Development of a latent variable model to capture the impact of risk aversion on travelers' switching behavior

Tsirimpa, Athena; Polydoropoulou, Amalia; Antoniou, Constantinos Article Development of a latent variable model to capture the impact of risk aversion on travelers' switching behavior Journal of Choice Modelling Provided in Cooperation with: Journal of Choice Modelling Suggested Citation: Tsirimpa, Athena; Polydoropoulou, Amalia; Antoniou, Constantinos (2010) : Development of a latent variable model to capture the impact of risk aversion on travelers' switching behavior, Journal of Choice Modelling, ISSN 1755-5345, University of Leeds, Institute for Transport Studies, Leeds, Vol. 3, Iss. 1, pp. 127-148 This Version is available at: https://hdl.handle.net/10419/66815 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc/2.0/uk/ Journal of Choice Modelling, 3(1), pp. 127-148 www.jocm.org.uk Development of a Latent Variable Model to Capture the Impact of Risk Aversion on Travelers’ Switching Behavior Athena Tsirimpa1,* Amalia Polydoropoulou1,+ Constantinos Antoniou2,∞ 1Department of Shipping, Trade and Transport, University of the Aegean, Korai 2a, Chios, Greece, 82100 2National Technical University of Athens, Laboratory of Transportation Engineering, School of Rural and Surveying Engineering 9 Heroon Politechniou st., 15780-Zografou, Athens, Greece Received 13 March 2008, received version revised 21 November 2008, accepted 19 September 2009 Abstract Advanced Traveler Information Systems (ATIS) are becoming increasingly available throughout the world. While the impact of the provided information on the switching behavior has been investigated in the past, an area of research that is less well understood relates to the effect of the travelers’ risk aversion (or riskseeking) in their travel behavior. The objective of this research is to examine the impact of information acquisition on travelers’ switching travel behavior and to identify and quantify the role of attitudes and perceptions on switching behavior. A combined choice and latent variable model has been developed, in which the individual traveler’s risk aversion has been modeled as a latent variable. The model has been estimated using data collected through travel diaries in the Puget Sound Region (PSRC) in 2000. As expected, travelers in general tend to maintain their habitual travel pattern. However, specific travel information –such as that regarding an incident or road closure– influences behavioral switches such as departure time change and route change. Keywords: Advanced Traveler Information Systems (ATIS), Risk Attitudes, Integrated Choice and Latent Variable Model * Corresponding Author, T: +30 22710 35263, F: +30 22710 35299, a.tsirim[email protected] + T: +30 22710 35263, F: +30 22710 35299, [email protected] ∞T: +30 210 7722783, F: +30 210 7722629, [email protected] Tsirimpa et al., Journal of Choice Modelling, 3(1), pp. 127-148 128 1 Introduction The continuous growth of demand for road traffic increases the delays that road users face and negatively affects the overall transportation system performance. Traffic information provision may offer significant benefits in terms of improving the travel experience of individuals and overall system performance. However, the impact of traffic information on travelers' behavior is difficult to predict and may result in additional issues such as overreaction. Information acquisition may update travelers' perceptions of travel alternatives in two ways. First, it may increase travellers’ awareness of alternatives by introducing them to the traveller or by explaining them and second, it may alter travellers’ perception of the characteristics of travel alternatives. Eventually, travel information may, through the updating of perceptions, influence travellers’ choice-behaviour (Chorus et al. 2006a). Bonsall (2004) discusses the nature and consequences of uncertainty in transport systems. The principal theories and models used to predict travellers’ response to uncertainty are presented and a number of alternative modelling approaches (including random error components) are outlined. The term ATIS has been used to describe a wide variety of services and systems with very different philosophy, scope, as well as operating characteristics. After many years of ATIS research, and many successful (and less successful) implementations, there is today a considerable amount of knowledge accumulated on the subject. Two thorough reviews on the subject (Lappin and Bottom 2003, and Chorus et al. 2006a) both agree on the fact that ATIS are a promising research direction. Chorus et al. (2006a) present a framework for a next-generation ATIS. Perhaps the biggest concern in the evaluation of ATIS is the difficulty associated with collecting high-quality, appropriate data sets, with the suitable parameters and size for their effective validation. This paper presents a case study for the Puget Sound Region (PSRC), where the Regional Council has been running, since 1989, the longest continuous survey in the United States, regarding travel behavior. A supplement has added to the travel diaries, since 2000, asking individuals about the traveler information sources consulted on each trip and how the information was used. Traveler information sources, available in the region, encompass both conventional forms of information, such as radio traffic