. C., Li, T., Trivedi, P. K., & Zimmer, D. M. (2004). Modelling the differences in counted outcomes using bivariate copula models with application to mismeasured counts. The Econometrics Journal, 7(2), 566-584. Nashad, T., Yasmin, S., Eluru, N., Lee, J., & Abdel-Aty, M. A. (2016). Joint modeling of pedestrian and bicycle crashes: copula-based approach. Transportation Research Record: Journal of the Transportation Research Board(2601), 119-127. Sklar, A. (1973). Random variables, joint distribution functions, and copulas. Kybernetika, 9(6), (449)-460. Yasmin. S., & Eluru, N. (2017). "A Joint Econometric Framework for Modeling Crash Counts by Severity," forthcoming Transportmetrica A: Transport Science.

ABSTRACT. The MDCEV modelling framework has established itself as a preferred method for modelling time allocation, with data very often collected through travel or activity diaries. However, standard implementations fail to recognise the fact that many of these datasets contain information on multiple days for the same individual, with possible substitution between days. This talk discusses a new possible solution to accommodate these effects with budget constraints at the day and multi-day level.

We first estimate the model proposed by Bhat et al (2015) that allows accommodating complementarity and substitution across activities. Results from this model are compared with our alternative approach. Here, we instead rely on additive utility functions where we accommodate correlation between activities at the within-day and between-day level using a mixed MDCEV model, with multi-variate random distributions. We put forward adaptations of the standard Pinjari and Bhat (2010) forecasting approach to allow us to make links across days also in model application. Finally, we illustrate the issue and the methods using the Mobidrive time use datasets, confirming our theoretical points and highlighting the benefits of allowing for correlation across days in estimation and substitution in forecasting.

ABSTRACT. Background and study objectives

The journey to work is a main target for transport policy interventions given the impacts that commuting has on the daily lives of individuals and on society in general. In England in 2016, 64% of commute trips were made by car (DfT, 2017). Policy interventions will be better informed if they are based on a good understanding of individual commuting behaviour. It is widely regarded that commuting, as a frequently repeated behaviour, becomes habitual and is repeated without conscious deliberation, unless there are changes in situational context (Verplanken et al., 2008). This has led to research investigating how contextual changes influence changes in commute mode choices in the longer run (for example, Clark et al., 2016). These have made the assumption that a single mode of transport is used at any time, such as prior to and after a contextual change. However, evidence has emerged that this is an over-simplification with a significant minority of commuters exhibiting day-to-day variability in commute mode choices (Kuhnimhof, 2009; Chatterjee et al., 2016). This suggests the need to take account of day-to-day variability in commute mode choices when investigating longer term commute behaviour changes. This paper reports on results from a unique panel survey which captured both day-to-day variability in commuting behaviour and longer term change over a 18 month period, as well as explanations provided by commuters for changes in their commuting behaviour. The study objectives were to: • Differentiate different types of change in commute mode taking into account the extent of change and longevity of change; and • Explain reasons for the identified changes in commute mode occurring.

Study approach

The North Bristol Commuter Panel (NBCP) survey collected longitudinal data on the commuting behaviour of workers at 24 employers located on the western and northern edges of the city of Bristol, south-west England. In both employment areas, interventions were being carried out to encourage the take up of alternatives to driving a car alone to work. The NBCP tracked the commuting behaviour of approximately 1900 commuters every three months between March 2014 and October 2015. In each wave, panel members were asked to provide (i) the form of transport they normally used to travel to work, (ii) reasons in their own words for change in normal form of transport (if applicable), (iii) a one-week diary of their commuting travel, (iv) any changes in their personal or work circumstances and (v) awareness of local transport initiatives and any influence of these on their commuting behaviour. The analysis of the data involved first identifying the overall amount of change in normal commute mode amongst the panel participants over the survey period. It then involved detailed examination of cases where the normal mode was reported to have changed to identify if changes were enduring (by referring to the series of observations available for the same participant) and to what extent the change represented a partial or full mode shift (by referring to one-week diary data). Finally, it involved qualitative analysis of explanations for changes made based on the reasons given by participants.

Findings

Table 1 shows the prevalence of stability and change in normal mode across the whole study, by combining 8,390 pairs of reported observations of normal mode made from wave to wave. Overall, the net totals for each mode remained relatively stable across the survey period, but the level of turnover was approximately 10%, i.e. there were 866 instances where a change in normal mode did place. Changes to and from each pair of modes were relatively symmetrical. For example, there were 148 wave-to-wave changes from car alone to car share across the seven time points and 155 changes from car share to car alone. There were 61 changes from car alone to cycling, and 64 changes from cycling to car alone.

Table 1: Transitions in normal commute mode from wave to wave

In order to better understand the nature of the change in behaviours occurring, and to explore the self-reported explanations for these changes, sub-sets of participants were selected for more detailed analysis. The sub-sets comprised the most commonly observed mode transitions (car alone to/from car share, car alone to/from cycle, car alone to/from bus) and panel participants who responded at each wave (and for whom therefore a good picture could be obtained on their commuting history over the survey period). The commuting histories, together with diary date, led to the identification of three types of change in normal mode. These were stable changes (enduring over the survey period), periodic changes (temporary and reversible changes) and routine modal mixing (using a variety of modes with the relative mix varying on survey occasions). Qualitative analysis revealed four distinct prompts for mode changes. Figure 1 depicts these four prompts plus two additional motivating factors (reasoning and emotion) which could contribute to decision-making in the case of any of the four prompts.

Figure 1: Prompts and factors for changing commute mode

Reference is made to exemplar cases to illustrate each of the ‘mode change types’ and the way the above prompts and motivating factors contributed to changes occurring. The exemplars highlighted how life events, such as job changes or children starting school, and day-to-day variations in work or family routines, were often given prime importance in participants’ accounts. Changes in access to vehicles also could be important prompts for mode changes (temporary and longer lasting). External conditions represent changes to the context in which personal (or family) trips and travel decisions take place. Seasonal changes in the weather and hours of daylight were the most commonly cited external triggers for change to and from cycling. However, this category also includes changes to transport systems and services, and measures taken by local authorities and employers to discourage car use and encourage use of other modes. Whilst changes to the transport context were crucial factors for some, they generally played a supporting rather than a decisive role in prompting mode change – that is, they were secondary to occurrences within the personal realm. Respondents’ accounts of mode changes prompted by these four sets of influencing factors in the personal or external realm often included statements of reasoning and intention, as well as emotional expression relating to the experience of the commute; these two types of explanation could apply across the four prompts.

Discussion and conclusions

The analysis of the commuter panel showed that what are recorded as commute mode changes are seldom complete and enduring shifts from one mode to another. Participants’ own accounts suggested that changes in their personal situation are more dominant than changes in external factors for commute mode changes. However, local transport interventions were an additional motivating factor for some respondents who changed mode use. A key conclusion from this analysis is that interventions to promote the use of alternatives to car driving for the commute should aim to encourage not just complete mode change from one mode to another, but also greater use of non-car modes within an individual’s modal mix, in a way which facilitates the decision to leave the car at home whenever personal, work, and family commitments permit. To further increase understanding of travel behaviour change it is recommended that efforts are pursued more widely to collect longitudinal data and to combine objective measurement of travel behaviour with subjective questioning of reasoning for travel behaviour.

References

Chatterjee, K., Clark, B. and Bartle, C. (2016). Commute mode choice dynamics: Accounting for day-to-day variability in longer term change. European Journal of Transport and Infrastructure Research, 16(4), 713-734. Clark, B., Chatterjee, K. and Melia, S. (2016). Changes to commute mode: The role of life events, spatial context and environmental attitude. Transportation Research Part A: Policy and Practice, 89, 89-105. DfT (2017). National Travel Survey: England 2016. Department for Transport Statistical Release, 27 July 2017. Available from https://www.gov.uk/government/statistics/national-travel-survey-2016 (06/11/17) Kuhnimhof, K. (2009). Measuring and modeling multimodal mode use in the longitudinal section. Paper presented at the Transportation Research Board 88th Annual Meeting, Jan. 2009, Washington DC. Verplanken, B., Walker, I., Davis, A. and Jurasek, M. (2008). Context change and travel mode choice: Combining the habit discontinuity and self-activation hypotheses. Journal of Environmental Psychology, 28(2), 121-127.

ABSTRACT. The question of mode choice is hardly a static one. With increasing work flexibility, commuting behaviour is becoming more variable. As workers decide to work from home or change they departure time some days of the week, their mode choice also varies. These decisions are hardly independent from one day to the next, making a week-long modelling horizon necessary. This approach, however, requires focusing not only on which mode is selected, but how many times each mode is used throughout the week.
In this study, we compare the performance of Multiple Discrete Continuous (MDC), and Multiple Ordered (MO) models when modelling commuting mode choice. While both approaches predict the number of times each mode is used throughout the week, MDC assumes the number of times to be a continuous variable, and MO assumes it to be ordered, i.e. a natural number. In particular, we compare Bhat’s MDCEV, a system of zero inflated count models, and Hasuman-Leonard-MacFadden’s mix of MNL+Count models. We discuss theoretical and practical differences between the approaches, and measure their predicting capabilities using a revealed preferences dataset.
Preliminary results indicate similar predicting capabilities between models under similar conditions. However, if the total number of trips is known a priori, then MDCEV does a better job allocating these trips to different modes.

ABSTRACT. A growing tendency was observed about three decades ago towards day-to-day variability of individual travel behavior, although it did not receive much attention until recent years. Recently, an interesting trend of research has been initiated about recognizing the nature of such variations with respect to different aspects of travel behavior, including mode choice. Mode choice is considered as an important aspect of travel behavior, as it directly affects the society by influencing traffic congestion, air quality, public health, etc. Many research efforts are focused on recognizing predictors of mode choice dynamics in short and long terms. Yet, more concrete studies are needed to implement findings of those studies and investigate how such dynamics could affect peoples’ everyday life. The present research is an effort to extend the Agent-based Dynamic Activity Planning and Travel Scheduling model (ADAPTS) to account for such variations. ADAPTS considers a variety of travel behavior dynamics, yet it needs further developments to account for interrelationships between the mode(s) chosen on one day and the mode(s) chosen on the following day. To this end, a mode choice model is developed in this paper.

ABSTRACT. The standard random utility model assumes fully rational and compensatory decision-making behaviour. Traditionally, choice set is defined deterministically based on the analyst’s knowledge of actual choice context and choice maker characteristics. However, the role of cognitive processing strategies at choice set formation stage has started getting attention in discrete choice modelling context and various research studies have shown that non-compensatory decision-making strategy reflects the choice set formation process more realistically. It was found that overlooking choice set formation process induces bias in parameter estimates of traditional models.

There are two general approaches to account for choice set formation in the literature: stochastic two-stage approach and implicit approach. This research argues that while the stochastic Independent Availability Logit model (IAL) takes into account the probabilistic nature of the choice set formation, but it does not consider the probabilistic process of alternative inclusion in the choice set when modelling the conditional choice model. This is a simplifying assumption that can limit the prediction capacity of the model. On the other hand, models based on implicit approach do not account for the probabilistic nature of choice set formation process. These semi-compensatory models are, at best, the approximation of the actual choice set formation. In this paper, we develop a non-compensatory mode choice model to account for the effect of system performance constraints on choice set formation process. The proposed approach combines the IAL and implicit constrained multinomial logit model (CMNL) to mitigate the shortcomings of both types of models. The model is applied to actual empirical data of household travel survey of the Ottawa-Gatineau metropolitan regions in Canada. The explanatory power and elasticity measures of the proposed model is then compared with IAL and MNL model.

ABSTRACT. The ecosystem for urban mobility nowadays is much more multimodal. In this study, we examined non-separable, social-influenced multimodal route choice with congestible link capacities. The proposed model is calibrated and validated using real data collected from Citi Bike in lower Manhattan, New York City. Such models can provide decision making support for mobility operators. This work is mainly contributing to the research fields of human interaction in social network, travel behavior in multimodal system, and route choice with congestion effects.

ABSTRACT. Recent developments in transportation, driven by advances in communication and connectivity technologies, the widespread adoption of smartphone technologies, emergence of transportation network companies and new mobility business models and advances in vehicle automation technology are fundamentally changing the transportation landscape and have the potential to radically alter it even further in the future. New modes of travel and intermodal travel options are emerging to provide travelers with increased travel options, potentially leading to increases in mobility, accessibility and utility, but at the possible cost of increases in congestion, emissions and fuel use (Brown et al. 2014, Stephens et al, 2016). Wide ranges in the estimates of potential impacts on key performance metrics from emerging and future technologies are emerged, largely driven by the uncertainty in traveler behavior surrounding the use of such modes (Stephens et al 2016). Therefore, it is increasingly important when using travel demand models for evaluating impacts of future mobility technologies to capture the modal choice profiles of travelers and how mode choice could be change as mobility technologies develop.

Significant challenges are observed in the literature on mode choice modeling regarding the consideration of intermodal alternatives, the difficulty of obtaining information on unselected alternatives in revealed preference choice set, and on the use of networked skimmed travel times in the estimation and application of the estimated mode choice models (Javanmardi et al. 2016). In this paper, we propose an approach to leverage a newly developed intermodal shortest path algorithm for generating mode alternatives and constraints. We will apply the detailed evaluation of location-to-location travel characteristics from the router to an expanded set of modal alternatives, including shared-ride, transit and intermodal (TNC/taxi to transit, bike to transit, etc) in a cross-nested logit model formulation. Finally, we will apply the newly developed mode choice model in an agent-based travel simulator called POLARIS (Auld et al. 2016), to simulate the decision process, choice constraints and modal characteristics, with a goal of demonstrating the impact of using exact mode travel characteristics in place of zonal skim travel characteristics on simulation outcomes.

Our mode choice model estimation and implementation process starts with the development of a new Time-Dependent Inter-Modal A* routing algorithm (Verbas et al. 2018). The algorithm works on a multi-modal network with transit, walking, and vehicular network links, and finds paths for transit, walking, driving and any feasible combination thereof (e.g. park-and-ride). Turn penalties on the vehicular network and progressive transfer penalties on the transit network are considered for improved realism. Moreover, thresholds to prevent excessive waiting and walking are introduced, as well as a threshold on driving for the park-and-ride mode. The algorithm has been developed and evaluated against a set of observed intermodal travel routes from household travel surveys and on-line router results (Verbas et al. 2018). The route selection in the TDIMA* algorithm is controlled by a set of behavioral parameters and heuristic assumptions which allow it to target a wide variety of intermodal trip-making, including park-and-ride, which are difficult to capture in many routing platforms but are key for representing a significant subset of observed transit, taxi and TNC trips.

The development of the intermodal mode choice model leverages several traditional and emerging data sources in the model estimation and calibration process. The primary source data is a traditional household travel survey conducted by the Chicago MPO, which includes detailed trip observations including the exact starting and ending points for each trip. For each trip in which one or more transit modes is used, additional detail about each transit leg is also provided. This dataset has been used to generate the set of mode choice observations, where the TDIMA* algorithm has been used to generate unobserved mode characteristics and to set availability indicators. Additionally, the ADAPTS planning process model (Auld and Mohammadian 2009) has been used to estimate planning constraints for the data set (i.e. the order in which each observed activity is planned to account for additional constraints imposed by this order). This data has been paired with big datasets that have been collected by the City of Chicago, regional transit agencies, and others that provide information on alternative mode use. The City of Chicago, in particular, has collected extensive information over the previous five years on taxi use, bike-share system use and transit ridership, which serve as both sources of input to define the mode alternatives (i.e. avg. taxi wait times, fares, bike availability at stations, etc.), as well as calibration totals for validating the model (i.e. passenger counts, bike-share usage, taxi rides, etc.).

The database of observed trips and appended modal characteristics for selected and unselected modes will be used to estimate a cross-nested logit model, with the structure shown in Figure 1. The model allows for a complicated nesting structure to capture modal correlations, especially for multi-modal options. We will additionally explore allowing random parameters to represent taste heterogeneity, and evaluate if this is more usefully handled in the CNL specification or within the route selection process.

Figure 1. Cross-nested logit mode choice model structure

Finally, the new mode choice model will be implemented within the POLARIS transportation system simulator, replacing the existing nested-logit mode choice model. The TDIMA* algorithm that has already been implemented in POLARIS will be used to estimate the modal properties within the simulator, in place of the current network skimming process, giving continuous up-to-date information on travel times, costs, transfers, etc. for each specific trip. As this process necessarily entails substantially increasing the calls to the routing routine, which tend to be the most computationally intensive portion of the simulation, great care will be exercised to ensure that the algorithm and mode alternative generation process are implemented efficiently. The simulation process will leverage the heuristics built in to the TDIMA* algorithm to constrain choices and reduce routing calls whenever possible, along with the planning constraints introduced from the POLARIS ABM model.

In this paper, we will demonstrate the usefulness of the TDIMA* routing algorithm, transportation big-data sources and activity planning constraints on accurately capturing mode choice considering alternative mode options like transit park-and-ride, taxi, TNC, etc. We will demonstrate a detailed mode choice model which captures these small share/emerging modes and reduce overfitting to auto modes. We also aim to demonstrate through simulation the benefits of using router in place of network skimmed travel characteristics.

References Auld, J.A., M. Hope, H. Ley, V. Sokolov, B. Xu and K. Zhang (2016). POLARIS: Agent-Based Modeling Framework Development and Implementation for Integrated Travel Demand and Network and Operations Simulations. Transportation Research Part C: Emerging Technologies, 64, 101-116.

Brown, A., J. Gonder, and B, Repac. 2014. “An Analysis of Possible Energy Impacts of Automated Vehicles.” In Road Vehicle Automation, edited by G. Meyer and S. Beiker. 137–153. Switzerland: Springer International Publishing. DOI:10.1007/978-3-319-05990-7_13.

Javanmardi, M., Fasihozaman, M., Shabanpour, R., and Mohammadian, A. 2015. Mode Choice Modelling Using Personalized Travel Time and Cost Data. In proceedings of the 14th International Conference on Travel Behavior Research, Winsdor, England, July 19-23, 2015.

Stephens, T., Gonder, J., Chen, Y., Lin, Z., Liu, C., Gohlke, D., 2016. Estimated Bounds and Important Factors for Fuel Use and Consumer Costs of Connected and automated Vehicles. Technical Report. National Renewable Energy Laboratory, U.S. Department of Energy, Golden, CO. Nov. 2016

Verbas, O. J. Auld, H. Ley, R. Weimer, S. Driscoll (Forthcoming). A Time-Dependent Intermodal A* Algorithm: Methodology and Implementation on a Large-Scale Network. Accepted for presentation at 97th Annual Meeting of the Transportation Research Board, Washington, D.C, January 2018.

ABSTRACT. Please see the attached PDF file.

ABSTRACT. Waiting, whether for services, for someone, or for something, is an inescapable part of life. This paper addresses a gap in the waiting time literature by examining previously sparsely studied relationships between individuals’ characteristics and attitudes toward waiting using a revealed preference dataset of Northern California commuters (N = 2849). Correlational analyses, followed by a trivariate seemingly unrelated regression equations (SURE) model, are developed for three waiting constructs: general tolerance toward waiting, and attitudes toward equipped and expected waiting. Attributes ranging from sociodemographic characteristics and personality traits, to multitasking attitudes, commute preferences, current time use, and general attitudes, are all seen to have significant effects on waiting attitudes. As this survey was executed on commuters, it also facilitates a unique simultaneous exploration of travel and wait time attributes, time uses that are often similarly viewed in day-to-day life. From this perspective, we see that longer commute times and distances are correlated with negative attitudes toward waiting, a relationship that is reversed for commuters with pro-transit, pro-density, and pro-active transportation attitudes. Additionally, we see that those with preferences for multitasking in general or at their jobs are less likely to mind waiting. Overall, this study constitutes an important contribution to the waiting time literature, capitalizing on the rich dataset to make important connections between related time uses, and a multitude of other variables, key among them multitasking, and its potential ability to reduce the negative perception and experience of waiting.