reports, as well as advanced traveler information systems, such as variable message signs (VMS) and web sites. In this paper, these data are used to model the impact of information acquisition on switching travel behavior (for more information regarding Puget Sound Regional Travel Survey, see Goulias et al. 2003). Besides the responses stated directly in the travel diaries, in this research latent variables capturing unobserved characteristics of the travelers are also modeled. The remainder of this paper is organized as follows. Section two presents a brief review of the state-of-the-art of modeling ATIS impact on travelers’ behavior. Section three presents a behavioral framework that incorporates the effect of information acquisition on the switching behavior of individuals from their usual travel pattern and the modeling methodology used. Section four presents the data used for the model development. Section five presents the model specification and estimation results and section six presents the conclusions. Tsirimpa et al., Journal of Choice Modelling, 3(1), pp. 127-148 129 2 State of the Art Many researchers have studied the impact of advanced traveler information systems on travelers' decision-making behavior. A review of the state-of-the-art on modeling travelers' response to ATIS is presented in the remainder of this section. This review does not aim to be exhaustive, but instead strives to provide the necessary background, motivate and support the subsequent methodology and application. An exhaustive review of the literature on the assessment and prediction of drivers' response to information may be found in Lappin and Bottom (2003). ATIS are likely to influence a variety of travel decisions (such as mode, route, and departure time), lifestyle decisions (location of residence, car ownership) and activity participation decisions (work, shopping) (Polydoropoulou and Ben-Akiva 1999). Toledo and Beinhaker (2006) evaluate the potential travel time-savings from Advanced Traveler Information Systems (ATIS) that provide drivers with travel time and routing information. A case study, using real-world data collected from a freeway network in Los Angeles, California, examines the potential travel time savings of ATIS as well as the implications on travel time variability and reliability and the sensitivity of the results to the accuracy of the information. Sources of information applicable to modeling the users’ response to ATIS include (a) data from field experiments, travel surveys and diaries, and (b) data from simulators. 2.1 Travel Surveys, Diaries and Field Experiments The most traditional method of obtaining travel data is through surveys and field experiments. Collected data may be either Revealed Preferences (RP) data or Stated Preferences (SP) data. Abdel-Aty et al. (1997) studied commuters’ route choice including the effect of traffic information. Two route choice models were estimated. The first model used five hypothetical binary choice sets collected in a computer-aided telephone interview. The results yielded important insights on the commuters’ route choice in general, and the tradeoffs involved in the choice between a route which is longer but has reliable travel time versus another route which is shorter but has uncertain travel time. The model showed that both expected travel time and variation in travel time influence route choice. Commuters’ attitudes toward several commute characteristics (e.g. distance and traffic safety) also influenced route choice, as well as socioeconomic factors, in particular gender. Receiving traffic information is found to have a significant effect in the model. Information might be used by commuters to reduce the degree of travel time uncertainty, and enables them to choose routes adaptively. The second model used data collected in a mail survey from three binary route choice stated preference scenarios customized according to each respondent’s actual commute route and travel time. The results of the second model asserted the significance of travel time reliability on route choice, and showed clearly that ATIS has great potential in influencing commuters' route choice even when advising a route different from the habitual. Several other commute factors were found to affect the route choice, including freeway use and travel time. The correlation among error components in repeated measurement data was also addressed with individual-specific random error components in a binary logit model with normal mixing distribution. Khattak and Khattak (1998) conducted mail surveys in the San Francisco and Chicago areas and asked respondents about the effects of en-route travel information on their trip-making decisions. About 16% of the respondents in San Francisco and Tsirimpa et al., Journal of Choice Modelling, 3(1), pp. 127-148 130 42% in Chicago had diverted within the past three months in response to their most recent unexpected traffic delay. It was found that automobile commuters’ cognitive maps are influenced by duration of residence, personality, and location characteristics. In addition, travel time on both usual and alternate routes, reception of delay information via radio, and personality aspects are important attributes in drivers’ enroute diversion behavior. Hato et al. (1999) conducted a survey targeting drivers traveling on the Tokyo Metropolitan Expressway network, where drivers can actually