ABSTRACT. see attached pdf

ABSTRACT. Travel behavior modeling aims at describing how individuals travel across space and/or use transportation systems. Different modeling approaches have been developed over the years. Behavioral modeling is concerned with identifying and representing the principles and mechanisms underlying individuals’ choice and decision-making processes. Since decades, the utility-maximization decision rule based on (expected) utility theory (e.g. Von-Neumann and Morgenstern, 1944) has been the major foundation of models of individual choice behavior. It asserts that individuals evaluate choice alternatives on the basis of the utility they derive from the attributes of the choice alternatives, and choose the alternative that maximizes their utility. Although the multinomial logit model can be derived from different theoretical principles, in transportation research random utility theory (e.g. McFadden, 1974) has been the main theoretical underpinning of this discrete choice model. Despite the dominance of utility maximization models in transportation research, their behavioral validity in particular decision contexts has been questioned, and over the years the performance of alternative decision rules has been studied (e.g. Fuji and Gärling, 2003; Arentze and Timmermans, 2007; Zhu and Timmermans, 2010; Chorus et al., 2008; Chorus, 2013; Leong and Hensher, 2014 to name a few). Results have been mixed as might be expected in the sense that individuals may apply different decision rules, and the same individual may use different principles in different decision contexts. Disciplines other than transportation such as psychology, decision sciences and consumer research traditionally have been more open to examine a much wider spectrum of consumer decision theories and associated models. Most theories that have been applied in travel behavior research are based on earlier developments in these disciplines. In this paper, we focus on regret-based models. Already back in 1982, Loomes and Sugden (1982) proposed the concept of regret as the basis for an alternative theory of how people evaluate choice alternatives and arrive at a choice. It helped explaining and predicting violations of conventional expected utility models. Regret is generated when people realize they made the wrong choice in the sense that the alternative they chose is outperformed by one or more foregone or non-chosen alternative(s). Subsequent research in consumer science (e.g. Zeelenberg, 1999; Humphrey, 2004; Connolly and Reb, 2005), economics (e.g. Quiggin, 1990; Simonson 1992; Irons and Hepburn, 2007; Sarver 2008), brain science (e.g. Coricelli et al., 2005; 2007; Chandrasekhar et al., 2008; Chua et al., 2009) provided substantial empirical evidence that the principle of regret minimization better reproduces observed choice data than the principle of utility maximizing behavior in many different choice contexts, providing support for the validity of the concept of regret in particular decision contexts. Regret theory also caught the attention of transportation researchers. Regret-based discrete choice models, so-called random regret minimization (RRM) models, appeared in the spotlight of travel behavior research a decade ago (Chorus, Arentze and Timmermans, 2008; Chorus 2010). These models share with the seminal work on regret the underlying theoretical notions, but differ in terms of the specification of the models and the importance researchers attached to particular methodological principles. Regret-based choice models currently represent an active and dynamic field of research in travel behavior analysis, strongly resembling the early days of discrete choice modeling, with the exploration of many different functional forms for the regret function, controversy about precise theoretical foundations and interpretations, and an increasing number of applications. A critical issue in the early development of regret theory concerned the intransitivity of pairwise choices. This implies that an individual with cyclical preferences can become locked into an endless chain of trades, called money pump (e.g. Loomes and Sugden, 1987). That is, if we assume individuals are willing to pay for their preferred alternative, then the manipulator can infinitely extract payment by adding inferior alternatives to the choice set. In line with this argument, Quiggin (1994) argued that individuals feel regret only against the best foregone alternative in their choice set, called the principle of Irrelevance of Statewise Dominated Alternatives (ISDA). Whereas classic regret theory mostly dealt with gambling experiments (single attribute) under binary choice conditions, travel behavior mostly involves multi-alternative choice sets. Therefore, this issue became more critical in the analysis of travel-related choice behavior. Chorus, Arentze and Timmermans (2008) used this appealing principle to define the amount of regret for the chosen alternative as a function of the attribute difference between the chosen and the best non-chosen alternative in the choice set. Later, Chorus (2010) assumed that regret is defined against all other choice alternatives. Rasouli and Timmermans (2014, 2016) argued that the specification of the regret function in multi-alternative choice sets is an empirical matter, and may depend on the decision context in terms of complexity of the choice set, cognitive effort, distinctiveness of the choice alternatives, long-term consequences of the decision, etc. In some decision contexts, the ISDA principle and therefore the classic model may better represent the amount of regret that individuals experience, whereas in other decision contexts the amount of regret is based on the comparison of two or even all foregone choice alternatives. Hensher et al. (2015) argued that the issue of the best specification of the regret function may be a matter of recognition of the dominant alternative in choice set, depending on relative differences between utilities (attribute values). More specifically, he postulated that if the difference is enough to recognize the dominant alternative, regret is generated according to the ISDA principle. Otherwise, when the difference is insufficient to recognize the dominated alternative, regret would be generated against all non-chosen alternatives. It seems this argument fails to consider the more general issue of discrimination between choice alternatives as a function of the distribution of choice alternatives in attribute space. Moreover, what is “difference enough” for some people may not be for others. The ability to discriminate between choice alternatives heavily depends on people’s sensitivity to attribute values and their mental ability to recognize/process differences. More specifically, either people cannot really distinguish between the values of attributes associated with their chosen alternative and the non-chosen alternatives (mentally-triggered threshold) or they do not mind that difference (indifference threshold). Moreover, it is realistic to imagine that both types of thresholds can be stretched or shrunken depending on the seriousness of the long-term and short-term consequences of the decision. Although Chorus and Van Cranenburgh (2017) also took the stance that the issue is mostly an empirical question, nevertheless, their more recent formulations of regret-rejoice models (e.g. Chorus, 2014; Van Cranenburgh et al., 2015) are all based on the assumption that regret is defined against all non-chosen alternatives. Paucity of literature, which systematically investigates the relevance of various regret specifications as a function of complexity of the choice task and the decision context, motivated us to set out the current research project. We are particularly interested in understanding how people perceive differences in attribute values and consequently incorporate their perceptions in assessments of regret. Do indifference thresholds exist and do they differ across individuals? Could the decision context be a possible covariate determining the threshold? Does the number of alternatives in the choice set, ceteris paribus, influence the way people assess regret and make decisions? How does attribute range play role in the specification of regret? To that end, we designed a dedicated choice experiment in which we systematically changed the number of alternatives, attribute range and attribute levels for two different decision contexts (with a potentially different period for the long-term effect of regret). More specifically, we designed two different choice contexts consisting of three attributes using an orthogonal design: i) route choice described by free flow time, congestion time, and variability of travel time. ii) residential (housing) choice described by rental price, travel time to go to work, and travel time to go to shopping. Each context consists of seven different scenarios. On the basis of the base scenario with three alternatives, intermediate attribute range, and four attribute levels, we defined three types of complexity: i) number of alternatives - five alternatives and seven alternatives. ii) attribute range – narrower range and wider range. iii) attribute levels – smaller attribute levels and larger attribute levels. Each scenario is composed of eight choice situations. Consequently, each respondent was asked one hundred and twelve questions (eight choice situations * seven scenarios * two choice contexts). To reduce burden, the survey was administered across two days. To decrease bias, we counterbalanced the order of the survey: i) While half of respondents was asked to complete the route choice tasks on the first day, and the residential choice on the second day, the remaining respondents were asked to complete the residential choice tasks on the first day and the route choice tasks on the second day. ii) Four sequences of scenarios were prepared for each choice context, which were randomly assigned to respondents. A total of 418 participants completed the experiment. The data collection has been completed. We plan to analyze the effect of complexity and attribution variation on the performance of the regret-based choice model. The results will provide empirical evidence about the relative performance of alternative specifications of regret functions in multi-alternative choice sets. Furthermore, we will propose a generalized definition of regret in multi-alternative choice sets including both current definitions.

References

Arentze, T. A., and Timmermans, H. J. P. (2007) Parametric actions trees: incorporating continuous attribute variables into rule-based models of discrete choice. Transportation Research Part B, 41(7), pp. 772–783.

Chandrasekhar, P. V. S., Capra, C. M., Moore, S., Noussair, C., and Berns, G. S. (2008) Neurobiological regret and rejoice functions for aversive outcomes. Neuroimage, 39(3), pp. 1472–1484.

Chorus, C. G., Arentze, T. A., and Timmermans, H. J. P. (2008) A random regret-minimization model of travel choice. Transportation Research Part B, 42, pp. 1–18.

Chorus, C. G. (2010) A new model of random regret minimization. The European Journal of Transport and Infrastructure Research, 10(2), pp. 181–196.

Chorus, C. G. (2013) Capturing alternative decision rules in travel choice models: a critical discussion. In. Hess, S., and Daly, A. (ed) Handbook of choice modelling. pp. 290–310. Edward Elgar Publishing, UK.

Chorus, C.G. (2014) A generalized random regret minimization model. Transportation Research Part B, 68, pp. 224–238.

Chorus, C. G., and van Cranenburgh, S. (2017) Specification of regret-based models of choice behaviour: Formal analyses and experimental design based evidence – Commentary. Transportation. doi:10.1007/s11116-016-9739-x.

Chua, H. F., Gonzalez, R., Taylor, S. F., Welsh, R. C., and Liberzon, I. (2009) Decision-related loss: Regret and disappointment. Neuroimage, 47(4), pp. 2031–2040.

Connolly, T., and Reb, J. (2005) Regret in cancer-related decisions. Health Psychology, 24(4), pp. 29–34.

Coricelli, G., Critchley, H. D., Joffily, M., O’Doherty, J. P., Sirigu, A., and Dolan, R. J. (2005) Regret and its avoidance: A neuroimaging study of choice behaviour. Nature Neuroscience, 8(9), pp. 1255–1262.

Coricelli, G., Dolan, R. J., and Sirigu, A. (2007) Brain, emotion and decision making: the paradigmatic example of regret. Trends in Cognitive Sciences, 11(6), pp. 258–265.

Fujii, S., and Gärling, T. (2003) Application of attitude theory for improved predictive accuracy of stated preference methods in travel demand analysis. Transportation Research Part A, 37(4), pp. 389–402.

Heiner, R. (1983) The origin of predictable behavior. The American Economic Review, 73, pp. 560-595.

Hensher, D. A., Rose, J. M., and Green, W. H. (2015) Applied Choice Analysis. 2nd ed. Cambridge University Press, UK.

Humphrey, S. J. (2004) Feedback-conditional regret theory and testing regret-aversion in risky choice. Journal of Economic Psychology, 25(6), pp. 839–857.

Irons, B., and Hepburn, C. (2007) Regret theory and the tyranny of choice. Economic Record, 83(261), pp. 191–203.

Loomes, G., and Sugden, R. (1982) Regret-theory: An alternative theory of rational choice under uncertainty. The economic journal, 92(368), pp. 805–824.

Loomes, G., and Sugden, R. (1987) Some implications of a more general form of regret theory. Journal of Economic Theory, 41, pp. 270-287.

McFadden, D. (1974) Conditional logit analysis of qualitative choice behaviour. In: Zarembka P. (ed) Frontiers in Econometrics. Academic Press, NY.

Quiggin, J. (1990) Stochastic dominance in regret theory. The Review of Economic Studies, 57(3), pp. 503–511.

Quiggin, J. (1994) Regret theory with general choice sets. Journal of Risk and Uncertainty. 8, pp. 153–165.

Rasouli, S., and Timmermans, H. J. P. (2014) Regret-based models of consumer choice of shopping center. In: Proceedings 21th RARCS Conference, Bucharest, Romania.

Rasouli, S., and Timmermans, H. J. P. (2016) Specification of regret-based models of choice behavior: Formal analyses and experimental design based evidence, Transportation. doi:10.1007/s11116-016-9714-6.

Sarver, T. (2008) Anticipating regret: Why fewer options may be better. Econometrica, 76(2), pp. 263–305.

Simonson, I. (1992) Choice in context: Tradeoff contrast and extremeness aversion. Journal of Marketing Research, 29 (3), pp. 281-295.

Van-Cranenburgh, S., Guevara, C. A., and Chorus, C. G. (2015) New Insights on Random Regret Minimization Models. Transportation Research Part A, 74, pp. 91–109.

Von-Neumann, J., and Morganstern, O. (1944) Theory of games and economic behavior. Princeton University Press, Princeton, NJ.

Zeelenberg, M. (1999) Anticipated regret, expected feedback and behavioral decision making. Journal of Behavioral Decision Making, 12, pp. 93–106.

Zhu, W., and Timmermans, H. J. P. (2010) Cognitive process model of individual choice behavior incorporating principles of bounded rationality and heterogeneous decision rules. Transportation Research Part B, 37(1), pp. 59–74.

ABSTRACT. Service reliability has an important impact in the satisfaction of public transport users. Its main source is found in headway variance which not only affects waiting time, but also distributes passengers unevenly across vehicles. However, even nowadays it is unclear how headway irregularity, with its impact in waiting, crowdedness and reliability, affect travellers’ service satisfaction. Different stated preference studies have tried to identify non-linear impacts produced by overcrowding, however, none of them is directly related with users’ satisfaction evaluation.

In this study, we try to demonstrate this non-linearity in users’ satisfaction caused by both the crowding level and the uncertainty about expected waiting time. A simultaneous Latent Class Ordered Logit Model was calibrated. Overall, there is a significant and negative impact for bus, meaning that users have a more negative perception of the level of service experienced inside a bus. Also, the impact of missing a vehicle is larger for metro while the impact of traveling seated is larger for bus users. Regarding the latent class model, woman under 35 years old are more probable to be sensitive to crowding.

The relationship presented between headway reliability and traveller’s satisfaction could lead to a change in the perspective public transport systems are planned and operated. Irregular headways generate heterogeneity in vehicles’ level of service, meaning that more travellers experience worst vehicles, getting lower than expected average satisfaction indexes when considering users’ point of view. When the focus is placed on traveller’s experience, reliability should play a role equally important as speed has today.

ABSTRACT. Background Public transport satisfaction surveys usually investigate customers’ satisfaction and the perceived performance of public transport (Abenoza et al., 2017; Efthimiou and Antoniou, 2017; Morton et al., 2016; Ettema et al., 2012) and then they are compared to the objective operational data. However, there are several arguments on the importance of linking the perceived and the objective data to maximize the outcomes of satisfaction and performance surveys (Carrel et al., 2016; Bordagaray et al., 2014; Tirachini et al., 2013; Friman and Fellesson, 2009). So far, there are only few surveys that take into account both data types to assess customers’ satisfaction.

Another characteristic of most of the available satisfaction surveys is that they typically use cross-sectional data to examine the factors affecting public transport customers’ satisfaction and infer the magnitude of their effects (Abenoza et al., 2017; Efthimiou and Antoniou, 2017; Morton et al., 2016; Ettema et al., 2012; Ravulaparthy et al., 2013; Trompet et al., 2013; Pedersen et al., 2011; Duarte et al., 2010; Ji and Gao, 2010). Satisfaction data is usually collected only at one point of time asking the public transport users to rate their last trip, or their last commute trip, or their satisfaction in general with the public transport services (Morfoulaki et al., 2007). When individuals are repeatedly involved in an event/activity, they tend to remember only the most intense pleasant and unpleasant moments overall and the most recent events (Kahneman et al., 1993). Lately, there are several studies that use text-mining techniques to derive information about public transport users sentiments via social media to assess customer satisfaction (Collins et al., 2013; Maghrebi et al., 2015). Although, this approach can offer spatio-temporal and longitudinal data, it has several limitations, such as the lack of information about the socio-demographic characteristics of the users, or the bias of the collected data as social media users tend to post their negative feelings about transport and not their positive thoughts (Collins et al., 2013). Against this background, there is a necessity to investigate public transport customers’ satisfaction using both perceived and objective data, but also longitudinal data to derive more and better information about the factors affecting satisfaction. The smartphone based travel survey tools have enabled the collection of longitudinal travel data, while the increasing availability of Open public transport data allows the acquisition of objective operational data. The aim of this study is twofold: 1. to describe the procedure of matching longitudinal smartphone based travel survey data with open operational data, and 2. to quantify the effect of both types of data on customers satisfaction with using rail-based public transport modes. Survey design & Data The data used in this study originate from the London Mobility Survey (LMS), which has been designed by the MaaSLab at University College London (UCL). LMS has been developed using a smartphone based travel survey tool, the Future Mobility Sensing (Cottrill et al., 2013; Zhao et al., 2015). LMS incorporates several parts of the London Travel Demand Survey (the official travel survey of London that takes place every year ) to allow for comparisons, while it has been enhanced with additional detailed questions about usage of new mobility services (for a detailed description see Matyas and Kamargianni, 2017a). LMS consists of 3 steps: • Step 1: The participants create an account and answer to the pre-questionnaire that includes questions about their socio-demographic and mobility tool ownership characteristics along with their attitudes towards private vehicle ownership and shared mobility. • Step 2: after the completion of Step 1, participants are asked to download the FMS app. As a prompted recall survey, respondents need to log back in online to validate their travel and non-travel activities and answer some additional questions for each of their activities. • Step 3: The exit section; when the 7-day tracking and validation are complete, respondents are presented with stated preference experiments about mobility-as-a-service monthly plans.

In the Step 2 of the survey, users are presented with their activity diary on the online platform, where they need to verify their trips. They are also asked to provide more information for each activity. Since the focus of this paper is on public transport and satisfaction, we will only focus on the questions that follow when the participants conduct a trip with a TfL rail-based mode (these are: tube, overground, TfL rail, DLR and tram). These questions are: i. How many people travelled with you? (options: 0 to 5+), ii. How did you pay for this trip? (options: Oyster card-Pay as you go, Contactless, Smartphone pay, Travel pass, Other), iii. Which line did you travel with? (options: the 15 rail-based lines, other), iv. Were you doing any of the following on your journey? (options: Using mobile, Working, Listening to music, Watching movies, Playing games on mobile device, Reading adverts/posters, Daydreaming, Nothing, None of the above-could choose more than one), and v. How happy were you by using this mode? (7-point Likert scale, with a frown and a smiling face on the extremes). In addition, a number of extra data from Open sources is linked to each recorded and validated trip to decrease the respondent burden during survey processes, while improving the accuracy, quality and amount of the collected data (Zhao et al., 2015). In the context of this study we focus on linking TfL data concerning the status of each rail-based public transport mode line with an ultimate purpose to investigate its effect on users satisfaction with using these modes. The status of a TfL line describes how the service is running at a given moment including information about possible disruptions. The request we use for our purposes receives as parameters a comma-separated list of the TfL rail-based modes and returns a status severity code for each PT line, which corresponds to a short description of the service status. By repeatedly making such requests in 10-minute intervals we collect the TfL status data into a MongoDB database. To match this open data to the LMS data we use KD trees machine learning method (Bentley, 1995). The data utilized in this paper originate from the smartphone-based London Mobility Survey (LMS) and was collected between November 2016 to February 2017. 157 individuals out of the 252 have used public transport modes and these individuals constitute the sample for this paper. These 157 individuals have conducted 1,323 stages by TfL rail-based public transport modes. The minimum number of validated days is 1, the maximum is 9, while the average number of stages per individual is 8.4. Public transport customer satisfaction model The ultimate goal of this study is to explore and model the magnitude of the factors that affect the satisfaction of the public transport users using both travel survey and open data. The dependent variable is “How happy were you by using this mode (for this specific stage)?”. Participants were requested to indicate their level of satisfaction to a 7-point Likert scale. Since the dependent variable is ordinal, the ordered logit model would be appropriate. The ordered logit model is a regression model for an ordinal response variable. The model is based on the cumulative probabilities of the response variable. In particular, the logit of each cumulative probability is assumed to be a linear function of the covariates with regression coefficients constant across response categories (Abdel-Aty, 2001). A regression model would not be appropriate as it assumes differences between categories of the dependent variable to be equal, whereas, the data are ordinal.

Due to the fact that our data has been collected over time and over the same individuals, an additional mixing coefficient σ is incorporated in our model to account for correlation across the responses given by a single individual i (panel effect). The model has been developed in SPSS 22 using also a tailored script written in Python. Preliminary Results Our results indicate that customer satisfaction is indeed associated with the open public transport status data. Activities while travelling and trip purpose also affect customers satisfactions, while these results provide insights for offering products that can advance customers experience in the MaaS and automated vehicles era that lies ahead. For example, listening to music, paying games and watching movies while travelling positively affects customers satisfactions. These are services that in the future could be included in MaaS subscription packages. In addition, these findings support the hypotheses that travel time could have a positive utility as it can be used productively for other purposes, such as working.

By comparing our results to other surveys, we identified both similarities and differences allowing us to conclude that the factors affecting customer satisfaction vary across cities as the cultural environments are different (and of course the samples). As such, it is probably not wise to transfer customer satisfaction survey results from one city to the other, and it is better each public transport authority or company to have each one customer satisfaction survey to manage to attract more customers or retain the existing.

One of the most worth noting findings is that customer satisfaction varies from trip-stage to trip-stage as each trip-stage has each one conditions and characteristics. When satisfaction is aggregated into overall satisfaction with a specific transport mode, significant information is missed hindering transportation planning. Given the rise of new mobility services, and especially ridehailing services, public transport authorities and operators should update the satisfaction data acquisition and evaluation processes to acquire better information about their most valuable asset, their customers.

ABSTRACT. Customer satisfaction is a well-known concept in private-sector business development and service industry research. Use of customer satisfaction measures allows audiences from a wide range of academic backgrounds to easily understand obtained data. Expanding such research from product-based industries and modifying it to fit the context of public transportation services could provide a valuable understanding of the use of customer satisfaction data as a decision-making tool. Additionally, level of customer satisfaction regarding the performance of public transportation services is an aspect that can influence consumers’ transportation choices. Customers who are satisfied with a particular mode of public transportation will likely continue to use it and promote its use in the market. Meanwhile, customers who are dissatisfied are unlikely to be repeat customers, and will seek a superior mode of public transportation on subsequent trips. Therefore, it is crucial to improve the performance of public transportation services. However, bringing about improved performance and increased customer satisfaction requires an understanding of the travel behavior of public transportation users.

The objectives of this research were to investigate (a) customer satisfaction regarding the performance of public transportation, and (b) customer behavior when choosing public transportation modes. To describe the relationship between customer satisfaction and mode choice, we performed multiple analyses. First, we explored the customer satisfaction of public transportation passengers. Second, we examined the passengers’ behavior of choosing a transportation mode. Finally, we applied an integrated choice and latent variable model. Based on the datasets from online questionnaires including a customer satisfaction survey and choice model survey, the relationship between the two was analyzed.

We applied this model to the Jakarta metropolitan area as a case study. Jakarta, the capital city of Indonesia, is in the process of developing its public transportation systems. This development has brought about changes in the share of passengers using each mode. According to the Final Report of Project for the Study on JABODETABEK (Jakarta, Bogor, Depok, Tangerang, Bekasi) Public Transportation Policy Implementation Strategy by JICA in 2012, urban transportation in the Jakarta area heavily relies on road-based systems with public transportation comprising just 27% of overall traffic. If no action is taken to increase the usage of public transportation, the traffic situation will worsen, causing widespread congestion on public roads. It is therefore an unavoidable conclusion that the number of public transportation passengers must be increased. Analyses of customer satisfaction will aid in improving public transportation performance to increase the share of travelers who choose public transportation. Furthermore, a better understanding of passengers’ travel behavior will contribute to setting policies aimed at an optimal distribution of public transportation resources in the Jakarta metropolitan area.

Three public transportation modes were analyzed in this research. The first was Transjakarta Bus Rapid Transit. Since opening in 2004, Transjakarta’s aim has been to become one of Jakarta’s highest-quality public transportation services in order to ease the area’s traffic problems. As the longest bus rapid transit system in the world, Transjakarta continues to improve their performance. In 2010, the Indonesian Consumers Protection Foundation (YLKI) administered a customer satisfaction survey to 3,000 of Tansjakarta’s passengers. Responses indicated that approximately 50% of Transjakarta’s passengers felt satisfied, 39% rated the services as fair, 8% felt dissatisfied, 2% felt highly satisfied, and less than 1% felt highly dissatisfied. These data indicate that most customers of Transjakarta feel satisfied despite heavy congestion on most transportation systems. For this reason, this study sought to investigate the impact of customer satisfaction with public transportation performance on travel behavior.

The second transportation mode examined was the kopaja minibus system. Established in 1971, kopaja are a major component of public transportation in the Jakarta metropolitan area because of their cheap fares and ability to stop anywhere along their route according to passengers’ wishes. Since 2012, some of the kopaja system has been integrated with Transjakarta BRT, and the two services now use the same bus lanes. While changes in the kopaja service have included increased fares, they have also increased its convenience.

The angkot (mikrolet) microbus system was the third mode examined. Angkot service was established in Jakarta in the late 1970, and each vehicle has a capacity of 10 seats. Angkot is an incredibly cheap and easy mode of public transportation in Jakarta. An angkot usually has a predetermined route, but shares kopaja’s ability to stop anywhere along the route according to passengers’ wishes.

This research employed datasets that were collected via online questionnaires. Screening questions were asked to filter out the respondents. The customer satisfaction survey consisted of 26 questions, each of which was scored on an 8-point Likert scale. Respondents were asked to rate the following: overall satisfaction, operating time, headway (operating frequency), punctuality, route coverage, information system (vehicles and stops), fare payment system, compatibility of fare and distance, security (vehicles and stops), cleanliness (vehicles and stops), comfort (vehicles and stops), driver's level of compliance with traffic regulations, travel speed, accessibility of stops, ease of transfers, ease of reaching staff, staff readiness, safety information, physical condition of vehicles, crowdedness inside vehicles, accessibility for disabled passengers, and environmental friendliness of vehicles. For the mode choice analysis, a survey consisting of questions related to access and egress was conducted. The mode choice survey was restricted to the same three modes of public transportation as the customer satisfaction survey (i.e., Transjakarta, angkot, and kopaja). This allowed us to analyze the effect of level of performance of the three modes of public transportation on travel behavior.

The results show that customers’ perceived value regarding each mode’s level of performance affects both customer satisfaction and choice of travel mode. Transjakarta had the highest customer satisfaction level, and was most likely to serve as a respondent’s primary mode of transportation. There was no significant difference between the kopaja minibus system and angkot microbus system with respect to customer satisfaction. Because those modes are likely used as a feeder to reach other public transportation modes, such as Transjakarta or commuter trains, respondents tended not to care about their service performance. Perceived value error also affects customer satisfaction. Even though objective data on Transjakarta, such as punctuality data, do not indicate high performance, the respondents still felt satisfied with it. This tendency suggests that passengers have become resigned to this attribute, and have no choice of alternative modes of transportation.

ABSTRACT. Bogotá is witnessing a steady decline in its Bus Rapid Transit (BRT) system’s user satisfaction. On a scale of 5.0, general satisfaction with the system, known as TransMilenio, has fallen from 4.60 in 2001 to 2.62 in 2016. This paper analyzes what is happening to TransMilenio user satisfaction by defining the “Value Concept” within the principles of the “Lean Thinking” approach. Qualitative research techniques were used at the point of ticket sales to identify factors that are important to users. Quantitative data was collected through an in-person survey of a representative random sample of TransMilenio riders. This data was used in a linear regression model to gain meaningful insights into value for Transmilenio users in terms of what they want to receive from the system. The results show that six factors explain user satisfaction. From these results, this investigation infers that the main efforts of TransMilenio may not be in line with riders’ needs. Our findings can help TransMilenio understand its shortcomings and shed light on how to prevent service declines in other BRT systems modeled on that of Bogotá.

ABSTRACT. *** Extended abstract is attached ***

Several surveys have been conducted all over the world to investigate the most important factors that influence public transport users’ satisfaction. In this research, we present the results of a public transport users’ satisfaction survey in Munich, which includes bus and metro (U-Bahn) service, as well as a number of exploratory and model-based analyses on the data, aiming at extracting the most influential factors in case of Munich. Weather clearly influences transport, and public transport is often more vulnerable to it. Munich has a very advanced public transport system, with a high ridership, and also rather adverse weather conditions. Therefore, it is considered a rather suitable case study for this type of analyses. Indeed, while there is extensive research on the investigation between the link of weather and other modes of transport (primarily road traffic), the case of weather and public transport has been less investigated. In this research, we aim to fill this gap, using a questionnaire-based data collection, and a number of analyses to investigate the relationship between weather and public transport users’ behavior.

ABSTRACT. In this paper we present a multilevel analysis of opportunity-based accessibility indicators for different ethnic groups in the greater Los Angeles region. The accessibility indicators capture congestion effects and reflect the jobs and services available to residents. The multilevel structure of the models capture the spatial dependency among the units of analysis (US Census block groups and US Census Tracts). This is the first step of an analysis that includes other indicators of quality of life and travel behavior.

ABSTRACT. Introduction Modelling of residential location choice is a key component of integrated land use and transport planning. It is a key determinant for urban sprawl and substantially affects travel behaviour. For example, households living far from facilities like employment, shopping and healthcare are more likely to produce long commute and non-commute trips and depend more on car travel (Næss 2009). On the other hand, households living in compact areas (i.e. close to city centre) with high public transport accessibility are likely to travel less and be more dependent on public transport (Farber and Li 2013). Thus, a misspecification in residential location choice models (either in choice set formation or estimation stage) may lead to bias in the estimation of model parameters and can potentially lead to inappropriate land use and transport policy formulation.

In most large-scale residential location choice models based on revealed preference data, the true choice set of an individual is unknown to the researchers. A review of the literature reveals that, in many of the residential location choice models, either the full choice set (Bhat and Guo 2004) or a random subset (Lee and Waddell 2010) has been considered for each respondent. Both are behaviourally unrealistic. In practice, households are neither aware of the full choice set of alternatives nor consider all alternatives they are aware of. Different households might thus have different consideration sets based on household preferences, sociodemographic characteristics and their knowledge of available alternatives. Therefore, it is expected that better ways to model the choice set will not only make the models computationally easier, but also, behaviorally more representative.

Several candidate approaches for choice set generation exist. A popular approach has been elimination by aspects which is based on the assumption that households use non-compensatory decision rules for screening of alternatives based on some behavioural constraints (Scott 2006). Constraints are applied on attributes level to eliminate alternatives. For instance, a household sensitive to commute distance does not consider residential location alternatives outside a threshold distance from their workplace. As researchers do not know the true preference of the decision maker, they may make a wrong assumpting inducing a high risk of eliminating alternatives that have a non zero probability, resulting in biased parameter estimation. Zolfaghari (2013) compared analytically the performance of alternative constraint-based exogenous choice set formation approaches and found poor performance compared to random sampling in both estimation and validation sample.

Few studied attempted to model the choice set explicitly in a two-stage framework where the first stage is for choice set generation for each individual and the second stage is for calculating choice probability conditional on choice set (Manski 1977; Swait and Ben-Akiva 1987). In this two-stage framework, the first stage is considered as a non-compensatory decision process and the second stage as a compensatory decision process. This technique is computationally infeasible for a large universal choice set because the number of possible choice sets explodes with the number of alternatives. For instance, the size of the universal choice set in this study is 498, the total number of the possible choice set is 2498-1.

Few single stage semi-compensatory approaches have been proposed recently as an alternative to the two-stage approach where the choice set of an individual is modelled implicitly as a function utility penalization of less attractive alternatives. Popular semi compensatory approaches are Implicit Availability Method (Cascetta and Papola 2001), Constrained Multinomial Logit Method CMNL (Castro, Martínez and Munizaga 2013), and rth order Constrained Multinomial Logit Method rCMNL (Paleti 2015).

In spite of a significant body of research on choice set generations, to the best of our knowledge, none of the previous research has compared the performance of compensatory, non-compensatory and semi-compensatory approaches of choice set construction in a systematic manner. Furthermore, it has not been tested if the same choice set generation scores better for long (ownership) and medium (rental) choice contexts. We aim to fill these two research gaps in this paper.

The specific research objectives are therefore as follows: • To compare the performance of state-of-the-art compensatory, semi-compensatory and non-compensatory choice set construction techniques in the context of large-scale residential location choice modelling. • To investigate the existence of underlying heterogeneity in the choice set formation between two major housing submarkets (owners and renters).

Data Our study used the London Household Survey Data (LDSD), the Ward Atlas Data (WAD) and an Origin-Destination (OD) matrix from the London Transport Studies (LTS) model as main data sources.

The LHSD consists of 8,158 households information from greater London area (GLA) where 4,491 households live in their owned properties, 3,576 households live in rented properties and 91 households live in shared accommodation. This research focused on households having at least one commute member and used the information 2,180 owners and 1,293 renters. Dataset consists of household socio-demographic characteristics (household size, income, etc), dwelling information (tenure type, size, price/rent, etc.), employment status, home and work location, car ownership, etc. Zone level aggregated demographic, land use and other information (land use pattern, population density, household composition, ethnic proportion, employment and economic activity, household income, crime rates, land use, public transport accessibility, green space, car use) have been extracted from WAD. The origin-destination matrix of GLA from the London Transport Studies (LTS) model has been used to extract the commute distances between the reported residential and work locations in the LHSD.

From the descriptive statistics of the data, it has been observed that there is a substantial difference between owners and renters sociodemographic characteristics, dwelling and location preference and travel behaviour. For example, the average income of owners is higher than renters. Due to the high rate of car ownership, owners are more dependent on car travel and commuting more compared to renters.

Methodology Zone level residential location choice models have been estimated where each ward in the GLA is considered as an alternative zone. Due to the heterogeneity in sociodemographic and preferences of owners and renters, separate models have been estimated for this two submarket to see the potential existence of heterogeneous preference in consideration set and estimation results. Parameters considered in the models are location characteristics (land use mix, land use type, employment opportunity, ethnic diversity, crime rate, school quality, public transport accessibility), dwelling attributes (dwelling density, dwelling type), distance from past home, distance from central business district (CBD) and commute distances.

The data of each submarket has been divided into five subsets where four subsets have been considered for estimation and the rest for validation and all possible combinations of the data sets have been tested. The choice set generation techniques tested in this research are divided into three categories 1. Compensatory approach a. Full choice set (size of individual choice set is 498) 2. Non-Compensatory approaches: a. Random elimination (size of individual choice set is 25) - Uniform random sampling b. Elimination by aspect (size of individual choice set is 25) - Fully constraint based approach (alternatives are chosen from threshold zone) - Partially constraint based approach (80% are chosen from threshold zone and rest from outside the zone) 3. Semi-compensatory approached (size of individual choice set is 498) a. Implicit availability based on dominance rule b. Constrained Multinomial Logit Approach (CMNL) c. rth order Constrained Multinomial Logit Approach (rCMNL)

Performance of the different choice set generation techniques has been evaluated based on the model goodness of fit and predictive availability in hold-out sample validation. Instruments used to evaluate the predictive ability in validation sample are average probability of chosen alternatives, observations having highest average probability of chosen alternative, percentage of outliers, root mean square error (RMSE), mean absolute deviation between predicted and actual share (DPAS), loss of likelihood and adjusted rho-square

Results Preliminary Estimation Results Distance from past home and commute distance have been tested as an instrumental variable to capture the true choice set. Either deterministic constraint or dominance rule has been applied in both parameters but the rule of commute distance for choice set construction has been found insignificant. Models of both owners and renters using different choice set generation techniques have given consistent parameters estimates. Semi-compensatory approaches (CMNL and rCMNL) have given significant improvement in the loglikelihood over the all other models in both the ownership and renting model. CMNL and rCMNL have given almost same performance because the improvement of rCMNL over CMNL decreases with the increase of the number of alternatives in the universal choice set. Among all the methods, elimination by aspect based approaches perform worse with a big drop in log likelihood.

In terms of parameter estimates, the results are intuitive. In both ownership and renting models, the estimated parameters have shown similar direction of sensitivities. Housing cost, household size, dwelling density, the proportion of commercial activities, crime rates, distance from past home and commute distance have negative signs. On the other hand, households are inclined to choose areas with high public transport accessibility, high proportion of residential activities, mixed land use pattern, good school quality, proximity to CBD. Due to the difference in the time scale of ownership and renting decision (ownership is considered as long-term decision due to high investment and relocation cost whereas renting is considered as medium-term decision due to lower relocation cost, shorter tenures and duration of agreements), owners and renters might have different level of preference on attributes. Thus, estimated coefficients of five parameters have been found significantly different at 95% confidence interval in ownership and renting model: commute distance, distance from CBD, school quality, the percentage of residential land-use (in inner London) and preference for ethnic similarity among the white ethnic respondents.

Model Validation and Comparison Compensatory and semi-compensatory approaches have been compared first and found that CMNL and rCMNL are the superior approaches in terms of all of the validation measures used in this research. Among the non-compensatory approaches, most of the measures except RMSE have shown better performance of random sampling over deterministic constraint based approaches, though partial constraint based approach has shown superiority over full constraint based approach. We cannot compare directly non-compensatory approaches with compensatory and semi compensatory approaches because the different size of choice set has been used in different approaches. To compare all the methods used in this research, models estimated using semi-compensatory approaches have been re-estimated with a random subset. Then final validation results, have also clearly shown better performance of CMNL and rCMNL over all other methods.

Conclusion In this research, separate models have been estimated for residential ownership and renting and a set of available choice set formation techniques have been tested. One of the appealing points of this research is the evaluation of predictive performance models estimated using different choice set formation techniques in a holdout validation sample. The results of estimation and hold-out sample validation of both ownership and renting models suggest that semi-compensatory approach like CMNL and rCMNL perform best and are capable of capturing the households’ underlying preference of consideration set whereas elimination by aspect based methods perform worse. No significant difference has been observed between ownership and renting in terms of capturing the consideration set. Testing the methods in the same dataset (though tested on two different contexts: long and medium term) is one of the limitations of this research. However, this comparison gives us a useful insight for estimating a model with a large choice set where the two-stage probabilistic approach is infeasible. In addition to residential location choice, the research can also provide a guideline for modelling other travel behaviour models with large choice sets like route choice, destination choice, activity choice etc.

References BHAT, C. R. and J. GUO. 2004. A mixed spatially correlated logit model: formulation and application to residential choice modeling. Transportation Research Part B: Methodological, 38(2), pp.147-168. CASCETTA, E. and A. PAPOLA. 2001. Random utility models with implicit availability/perception of choice alternatives for the simulation of travel demand. Transportation Research Part C: Emerging Technologies, 9(4), pp.249-263. CASTRO, M., F. MARTÍNEZ and M. A. MUNIZAGA. 2013. Estimation of a constrained multinomial logit model. Transportation, 40(3), pp.563-581. FARBER, S. and X. LI. 2013. Urban sprawl and social interaction potential: an empirical analysis of large metropolitan regions in the United States. Journal of Transport Geography, 31, pp.267-277. LEE, B. H. and P. WADDELL. 2010. Residential mobility and location choice: a nested logit model with sampling of alternatives. Transportation, 37(4), pp.587-601. MANSKI, C. F. 1977. The structure of random utility models. Theory and decision, 8(3), pp.229-254. NÆSS, P. 2009. Residential self‐selection and appropriate control variables in land use: Travel studies. Transport Reviews, 29(3), pp.293-324. PALETI, R. 2015. Implicit choice set generation in discrete choice models: Application to household auto ownership decisions. Transportation Research Part B: Methodological, 80, pp.132-149. SCOTT, D. M. 2006. Constrained Destination Choice Set Generation: Comparison of GIS-Based Approaches. In: Transportation Research Board 85th Annual Meeting. SWAIT, J. and M. BEN-AKIVA. 1987. Incorporating random constraints in discrete models of choice set generation. Transportation Research Part B: Methodological, 21(2), pp.91-102. ZOLFAGHARI, A. 2013. Methodological and empirical challenges in modelling residential location choices.

ABSTRACT. Please see the attached pdf

ABSTRACT. The integration of discrete choice models in mixed integer linear programming (MILP) is appealing to operators and policy makers (the supply) because it provides a better understanding of the preferences of customers (the demand) while planning for their systems. These preferences are formalized with discrete choice models, which are the state-of-the-art for the mathematical modeling of demand, whereas MILP models are considered to design and configure the systems. However, the complexity of discrete choice models leads to mathematical formulations that are highly nonlinear and nonconvex, and makes them difficult to be included in MILP. In this research, we present a general framework that overcomes these limitations by relying on simulation in order to integrate advanced discrete choice models in MILP. A concrete application on benefit maximization from an operator selling services to a market is used to illustrate the employment of the framework.

ABSTRACT. A fundamental assumption for modeling discrete choices is that the researcher knows the set of alternatives considered by the decision maker. This assumption is easily questionable, especially when the universal set is large, as in route choices. Manski (1977) offered a theoretical solution for this problem, which practical implementation requires, however, additional ad-hoc assumptions, e.g. that the consideration set is built considering the best k-routes in terms of a generalized cost approach. Violation of these ad-hoc assumptions inevitably results in inconsistent estimators and sever prediction errors.

This research makes three contributions. First, using Monte Carlo simulation, we explore the robustness of several practical methods used to build the consideration set. Second, we analyze three methods to collect data on the consideration set: passive data, online surveys and map-based surveys. Finally, we develop a stated preferences (SP) experiment that emulates the process of generating the consideration set, with which we study the characteristics of the set and assess the robustness of methods to infer it.

Monte Carlo analysis showed that the widespread k-routes method can generate large biases, while a method based on historical choices may be more robust. Regarding data collection, preliminary evidence suggests that passive data is promising but still has considerable limitations on processing and measurement errors. Data obtained from surveys and maps showed to be feasible to study, e.g. the size of the consideration set, the heuristics used to build it, to model the generation process and to identify the variables that impact it. Finally, the SP model confirmed that widespread k-routes method can generate large biases, even resulting in parameters with incongruent signs.

ABSTRACT. For over a century, road pricing has been advocated by transport economists. It is regarded as the best way to manage congestion problems when infrastructure expansion is impossible or unprofitable. Despite the longstanding theoretical argument, only a few schemes have been implemented and most attempts were unsuccessful. The lack of the necessary public and political backing is often mentioned as the main reason for the failures (Vonk Noordegraaf, Annema, & van Wee, 2014). Because of the longstanding issues around implementation of road pricing and the increase of car related problem such as congestion and emissions research has been done on the acceptability of road pricing (see an overview in (Schade & Schlag, 2003). Especially the opinions of road users have extensively been researched. Many of these studies investigated aspects of a road pricing scheme that affect the acceptability such as ‘revenue allocation’ (Schuitema & Steg, 2008; Welch & Mishra, 2014) or ‘perceived fairness’ (Di Ciommo & Lucas, 2014; Eliasson & Mattsson, 2006). Also variables related to beliefs and attitudes such as ‘infringement on freedom’, ‘problem perception’ (Bamberg & Rolle, 2003) and the interaction with ‘social and personal norms’ have been researched (Sun, Feng, & Lu, 2016). These studies usually use quantitative techniques, seeking for the variables that predict the acceptability of certain road pricing schemes. This approach has revealed a wide range of variables empirically related to acceptability of road users regarding road pricing.

However, the implementation of a road pricing scheme does not only concern road users but it would affect the wider public. This wider public is very heterogeneous in worldviews, preferences, values and beliefs. We hypothesize that part of this heterogeneity is caused by the fact that different persons adopt different ‘frames’ when they evaluate a road pricing instrument. We define frame as: “a discourse that operates at the individual level, hence a coherent set of beliefs and attitudes that people use to observe and give meaning to reality” (Kroesen & Bröer, 2009). These frames give more insights in the subjective opinion of individuals: the ‘why’ and ‘how’ people think the way they do. Standard quantitative research approaches applied in previous research typically assume that respondents all have the same objective ‘frame of reference’ when they are answer survey questions. However, it is more likely that frames differ among people and that this causes that people have different understandings of same topic depending on their frames. To illustrate this, two different persons can both score “neutral” on the statement “I find road pricing a good idea” for different reasons: one person because the personal financial consequences are yet unclear to him/her and the other person because she/he has strong environmental beliefs and therefore prefers the stimulation of cycling and public transport. In standard quantitative research approaches these persons would end up in the same category and the different meaning given to the same statements would not be revealed. Although the subjective opinion of road users has been examined previously (e.g. Hermans & Koomen, 2006), to the best of the authors knowledge, a systematic research of how the wider public talks and thinks about road pricing that gives insights in whether different frames of reference can be identified is still missing.

In this paper, we aim to the take a first step to fill this knowledge gap by identifying the frames among the general public regarding road pricing. For this, we apply Q-methodology, a research method that pays more attention to differences in these frames. In a Q-study, respondents are presented with a sample of statements that defines the concourse: the whole of statements of opinions related to a certain topic (Brown, 1980). In a Q-study respondents are asked to rank order these statements in a specific way. By doing this, they reveal their subjective meaning to the statements which reveals their subjective viewpoint or personal profile (van Exel & Graaf, 2005). This method encourages that respondents actively construct their opinion on the topic. By rank-ordering the statements, respondents have to evaluate and interpret them in relation to each other. Thus in contrast to regular survey research, the meaning of the ranking of the statements is relational. To get back to the example of two respondents strongly agreeing with the statement ‘I find road pricing a good idea’: the relational perspective of a Q-study can relate the statement to other statements given by the two respondents and can therefore reveal two different frames on road pricing.

The statements used in this study were retrieved from (reactions to) news articles and Twitter. This provided us with initially 2060 ‘real-world’ Dutch statements about road pricing. To arrive at a limited but representative sample of statements, the so-called Q-set, we used the policy adoption framework of Feitelson and Salomon (2004) as a guide. The statements were all assigned to 5 categories retrieved from this framework: Problem perception, Suggestion innovation, Technical requirements, Perceived effectiveness, Distribution of benefits and costs. A sixth category was added that comprises the interplay between actors. Within each category, 7 statements were selected that together cover all subcategories and thus the concourse. We selected the most clear and simple statements to ensure that respondents with all kinds of educational backgrounds could easily understand them. The statements were balanced in such a way that the sample contained equal numbers of negative as positive statements about road pricing. The Q-set with the 42 statements was send via an online application (Aproxima, 2015; Hackert & Braehler, 2007) to the respondents: the P-set. In Q-methodology, the P-set is not a random sample from a population but a selection of persons that are expected to have different viewpoints on the topic. In this study, the respondents represent the viewpoints of the wider public in the Netherlands and were selected on car possession, employability and are of living.