make use of traffic information from multiple sources when choosing their route. Information acquisition behavior seems to be influenced by the following latent variables: (1) drivers aggressiveness towards route choice matters, (2) drivers attitude towards warnings concerning traffic condition and (3) familiarity with the road network and ability to use traffic information efficiently, which are determined by individual's driving experience and individual characteristics. Khattak et al. (1999) conducted telephone surveys in the San Francisco area and asked respondents about the effects of pre-trip travel information. The propensity to adjust pre-trip travel decisions based on travel information is highest for work trips, compared to a base of trips for school, shopping and other personal trips. Individuals who experience higher travel time uncertainty and reported occurrence of unexpected delays during the past month have a higher propensity for pre-trip decision changes in response to travel information. Non-commuters and radio listeners have a higher tendency to cancel their trips in response to information. Dia (2002) conducted a survey with mail-back questionnaires that were distributed to peak-period automobile commuters traveling along a traffic corridor in Brisbane, in order to investigate the individual driver behavior under the influence of real traffic information. Decisions that were investigated included pre-trip response to unexpected congestion information, en-route response to unexpected congestion information, and willingness to change driving patterns. On average, respondents took an alternative route 3.5 times per month. Response to information seemed to be influenced significantly by the information type and the way that it is provided to the individuals. Kyoung-Sik (2003) conducted a survey for morning commuters in Daegu (Korea) in order to investigate the effect of pre-trip information acquisition. A large portion of drivers change route (43%) or departure time in response to the acquired information. Drivers who have university degrees, professional or self-employed jobs, are more likely to change their route in response to pre-trip traffic information. Mahmassani et al. (2003) conducted web-based stated preference experiments to study the impact of ATIS on travelers’ decisions about a shopping trip. ATIS could be provided to the user at a pre-trip and en-route level. Discrete choice models were developed to examine all factors that make commuters change to an alternative destination or route. The fundamental difficulty in modelling this problem is due to the structure of the survey where the information provided and user choices are interdependent. The results indicated that gender, age, level of education and high income is not statistically significant in explaining either route or destination switching. Familiarity with the area, and number of frequent visits to a specific mall, are less related factors that influence the change of destination or road. On the other hand, provided information which concerns delays or traffic jams can make tripmakers to switch road. The model developed by Mahmassani et al. (2003) also accounts for heteroskedasticity and correlation between the decision states and trip dimensions. While the data suggests there is no difference in variances in the error Tsirimpa et al., Journal of Choice Modelling, 3(1), pp. 127-148 131 terms, with respect to the first choice, correlation exists among the decision states and along the trip dimensions. The inclusion of such correlation significantly improves the model. Bierlaire and Thémans (2005) present (a) a Mixed Multinomial Logit model with panel data to analyze the drivers’ decisions when traffic information is provided during their trip through Radio Data System (RDS) or variable message signs (VMS) and (b) a Nested Logit model capturing the behavior of drivers when they are aware of traffic conditions before their trip. The models are estimated using data from a twoyear national survey in Switzerland during which both Revealed Preferences (RP) and Stated Preferences (SP) data about choice decisions in terms of route and mode were collected. It was found that people who use the Internet to access traffic information and those who are aware of alternate routes have a propensity to switch routes. Van de Horst and Ettema (2005) conducted an internet survey to study travel information acquisition and mode choice decisions. The results of the survey showed that younger travelers are more inclined to retrieve any kind of information than older travelers. Between 20% and 57% of the public transport travelers and 50% to 62% of the car travelers sought information about aspects of the competing mode. Many travelers make the choice to use the car deliberately and the availability of information appears to play an important role in the decision process of non-captive travelers. Jou et al. (2005) interviewed freeway travelers from Taichung to Taiwan in order to study route-switching behavior. Real-time traffic information with guidance was preferable to freeway travelers, as was the quantitative type of real-time traffic information. The model results indicate that male travelers and travelers with a higher income would be more likely to switch to the best route, while elderly travelers would be less likely to switch due to habitual and risk-aversive