In the survey the respondent were first asked to read all 42 statements and to place them on one of three piles: disagree, neutral and agree. In the following step, the respondents were asked to place the statements on the scorecard, which has the shape of a normal distribution, starting with the extreme values (-5 and +5) before working towards the central part of the card for the neutral statements. We argue that reading and interpreting a statement would take at least five seconds (5x42=210 seconds). Assuming that respondents compare at least the 9 most positive statements with each other and the 9 most negative statements (1/2(9)(9-1)x2 = 72 judgements) this would also take at least 200 seconds. In total, 132 completes were realized. 9 of them were removed because these respondents completed the previous mentioned steps within 410 seconds. Of the remaining respondents, 42 own a private car, 41 a lease car and 40 no car. The sample was about equally spread in terms of employability and area of living (rural or urban areas). The Q-sorts (n=123) were factor analysed using the method of centroid factor analysis. This led to 4 factors that had an eigenvalue greater than 1 and at least two or more high loadings. They explain 41% of the total variance. Next, the varimax rotation method was used to approximate a simple structure. 91 participants had a loading greater than 0.400 on one factor. 7 respondents load on two factors and 25 load on none of the factors which were therefore excluded. Hence, 69 % of the data were used in the interpretation of the factors. The four factors are interpreted as follows: - Frame A (43 respondents load on this factor): The progressive environmentalist. The people in this frame strongly subscribe the problems caused by congestion and car use, both economically as environmentally. They support the idea of road pricing and trust both the technology as the decision makers who are responsible for implementation of an instrument. - Frame B (26 respondents load on this factor): The social environmentalist. The persons in this frame subscribe the environmental impact of car use but find road pricing an unfair solution for those with a lower income while especially lease car owners benefit from it. The green socialists seem to prefer public transport as a solution for the problems they perceive. - Frame C (12 respondents load on this factor): The mistrusting conservative. In this frame, road pricing is perceived as another way of the government to ask for more taxes. The people in this frame distrust politicians and the technology. Current car use is not seen as a threat to the environment and the current public transport system is not considered as a good alternative to road users. They think road pricing would harm the economy and would not reduce congestion or pollution. - Frame D (10 persons load on this factor): The cautious sceptic. The persons in this frame find personal financial consequences of road pricing very important. The cautious sceptic wants to have more clarity on the personal and societal implications of road pricing and intends to support implementation of a road pricing scheme if research shows that it would be effective. At the same time the cautious sceptic does not believe that the government is capable of implementing a fair road pricing instrument. These preliminary results confirm that the public is very heterogeneous in worldviews, values and beliefs. Although frame A seems to support the concept of road pricing and frame B and C seem to reject it, it can be questioned how these people would react on a package of mobility policies or innovative road pricing schemes such as rewarding for peak hour avoidance or a budget neutral instrument. Therefore, the authors will extend this study by investigating the distribution of the frames over the population and how the frames embed preferences for different policies. If the relation between frames and preferences is found, the results of this study can be used by policymakers in order to design policies and tailor implementation that take the needs and worries of specific subgroups into account to arrive at robust and broadly accepted road pricing instruments. References Aproxima. (2015). htmlQ: Q-method surveys in pure HTML5. Retrieved from https://github.com/aproxima/htmlq Bamberg, S., & Rolle, D. (2003). Determinants of people's acceptability of pricing measures: Replication and extension of a causal model. In J. Schade & B. Schlag (Eds.), Acceptability of transport pricing strategies (pp. 235–248). Amsterdam, London: Elsevier. Brown, S. R. (1980). Political Subjectivity: Applications of Q Methodology in Political Science. New Haven and London: Yale University Press. Di Ciommo, F., & Lucas, K. (2014). Evaluating the equity effects of road-pricing in the European urban context – The Madrid Metropolitan Area. Applied Geography, 54, 74–82. https://doi.org/10.1016/j.apgeog.2014.07.015 Eliasson, J., & Mattsson, L.-G. (2006). Equity effects of congestion pricing. Transportation Research Part A: Policy and Practice, 40(7), 602–620. https://doi.org/10.1016/j.tra.2005.11.002 Feitelson, E., & Salomon, I. (2004). The Political Economy of Transport Innovations. In M. Beuthe, V. Himanen, A. Reggiani, & L. Zamparini (Eds.), Transport Development and Innovations in an Evolving World (pp. 11–26). Hackert, C., & Braehler, G. (2007). Flash-Q. Retrieved from http://www.hackert.biz/flashq/home/ Hermans, J., & Koomen, M. (2006). Kilometerprijs. Amsterdam. Kroesen, M., & Bröer, C. (2009). Policy discourse, people's internal frames, and declared aircraft noise annoyance: An application of Q-methodology. Journal of the Acoustical Society of America, 195–207. Schade, J., & Schlag, B. (Eds.). (2003). Acceptability of transport pricing strategies. Amsterdam, London: Elsevier. Schuitema, G., & Steg, L. (2008). The role of revenue use in the acceptability of transport pricing policies. Transportation Research Part F: Traffic Psychology and Behaviour, 11(3), 221–231. https://doi.org/10.1016/j.trf.2007.11.003 Sun, X., Feng, S., & Lu, J. (2016). Psychological factors influencing the public acceptability of congestion pricing in China. Transportation Research Part F: Traffic Psychology and Behaviour, 41, 104–112. https://doi.org/10.1016/j.trf.2016.06.015 van Exel, J., & Graaf, G. de. (2005). Q methodology: A sneak preview. Retrieved from www.jobvanexel.nl Vonk Noordegraaf, D., Annema, J. A., & van Wee, B. (2014). Policy implementation lessons from six road pricing cases. Transportation Research Part A: Policy and Practice, 59, 172–191. https://doi.org/10.1016/j.tra.2013.11.003 Welch, T. F., & Mishra, S. (2014). A framework for determining road pricing revenue use and its welfare effects. Research in Transportation Economics, 44, 61–70. https://doi.org/10.1016/j.retrec.2014.04.006

ABSTRACT. Traffic congestion, local and global air pollution and noise pollution are some of the negative externalities generated by the increased use of private vehicles in many urban areas. Significant shifts of citizen’s travel habits towards sustainable modes of transport encouraged by long- or short-term policies are needed to address these externalities. One short-term solution is to enhance individuals’ awareness of the environmental impact of their travel choices and to use behavioural change interventions to motivate urban travelers towards the use of environmentally friendly multimodal options. In this context, this paper concerns the implementation of two motivating behavioural change strategies, namely i) persuasive technologies in the form of persuasive messages and ii) financial and non-financial rewards, tailored for and integrated in a route planning mobile application, and investigates their effect on changing travelers’ behavior towards green mobility. In fact, persuasive systems supporting behavior change in the context of personal urban mobility is an active area of research; see e.g. the recent review of Anagnostopoulou et al. (2016). Particularly, several approaches including behavior feedback, social comparison, goal-setting, gamification, personalized suggestions and challenges have been developed the last years (Bothos et al., 2014; Gabrielli and Maimone, 2013; Jylhä et al., 2013; Meloni et al., 2014; Pucher J. and Dijkstra, 2003). Overall, these studies identify positive changes in users’ perception of sustainability in urban mobility and increased concern regarding the impact of their choices on the environment. However, based on Anagnostopoulou et al. (2017), people differ in their susceptibility to different persuasive strategies, and, thus, personalized approaches can be more successful than “one size fits all” approaches. With regards to offering rewards to the urban travelers, various rewarding schemes have been implemented, the vast majority of which concerns rewards offered to car drivers with the aim to avoid rush-hour driving (Ben-Elia and Ettema, 2011; Khademi and Timmermans, 2014; Merugu et al., 2009) and to change their behavior towards eco-driving (Lai, 2015). A growing body of research also focuses on reward schemes that encourage modal shift and the use of public transport or non-motorized modes (Bamberg et al., 2003; Fujii and Kitamura, 2003; Koo et al., 2013; Thøgersen, 2009; Voon et al., 2017). On the one hand, research in travel behavioral psychology suggests that such rewarding systems could be effective in changing individuals’ travel habits and supporting green travel behavior, while Khademi et al. (2014) state that the reward schemes may lose their effectiveness over time, since participants may return back to their habitual travel patterns. In this paper, the simultaneous impact of the aforementioned behavioural change strategies (the provision of persuasive messages and the offer of rewards) on individuals’ travel behavior in a multimodal transportation network is explored. For this purpose, both of the aforementioned behavioural change strategies are communicated to individual travelers through a route planning mobile application, which has been developed in the context of the EU-funded OPTIMUM project . More specifically, the mobile application provides personalized persuasive features and offers financial and non-financial rewards with the aim to nudge users towards the selection of environmental-friendly multimodal routes. With respect to nudging, in our case it means to make individuals who mostly use their car to begin using public transportation, those who already use public transportation to consider cycling and walking as well as sustain their current habits and so on. In addition, both behavioural change strategies aim at promoting multi-modal travel activities which refer to using more than one means of transportation to reach a destination (i.e. a combination of public transport and bicycle or walking). The methodological steps to develop the persuasive and reward features on the OPTIMUM mobile application and evaluate their effectiveness are as follows: The first step is concerned with the pre-pilot phase, where two web-based surveys are conducted, with the following specific aims: i) to collect early opinions on the potential modal shift of individuals in the presence of persuasive messages and reward schemes and acquire insights on their preferences on the type of persuasive messages and rewards, ii) to collect socio-demographic data, data on existing travel habits and attitudinal data from a set of respondents and (iii) to collect data about users’ personality traits, mobility types and susceptibility to different persuasive strategies. The first set of data is used to define the requirements of the OPTIMUM mobile application regarding the persuasive and rewards features. The second set of data along with data collected from stated preference experiments on multimodal choices are employed to develop a travel behavioral model which estimates the probability of an individual to choose a certain multimodal option among various alternatives with differentiated attributes, such as travel time, travel cost, weather conditions, type and level of the rewards offered. The model corresponds to a discrete choice model derived in the random utility framework in which decision makers are assumed to be utility maximizers (Domencich and McFadden, 1975; Ben-Akiva and Lerman, 1985; Ben-Akiva and Bierlaire, 2003). The developed model is exploited during the next methodological steps where the persuasive messages and the reward systems are featured on the OPTIMUM mobile app. Finally, the third set of data were used to determine the perceived persuasiveness of the various persuasive strategies on users of different personality and mobility types. This formed the basis for developing the so called persuadability model, which determines each individual’s susceptibility to the various persuasive strategies that are incorporated in the persuasive messages, on the basis of his/her personality and mobility type. The developed persuadability model (see Anagnostopoulou et al., 2017), guides the presentation of the persuasive messages, since each individual is presented with a message implementing the persuasive strategy s/he is more susceptible to. The next step corresponds to the first pilot phase where the message based persuasive feature of the mobile application is developed and implemented. Ninety eight persuasive messages have been designed, with each one of them implementing a single persuasive strategy. For the purposes of our work we have selected the persuasive strategies of self-monitoring, comparison and suggestion among the 10 suggested by Orji et al. (2014), by taking into account the appropriateness of the strategy for message-based persuasion and the suitability of the strategy to the overall scope of our approach. Multiple messages have been designed per persuasive strategy. The messages are context-aware, in the sense that they are valid in specific contexts. Making the messages context-aware enhances the ability of our approach to provide tailored messages, since only messages with a context that is valid for a particular user with a particular trip profile, who is planning for a particular trip made under specific environmental conditions, are selected. In a later step, the app users are additionally presented with the reward scheme where they are rewarded when using sustainable means of transport such as public transport, bicycle, walking or a combination of the above. In this pilot phase, both persuasive messages and rewards are provided to the app users so that their impact on travelers’ behavior will be simultaneously assessed. The reward types that are presented to the app users are partially influenced by the preliminary results of the travel behavioral model developed in the first step of our research. Specifically, two reward types are offered to the users when conducting a sustainable travel activity, namely (i) monetary rewards, where the app users receive cash back rewards and (ii) credits, where the users earn credits (in the form of points) which can be redeemed for vouchers or free public transport tickets. Finally, the last step concerns the evaluation of the developed app features in terms of their effectiveness towards altering individuals’ travel behavior to more sustainable multimodal options.

Figure 1. Methodological steps to develop the persuasive and reward features of the OPTIMUM mobile app and evaluate their effectiveness Within the OPTIMUM project, the on-line surveys and the pilot phases were conducted in three European cities, namely Ljubljana, Vienna and Birmingham. A web-based questionnaire was distributed, while the mobile app users were recruited via social networks and via the communications channels of the OPTIMUM partners. In total, approximately 80 users in the three regions used the OPTIMUM mobile application during the first pilot phase. Quantitative travel behavior data collected by the use of the OPTIMUM app was exploited to evaluate the effectiveness of the persuasive messages. We identified that messages which nudge users to bike&ride were the most influential. In addition, user-experience data collected via on-line questionnaires, which were administered during the first pilot phase, indicated that, although the personalized persuasive messages were somewhat convincing, the users’ experience is differentiated in the three pilots. The second pilot phase is undergoing and will simultaneously evaluate the impact of the persuasive messages and the reward schemes. Indicative evidence on the impact of the rewarding schemes on travelers’ behavioral change have been so far obtained by the estimated multimodal choice model. In specific, the model estimation results indicate that both monetary- and credit-based rewards can be effective in altering travelers’ behavior towards greener multimodal options. Instead, other types of rewards, such as the provision of guaranteed parking space for the car or the bicycle, free wifi access on board or reserved seat on the bus or the metro, are not found to affect travelers’ choices. The quantitative travel behavior data which will be collected by the use of the OPTIMUM app during the second pilot phase will further evaluate the actual travel behavioral changes of the app users in the presence of the persuasive messages and the reward schemes. Keywords: reward schemes, persuasive technology, mobile application, travel behavior, discrete choice model, stated preference data, behavioral change.

Acknowledgments This research is part of the Project “Multi-source Big Data Fusion Driven Proactivity for Intelligent Mobility” - OPTIMUM. This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 636160. This publication only reflects the author’s view and the European Union is not liable for any use that may be made of the information contained therein.

References Anagnostopoulou, E., Bothos, E., Magoutas, B., Schrammel, J., Mentzas, G., 2016. Persuasive Technologies for Sustainable Urban Mobility. arXiv preprint arXiv:1604.05957. Anagnostopoulou, E., Magoutas, B., Bothos, E., Schrammel, J., Orji, R. and Mentzas, G., 2017, April. Exploring the Links Between Persuasion, Personality and Mobility Types in Personalized Mobility Applications. In International Conference on Persuasive Technology (pp. 107-118). Springer, Cham. Bamberg, S., Rolle, D, Weber, C., 2003. Does habitual car use not lead to more resistance to change of travel mode? Transportation, 30, 97-108. Ben-Akiva, M.E., Lerman, S.R., 1985. Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press, Cambridge, MA. Ben-Akiva, M., Bierlaire, M., 2003. Discrete choice models with applications to departure time and route choice, in: Hall, R. (Ed.), Handbook of Transportation Science, 2nd edition, Kluwer. pp. 7–38. ISBN:1-4020- 7246-5. Ben-Elia, Ε.,and D. Ettema, 2011. Changing commuters’ behavior using rewards: A study of rush-hour avoidance. Transportation Research Part F, 14, (5), 354-368. Bothos E., Prost S., Schrammel J., Röderer K., & Mentzas G., 2014. Watch your Emissions: Persuasive Strategies and Choice Architecture for Sustainable Decisions in Urban Mobility. PsychNology Journal, 12(3), 107-126. Domencich, T.A., McFadden, D., 1975. Urban Travel Demand-A Behavioral Analysis. New York: American Elsevier Publishing Company, Inc. Elvik, R., Ramjerdi, F. (2014). A comparative analysis of the effects of economic policy instruments in promoting environmentally sustainable transport. Transport Policy, 33, 89-95. Fujii, S., Kitamura, R., 2003. What does a one-month free bus ticket do to habitual drivers? An experimental analysis of habit and attitude change. Transportation, 30, 81-95. Gabrielli S., Maimone R., 2013. Digital Interventions for Sustainable Urban Mobility: A Pilot Study. UbiComp’13, Zurich, Switzerland. Jylhä A., Nurmi P., Sirén M., Hemminki S., Jacucci G., 2013. MatkaHupi: a Persuasive Mobile Application for Sustainable Mobility. UbiComp’13, Zurich, Switzerland. Khademi, E., Timmermans, H., Borgers, A. (2014). Temporal Adaptation to Reward Schemes: Results of the SpitsScoren Project. Transportation Research Procedia, 3, 60-69. Khademi, E., and H. Timmermans, 2014. The Long-Term Effectiveness of a Reward Scheme in Changing Daily Travel Choices. Procedia-Social and Behavioral Sciences, 111, 380-389. Koo, Y., Lee, M., Cho, Y. (2013). A point card system for public transport utilization in Korea. Transportation Research Part D, 22, 70–74. Lai, W.T., 2015. The effects of eco-driving motivation, knowledge and reward intervention on fuel efficiency. Transportation Research Part D, 34, 155–160. Meloni I., Sanjust B., Delogu G., Sottile E., 2014. Development of a technological platform for implementing VTBC programs. Transportation Research Procedia 3 129–138. Merugu, D., B.S. Prabhakar, and N.S. Rama, 2009. An Incentive Mechanism for Decongesting the Roads: A Pilot Program in Bangalore. In: Proceedings of ACM NetEcon Workshop on the Economics of Networked Systems. Orji, R., Regan, L.M., Julita, V., 2014. Gender and Persuasive Technology: Examining the Persuasiveness of Persuasive Strategies by Gender Groups. Persuasive Technology Adjunct Proceedings, University of Padova, Padova 48-52. Pucher J., Dijkstra L., 2003. Promoting safe walking and cycling to improve public health: lessons from the Netherlands and Germany. American Journal of Public Health, 93 (9), 1509-1516. Thøgersen, J., 2009. Promoting public transport as a subscription service: Effects of a free month travel card. Transport Policy, 16, 335–343. Voon, N.H., Kadir, S.N.H.A., Belayan, M.A., Poon, S.H., Zahran, E.S.M.M. (2016). Rides for Rewards (R4R): A Mobile Application to Sustain an Incentive Scheme for Public Bus Transport. In Computational Intelligence in Information Systems, Advances in Intelligent Systems and Computing 532, Springer International Publishing AG.

ABSTRACT. Based on an innovative field experiment involving ca. 500 commuters, the paper analyzes the effect of spatially and temporally differentiated pricing instruments on the travel behavior of commuters. The experiment uses GPS-based tracking technology together with automatic mode-detection and user profiling. The behavior of the participants is recorded over a period of five weeks. The objective of the experiment is to gain a better understanding of the underlying preferences and trade-offs faced by participants.

ABSTRACT. Traffic congestion is a significant problem in many cities in the world. Jakarta, one of the most populous cities, faces this problem. There are several policies that have been implemented to reduce traffic congestion, such as improving public transport and motorcycle restriction on several roads. One of the policies that will be implemented is congestion charging. In this study, we would like to measure the impact of congestion charging in Jakarta. This study will use two modelling approaches. We estimate the model using MNL (Multinomial Logit) and MXL (Mix logit), which both integrate latent variables. There are 5.879 observations for this research. The initial finding of this research found that MXL logit outperforms MNL model. The VOT of congestion charging and parking cost from the MXL is larger than the value found in the MNL model. The VOT of fuel/ticket cost from the MNL is larger than the value found in the MXL model.

ABSTRACT. Please refer to the attached file.

ABSTRACT. Researchers have long been divided on what is the appropriate method for measuring attitudes, Likert scale or open-ended questions. The simplicity in use and measurement, has favoured the use of Likert scales. Use of open-ended questions posed serious challenges to the researchers in analysing the textual data. However, advances in Machine Learning, now enables researchers to interpret and analyse open-ended responses. This research intends to analyse the relative merits and demerits associated with the use of these two approaches in the context of autonomous shuttle services. In the paragraphs to follow, we present an overview of research related to use of autonomous vehicles, Theory of Planned Behaviour and on analysis of open-ended questions. The proposed methodology to compare the two approaches, in the context of use of autonomous shuttle service, is also presented.

Innovations in technology, which helps vehicles to communicate with each other and with the system, is set to revolutionize the way in which people travel. Autonomous vehicles are classified into five levels, based on the level of automation, by National Highway Traffic Safety Administration [1]. The dissemination of AVs could have implications in road capacity, vehicle ownership, vehicle miles travelled and on parking infrastructure needs. Considering that human errors account for a significant proportion of crash [2, 3], autonomous vehicles might make travel safer [4]. Autonomous vehicles could also mitigate congestion, reduce pollution, improve traffic operations and mobility of people (young, elderly and disabled), change parking patterns and save fuel. Vehicle costs, challenges associated with the licensing of AV, litigation and liabilities during accidents, security risks associated with hacking of the system and system failure, and issues related to privacy could pose serious threats to the use of AVs [3]. Improved mobility could be positively associated with an increase in vehicle miles travelled [3, 5], which, in the absence of carefully devised travel demand management strategies, has the potential to increase congestion and accidents [3].

Familiarity with Google Car, ride sharing, ABS form of automation, socio-demographics (gender, age, household size, employment status) and zonal characteristics (density of population, employment and housing) has influenced the adoption rates of shared autonomous vehicles (SAVs) [6]. Technologically savvy individuals and those inclined towards green lifestyles are more likely to adopt car-sharing services, use ride-sourcing services and embrace SAVs in the future. The choice of SAVs is more popular among individuals with higher educational qualification and younger individuals [7]. Socio-demographic characteristics, zonal characteristics, technological savviness, environmental consciousness, and current travel characteristics, thus play an important role in use of AVs.

The theory of planned behaviour (TPB) is one of the social psychology theories used most in travel behaviour analysis. According to TPB, the intention to perform some specific behaviour is an indicator of the determination of individuals to pursue that behaviour. Intentions can be predicted by understanding the attitudes towards the behaviour, subjective norms and perceived behavioural control [8]. Examples of application of TPB include, among many other studies: intentions to use bike-sharing during holidays in Copenhagen [9], intention to use transit, related with perceived fairness and perceived transit service spatial equity [10]. TPB has also been used in the context of autonomous vehicles: the intention to purchase AVs [11], propensity to use of AVs for intercity travel [12], and to understand the intention to use autonomous shuttle service [13].

Applications of TPB as well as other studies, where attitudes have been collected, relied extensively on Likert scales. The extensive use, is motivated by the ease of use (to the respondent and analyst), particularly the easiness to transform responses into quantitative variables to be included in statistical models. Likert scales have been widely used to measure attitudes, social norms and perceived behavioural control. They present several potential disadvantages, in that, it may force individuals to not convey their attitudes to their satisfaction, create levels or categories that appears to be self-evident, but are inappropriate in reality [14], and systematically overestimate the attitudes [15]. On the contrary, open-ended questions allows respondents to express freely, without being influenced by the analyst. Open-ended questions are often not used, due to difficulties associated with the specificity of the information required, unfamiliar nature of the open-ended style, and complex distinction between open-ended and specifically targeted questions [16]. Advances in machine learning, now enables analysis of open-ended responses. Latent Dirichlet Allocation (LDA) is used to analyse open-ended responses. Each document (open-ended response) is assumed to be represented as random mixtures over latent topics (in the present work, topics about AVs), where each topic is characterised by a distribution of words [17]. In simpler terms, words will be extracted from the open-ended responses and depending on how words relate to topics related to AVs, information in the responses can be inferred.

This paper aims to compare the Likert scale questions with open ended questions, using a TPB framework aimed at measuring intention to use autonomous shuttle buses. AV shuttle bus services are being mooted in a number of cities and this study is undertaken to evaluate the stated intention to use the service in Copenhagen. Considering the lack of readily available technology, only the intention for use of the technology can be analysed at this point of analysis and TPB is used to measure this.

An online survey has been designed to assess the intention of use of service. The survey has two versions. In the first version, all questions related with attitudes, perceived behaviour control and subjective norms are in Likert scales, while in the second version, a combination of Likert scales and open-ended questions is used. Respondents are randomly assigned a version, based on the “Page randomization” feature in SurveyMonkey.

The questionnaire is divided into five sections. The first section captures attitudes related with technological savviness of the individual. To capture this, questions related to frequent upgrade of ICT gadgets, testing of mobile applications and following of news related to AVs were asked. The use of smartphone for travel related needs is presented, both as a Likert scale and open-ended question to the respondents. Individuals are then encouraged to share their views on the detrimental role of transport on environment, role of technology in making travel more environment friendly and how alternative mobility solutions contribute towards reducing the environmental footprint. The question on detrimental role of transportation on environment, is presented as Likert scale and open-ended question. This constitutes the second section which aims to capture attitudes towards environment and environmental impacts of transportation. The attitudes can arise from the understanding of the potential benefits of the system and on whether the behaviour is enjoyable or not. AVs could impact society positively by reducing congestion, accidents, pollution, eliminating the need for parking spaces, but could adversely affect the livelihood of drivers. The positive impacts and the negative impacts on society, due to AVs, as perceived by the respondent, are captured using both Likert scale and open-ended questions. From an individual’s standpoint, AVs could be beneficial in that, they eliminate stress during driving, and allow individuals to pursue other activities during travel. It might however remove the pleasure associated with driving and may also force individuals to share vehicles with others, which may not be appreciated by many. Subjective norms, which emerge from the influence of family and friends associated with the individual under consideration, might influence his/her attitude. The perceptions of friends and family members, regarding use of AV, its safety and benefits are captured using Likert scales. The merits associated with AVs might be of no influence to the individual, if the individual lacks confidence in AVs. The following factors can be attributed to perceived behavioural control. Potential factors that may control this behaviour include the affordability of vehicles, confidence in technology, etc. Questions related to each of these aspects are presented to the respondent, in the third section. In relation to the current travel characteristics of the individual, questions regarding the choice of mode, sharing of rides in the past, travel time and accidents in the past are asked. In the final section, socio-demographic characteristics of the individual, including age, gender, income and role in the university are collected.

The questionnaire was tested using a pilot survey, administered among general public. Based on the responses, the questions were fine-tuned before launching the survey among the staff (academic, non-academic and research staff) and students of “Technical University of Denmark-DTU.” It was decided to launch the survey in DTU campus considering the university’s ambitious plan to launch an autonomous shuttle service in the campus.

The intention to use an autonomous shuttle service will be modelled in the first step using an ordered ‘Probit’ model followed by Structural Equation Models (SEM). The model specification will include variables that depict the socio-demographic characteristics of the individual, technological savviness, environmental consciousness, attitudes and subjective norms towards AVs, perceived behavioural control variables related to intended use of AV and, his/her current travel characteristics. The Likert scale responses from the survey will be converted into factors before inclusion in the ‘Probit’ model. For open-ended responses, LDA will be used to extract information from the responses. This information will later be converted into factors. The models estimated using factors built from open ended questions and the Likert scales factors will be compared based on the goodness-of-fit measures and prediction capabilities. This will aid the evaluation of the relative performance of Likert scale and open-ended responses. The structural equation models will allow the inclusion of the factors that rely only on Likert scales (common in both cases) as latent variables, whereas the factors using alternatively the Likert scales and open ended questions will be included in the SEM models as observed variables, to allow for a more direct comparability. As with the ‘Probit’ models, the results will be compared and used for recommendations about the usefulness and adequacy of open ended questions versus Likert scales, to measure attitudes.

Keywords: Theory of Planned Behaviour (TPB), Likert Scales, Open-ended questions, Topic Modelling, Latent Dirichlet Allocation, Autonomous Shuttle  REFERENCES

[1] N. H. T. S. A. (NHTSA), “Automated Vehicles for Safety,” [Online]. Available: https://www.nhtsa.gov/technology-innovation/automated-vehicles-safety. [Accessed 15 October 2017]. [2] N. H. T. S. A. (NHTSA), “National Motor Vehicle Crash Causation Survey”. [3] D. J. Fagnant and K. Kockelman, “Preparing a Nation for Autonomous Vehicles: Opportunities, Barriers and Policy Recommendations,” Transportation Research Part C: Policy and Practice, vol. 77, pp. 167-181, 2015. [4] P. Tientrakool, Y.-C. Ho and N. F. Maxemchuk, “Highway Capacity Benefits from Using Vehicle-to-Vehicle Communication and Sensors for Collision Avoidance,” in Vehicular Technology Conference (VTC Fall), San Francisco, 2011. [5] D. J. Fagnant and K. M. Kockelman, “The Travel and Environmental Implications of Shared Autonomous Vehicles, using Agent-based Model Scenarios,” Transportation Research Part C: Emerging Technologies, vol. 40, pp. 1-13, 2014. [6] P. Bansal, K. M. Kockelman and A. Singh, “Assessing Public Opinions of and Interest in New Vehicle Technologies: An Austin Perspective,” Transportation Research Part C: Emerging Technologies, vol. 67, pp. 1-14, 2016. [7] P. S. Lavieri, V. M. Garikapati, C. R. P. R. M. Bhat, S. Astroza and F. F. Dias, “Modelling Individual Preferences for Ownership and Sharing of Autonomous Vehicle Technologies,” Transportation Research Record: Journal of Transportation Research Record, vol. 2665, pp. 1-10, 2017. [8] I. Ajzen, “The Theory of Planned Behavior,” Organizational Behavior and Human Decision Processes, vol. 50, no. 2, pp. 179-211, 1991. [9] S. Kaplan, F. Manca, T. A. S. Nielsen and C. G. Prato, “Intentions to Use Bike-sharing for Holiday Cycling: An Application of the Theory of Planned Behavior,” Tourism Management, vol. 47, pp. 34-46, 2015. [10] S. Kaplan, J. e. Abreu e Silva and F. Di Ciommo, “The Relationship Between Young People's Transit Use and Their Perceptions of Equity Concepts in Transit Service Provision,” Transport Policy, vol. 36, pp. 79-87, 2014. [11] R. Kelkel, “Predicting Consumers' Intention to Purchase Fully Autonomous Driving Systems- Which Factors Drive Acceptance?,” 2015. [12] Technologies for Safe and Efficient Transportation University Transportation Center; Research and Innovative Technology Administration, “Impact of Personal Attitudes on Propensity to Use Autonomous Vehicles for Intercity Travel,” 2016. [13] L. Moták, E. Neuville, P. Chambres, F. Marmoiton, F. Monéger, F. Coutarel and M. Izaute, “Antecedent variables of intentions to use an autonomous shuttle: Moving beyond TAM and TPB?,” European Review of Applied Psychology, vol. 5, no. 269-278, p. 67, 2017. [14] W. Foddy, Constructing Questions for Interviews and Questionnaires: Theory and Practice in Social Research, Cambridge University Press, 1994. [15] G. Gaskell, K. Hohl and M. M. Gerber, “Do Closed Survey Questions Overestimate Public Perceptions of Food Risks?,” Journals of Risk Research, vol. 20, no. 8, pp. 1038-1052, 2017. [16] R. Wright and M. B. Powell, “Investigative Interviewers´ Perceptions of their Difficulty in Adhering to Open-ended Questions with Child Witnesses,” International Journal of Police Science & Management, vol. 8, no. 4, pp. 316-325, 2006. [17] D. M. Blei, Y. N. Andrew and M. I. Jordan, “Latent Dirichlet Allocation,” Journal of Machine Learning Research, vol. 3, pp. 993-1022, 2003.

ABSTRACT. INTRODUCTION Autonomous vehicles (AVs) run automatically through use of various Artificial Intelligent (AI) technologies. From such a technological perspective, jobs of drivers will disappear eventually. Considering that human errors account for about 94% (±2.2%) of the traffic accident occurrence (National Highway Traffic Safety Administration, 2015), AVs will reduce traffic accidents dramatically. At the same time, such new technological developments will provide people with more mobility options. However, how people will adapt their behaviors to the birth of AVs has remained unclear. Such behavioral adaptations may be observed in every aspect of travel-related behavior: to drive or not to drive; how to make use of time inside a vehicle; how to change travel mode choices; how residential location choices will be altered, etc.

RESEARCH PURPOSE Because AVs will dramatically reduce human interventions into driving tasks, which will allow human beings to make more use of time inside vehicles, this study only focuses on in-vehicle time use, leaving other behavioral aspects for future research. The purpose of this study is to clarify how people perceive potential changes in in-vehicle time use under different AVs deployment scenarios.

SURVEY: A NEW TYPE OF STATED PREFERENCE SURVEY For the above purpose, we implemented a new type of stated preference (SP) survey in September 2016, which is based on an improved SP-off-RP (RP: revealed preference). The SP-off-RP (Train and Wilson, 2008) allows us to mitigate the biases due to making unrealistic scenario assumptions in the SP survey part, by designing the SP survey based on information revealed from actual market setting denoted from respondents' RP responses. As for the conventional contents of the SP-off-RP survey, choices alternatives include vehicle types of AV with conditional automation (AV_CA), AV with high automation (AV_HA), AV with full automation (AV_FA), and current vehicle type owned by respondent him/herself. The above three types of AVs were presented to respondents based on SAE International’s levels of driving automation for on-road vehicles (SAE International, 2014) (Table 1). What we improved the conventional SP-off-RP survey is that we further introduced a section, which allows respondents to report their perceived changes of in-vehicle time use (i.e., focus on driving, mind distraction, hand distraction, and mixed distractions, see Table 2), by referring to their actual time use inside their vehicles. For reporting the perceived changes of in-vehicle time use, we adopted the same sets of SP attributes as those used to investigate vehicle type choices. Details of these SP attributes are as follows, each of which has two or three levels. (1) Penetration rates of AVs (three attributes): Usually, social interactions play a critical role in encouraging or discouraging customers’ choice decisions (Manski, 1993). To reflect such a phenomenon, penetration rates of Conditional AVs, High AVs, and Full AVs are assumed, each of which has three levels, defined based on the diffusion of innovation theory (Rogers, 2003). Considering the technological advantages and the resulting cost, there should be an increasing trend from Conditional AVs to Full AVs. Therefore, first, the penetration rate of Full AVs is fixed to the following three levels: 5%, 10%, and 20%, and then, the levels for the rest two AVs are determined based on the additional increase in the rate corresponding to the above three levels, separately. Table 1 insert here (2) Additional cost for AVs (3 attributes): Additional costs for Conditional and High AVs were calculated based on Full AVs, which additional costs were fixed to be three levels: 700,000, 850,000, and 1000,000 Yen, corresponding to its penetration rates of 20%, 10%, and 5% in the future market. Levels of the additional costs for the rest two AVs are calculated based on the additional reductions corresponding to the above three levels, separately. (3) Insurance discount rate for AVs (2 attributes): There was no insurance discount policies released by any insurance companies at the timing of the survey in Japan. Here, the insurance discount rate of Conditional and High AVs are assumed to be the same with three levels of 10%, 30%, and 50%, and Full AVs are expected to enjoy a higher discount rate (from 15% to 70%), considering its highest safety level. Table 2 invert here (4) Parking cost for AVs (1 attribute): With self-driving/self-parking functions, it is expected that AVs could contribute to the parking cost reduction by parking itself to a cheaper parking lot, a little bit far from users’ homes. Here, two levels of parking cost reduction are introduced, 50% and 0% (i.e., no reduction). The 50% reduction is assumed based on the calculation by Litman (2012), who compared parking costs for moving a parking space to outside central business district (CBD) or suburbs. (5) Release timing of AVs to the market (1 attribute): Respondents’ choices of AVs are made by assuming that all types of the AVs will be available in the future market. However, when the AVs would be released in the future may matter to choices of AVs. To this end, release timing of AVs in the future market is further introduced into the SP survey. Three levels are assumed: 5, 10, and 15 years from the present time. To reflect the influence of income in the future properly, each respondent was asked to report a change (increase, decrease, or no change) in his/her income in the past five years, and such a change is assumed to continue in the future when they have to make a choice of buying a new car or not, and what types. Such treatment is expected to allow respondents to answer SP questions in a more realistic way. In total, 18 SP profiles were obtained by employing an orthogonal fractional factorial design. In the survey, they are divided into 6 groups, each of which has three SP profiles. In addition to the above SP part, each respondent was further asked to report his/her actual travel behavior and driving experience, for short- and/or long-distance driving; occurrence of unsafe driving incidents during driving; self-cognition about behavioral changes toward safe driving measured in terms of whether and how much he/she wants to improve his/her current driving safety level. Finally, individual attributes were also investigated. As a result, we conducted valid data from 1,002 respondents across the whole Japan.

MODELING APPROACH To further delve the above phenomenon, we build a resource allocation model based on multi-linear function to jointly represent drivers’ in-vehicle time use behaviors with respect to actual cases and future hypothetical cases. Using the multi-linear utility function allows us to capture the interactions between in-vehicle activities in a comprehensive way (see Equation 1). Zero-consumption (i.e., no time spent on some specific in-vehicle activity by an individual) is also reflected in the modeling process. Concretely speaking, individual n is assumed to allocate its available in-vehicle time T_n to several activities (i) so as to maximize the total utility U_n. Maximize Equation(1)insert here Subject to Equation(2)insert here Equation(3)insert here Where, u_ni indexes the utility obtained from spending a length of time t_ni on activity i (i=1,2,⋯,I). λ describes the interaction between different activities, ω_ni is a weight parameter of activity i with all ω_nito be 1.ρ_ni signifies the baseline preference for the time spent on activity i and it is defined as a function of explanatory variables x_ni, whereβ_i is the corresponding coefficient of x_ni, andε_ni is an error term. Equation(4)insert here As a result, the probability of spending time to the first M out of I activities can be calculated as: Equation(5)insert here where ϕ and Φ denote the probability density function and cumulative density function of standard normal distribution, respectively.Ω_ni is the exponential form of the Kuhn-Tucker first order conditions for the optimal time spending allocation under the total in-vehicle time T_n, and error term ∆ε ̂_ni is assumed independent with each other and normally distributed with mean zero and variances of (σ_i)^2. Equation(6)insert here The above modeling process follows Yu and Zhang (2015). AVs provides a completely new type of mobility in the future and consequently there are probably more unknown behavioral variations among individuals’ responses. To accommodate such variations, the above modeling process needs to be improved by incorporating individual unobserved heterogeneity. To do so, we assume that the parameters (β) of major factors x_ns (mainly SP attributes) follows a specific distribution with the density function f( β_s |θ_s), where θ_s are parameters of the distribution. As a result, the above probability is re-written as follows. Equation(7)insert here

EXPECTED OUTCOMES AND SIGNIFICANCE As a behavioral research, this study explores adaptive in-vehicle time use due to the introduction of AVs in the future, in comparison with actual in-vehicle time use in the present, and further compare differences between the present drivers and non-drivers. The expected findings will not only provide unique insights into AVs practices (e.g., policymaking, product design, and marketing), but also advance time use research associated with travel behavior. Understanding in-vehicle time use can also provide new insights into transport policy in mitigating negative impacts of travel time, at least because some parts of in-vehicle time use generate positive utility (e.g., music listening, and watching TV and/or sleeping in case of AVs with full automation function). Positive utility derived from the use of AVs may further increase the car use share in the whole travel choice set, to which transport policy makers should pay more attention. As general scientific merits, this paper not only explores the generality of multi-linear utility function in presenting various interactions between activities associated with travel behavior, but also explores the generality of the above model as a new type of multiple discrete-continuous choice model. The findings from such a perspective will advance the activity-based approach for travel behavior research.

REFERENCES Litman, T. (2012) Parking Management: Strategies, Evaluation and Planning. Victoria Transport Policy Institute. Victoria, B.C. Manski, C.F. (1993) Identification of endogenous social effects: the reflection problem. Review of Economic Studies, 60, 531-542. National Highway Traffic Safety Administration (2015) Traffic safety facts: a brief statistical summary. U.S. Department of Transportation. NHTSA’s National Center for Statistics and Analysis, 1200 New Jersey Avenue SE., Washington, DC 20590. (Access on Nov. 4, 2017: https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812115). Rogers, E. M. (2003). Diffusion of innovation (5th Ed.). New York: The Free Press. Society of Automotive Engineers (SAE) International (2014) Summary of Levels of Driving Automation for On-Road Vehicles (http://cyberlaw.stanford.edu/files/blogimages/ LevelsofDrivingAutomation.pdf; Accessed October 1, 2017). Train, K., Wilson, W.W. (2008) Estimation on stated-preference experiments constructed from revealed-preference choices. Transportation Research Part B 42, 191-203. Yu, B., Zhang, J. (2015) Modeling household energy consumption behavior: A comparative analysis. Transportation Research Part D: Transport and Environment, 39, 126-140.

ABSTRACT. Aims and objectives

Australia’s population is rapidly ageing: the proportion of Australians aged 65 years or older is predicted to nearly double in the next fifty years, from 14 percent in 2012 to 25 percent in 2061 (ABS, 2012). The most commonly reported problems facing older Australians are a need for assistance with mobility and transport (Whelan et al., 2006), and driving cessation is routinely cited as a cause for increased depression and lower quality of life (Musselwhite and Shergold, 2013). The effects of driving cessation have typically been mitigated through the use of strategies that foster the creation of more walkable environments, the provision of alternative modes of motorized transportation, and the dissemination of information that can help older adults plan and prepare for future transport needs (see, for example, Ryan et al., 2015; Zeitler and Buys, 2015; Liddle et al., 2014). However, none of these strategies offer the same level of mobility or accessibility as the continued ability to drive.

Driverless cars are motorized vehicles that are capable of navigating without human input. The first fully autonomous car has been promised as early as 2020, and the technology holds irremediable implications for urban travel and land use, particularly for older adults. Unlike other mitigation strategies mentioned in the preceding paragraph, driverless cars could offer similar levels of mobility and accessibility as the continued ability to drive, without requiring major shifts in existing lifestyles. For example, a large fraction of Australians have lived for most of their lives in largely suburban or rural neighborhoods built around the use of the car. When driving is no longer a feasible option, they have chosen either to continue living in the same neighborhood, relying upon their friends and family to fulfill their mobility needs, or to relocate to a new neighborhood with better access to amenities, at the risk of being further removed from their support system (Alsnih and Hensher, 2003). The arrival of driverless cars could obviate the need to choose between these different but similarly imperfect scenarios, offering that ideal where those unable to drive could both fulfill their mobility needs independently and maintain physical contact with their social circle.

Despite their promise, a number of unresolved technological and legal concerns still surround driverless cars. From a technological standpoint, potential concerns include risk of equipment failure, fear of relinquishing control and threat from online hackers (see, for example, Bansal et al., 2016). From a legal standpoint, potential concerns include liability in case of collision with other cars, pedestrians or property (Duffy and Hopkins, 2013) and ethicality should an unavoidable crash situation involving other individuals arise (Kirkpatrick, 2015). The technology’s dependence on passengers’ willingness to trust a computer to safely navigate a wide variety of driving situations will make public attitudes toward driverless cars one of the most important determinants of the adoption and diffusion of automated vehicle technology (Duncan et al., 2015). Older adults are typically the slowest to adopt new technologies (Pew Research Center, 2014), and there is the additional concern that the segment that could benefit the most from automated vehicle technology may likely be the last to gain from it. And yet, not much is known about the relationship between the attitudes of older adults towards driverless cars, and their willingness to pay for access to the same. And even less is understood about how changes in these attitudes over time, as may be the case once the technology matures and a robust legal framework falls into place, are going to impact future willingness to pay.

The objectives of this study were two-fold: (1) to understand current barriers to the adoption of driverless cars by older Australians; and (2) to predict the diffusion of driverless cars over time among older Australians, as the technology matures and a robust legal framework is established. Findings will help inform if and how driverless cars can be leveraged to offer greater mobility and higher quality of life to an ageing national population.

Study Design

The study comprised two broad phases. In April and May 2017, we conducted six focus groups with 8-12 participants each. Participants were recruited from the members’ database of the Royal Automobile Association (RAA) of South Australia. In view of the study objectives, all participants were 65 years or older. Three focus groups were conducted in the inner Adelaide city, and three focus groups were conducted in exurban communities in Mawson Lakes, Victor Harbor and Stirling. Given both public unfamiliarity and contextual uncertainty with how driverless cars will function, technologically, legally and logistically, any quantitative analysis of the potential impacts must necessarily be preceded by appropriate qualitative research.

In November 2017, we collected quantitative data from 481 older Australians (ages 65 and older). Respondents were asked about their attitudes and opinions about driverless cars, and their willingness to use different shared driverless car services. In particular, each respondent was presented four hypothetical scenarios, such as the one shown in Figure 1. For each scenario, they were told, “In addition to all the transportation services that you currently own or have access to, imagine that you had access to self-driving cars as well, through a shared service such as Uber.” We systematically varied the following six attributes of the shared driverless car service across scenarios: (1) presence of contingency driver; (2) presence of other passengers; (3) legal liability in case of crash; (4) approximate fare per unit distance; (5) average waiting times in case of real time requests; and (6) driving conditions. Respondents were asked to indicate how frequently they think they’d use the service to make particular trips.

Over subsequent sections, we briefly summarize some early findings based on descriptive analyses of the qualitative and quantitative data. By July 2018, we expect to share findings from more sophisticated analyses of the data. In particular, we are employing a thematic approach for the analysis of the qualitative data, where any conclusions are emergent and interpretive in nature. And we are using tools for multivariate analyses to describe, explain and predict older Australians’ willingness to pay for access and use of shared driverless car services, as a function of service attributes and their own attitudes and opinions regarding the same.

Overall use

In general, stated use of shared driverless cars is low. The large majority (258 individuals, or 54 per cent of the sample) indicated that they would never use the service, regardless of the service attributes. On average, individuals were willing to use the service for roughly 10 per cent of their total travel.

These findings are supported by the discussion during the focus groups. The gaining of a driver’s licence in one’s teenage years was viewed as an important rite of passage and part of the Australian way of life, which would potentially be lost with the advent of driverless cars. The gaining of a driver’s licence is often accompanied by the rite of passage of owning a vehicle. As one participant asked “Can you see a time when Australians give up the right to drive” suggesting that driving a car is such an embedded part of life in Australia that a change to not driving just won’t happen. Many of the focus group participants were car and driving enthusiasts and could not see themselves easily giving up the opportunity to drive. Comments such as “I have a car with gears and I love to drive” and “open road and pedal to the metal” were typical of those participants.

Given that the majority of Australians own their own car or indeed are part of a two car household, thoughts immediately went to ownership of a driverless car and the associated costs. The suggestion that driverless cars may well not be owned by individuals but by collectives and perhaps booked and called up using an app on a mobile phone was met with a mixed response. There would need to be a range of options available from owning to sharing for individual use and sharing of rides to cater for individual preferences. Driverless cars posed questions about lifestyle such as towing a boat, caravan or horse float for leisure and for grey nomads.

Sensitivity to service attributes

Stated use was not found to vary significantly across service attributes. Price was the only attribute that seemed to have any effect. Average stated use declined, from roughly 20 per cent of total travel at a service cost of $0.5 per km, to roughly 10 per cent of total travel at a service cost of $4.0 per km. Surprisingly, attributes such as the presence of a contingency driver who can take over the self-driving system at any point, or the creation of separated traffic conditions where driverless cars are only allowed to drive on protected lanes, did not appear to have an effect on stated use of the service.

These findings suggest that the technology is still in its incipient stages, and most individuals do not have a fully developed understanding of the technology, and how they might respond to particular attributes. When asked their level of familiarity with driverless car technology, 95 per cent of the sample reported no to moderate familiarity, and only 5 per cent of the sample indicated being very or extremely familiar with driverless cars.

Sensitivity to attitudes and concerns

Stated use was found to vary as a function of current attitudes and concerns about the technology. In particular, stated use was found to be sensitive to the following five concerns: (1) legal liability for drivers and owners; (2) system security from hackers; (3) vehicle security from hackers; (4) data privacy (location and destination tracking); and (5) system performance in poor weather.

Stated use was not found to be sensitive to the following six concerns: (1) safety consequences of equipment or system failure; (2) interacting with non-self-driving vehicles; (3) interacting with pedestrians and bicyclists; (4) learning to use self-driving vehicles; (5) self-driving vehicles getting confused by unexpected situations; and (6) self-driving vehicles not driving as well as human drivers in general. This is not to imply that respondents are not concerned about these factors. Rather, that concerns about these factors do not necessarily limit willingness to use these services. And in general, public information and education campaigns might be better served if they focus on the factors identified in the first paragraph, and not the second.

Each of the focus groups devoted considerable time to the discussion of potential fears and safety concerns as well. Many of the findings are consistent with those from the quantitative survey. For example, the thought of not having to operate the accelerator, brakes and steering wheel raised fears of not being in control. Thoughts of being potentially hijacked or the car breaking down in the middle of nowhere raised issues of being in danger. There was concern about the lack of testing in Australian conditions and whether driverless cars can account for wildlife and rural and remote locations with long distances. Many questions were raised about licensing issues and legalities. Thoughts of the algorithms that would need to be developed in order to instruct the cars on what to do in certain situations raised concerns about ethics and law. Participants sought reassurance that these issues would be resolved before any major introduction of driverless cars. Indeed some thought that the technology was available and it was the legal and ethical issues that were delaying introduction of driverless cars.

ABSTRACT. I uploaded the extended abstract in pdf instead of pasting the text here.