effects. Tsirimpa et al. (2007) developed a mixed Multinomial logit model to capture the impact of ATIS on travelers' switching behavior that accounts for correlation among observations from the same individuals in the data set. The model was applied in a data-set from the Puget Sound Region (PSRC). The estimated models show that travel pattern characteristics, the time of information acquisition (pre-trip vs. en-route), the source and the content of provided information significantly affect commuters’ response to information. Farag and Lyons (2008) studied the factors that influence the pre-trip use of public transport information services (via different media - Internet, phone, paper timetables, asking staff). A social-psychological perspective was adopted which takes habit, attitudes, anticipated emotions, and perceived behavioural control into account. The results showed that social-psychological factors are important determinants of travel information use, while external factors such as trip context could affect these determinants. Zhang et al. (2008) examined the effect of real-time transit information on travellers’’ behaviour and psychology, using the data of a campus transportation panel survey. Two fixed-effects models and five random-effects ordered probit models were estimated to sort out causal relations between information use and two behavioural and five psychological indicators respectively. The results showed that the use of realtime information significantly increased rider’s feeling of security about riding bus after dark and boosted their overall satisfaction level. Choocharukul (2008) studied the interrelationships among the likelihood of making route diversion, attitudinal variables, and several exogenous factors such as socioeconomic and travel characteristics of the motorists. A structural equation model was developed based on empirical data of road users in Bangkok. Modelling results Tsirimpa et al., Journal of Choice Modelling, 3(1), pp. 127-148 132 indicate a direct relationship between stated route diversion and two of the attitudinal variables, i.e. VMS comprehension and perceived usefulness of VMS, while the awareness of VMS is not found to be a direct determinant for route diversion decision. Unlike past studies, none of the socioeconomic variables appeared to directly influence route diversion intentions. Gan et al. (2008) conducted a quantitative assessment of the potential effects of Variable Message Signs (VMS) information, displaying travel times on both original and alternate routes, on drivers’ en-route diversion behaviour. Based on a stated preference (SP) survey on freeway drivers in Shanghai urban area, three types of binary probit models were estimated. The results showed that the impact of VMS information varies significantly across the characteristics of driver, route and VMS message of travel time. Travel time saving and drivers’ driving age serve as positive factors in drivers’ diversion behaviour. 2.2 Simulators A number of studies have focused on the use of travel simulators to obtain data, which indicates how drivers would behave under various scenarios of information provision. Mahmassani and Liu (1999) investigated departure time and route switching decisions made by commuters, in response to ATIS. The data was collected using a dynamic interactive travel simulator for laboratory studies of user response under realtime information. The experiment involved actual commuters who simultaneously interacted with each other within a simulated traffic corridor that consisted of alternative travel facilities with differing characteristics. In the pre-trip departure time switching decision, older commuters tended to tolerate greater schedule delay than younger ones. Also, female commuters exhibited a wider mean indifference band than male commuters for pre-trip departure time and route decisions as well as en-route path switching decision. The reliability of the real-time information is a significant variable that influences commuters' pre-trip departure time and route switching decisions as well as en-route path switching decisions. Travelers become more prone to switch routes when they perceive late arrival by following the current path than when they perceive early arrival by following the current path. Abdel-Aty and Abdalla (2004) used a travel simulator (OTESP) to collect dynamic pre trip and en-route route choice data, in order to model the factors that affect drivers’ compliance with a long-term pre trip advised route and drivers’ usage of en route short-term traffic information. OTESP provided five different scenarios (levels) of traffic information to the subjects (pre trip information with and without advice, en-route information, keeping the pre trip information, with and without advice), besides the base-case of no traffic information, Each subject was required to choose his/her link-by-link route from a specified origin to a specified destination. A real network with historical congestion levels and weather conditions were used, and two models were developed. Generalized estimating equations (GEEs) with repeated observations and a binomial probit link function model were developed. The analysis showed that: (a) the pre trip advice has a good chance to be followed and is more beneficial compared with advice-free pre trip information and (b) the en-route shortterm information provision