ABSTRACT. This proposed study aims to provide pedestrian-centered insights that can help to propose primary standards for some features of autonomous vehicles in urban areas. We quantify the effects of specific autonomous vehicle traffic parameters and streets' geometric designs on pedestrians' walking behaviour and safety measures. To do so, an optimal experimental design is generated and the experiment is implemented for more than 300 people using virtual reality.

ABSTRACT. This paper presents a generic method for context-aware SP surveys leveraging state-of-the-art smartphone-based RP methods, and presents its application to mode choice of future smart mobility solutions. The context is coming from the observed RP data, e.g., weekly activity pattern or a selected trip for a given day, together with individual specific information, e.g., vehicle ownership, usage of car/bike sharing services etc. In addition to the direct information obtained from the individual, we collect external contextual data such as the available activity or transportation alternatives for the user through online sources. The experimental design uses those data in order to generate SP choice experiments with reasonable alternatives and attributes. The preferences can then be estimated based on both RP and SP data.

ABSTRACT. Great efforts are made to introduce automated vehicles in conventional traffic. Some expect these vehicles to be safer compared with human drivers. However, less attention has been put on the other road users which will have to interact with the automated vehicles. The objective of this study is to investigate how a combination of expected AVs’ characteristics could affect pedestrians’ crossing intention and test the usefulness of smartphone-based Virtual Reality (VR) by using 360° video recordings for pedestrian crossing behavior research purposes. We studied the crossing intentions of pedestrians when interacting with (automated) vehicles and how these intentions were affected by vehicle type, speed, gap size, the presence of crossing facilities, and presence of communications displays by making a VR simulation of a crossing situation involving an AV. The results show that the participants' crossing intention was not affected by the physical appearance of the vehicles. The presence of communication display did have the expected effect. The participants crossed more when they were confronted with an AV with a green sign as compared to a red sign. They also did cross less when they saw a red sign as compared to no sign. This type of VR simulation proved to be immersive and to not induce sickness which makes it a useful tool for this kind of research.

ABSTRACT. Autonomous driving technologies are currently penetrating the market, and the coming fully self-driving cars will have far-reaching, yet largely unknown, implications. For example, the literature has hypothesized that VMT will increase and activity patterns will change, but little knowledge exists on how these changes will likely happen. A critical unknown is the impact on traveler behavior, which in turn impacts sustainability, the economy, and wellbeing. Rather than explicitly focusing on understanding the travel behavior impacts of self-driving vehicles, most behavioral studies, to date, either focus on safety and human factors (driving simulators; test beds), assume travel behavior implications (microsimulators; network analysis), or ask about hypothetical scenarios that are unfamiliar to the subjects (stated preference studies). This study helps bridge this gap. We implement a naturalistic experiment where we mimic potential life with an autonomous vehicle by providing 60 hours of free chauffeur service for each subject household for use within a seven-day period. We seek to understand the changes in travel behavior observed in the subjects as they adjust their travel and activities during the chauffeur week. In a first pilot application, our sample consisted of 13 individuals from three main cohorts: millennials, families, and retirees. We tracked each subject’s travel for three weeks (the chauffeur week, one week before and one week after) and conducted surveys and interviews. During the chauffeur week, we observed a sizable increase in number of trips, VMT, and zero-vehicle miles, with a more pronounced increase in trips made in the evening and for longer distances.

ABSTRACT. In pursue of greater realism and accuracy, stated preference surveys have increasingly chosen to present alternatives through images along with, or sometimes instead of, text. This decision is especially useful when the measured attributes are complex. As questions related to users' choices in the transportation context become more complex, we believe the visual representation of alternatives will be increasingly relevant. However, it is not clear how realistic an image must be in in order to correctly convey the nature of the displayed alternatives.

The growth of new technologies has the potential to change this and allow to portray complex alternatives in a more realistic way. We believe this is especially the case for immersive images and videos. 

Our main objective is to determine if stated preference experiments presented through static or immersive images produce models with significantly different parameters. To do this, a survey was deployed related to pedestrians’ evaluation of specific streets. We specifically tested three streets in Santiago, Chile, in three ways: by interviewing pedestrians walking on those streets, by presenting these streets through immersive videos, and by presenting these streets through static images.

Results show neither immersive videos nor static images can effectively mimic the results obtained by viewing these streets on-site, but that immersive videos produce a weaker distortion of perceptions. This suggest researchers should prefer these technologies over static images when seeking to assess perceptual or complex qualities of public spaces.

ABSTRACT. The aim of the present paper is first of all to design a survey for collecting longitudinal data (SE and psycho-attitudinal) in order to be able to evaluate, on the one hand, the short and long term effects of the combined implementation of hard and soft measures on travel behaviour, on the other the psycho-attitudinal factors and socio-economic characteristics both before and after implementation of those measures. Using these data it was possible to provide a contribution to the research, analysing, from a statistical perspective the evolution of travel behaviour over time as well as the individual’s intrinsic characteristics and from a modelling perspective, through the specification and estimation of HCMs that use for the same sample the data for these two moments in time.

ABSTRACT. Panel data models provide information on individual behavior, both across individuals and over time. In this paper three different approaches of panel data analysis are explored to examine dynamics in mode choice behavior in the Netherlands: a hybrid choice model, a generalized linear mixed model and the generalized estimation equations approach. We use data from the first four waves of the Netherlands Mobility Panel, unique for its survey design, large sample size and characteristics.

Background Although cycling is very popular in the Netherlands, like in the most developed countries the car is by far the most used mode of transport in everyday life: 47 percent of all trips are made by car. Extensive car use can have negative consequences on several levels, with congestion, noise and pollution the most evident results. Therefore, transport policies mainly focus on reducing car use in favor of more sustainable transport modes, such as public transport, cycling and walking. In the last five years, policy measures and campaigns focus on increasing these modes of transport, motivated mainly by the intention of reducing congestion and pollution, but also improving people’s health.

Due to burgeoning economy, by 2022 traffic congestion is expected to have increased by 9 percent, and travel delays by 28 percent, as compared to 2016 levels (KIM Netherlands Institute for Transport Policy Analysis, 2017). Increasing congestion and travel delays are not solved solely by expansion of the road network. In the next years, policy measures should continue to stimulate the shift towards more sustainable modes of transport. Our knowledge of changes in travel behavior, such as changes in mode choice, is mostly based on data at aggregated level. This is not enough to gain a proper understanding of the dynamics in travel behavior and changes in behavior needed to reverse the worrying long-term trends of congestion, increasing oil consumption and greenhouse gas (GHG) emissions (Ortúzar et al., 2011). In the development and implementation of policies and measures to facilitate these mode shifts, it is important to understand changes at individual level over time.

In this paper we analyze the dynamics in mode choice behavior. Panel data models provide information on individual behavior, both across individuals and over time. Different types of panel data models are estimated to examine changes in mode choice behavior. The interest around the application of panel data analysis in transportation research is increasing (for example Oakil, 2016; Scheiner et al., 2016; Klinger, 2017). The choice of method often depends largely on what the researcher wants to investigate and is not always straightforward because a wide range of legitimate methods are available.

In this paper, the following research questions are addressed: - what are the dynamics in mode choice behavior on household and individual level over time in the Netherlands? - which socio-demographic and socio-economic characteristics, attitudes and preferences, and life-events are associated with changes in mode choice behavior? - what are advantages and disadvantages of different methods for the analysis of panel data?

The contribution of our study to the existing body of knowledge on mode choice behavior is twofold: (1) a better understanding of dynamics in mode choice behavior and (2) the explanatory power and capability of different approaches for longitudinal analysis of mode choice behavior.

Data To examine temporal dynamics in mode choice behavior of individuals imposes specific requirements on the data. First, to describe and understand these dynamics over time longitudinal data is needed. Panel surveys provide information on how travel behavior of individual sample members changes over time in response to changes in the built environment, household and personal situation, or other travel related factors. Secondly, different types of determinants affect mode choice behavior, for example household and individual characteristics, attitudes and preferences, spatial and land use variables, and trip characteristics (Olde Kalter et al., 2015). Including all types of determinants, and also the impact of life-events and several causal relations puts high demands on the data.

We use data from the first four waves (2013, 2014, 2015 and 2016) of the Netherlands Mobility Panel (in Dutch: Mobiliteitspanel Nederland, MPN). The MPN is a state-of-the-art household panel of which the main objective is to examine short-term and long-term dynamics in the travel behavior of households and individuals. Another goal of the MPN is to measure how changes in household and personal characteristics and in other travel-related factors correlate with changes in travel behavior (Hoogendoorn-Lanser et al., 2015). The MPN uses questionnaires to collect a large amount of background information on households and their members, such as sociodemographic characteristics, infrastructure and land-use variables, attitudes and preferences, and life-events. Participants with a completed questionnaire are invited to record their travel behavior for three successive days. It is the unique combination of longitudinal information on household and individual level, other travel-related factors and (changes) in travel behavior that enables us to study the dynamics in mode choice behavior.

Before the start, over 9,000 households were approached to participate in the MPN (screening phase). In total, 6,781 households and 13,690 individuals participated in one or more waves of the MPN; 5,003 respondents did not complete the travel diaries. Overall, 4,832 respondents participated in multiple waves. In our analysis, we only include individuals aged over 17 (the age at which it is legal to drive a car in the Netherlands) and who participated every year. In total, we use data from 1,192 individuals to examine the dynamics in mode choice behavior. The average age in our sample is 49 years, with 2% being younger than 20 and 21% being older than 64 years. Overall, individuals aged from 20 to 64 years are slightly overrepresented in our sample. Also, the share of female persons is almost 5% higher than in the entire population. Regarding household composition, the share of single households and single households without children is higher and the share of couples with children is lower.

Methodology In travel behavior research, a great deal of focus has been given to understand mobility choices. Many different approaches exists to the analysis of travel behavior with panel data. In this paper we explore three approaches to examine the impact of (changes in) sociodemographic characteristics, infrastructure and land use variables, attitudes and preferences, and life-events on transport mode choice: a hybrid choice model (HCM), a generalized linear mixed model (GLMM), and the generalized estimating equation (GEE) approach. Hybrid choice models have been developed during the last decades to capture the impact of subjective factors over the decision process (Ben-Akiva et al., 2002; Walker and Ben-Akiva, 2002). A hybrid choice model integrates latent variables and classes on the choice process. The interest around the application of hybrid choice models in transportation research is increasing and recently these type of models have been applied in modelling choice behavior (for example Yánes et al., 2009; La Paix Puello & Geurs, 2014, Cherchi et al., 2017). Both generalized linear mixed models and the generalized estimation equation approach are an extension of general linear models to handle the random effects. GLMMs provide a broad range of models for the analysis of grouped data, for example multilevel models, since the differences between groups can be modelled as random effects. GEE, introduced by Liang and Zenger (1996), provides a semi-parametric approach to longitudinal analysis of categorical response. Both methods have been applied recently to model dynamics in mode choice behavior (see for example Scheiner, 2014; Olde Kalter et al., 2016).

In the discrete choice part of the models the dependent variable will be mode choice, and includes four alternatives: car driver, car passenger, public transport and slow modes. Home-based tours are the unit of analysis. All trips with home as starting and end point form one tour. Each tour is assigned a main purpose based on hierarchy, in which work has the highest priority, followed by education, shopping and personal services, and social and leisure. Similarly, tours involving more than one travel mode are assigned a main mode, which is the mode with the highest share of travel time. To study the dynamic effects in mode choice behavior we include inertia effects. Inertia may be reflected in the tendency to regularly using the same transport mode for the same purposes (Yang et al., 2015; Cherchi et al., 2013). We assume that the current choice may be affected by the choice made in the previous period (state dependency) and lagged effects. In this case habit is measured as the frequency of past behavior, operationalized by both revealed behavior (mode choice) and attitudes (preferred mode). Furthermore, we include life-events and several ‘change’ variables in the discrete choice model to examine the effect of changes between the different time-points in individual, household and other travel-related characteristics.

Results and discussion Preliminary descriptive analyses show us that at an aggregated level no major changes in mode choice seem to occur. For the whole sample, car driver is the most used mode of transport, followed by cycling and walking. In addition, the results indicate that, also, preferences for the car, public transport, cycling and walking seem to be stable between 2013 and 2016. At the individual level, however, changes in mode choice and preference exists between the different waves of the MPN. About 25% of the individuals in our sample uses the car less frequently, while another 25% uses the car more frequently. Changes in individual cycling behavior are of the same size. For public transport, individual changes are less, but still 10% uses public transport less frequently and another 9% uses public transport more frequently. Also, about 25% of the individuals changes their preferred commuting mode. The largest shift in preferences has taken place between the car and bicycle (in both directions). From previous analyses with the MPN we know that especially major life-events, such as changing job or moving home, are reasons for changes in mode use and mode preferences.

In this study three different panel data models will be estimated to explore patterns in changes in mode choice. In the full paper, the results of the dynamic analyses will be presented: who changes mode, why and when? Furthermore, we will provide insight in the contribution of these different approaches to study dynamics in mode choice behavior. The performance of the hybrid choice model will be compared with generalized linear mixed modeling (GLMM) and the generalized estimating equation approach (GEE). A hybrid choice model has the advantage that latent variables and classes are integrated in the choice process. The GEE approach and GLMM are the two most commonly used techniques to analyze longitudinal data. Coefficients from GEE regressions are marginal effects, i.e. the effects average across all the subjects in longitudinal studies. On the other hand, the coefficients from the GLMM are conditional over the source of intra-class correlation, i.e. the effect for a particular subject. We will analyze the differences in the interpretation of the significance of the factors that contribute to changes in mode choice over time. The hypothesis is that all approaches are capable in capturing and dealing with repeated measures data and to model travel patterns over time. However, we expect that integrating latent variables and classes in the choice process (hybrid choice modelling) will increase explanatory power. At the end, we will discuss the advantages and disadvantages of marginal and conditional models in panel data analysis.

References Ben-Akiva, M., Mcfadden, D., Train, K., Walker, J., Bhat, C., Bierlaire, M., Bolduc, D., Boersch-Supan, A., Brownstone, D., Bunch, D.S., Daly, A., De Palma, A., Gopinath, D., Karlstrom, A., Munizaga, M.A. (2002). Hybrid choice models: progress and challenges. Market Lett. 13, 163–175.

Cherchi, E., Börjesson, M., Bierlaire, M. (2013). A hybrid choice model to account for the dynamic effect of inertia over time. International Choice Modelling Conference, Sydney, Australia.

Cherchi, E., Cirillo, C., Ortúzar, J.d.D. (2017). Modelling correlation patterns in mode choice models estimated on multiday travel data. Transportation Research Part A: Policy and Practice 96, 146-153.

Hoogendoorn-Lanser, S., Schaap, N.T.W., OldeKalter, M.J. (2015). The Netherlands Mobility Panel: An Innovative Design Approach for Web-based Longitudinal Travel Data Collection. Transportation Research Procedia 11, 311-329.

KIM Netherlands Institute for Transport Policy Analysis (2017). Mobiliteitsbeeld 2017. Den Haag, The Netherlands.

Oakil, A.T.M. (2016). Securing or sacrificing access to a car: Gender difference in the effects of life events. Travel Behaviour and Society 3, 1-7.

Klinger, T. (2017). Moving from monomodality to multimodality? Changes in mode choice of new residents. Transportation Research Part A: Policy and Practice 104, 221-237.

Olde Kalter, M.J., Geurs, K., Hoogendoorn-Lanser, S. (2015). Vervoerwijzekeuze in woon-werkverkeer. Eerste analyses met het nieuwe Mobiliteitspanel Nederland. Tijdschrift Vervoerwetenschap 51, 107-127.

Olde Kalter, M.J.T., Geurs, K.T. (2016). Exploring the impact of household interactions on car use for home-based tours: a multilevel analysis of mode choice using data from the first two waves of the Netherlands Mobility Panel. European Journal of Transport and Infrastructure Research 16, 698-712.

Ortúzar, J.d.D., Armoogum, J., Madre, J.L., Potier, F. (2011). Continuous Mobility Surveys: The State of Practice. Transport Reviews 31, 293 - 312.

Puello, L.L.P., Geurs, K. (2014). Modelling observed and unobserved factors in cycling to railway stations: application to transit-oriented-developments in the Netherlands. European Journal of Transport and Infrastructure Research 1, 1-25.

Scheiner, J. (2014). Gendered key events in life course: effect on changes in travel mode choice over time. Journal of Transport Geography 37, 47-60.

Scheiner, J. (2016). Time use and the life course: a study of key events in the lives of men and women using panel data. European Journal of Transport and Infrastructure Research 16, 638-660.

Walker, J., Ben-Akiva, M. (2002). Generalized random utility model. Math. Soc. Sci. 43, 303–343.

Yánez, F., Raveau, S., Rojas, M., de D Ortuzar, J. (2009). Modelling and forecasting with latent variables in discrete choice panel models. European Transport Conference, Leiden, The Netherlands.

Yang, D., Timmermans, H. (2015). Analysis of consumer response to fuel price fluctuations applying sample selection model to GPS panel data: Dynamics in individuals’ car use. Transportation Research Part D: Transport and Environment 38, 67-79.

ABSTRACT. Please see pdf attached

ABSTRACT. Background: Recent intervention studies provide new support to the hypothesis that interventions in the built environment can change travel behaviour in a societally favourable direction, the patterns of change have not yet been explored in detail. The measured associations may be a result of small changes in travel behaviour in most individuals, or may be a result of large changes of fewer individuals. Moreover, the existing studies mainly show whether modal shifts took place towards (or away from) active travel. Changes in travel behaviour for individuals whose active travel mode share or active travel time remained stable are mostly unexplored. To illustrate this with an example: suppose an individual changes from commuting by bicycle only to the combined use of public transport and walking. This change may not necessarily affect the overall active commuting time, neither would it result in a change in the share of trips involving active travel. Nonetheless, large changes in individual travel behaviour have occurred in terms of (a) the primary mode (from bicycle to bus) as well as (b) the form of active travel (from cycling to walking). Neither of these changes are fully captured in most existing intervention studies. It is important to glean a deeper understanding of such changes in order to project consequent impacts (e.g. on air pollution, congestion or physical activity) and longer-term behavioural changes (e.g. the development or breaking of habits, self-efficacy to use certain modes) as well as to estimate the potential modal shift impacts of similar interventions in the future.

Method: We aimed to characterise patterns of change over time using data from the four-year Commuting and Health in Cambridge quasi experimental study cohort, and tested whether exposure to the intervention was associated with certain individual behavioural patterns. We paid particular attention to all modes used in the modal mix and explore the modal shifts, the patterns of mode choice and the patterns of changes in mode choice.

The Commuting and Health in Cambridge study aimed to understand the impacts of a major transport infrastructural intervention in Cambridgeshire, UK, on travel behaviour, physical activity and related health outcomes (Ogilvie et al., 2016). The city of Cambridge (123,900 inhabitants) has a comparatively affluent and well-educated population. In 2011, 45%ofits commuting population travelled to work by car or taxi, 28% by bicycle, 15% on foot, and 9% by public transport (ONS, 2011). The physical intervention comprised the Cambridgeshire Guided Busway, a 25km guideway (separate off-road track) for specially adapted buses, with a parallel service path that offers high-quality segregated infrastructure for non-motorised transport such as walking and cycling. The busway connects the city centre and several major employment sites with surrounding towns and villages, and was opened in 2011. The principal aim of the busway was to change travel behaviour in order to reduce traffic congestion (Atkins, 2004; BBC, 2014) by offering improved facilities for public transport, walking, and cycling.

Adult commuters working in Cambridge(UK) completed annual questionnaires between 2009 and 2012. Commuting was assessed using a validated seven-day travel-to-work record. For each day, respondents were asked to report the day of the week, their working hours and their mode(s) of travel to and from work, or to positively indicate that they had not travelled to work that day. The respondents were explicitly encouraged to report all travel modes used for each trip. No imputations or deletions were made, even if travel data appeared incomplete or incorrect. Exposure to the intervention was defined as the negative of the square root of the shortest road distance from home to the busway. We investigated the association between exposure to the intervention and specific modal shifts and patterns of change, along with individual mode choice patterns over the entire four-year period.

Results: Approximately 20% of the cohort always used a car for the entire commute trip in all four waves, and 20% did not use a car at all. Nevertheless, a larger share of the population showed a more complicated pattern. About 8% of our respondents showed a change from using the car as the only mode of transport in all trips, to a pattern of less frequent or less ‘intense’ car use. The opposite trend was, however, more prevalent: a proportion of respondents increased their intensity of car use by changing from partly using the car to driving the entire journey to/from work, and another group changed from being non-drivers to occasional drivers.

Almost 75% of the respondents travelled by active travel in at least one of the waves (Appendix B). Approximately 20% made a trip entirely by an active travel mode (and not in combination with other modes) in the first wave and continued this behaviour in consequent waves. Cycling appeared to be more commonly used for the entire trip, whereas walking was mostly done in combination with other modes of transport.

Five groups of patterns of change were found in our in-depth explorations: (1) no change, (2) a full modal shift, (3) a partial modal shift, (4) non-stable but patterned behaviour, and (5) complicated or apparently random patterns. A minority of participants had a directed change of either a full modal shift or, more commonly, a partial modal shift, whereas a large proportion showed a highly variable pattern.

Although previous analyses have connected exposure to the intervention with a higher likelihood of an increase in active travel and a decrease in car use (Panter et al., 2016; Heinen et al., 2015), in the present analysis, we did not find a significant association between the level of exposure to the intervention (residential proximity to the new infrastructure) and specific modal shifts, nor did we find evidence that exposure to the intervention was associated with belonging to a group that showed a full or partial modal shift.

Conclusion: Our analyses revealed a large diversity in (changes in) travel behaviour patterns over time, and showed that the intervention did not result in one specific pattern of behaviour change or produce only full modal shifts. It therefore appears presumptuous to conclude that individuals who were more exposed to the intervention were more likely to change their travel behaviour by making a full shift from a car to a bicycle. Our insights are important for improving the measurement of travel behaviour, improving our understanding of how changes in travel behaviour patterns occur, and fully capturing the potential impacts of interventions.

ABSTRACT. See attached

ABSTRACT. A hallmark of market development for new energy technologies is the number and nature of complex interrelated factors including: the existence of a well-established status quo market with products that already provide the essential services, volatility of energy prices that frequently include periods where status quo options have low operating costs, high prices for initial offerings of using new technologies that require a combination of learning and economies of scale to bring prices down, a highly differentiated market with heterogeneous preferences across many consumer segments, and public policies intended to support and subsidize entry. These issues occur for a wide range of product categories, including lighting, heating/cooling, and personal transportation. For personal vehicles, until recently the focus has been on developing an improved understanding of what consumer preferences might be for alternative energy vehicles using, by necessity, a variety of stated preference techniques. Generally speaking, these approaches are limited to insights in a static environment and have limited capabilities for addressing important dynamic effects.