increases the likelihood of making a positive link choice. Chorus et al. (2006b) present a computer-based travel simulator for collecting data on the use of ATIS and their effects of travelers' decision making in a multimodal travel environment. The decision maker is presented with an abstract multimodal transport network, where knowledge levels are fully controlled for in terms of Tsirimpa et al., Journal of Choice Modelling, 3(1), pp. 127-148 133 awareness of mode-route combinations as well as in terms of characteristics of known alternatives. Difference types of information provision with varying levels of reliability can be provided. The validation results suggest that the simulator succeeds in collecting data with a high validity on multimodal travel choice making under provision of advanced types of travel information. Pan and Khattak (2008) explored the impacts of electronic traveler information on commercial and non-commercial users. By combining a behavioral model with a simulation tool, they studied whether traveler information can impact travel costs when (a) commercial truck percentages increase in traffic, (b) truck drivers divert to alternate routes in the same way as motorists do, and (c) when commercial trucks have relatively higher values of time. The analysis showed that by increasing the percentage of electronic traffic information dissemination in incident conditions, the network average travel time and total travel cost can be reduced up to 9%. The study also found that savings associated with electronic information are highly context dependent, i.e. they can be almost wiped out if drivers are able to observe traffic congestion. Marchal and de Palma (2008) proposed a new method for measuring the impacts of non-recurrent congestion on travel costs by taking risk aversion into account. The traffic model used was based on the dynamic traffic simulations model METROPOLIS. Incidents were generated randomly by reducing the capacity of the network, while users could instantaneously adapt to the unexpected travel conditions or change their behavior via a day-to-day adjustment process. The main finding of this study is that the utility loss due to uncertainty is of the same order of magnitude as the total travel costs. Han et al. (2008) investigated the effects of recommendation with different underlying control objectives on route choice under uncertainty via a computer experiment. The results indicated that when anticipating potential congestion, travelers use the provided recommendation as an indicator of the choices of other travelers as they conjecture the compliance rate to reduce the uncertainty when making decisions. While actual results tend to vary among case studies, the information acquired by travelers (whether pre-trip or en-route) seems to play an important role on travelers switching behavior in almost all cases, especially when it involves traffic congestion and delays. In addition, the reliability and comprehension of the provided information is also crucial for ATIS usage and switch related decisions. Travelers’ socioeconomic characteristics are not always found significant (e.g. Mahmassani et al. 2003; Choocharukul 2008), but in the cases that they are, the most important are gender, age, education level and income. In most of the cases, young professional males with university degrees and high incomes are more inclined to retrieve traffic information than others. In addition, minimizing travel time uncertainty seems to be one of the most important stimulants for ATIS usage, while the propensity of switching increases when travel time uncertainty is combined with reported delays from ATIS. According to Marchal and de Palma (2008), the utility loss due to uncertainty has the same order of magnitude with the total travel cost. Moreover, past travel experiences as well as attitudes and perceptions were found to significantly affect commuters’ behavior. Recent studies (e.g. Farag and Lyons 2008; Choocharukul 2008) showed that social and psychological factors affecting decisions and choices play an important role both in, ATIS usage and switching behavior, necessitating further research in this field. As analysed by Schwartz (2004) information gathering, quality and quantity of information, evaluation of alternatives, availability of alternatives, anchoring and frames, as well as Tsirimpa et al., Journal of Choice Modelling, 3(1), pp. 127-148 134 8 134 comparisons of alternatives are notions to be taken into account when studying peoples’ choices and decision making. Furthermore, it should be noted that most people tend to be risk averse when they are contemplating a choice between a certain small gain and an uncertain large one. Risk will be studied and modeled in detail in this research. comparisons of alternatives are notions to be taken into account when studying peoples’ choices and decision making. Furthermore, it should be noted that most people tend to be risk averse when they are contemplating a choice between a certain small gain and an uncertain large one. Risk will be studied and modeled in detail in this research. 