More recently, vehicles using alternative energy technologies have been developed and introduced into the market, so that revealed preference (RP) data are becoming increasingly available. There are many well-known challenges associated with RP data. In addition to the usual econometric issues (price endogeneity, limited information on differences across sub-markets), the sales levels for these vehicles are still relatively low, and are limited to consumer segments that are highly non-representative of the mainstream market (e.g., very high income, highly educated early adopters, many with specific environmentally-related attitudes and motivations). At the same time, if sufficiently large data sets over multiple time periods and geographic regions is available to the analyst, statistical identification of consumer preferences can be improved due to intertemporal and spatial variation (e.g., changes in fuel prices, expansion of vehicle offerings, vehicle subsidies, various policy changes). If, in addition, detailed data on consumer attitudes, knowledge, and perceptions is also available, it can become feasible to investigate potentially important dynamic effects associated with the emerging market for these vehicles (e.g., evolving attitudes, improved awareness and knowledge of new technologies).

We will be reporting new modeling results, findings and insights from an ongoing modeling effort that uses longitudinal consumer survey data from the US new vehicle market collected by Strategic Visions, Incorporated, through their New Vehicle Experience Survey (NVES). The data have been collected as repeated cross-sections on a quarterly basis over a period of many years (going back as far as 2008 or earlier) up to the present time. The sample sizes are much larger than those typically available to academics (e.g., in the 100K to 300K per model year range, depending on the year) and have essentially all of the advantages described above. The NVES includes a high level of detail on vehicle choices and transactions, as well as household demographics and responses to a variety of psychographic scales that are consistent with behavioral theory and results that have appeared in the academic literature. In particular, there are relatively large sample sizes for recently introduced vehicles that employ new technologies [e.g., battery electric vehicles (BEVs) and plugin hybrid electric vehicles (PHEVs)].

The analytical approaches being developed employ state-of-the-art methodologies in discrete choice modeling that support highly detailed choice alternatives at the individual household level. Specifically, the new vehicle market is defined at the ‘vehicle configuration level,’ so as to capture variation in policy-relevant operating characteristics (e.g., fuel efficiency), i.e., vehicle options are defined at the level of Make-Model-FuelType-Engine-Drivetrain-Transmission-VehicleTypeSize-BodyStyle (over 1,000 options at any point in time). Nested Logit models for specific quarters (e.g., April-June 2015) have been estimated that extend previous RP-based approaches for identifying variation in consumer preferences on the basis of demographic segmentation, locational, and regional effects, including policy-relevant interaction effects for alternative-fuel vehicle options. Preliminary work has also been completed on incorporating consumer attitudes, with next steps being the extension of the framework to utilize data spanning multiple quarters/model years simultaneously. The goal is to identify evolving dynamic effects and identifying attitude-based consumer segments through the development of hybrid choice models.

ABSTRACT. Increasing the share of electric vehicles (EVs) within the vehicle fleet is a goal shared by many governments around the globe. This goal is motivated primarily by the desire to reduce the transport sector’s greenhouse gas emissions, and in turn, the sector’s contribution to anthropogenic climate change. Some governments also seek to leverage EV sales to reduce air pollution, assist in the transition towards renewable energy sources, reduce dependence on foreign oil, and support innovation and jobs within the vehicle manufacturing sector. Whilst having the advantage of low or no tail-pipe emissions and lower operating costs, EVs are often disadvantaged by higher purchase costs, operational and technological uncertainty amongst consumers, as well as perceived recharging inconvenience (relative to petrol refueling) –that largely arises from a current lack of ubiquitous public EV charging infrastructure.

Given these known challenges, governments have implemented various types of policies in order to increase consumer demand for these vehicles. These incentives have been found to influence consumer demand for EVs in a variety of different ways (Batley et al. 2004, Beck et al. 2013, Beresteanu & Li 2011, Brownstone et al. 1996, Bunch et al. 1993, Chandra et al. 2010, Dagsvik et al. 2002, Diamond 2009, Ewing & Sarigöllü 1998, Gallagher & Muehlegger 2011, Hackbarth & Madlener 2013, Mabit & Fosgerau 2011, Martin 2009, Musti & Kockelman 2011, Riggieri 2011, Sierzchula et al. 2014, Whitehead et al. 2014, Ziegler 2012).

It might be expected, however, that EV vehicle pricing could also be influenced by these same government policies—with vehicle retailers adjusting pricing arising from the increased demand. Research into the combined effects of government incentives on both the demand for and pricing of EVs is more limited. Sallee (2011) assessed the effect of government incentives on the price of EVs using a representative sample of 15% of Toyota Prius transactional sales in the United States between 2002 and 2007. Contrary to initial expectations, he found that government subsidies did not affect the prices paid for a Toyota Prius during this period. He explained the lack of price response by suggesting that Toyota purposefully did not absorb the value of the government subsidies in order not to stifle future demand for their vehicles.

In this research we further explore this issue by examining not only the impact of government policies on the demand and pricing of hybrid-electric vehicles, but also on plug-in electric vehicles, such as the Nissan Leaf and Tesla Model S, across several international markets.

Considering the number of EV related policies that have been implemented around the world to date, and the current interest in EVs internationally, this study aims to: 1. Identify which factors have affected the demand and price of EV’s; 2. Understand whether EV demand and price are endogenous; 3. Estimate the effects of different government policies on EV demand; and, 4. Estimate the effects of different government policies on EV price premiums.

In order to achieve these four aims, we utilize two sets of panel EV sales data from 40 international markets with a range of government policies. The first set of panel data includes annual sales of hybrid-electric vehicles, such as the Toyota Prius, across 15 regions between 2008 and 2012. The second set of panel data includes monthly sales of plug-in electric vehicles (both plug-in hybrid and battery electric vehicles) across 40 markets between 2013 and 2018.

The analysis is focused on both hybrid-electric and plug-in electric vehicle sales in order to gain further insight into whether similar government incentives have impacted both the demand and pricing of these similar vehicle types, with the intention of providing greater insight for policy makers in the development of future transport incentive policies. The difference in the analysis of annual sales for hybrid-electric vehicles versus monthly sales for plug-in electric vehicles is simply due to data availability.

Using the two panel datasets described previously, we construct similar econometric models for each, comprised of two equations and two dependent variables: Equation 1: EV marginal demand i.e. EV sales as a proportion of total vehicle sales (annual sales for hybrid vehicles; monthly sales for plug-in electric sales) as a function of: economic and socio-demographic factors, government incentive policies and EV price premium. Equation 2: EV price premium i.e. the normalized difference between the dealer-listed price of a new EV (hybrid = Toyota Prius; plug-in electric = Nissan Leaf) and its internal combustion engine vehicle (ICEV) equivalent (hybrid = Toyota Corolla; plug-in electric = Nissan Pulsar/Tiida) as a function of: economic and socio-demographic factors, government incentive policies and EV marginal demand.

To address aim 1 a number of exogenous factors are included in Equations 1 and 2, including: socio-demographic characteristics of the consumer population, economic factors, and incentive policy indicator variables. Additionally, to test for potential endogeneity between EV price premium and demand (aim 2), endogenous variables are included in each equation i.e. price premium in the demand equation and demand in the price premium equation. In order to address aims 3 and 4, we aggregate incentive policies into four categories based upon how and when they affect consumers.

In order to make policies across regions comparable, and representing a unique contribution to the literature, incentives are aggregated into four categories depending on how and when the government policy affects consumers: A. Upfront subsidies (cash rebates, income tax credits); B. Purchase cost reductions (reduced sales tax, import duty, registration tax); C. Running cost reductions (reduced annual vehicle tax, emissions tax); and, D. Usage-based benefits (reduced tolls; congestion charges; parking fees).

The main motivation behind this grouping was to facilitate a better understanding of how different types of policies affect EV markets. Gallagher and Muehlegger (2011) suggest that different types of incentives indeed have different market effects. This specific categorization differentiates between one-off and ongoing cost savings, and operating versus non-operating benefits. Other variations of this categorization of incentive policies continues to be explored as part of this ongoing analysis.

A conceptual overview of our econometric model is shown in Figure 1 (see pdf version).

To capture pricing heterogeneity across EV markets and to facilitate inter-market price comparisons, we defined a price premium variable which captures the normalized cost ratio between a new EV and an equivalent ICEV. In the case of hybrid-electric vehicles, this price premium variable is the ratio of Toyota Prius and Toyota Corolla retail prices. For plug-in electric vehicles, the price premium variable is the ratio of Nissan Leaf and Nissan Pulsar/Tiida retail prices.

The price premium is intended to capture the heterogeneity in supply chain and local market factors (4) that may affect EV pricing, as well as to see whether different government policies have assisted in reducing the gap between EV and ICEV prices, or have exacerbated it.

Figure 1 shows the potential endogenous relationship between marginal demand for EVs (2) and EV price premium (3) – see dotted lines. If different government policies (1) have affected the EV price premium (3) whilst affecting the marginal demand for EVs (2) – it could also be true that these two factors interact. Ignoring this potential endogeneity would lead to biased parameter estimates.

Based on the relationships depicted in Figure 1, the two panel datasets required a system of equations modelling approach, with EV marginal demand (2) and EV price premium (3) as the dependent variables. To account for the system of equations and endogeneity, the analysis required the adoption of an instrumental variable (IV) approach known as Three-Stage Least Squares (3SLS). Generally this method is sufficient for such an analysis, however, given the nature of the panel data (multiple observations within each market), the standard 3SLS approach could not account for correlation across time periods. To account for this, Error Component Three-Stage Least Squares (EC3SLS) was adopted – see Baltagi (1981).

The EC3SLS estimator, being a weighted combination of three 3SLS estimators (between groups, between time-periods and within-groups-and-time-periods) yields efficient parameter estimates of a system of interrelated equations, with cross-correlated error terms, and accounting for serial correlation and heteroscedasticity arising from EV market trend data within markets observed across multiple years. Although this estimation procedure does enable estimation of a system of equations based on panel data, it is restricted to balanced datasets. This restriction arises from the matrix transformations that are performed during the EC3SLS procedure, and resulted in a reduction of our panel dataset to only include markets with full sales data for each time period analyzed.

Initial results from our model of annual hybrid-electric vehicle sales suggest that certain incentive policies affect both the demand for and price of hybrid EVs. Specifically it appears that Type A incentives such as cash rebates are associated with an increase in the marginal demand for hybrid EVs; however, they also appear to have widened the price gap between hybrid EVs and comparable ICEVs. As expected, higher hybrid EV price premiums are associated with lower marginal demand for hybrid EVs, whilst lower marginal demand for hybrid EVs is associated with increased hybrid EV price premiums. Marginal demand for hybrid EVs is highest in markets with higher Gross National Incomes (GNI) and lower rates of inflation. Our models also reveal that increased fossil fuel prices are associated with increased marginal demand for hybrid EVs.

Further work is currently underway to analyze the monthly sales of plug-in electric vehicles, and to compare these results to those outlined above for hybrid-electric vehicles. These results are intended to be included in the final manuscript, along with a discussion of the implications of these results for the design of future transport incentive policies.

ABSTRACT. In much of Switzerland, the road network as well as the public transport network offer competitive accessibility levels that enables commuters to choose from a variety of mobility tools such as cars and season tickets. While one can drive the car everywhere, season tickets typically offer unlimited use of public transport within a local municipal service area. However, all public transport operators in Switzerland jointly offer a nation-wide season ticket called Generalabonnement (GA) that offers unlimited travel within the entire country. In general, the GA is more expensive than any combination of local season tickets, but it provides larger discounts for longer commutes (as long-distance train services are included as well) in terms of price per kilometer. Arguably, commuters with a longer commute will then subscribe to the GA, while those with a shorter commute will choose local season tickets. Understanding this somehow unique choice environment in Switzerland of cars and season ticket is important for modeling and policy making.

Many studies suggest a strong relationship between the built environment and travel behavior summarized as that dense environments encourage less driving and more use of public transport and active modes (Ewing and Cervero, 2010). However, several authors have pointed out that residential self-selection, i.e. individuals who like to drive will locate themselves in the long-term in car friendly locations, should not be ignored. Following Bhat and Guo (2007), three approaches for considering residential self-selection exist: (i) controlling with personal factors that jointly affect residence and travel choices, (ii) instrumental variable methods to accommodate the potential endogeneity of residential choice decisions, and (iii) using panel household moving data. Therefore, when analyzing the choices of mobility tool ownership (for commuters) one should not simply ignore the influence of the choices of home (and workplace) location. The recent literature review by Anowar et al. (2014) provides a summary on modeling approaches for car ownership with other endogenous choices such as residence location.

So far, there is general understanding on the factors regarding the choice of mobility tools in Switzerland (Simma and Axhausen, 2001; Kowald et al., 2017; Becker et al., 2017), but less with focus on commuters and arguably no focus on the relationship between the choices of location and mobility tools with respect to self-selection. In this paper, we use the recently introduced framework of the Generalized Heterogeneous Data Model (GHDM) by Bhat (2015) together with the 2015 Swiss Transportation Travel Survey (Bundesamt für Statistik BFS and Bundesamt für Raumentwicklung ARE, 2017) to address this gap by providing a comprehensive model of mobility tool ownership and location choice for commuters.

The available dataset includes detailed information on households’ socio-demographics and vehicle holdings, along with a one day travel diary and detailed information on education, labor market status and mobility tool ownership of one person in the household. For each location, the coordinates are available allowing a highly detailed spatial classification. In this paper, we use the three level spatial classification by Eurostat based on population density: high (cities), medium (suburbs) and low (countryside).

Of the in total 57,090 respondents, who answered the travel diary, 4,677 (8.19 %) answered an additional questionnaire with 19 items on their attitudes towards transport policy. We include only these observations in our sample because we use attitudes to control for residential self- selection (Cao et al., 2009). Finally, as our focus is on commuters that drive to work or education place locations, we exclude all observations from the sample that are not commuters, e.g. unemployed, stay-at-home parents, or retired persons. This leaves us with a final sample size of 3,023 observations.

Monthly household income was assessed in nine groups, but only available for 71 % of the total 57,090 respondents (and of 83.96 % of the 4,677 that were asked the mobility questions). Therefore, we imputed the income using Stata’s mi impute ologit command with all available socio-demographic information of the households. The variable of commuting distance is obtained as the crow-fly distance between the primary home to the primary work or education location multiplied by a detour factor, which we obtained from the trip diary in the same data set.

We use the 19 items from the questionnaire on transport policy attitudes to carry out a factor analysis to recover the underlying structure that best describes the observed variance. Finally, we included 15 items and identified three factors that reflect (i) the support increase in car prices, (ii) the demand for innovation and subsidy thereof and (iii) road safety explain most of the underlying variation in the data. As the items have several missing items, we use the imputation strategy by Becker et al. (2017) in order not to decrease the sample size even further due to missing values.

In this model, the choices of car ownership, season ticket ownership and season ticket type (local season ticket or GA) are the main outcomes of interest, while the home and workplace location as well as the commuting distance are endogenous variables in the model. We model the mobility tool ownership choice in the GHDM framework as a multivariate probit with sample selection (see Becker et al., 2017). The sample selection means that each individual has an observed outcome for car ownership and any kind of season ticket, but the outcome of season ticket type is only observed for those individuals that have any kind of season ticket. Here, those who have a GA are coded as 1 and those who have a local season ticket as 0. The location choice outcomes are nominal outcomes with three levels of population density (low, medium and high) and are modeled in the GDHM as a typical multinominal probit model. The outcome of commuting distance is modeled with linear regression. We are accounting for self-selection by including three attitudinal variables in the GHDM model, also with linear regression.

We establish jointness across all outcomes by a common underlying covariance matrix. Please see Bhat (2015) for details on the measures for identification of these matrix. The model is pro- grammed in Stata 14 and the parameters are estimated using maximum simulated likelihood.

The model specification has eleven outcome equations which results in a covariance matrix that requires more than fifty Cholesky factors to be estimated. Therefore, we restrict the correlation among unobserved factors to the three latent variables in the same way as Bhat et al. (2016). This allows us to use the available degrees of freedom to estimate the parameters for the explanatory variables of interest. All entries in the covariance matrix are positive and significantly different from zero, which is reasonable as one would expect to some extent a positive correlation between persons that favor increase in car pricing, road safety and the subsidy of new and environmental friendly technologies.

The results of our model show that the choice of season ticket type is as expected a matter of commuting distance, while a good local access to public transport increases the likelihood of subscribing in general to a season ticket and reduces car ownership. Also, as expected, the home and work location affects the commuting distance, while we do not find strong and significant influence of the type of work-location on the choice of mobility tool ownership, while the home location still influences the choices of mobility tools. The latter findings are in line with the main conclusions by Ewing and Cervero (2010).

This paper contributes with the first application of the GHDM framework to the model the mobility tool ownership choices of commuters while considering the choice of residential and workplace location. The special case of Switzerland with its variety of mobility tool choices requires a joint modeling approach such as the GHDM framework, but traditional structural equation models could be appropriate as well. In future research, we have to investigate in detail the causal relationships of the endogenous variables in the model, but due to the rather small sample and only limited number of exogenous covariates, this must be done carefully. Moreover, robustness analyses with full and restricted covariance matrix as well as with and without latent variables need to be carried out to understand the wider picture in the decision making process.

ABSTRACT. Motivation A significant reduction in greenhouse gas emissions is the declared political goal in Germany. To this end, the transport sector must also contribute. At present, the car is used for 70% of the vehicle mileage traveled (VMT) in everyday travel (Weiss et al., 2016a). Technical (e.g., electric vehicles), modal (e.g., shifting traffic demand to rail, bicycle or pedelec) or organizational (e.g., carsharing) alternatives available to substitute the use of conventional passenger cars and to reduce traffic emissions. Though, a comprehensive data base is needed to examine the potential and environmental impacts of policy measures and transport services that contribute to the substitution of conventional private cars. In particular, these analyzes require representative passenger car usage data over longer periods of time, as cars are not used in the same way each day, but the usage characteristics vary considerably over a longer period of time (e.g., daily commuting, summer holiday trips). However, original, representative and longitudinal data on the use of the German car fleet are not available. National surveys on the mobility of persons (e.g., the German Mobility Panel (MOP), Mobility in Germany (MiD)) and the use of motor vehicles are only carried out over short periods of time (e.g., one day, one week) or other studies on driving performance are limited to the detection of annual mileages (e.g. German Vehicle Mileage Survey) (Bäumer et al., 2017; infas and DLR, 2010; Weiss et al., 2016a; Wermuth et al., 2012). Existing data sources are however not sufficient to analyze how annual passenger cars are distributed over individual trips during the course of the year.

Modeling car use profiles In order to close this gap, we developed the car utilization model CUMILE (Car Usage Model Integrating Long Distance Events), which models car trips over a full year for a representative passenger car fleet (Chlond et al., 2014; Weiss et al., 2014; Weiss et al., 2016b; Weiss et al., 2017). Input data for CUMILE are everyday mobility survey (MOP-EM) and the car usage and fuel consumption survey (MOP-FCOR) of the German Mobility Panel (MOP). Information on the characteristics of long-distance trips was obtained from the long-distance travel survey INVERMO (Zumkeller, Chlond et al., 2006). The travel surveys used reflect the travel characteristics of persons. However, we modelled the longitudinal usage characteristics of cars. The CUMILE car fleet consists of cars whose owners participated in the MOP-EM and the MOP-FCOR surveys. The representativeness of the sample is ensured by weighting the data according to the socio-demographic characteristics of the car owner and the car’s features. The CUMILE algorithm consists of four steps. In the first step, we analyzed the individual car travel data of survey participants during the MOP-EM week. Since MOP does not allocate specific cars to single trips, we developed a heuristic assignment of cars to individuals to approximate the mileage of a specific car for every day of the MOP week from the travel diaries. In the second step, we estimated the car usage for typical days of the year. The MOP participants report whether it was a rather typical or an atypical day for every survey day, that is, whether the car was under repair or the car user was ill or on holidays. Since most trips made on a typical day are frequently repeated, such as commuting trips, we assumed that every weekday was representative for the same day of the week throughout the course of one year. Though, not every day in the course of one year is typical in terms of travel behavior since. This is taken into account in step 3 and 4 by the use of additional information from the input data. In the third step, we calculated the car mileage per day during the period of the MOP-FCOR survey. The algorithm compares the typical daily mileages calculated in the second step with the actual metered mileages between refueling procedures. When the actual car usage was overestimated, the car mileage was set to zero on randomly selected days. When the actual car mileage was underestimated, the model assumes the car was used for an additional long-distance trip (LD-trip) and draws a feasible LD-trip from the INVERMO survey. In the fourth step, we modelled the car mileages per day for the remaining days of the year. As in step three, the algorithm detects whether the typical mileages explain the reported annual mileages. In case of an overestimated car usage, it is assumed that some typical trip chains did not take place in the model period. When the actual car mileage was underestimated, the LD-trip drawing procedure was run again. Steps three and four procedure we repeat until the daily car mileage of all days was calculated, given that the annual car mileage was correctly represented. As the result of CUMILE, information about departure and arrival time, trip purpose and the distance of the modeled car trips are available for each modeled vehicle. The sample covers around 6,300 cars and 4.6 million trips. In addition, information on socio-demographics of car owners and characteristics of passenger cars (e.g., car class, fuel consumption) is recorded in the MOP.

Modeling environmental impacts In order to be able to determine the energy consumption and the emissions of the modeled car trips, we supplemented the car user profiles with further information. In addition to the vehicle characteristics (car class, drive, age), other factors, such as the driving behavior, the trip length, stationary periods between any two trips, the weather conditions and road categories do also impact the emission behavior. Information on the outside temperature, rainfall and lighting conditions are supplemented by data from the German Weather Service (DWD, 2016). The results of the transport demand model VALIDATE are integrated into CUMILE (Vortisch and Waßmuth, 2007) for the determination of the driving performance on different road categories (freeway, inner city, out of town). The emission model TREMOD (Transport Emission Model) contains a set of instruments, which allows taking into account many of the relevant parameters (ifeu, 2016). TREMOD uses the highly differentiated database "Manual Emission Factors for Road Traffic" (Keller et al., 2017) as a basis for the emissions calculation, supplemented by specific emission values, derived specifically for the German fleet. Input data come from the differentiated inventory and driving performance data for Germany. For the CUMILE-TREMOD interface we adapted the emission reduction concept composition (emission standards) of the CUMILE fleet to the TREMOD passenger cars fleet, which is described in (ifeu, 2016). We furthermore calculated the cold-start and warm-phase emissions for different traffic situations taking into account the emission standard, standing time, ambient temperature, trip length or total vehicle mileage (see Keller et al., 2017). Thanks to the CUMILE-TREMOD interface, we are able to allocate differentiated energy consumption and emission characteristics to air pollutants (e.g., PM, NOx) and greenhouse gases (e.g., CO2, CH4) to the car trips modelled in CUMILE.

Model application By the use of the CUMILE-TREMOD interface, we analyzed the economic potential and the resulting environmental effects of a substitution of private cars by station-based carsharing. Therefore, we determined the costs of private car use and ownership for every car in the CUMILE fleet, see Kuhnimhof and Eisenmann (2017). We then assessed annual costs in case the private car was abandoned and carsharing is used for trips, which were performed by the private car before under different carsharing scenarios. By comparing annual private car costs and annual carsharing costs, we determined cars, which were feasible for substitution from a financial perspective. Our results show that a substitution of private cars by carsharing is cost-effective for 18% of the German private car fleet, assuming that the same car type as before is used under carsharing and that carsharing members use active modes of transport for short-haul tours (below 5 km). Cars suitable for substitution are often owned by city-dwellers, retirees, young adults without kids and they are often the second or third car in the household. Replacing carsharing by private cars under the assumptions described above results in an energy consumption reduction potential of 1.4%. We found the biggest potential savings for hydrocarbon emissions (9.1%) due to the avoided short-haul car routes. Emission reduction potentials of PM, NOx, CO2 and CO range from 0.9% to 4.4%.

References Bäumer, M., Hautzinger, H., Kuhnimhof, T. and Pfeiffer, M. (2017) The German Vehicle Mileage Survey 2014: Striking the balance between methodological innovation and continuity, 11th International Conference on Transport Survey Methods ed International Steering Committee for Travel Survey Conferences (ISCTSC). Chlond, B., Weiss, C., Heilig, M. and Vortisch, P. (2014) Hybrid Modeling Approach of Car Uses in Germany on Basis of Empirical Data with Different Granularities. Transportation Research Record: Journal of the Transportation Research Board, 67–74. DWD (2016) Climate Data Center FTP-Server. http://www.dwd.de/DE/leistungen/cdcftpmesswerte/cdcftpmesswerte.html?nn=17626. ifeu (2016). Aktualisierung „Daten- und Rechenmodell Energieverbrauch und Schadstoffemissionen des motorisierten Verkehrs in Deutschland 1960-2030“ (TREMOD) für die Emissionsberichterstattung 2016 (Berichtsperiode 1990-2014), 2016. Heidelberg, Im Auftag des Umweltbundesamtes. In Zusammenarbeit mit dem Öko-Institut. infas and DLR (2010). Mobilität in Deutschland (MiD) 2008: Ergebnisbericht Struktur - Aufkommen - Emissionen - Trends. Bonn und Berlin. Keller, M., Hausberger, S., Matzer, C., Wüthrich, P. and Notter, B. (2017). HBEFA Version 3.3. Berne, Background documentation. Berne, 25. April 2017. Vortisch, P. and Waßmuth, V. (2007) VALIDATE - A Nationwide Dynamic Travel Demand Model for Germany, Proceedings of the 11th National Transportation Planning Application Conference of the Transport Research Board. Weiss, C., Chlond, B., Behren, S. von, Hilgert, T. and Vortisch, P. (2016a). Deutsches Mobilitätspanel (MOP) - Wissenschaftliche Begleitung und Auswertungen Bericht 2015/2016: Alltagsmobilität und Fahrleistung. Karlsruhe. Weiss, C., Chlond, B., Heilig, M. and Vortisch, P. (2014) Capturing the Usage of the German Car Fleet for a one Year Period to Evaluate the Suitability of Battery Electric Vehicles - A Model Based Approach. Transportation Research Procedia, 1, 133–141. Weiss, C., Chlond, B., Heilig, M., Wassmuth, V. and Vortisch, P. (2016b) Who Uses Freeways and Who Pays for Them? – A Model Based Analysis of Distribution Effects of Different Toll Tariff Systems in Germany, TRB 95th Annual Meeting Compendium of Papers ed Transportation Research Board. Weiss, C., Chlond, B., Knörr, W., Bergk, F., Kämper, C., Kräck, J. and Vortisch, P. (2017) Modellierung von Nutzungsprofilen und resultierenden Umweltwirkungen der deutschen Pkw-Flotte über ein Jahr. Straßenverkehrstechnik, 523–532. Wermuth, M., Neef, C., Wirth, R., Löhner, H., Hautzinger, H., Stock, W., Pfeiffer, M., Fuchs, M., Lenz, B., Ehrler, V., Schneider, S. and Heinzmann, H.-J. (2012). Kraftfahrzeugverkehr in Deutschland 2010. Schlussbericht. Forschungsprojekt 70.0829/2008 im Auftrag des Bundesministeriums für Verkehr, Bau und Stadtentwicklung. Braunschweig, Heilbronn, Berlin, Flensburg.

ABSTRACT. Aiming at identifying generational differences in the attitudes towards cars, this study developed a latent class model using data collected in Beijing, China. Results suggest young adults have less favorable responses to questions about attitudes towards private cars compared to middle-age adults and senior adults have the least favorable mean responses. Three classes are identified through modeling and are labeled “car enthusiasts”, “car critics”, and “car pragmatists”, respectively. Middle-age adults are found to be the most likely groups to be classified as a “car enthusiast” who report favorable responses to attitudinal questions. Young adults have significantly greater probabilities to be classified as a “car pragmatist” that have generally positive but relatively conservative evaluations on private cars. Few respondents are classified as “car critics” and senior adults are more likely to be labeled into this class. Results are largely in agreement with some previous literature who reported declining attractiveness of cars among younger generations and they suggest the enthusiasm to private cars may also start to cool off in a less car-dominant but rapid developing society.

ABSTRACT. Introduction

This paper develops a methodology for testing and implementing differences in preferences for a set of public transport modes, relating to observed and unobserved attributes, in state-of-practice large-scale travel demand models. Results of a case study for commuters in the Stockholm public transport system suggest that there are preference differences among public transport modes, and that they are captured by unobserved attributes. Surprisingly, we found no evidence for differences proportional to the in-vehicle time, suggesting that characteristics of in-vehicle time in different public transport modes, such as comfort, are valued equally by the travellers. We also found that the value of time is higher for auxiliary modes than for the main mode, and that the unobserved preference for metro is highest and the preference for light rail lowest.

There is a political preference for rail-based Public Transport (PT) modes over bus services. Decision makers in many European countries seem to prefer rail over bus services, claiming that travellers prefer services operating on tracks, referring to this as a “rail factor”. Moreover, property developers often claim that metro investments increase the land values over and above what a bus system with equal capacity and travel times would. If travellers’ preferences differ between PT modes, the treatment of them as the same mode in transport models translates into biased parameter estimates and model predictions. Such bias would then propagate to all types of policy analyses including Cost Benefit Analysis (CBA).

A higher preference for rail-based PT modes is to some extent supported by studies in psychology, transport modelling and economics. For instance, Eliasson (2016) found that accessibility by metro increases the property prices of apartments in Stockholm more than accessibility by bus. In the transport modelling field, Ben-Akiva and Morikawa (2002) explore differences in the preference for rail and bus services in two case studies. They estimate choice models on combined revealed preference (RP) and stated preference (SP) data. The utility function for PT includes dummy variables for each PT mode. The alternative specific constants indicate that metro is most preferred, followed by bus and commuter train. However, level-of-service (LoS) parameters do not differ significantly across PT modes. Scherer et al. (2012) includes a meta-analysis of German and Swiss studies focusing on user perceptions and mental representation of train, tram, and bus. They conclude that there is a rail factor, loaded with emotional and social attributions.

In this study, we develop a method for testing whether travellers’ preferences for PT differ between modes in a state-of-practice large-scale transport model. We show how to implement such differences in the transport model. We find that in the Stockholm public transport system, travellers’ preferences vary across PT modes and travel time components. We also find that the value of time is higher for auxiliary modes than for the main mode, consistent with earlier literature (see the meta-study by Wardman, 2004).

We expand the analysis by Ben-Akiva and Morikawa, by introducing a more flexible model specification tailored to capture differences in observed and unobserved preferences between PT modes. The models are specified to capture the systematic difference in unobserved preferences among PT modes (alternative-specific constants), the correlation of the random errors across the PT modes, and the systematic differences in preference for LoS variables and travel time components. We introduce the definition of main mode, in cases where more than one PT mode is used within the same trip. To our knowledge, no previous study on large-scale models has in this way systematically explored how the preference differs across PT modes in all these dimensions, although large-scale transport model prediction is the cornerstone of transport appraisal.

We stress that the aim of this paper is primarily to develop an approach for testing and implementing systematic differences in preferences among PT modes. Such differences are likely to vary across transport systems, since customer preferences derive from perceptions and beliefs which are influenced by local conditions and cultures (Scherer 2010). Hence, empirical evidence might differ across transport networks and over time.

Method In order to properly model different PT modes, a main mode needs to be defined. Unless one defines a main mode, then, as in Ben-Akiva and Morikawa 2002, the model estimation needs to be constrained to use “pure” observations only – observations where only one PT mode is used – neglecting all PT intermodal observations. Hence, in order to use as much information as possible to investigate nuances in the preference for different PT alternatives, we need to define a main mode.

In this study we defined the main mode of each trip as the mode used for the longest distance. We applied this definition for two reasons. First, a definition based on distance, as opposed to travel time, will not change with traffic conditions. Second, it is consistent with how the main mode was defined in the national travel survey data used to estimate the model, allowing us to cross-check whether the user-reported main mode and the main mode imputed from network attributes are the same. Observations where the main modes differ are discarded for this study.

As a starting point, we define a state-of-practice large-scale mode choice model including commuting and education trips, with utility function defined by (1). It includes the alternatives PT (with the sub-alternatives metro, commuting train, bus and tram, all four as one single alternative), car driver, car passenger, walk and cycle. The observed part of the utility function for PT includes In-Vehicle Time (IVT), initial waiting time, access/egress time, transfer penalties and trip cost. The utility functions for the car alternatives include travel time and travel cost, and dummy variables indicating car ownership and gender. The car driver alternative also includes a variable indicating car competition in the household.

Now, we formulate a sequence of hypotheses regarding the model specification and set up models to test them using statistical tests. A key issue is whether to model the different public transport as a single alternative or as separate alternatives. This depends on whether all or some of the unobserved preferences for the PT modes, captured by the random error, are equal. If they are independent, the PT modes should be modelled as different alternatives in an MNL model, if they are identical, they should be modelled as a single alternative, and if they are correlated, they should be modelled as nested alternatives in a NL model. There are reasons why the error terms of the PT modes should be equal: they share several unobserved attributes, comfort and characteristics, they are all public transport. However, it is also possible that they are different, for instance since the comfort, crowding and reliability may be different.

To explore the correlation of the random errors, we set up three hypothesis tests concerning the error term assumptions. Hypothesis 1 - test 1, tests the assumption that the error terms of the PT modes are correlated against the assumption that they are identical. It is tested by estimating a Mixed Logit (ML) and comparing it to a MNL model. Hypothesis 1 – test 2 is similar but tests the assumption that the random errors are independent against the assumption that they are identical, applying MNL models only. Hypothesis 5 tests whether the random error terms of the PT modes are correlated against the assumption that they are independent. It is tested by testing a NL model against an MNL model.

Conclusion

We find evidence that mode choice constants are significantly different among the PT modes. Hence, the average effect of the unobserved attributes differs significantly across the modes. Results suggest that, ceteris paribus, the preference for metro is highest and the preference for tram lowest.

Furthermore, we find that the value of time is higher for auxiliary modes than for the main mode. This finding suggests that comfort differences, as for instance the ability to use the time productively for other activities when travel time is short, penalises the perceived utility. Nevertheless, we also find that systematic preferences among the PT alternatives are not proportional to the travel time. This suggests that the comfort is similar across the PT modes. SLL (2015) shows that all public transport modes suffer from crowding during the peak hour, although the crowding levels vary substantially between services within the system.

Regarding the existence of a rail factor, we find evidence to support the hypothesis that some rail-based modes - metro and commuter trains - have in fact a higher alternative specific constant, indicating that the average effect of the unobserved attributes makes them preferable to bus and tram, ceteris paribus.

We can reject the hypothesis that the PT modes have the same random error. This implies that, as long as we use MNL models, the model assuming that the public transport main modes are different alternatives outperforms the model assuming that they are the same alternative. In addition, we observe how the nested model outperforms the logit model with an identical utility specification, indicating that the error components of the PT alternatives are in fact correlated, confirming a priori expectations.

The simulated PT price elasticities yield values consistent with the international literature for bus, metro and train. The resulting values of time are also consistent with previous findings, suggesting the adequacy of the models.

Models able to capture and exploit nuances in the preferences for different PT alternatives may be important for policy analysis and detailed evaluation of infrastructure investments. Assuming travellers’ preferences for different modes are equal, when in fact they are different, will translate into biased parameter estimates and forecasts. This study shows how current state-of-practice large-scale transport models can be enhanced to investigate these issues in further detail, and we find empirical evidence in our case study that supports both the manifest political preference for rail-based PT modes and Eliasson’s (2016) finding that accessibility by metro increases the property prices of apartments in Stockholm more than accessibility by bus.

ABSTRACT. Although many papers about conducting onboard activities start from the assumption that these reduce the value of time (VoT), surprisingly limited empirical evidence is provided for this assumption. This evidence is hard to find in regular cross sectional research due to a self-selection problem: Travelers that are time pressured are more likely to work on board, which may reduce their VoT, but their VoT may remain relatively high. As a solution for this problem, Wardman and Lyons (2016) propose to compare the VoT estimated from choices made by the same persons observed under two different conditions: travelers can conduct their favorite activity or they cannot. The difference between both VoT estimates can be regarded as the value of activities (VoA). Our paper is the first to apply this approach. The results provide evidence that conducting onboard activities indeed reduces the VoT. Furthermore, the VoA has the expected positive value. The paper presents VoA values for three common in train activities separate for commuters and leisure travelers as well as the VoT in the activity and in the non-activity condition. Furthermore, VoA and VoT relations with socio-demographic and trip characteristics are explored. Finally, an endowment effect is discussed: VoA is found to be higher for travelers that first make choices in the activity condition compared to travelers who first make choices in the non-activity condition. Implications for policymakers, transport providers and scholars are discussed as well implications for the Automated Vehicle (AV) area, because just like trains, AV’s allow conducting onboard activities.

ABSTRACT. Prospect Theory (PT) has been successfully applied to represent monetary decisions under risk but it has repeatedly failed in time-related decisions. In monetary gambles, risk aversion can be easily elicited by asking individuals to choose between a prospect with two outcomes that are greater than a certain reference point together with the certainty equivalent of that prospect. Risk aversion can be then explained with the concave gain function proposed by PT and - without loss of generality - a reference point set to zero. In time-related decisions, in contrast, we posit that risk aversion arises from reference points that are formed while individuals make risky choices. Using data from a route choice experiment, we estimated via maximum log-likelihood all the PT components under different specifications for the reference point. The PT model was developed for time prospects with two attributes (waiting and in-vehicle times), where each attribute can take at most two values (and with the same probability of occurrence). Our results support our prior expectations about the role of reference points as drivers of risk aversion in time-related decisions.

ABSTRACT. Priced freeway lanes, which now exist in ten states and are becoming increasingly prevalent, can improve drivers’ welfare overall by addressing heterogeneity in their willingness to pay for time savings. (Small and Yan 2001, Arnott et al. 2002). Users who value time savings more highly can pay to access a lane(s) with faster service, while those who value them less get longer travel times but avoid tolls (Small and Yan 2001). In practice, however, welfare-maximizing pricing is not implemented. Instead, express lanes are dynamically priced to maintain a minimum free-flow speed. This pricing regime often results in significant underutilization of the express lane, and may be less efficient than no pricing at all (Small and Yan 2001). Welfare-maximizing tolls have been the subject of a wide body of theoretical work including studies by Small and Yan (2001), Verhoef and Small (1999), and Arnott et al. (2002). They consist of single price applied to one or more lanes that achieves an allocation of drivers between the toll and free lanes such that the sum of congestion costs across all users is minimized. A crucial assumption made in these works is that toll operators know the exact distribution of values that each user places on travel time savings. In reality, however, welfare-maximizing tolls are not implemented because transportation agencies lack the means to determine each drivers’ willingness to pay. Without accurate estimates for these values, a toll will likely be set either too high or too low, and result in inefficient utilization of the fast lane(s). Currently, achieving sufficient knowledge of these driver preferences is exceedingly difficult. Values of time can vary dramatically by person, trip purpose, time of day, day of week, and locale (Koppelman 2012, Wardman 2007, Parkany 1999, Sullivan 1998). Only limited survey and field data are available for determining driver values of time (VOT) for a particular corridor served by a high-occupancy toll (HOT) lane, and VOT estimates can vary significantly for a specific locale based on data collection and model specification (Koppelman 2012). Furthermore, VOT and corresponding parameter estimates, from which they could be forecasted, vary from region to region (Koppelman 2012). A significant amount of effort to determine a VOT distribution is required anytime HOT lanes are introduced to a new area. Lastly, VOT has been shown to vary based on congestion levels (Koppelman 2012). This means that a typical VOT distribution may be invalid for HOT lane pricing during periods when non-recurring incidents or events result in local congestion; posing an additional challenge for incident management using lane. An alternative pricing scheme that uses Vickrey-Clarke-Groves (VCG) mechanism to determine tolls and lane assignments could obviate the need to estimate valuations of time savings by eliciting them directly from users. VCG mechanisms select socially optimal outcomes by incentivizing participants to truthfully reveal their preferences. Their design ensures that no matter which outcome is selected, no participant can increase their utility by providing non-truthful preferences. If the mechanism was used to manage a toll lane, it would be in each driver’s best interest provide their truthful travel time preferences – remotely relayed through connected vehicle technology which is becoming increasingly prevalent. These preferences could then be to calculate (and assign to drivers) a lane allocation that achieves essentially the same welfare benefits as second-best tolling. The use of VCG mechanisms in practice is limited, in part because there is mixed evidence regarding the extent to which these mechanisms able to elicit truthful preferences from the average user. One notable application of the mechanism is by the application Facebook, which uses it to sell ads, but its use in real-world transportation settings is non-existent. This work explores the feasibility of Vickrey-Clarke-Groves mechanisms for allocating freeway lane capacity to drivers. The primary aim is to understand the extent to which actual human users will provide truthful travel time preferences in practice, which will in turn determine the extent to which the theoretical welfare benefits of mechanism implementation can be captured. Although VCG mechanisms incentivize truthful preference revelation, no experimental study of them thus far has achieved perfect revelation from subjects. VCG mechanisms are extremely difficult to grasp conceptually, especially the strategic dominance of truth-telling. Furthermore, most new processes require some degree of learning. Travelers and participants in other strategic endeavors in general have a strong tendency to experiment and explore, which would result in misrevelation - at least initially. This misrevelation could also persist, however, because sub-optimal equilibria are possible in many applications of the mechanism. Perfect truth-telling is not required to achieve near-optimal outcomes, however; the mechanism outperforms alternative allocations schemes in experiments by Healy (2006), Brenner and Morgan (1997), and McLaughlin & Friedman (2016), despite some degree of misrevelation by subjects. In other studies, however, misrevelation is so substantial that the mechanism fails to increase user welfare over that of other schemes (Attiyeh et al. 2000, Kawagoe and Mori 2001). Freeway capacity allocation shares some traits with applications in prior experiments where the VCG outperformed other alternatives, and traits with those where it didn’t. This is discussed in greater detail in the literature review, however the takeaway is that there is uncertainty a priori regarding the potential empirical performance of the mechanism. Furthermore, this application is unique in that travel time savings are allocated rather than goods. It is possible that humans have a different understanding of their willingness to pay for a good or service as opposed to travel time. In addition, there are significant impediments to myopically learning truth-telling in the context of lane assignment. One is that the mechanism relies on congestion functions to predict average travel times for a lane as a function of volume, but deviations may occur due to idiosyncratic driver behavior. This random variation in travel times that will distort drivers’ perceived relationship between the willingness to pay they provide and the travel times they experience. Another is that misrevelation resulting in sub-optimal outcomes can be an equilibrium, which is explained in greater depth in the literature review. To study the workability of using a VCG mechanism to allocate freeway lane capacity, a traffic experiment was developed consists of an interactive, multi-user driving simulator where human subjects travel a freeway with an express lane. I attempt to allocate the experimental subjects between the express and general lanes using an optimal tolling scheme where users reveal their valuation of the road through a VCG mechanism before driving. I work from a framework where individuals have heterogenous preferences that are unknown to the regulator, preventing the optimal allocation of individuals across lanes. While real-world travel time preferences consist of many factors including value of time, value of reliability, and urgency costs, the experiment incentivizes subjects such that the only relevant travel time preference is value of time. This simplification greatly enhances the ease of observing the truthfulness/accuracy of revealed preferences, as well as calculating the efficiency of the resulting allocation. This work is the first to empirically demonstrate that a VCG mechanism can elicit truthful preferences in setting where travel time is the allocated good. Despite the strong overall correlation between subjects’ true and stated VOT, however, only 34% of bids were truthful, and the same degree of misrevelation persisted after 10 rounds. This work finds strong evidence that these deviations from truth-telling were due in part to limited subject understand of the mechanism, as well as stochasticity in travel time outcomes. Nevertheless, I show that given the amount of truth-telling observed, the mechanism may still dominate alternative pricing schemes in terms of driver welfare.

ABSTRACT. Over the past decades, the quality of time spent travelling has been shaped more and more by the quality of on-board/in-vehicle facilities enabling convenient use of mobile devices and participation in digital activities. However, due to nature of the available technology, in the contexts examined in existing studies the provision of on-board connectivity has been limited up to about 40Mbps. This is clearly insufficient for the growing needs resulting from data hungry services such as media streaming or cloud-based storage, requiring ca. 1.5-2 Mbps, especially in the context of transport modes carrying hundreds of individuals simultaneously.To address the challenges of poor connectivity on-board trains, a collaborative project SWIFT led by Cisco and with ScotRail to trial the technology of achieving consistent and reliable bandwidth above 300 Mbps on a train moving at 100 mph and above. Part of Project SWIFT looks at collecting pre- and post-implementation on-board survey which includes a detailed set of questions regarding participation in digital activities and the experienced productivity. This offers a unique opportunity, to date absent from the literature, for understanding how a step-change in connectivity enables more sophisticated digital behaviour and what its implications on productivity are.

In particular, this paper aims to analyse impact of providing connectivity with bandwidth of an order of magnitude higher to passengers, by using a combination of on-board survey data collected pre- and post-implementation and secondary data from NRPS. To achieve this aim, two research objectives have been defined. In the first one, a model of overall trip satisfaction as a function of data and voice connectivity as well as controlling variables (journey context, delays, crowding, geographical region) is estimated using NRPS Autumn 2015 data. Given the national extent of the NRPS data, realisation of this research objective will provide insights into the role of voice and data connectivity in determining travel satisfaction across the entire UK’s rail network, controlling for the aforementioned factors. The resulting benefit is two-fold. Firstly, the analysis will reveal relative importance of various factors, including voice and data connectivity, in influencing the overall travel satisfaction, and hence possibly guide rail industry’s efforts aimed at improving passenger satisfaction. Secondly, the analysis will serve as a benchmark for assessing the degree to which the dedicated data collection resulting from Project SWIFT and the associated analysis can be representative of the conditions observed across the entire UK rail network. The second objective of the study, relying on data collected from Project SWIFT is to develop joint travel time use, productivity, and ridership model, and thereby obtaining additional insights with respect to interaction between those three aspects of travel behaviour. This will draw upon methodological developments in this space proposed recently by the authors, including an underlying microeconomic framework as well as the accompanying econometric formulation using a copula approach to modelling travel time use and productivity jointly. In addition, the on-board survey contains a much richer set of online activities for a respondent to choose from, as compared to surveys to date, including NRPS or time use surveys. Consequently, the analysis will contribute to the broader field of analysing the relationship between participation in digital activities and productivity, to date explored almost exclusively in the context of adoption of information and communication technologies by companies.