3 Methodological Framework 3 Methodological Framework Travelers’ response to traffic information consists of a series of actions and decisions occurring over time. Figure 1 describes the modeling framework of the switching from the habitual travel behavior of commuters in the presence of traveler information. The attributes that influence travelers’ decision-making patterns can be broadly categorized into four groups. The first group consists of socio-economic characteristics, such as gender and age. The second group includes variables, which express habitual travel pattern, such as travel time, trip purpose and number of trips per day. The third group includes technology characteristics, such as access to the Internet, high-speed connections, and cellular phone ownership of the travelers. Finally the fourth group includes attitudes and perceptions, e.g. individual attitudes towards willingness to change travel patterns. It has been found that these are the main categories of independent variables that strongly affect travel decisions under the influence of ATIS (see for example Polydoropoulou et al. 1994; 1996; and Polydoropoulou and BenAkiva 1999). Travelers’ response to traffic information consists of a series of actions and decisions occurring over time. Figure 1 describes the modeling framework of the switching from the habitual travel behavior of commuters in the presence of traveler information. The attributes that influence travelers’ decision-making patterns can be broadly categorized into four groups. The first group consists of socio-economic characteristics, such as gender and age. The second group includes variables, which express habitual travel pattern, such as travel time, trip purpose and number of trips per day. The third group includes technology characteristics, such as access to the Internet, high-speed connections, and cellular phone ownership of the travelers. Finally the fourth group includes attitudes and perceptions, e.g. individual attitudes towards willingness to change travel patterns. It has been found that these are the main categories of independent variables that strongly affect travel decisions under the influence of ATIS (see for example Polydoropoulou et al. 1994; 1996; and Polydoropoulou and BenAkiva 1999). Figure 1 presents the modeling framework of individual’s switching behavior. In this figure ellipses represent variables that are not directly observable and therefore called latent variables. Rectangles represent observable variables, either explanatory or indicators of the latent variables. Figure 1 presents the modeling framework of individual’s switching behavior. In this figure ellipses represent variables that are not directly observable and therefore called latent variables. Rectangles represent observable variables, either explanatory or indicators of the latent variables. The selected model specification is a standard linear-in-the-parameters specification, used in the vast majority of such models. The actual choice of variables is limited by data availability and postulated based on a priori expectations. The model specification has been refined based on statistical tests on estimation results of alternative considered models. The integrated model consists of two parts: a discrete The selected model specification is a standard linear-in-the-parameters specification, used in the vast majority of such models. The actual choice of variables is limited by data availability and postulated based on a priori expectations. The model specification has been refined based on statistical tests on estimation results of alternative considered models. The integrated model consists of two parts: a discrete Socioeconomic Characteristics, Travel Pattern, Indicators of Risk, Risk, * 2 Z Figure 1. Modeling Framework for Switching with Latent Attributes Figure 1. Modeling Framework for Switching with Latent Attributes n X Z I Utility of Switching, U Revealed Preference y, (Switch) Tsirimpa et al, Journal of Choice Modelling, 3(1), pp. 127-148 Table 5. Specification Table of the Choice Model α01 α02 β1 β2 β3 β4 β5 β6 β7 β8 β9 Utility of No Change 0 0 1 if Information Source is Internet 0 0 0 0 0 0 0 0 Utility of Change Departure Time 1 0 0 1 if Consult Travel Information Prior to Departure 1 if Reason for consulting information was: I wanted to be sure I would arrive on time 0 0 0 0 0 0 Utility of Change 0 1 0 0 0 1 if Usual Travel Pattern: At least twice a week there is an unexpected delay on my route 1 if Information provided was: There was an incident on my route such as a car accident or overturned truck 1 if Information provided was: Some part of my route was closed or out of service for repairs or construction 1 if Primary benefit for seeking information was: Reduced trip time 1 if Travel Mode Used: Car Risk Aversion 141 142 The positive value of the estimated coefficient associated with the access to information (specific to the change of departure time) indicates that travelers are more likely to change departure time when they receive information. The significance of the internet, as the source of information in no changing alternative, is consistent with the findings of Bierlaire and Thémans (2005) and Tsirimpa et al. (2007), who also identified the effect of internet as an information source in individuals’ route switching behavior. Individuals that consulted information because they wanted to be sure they would arrive on time have a higher tendency towards changing their departure time. In addition, those who acquired information having as primary benefit the reduction of travel time are more prone to change route. Drivers who face unexpected congestion at least twice a week are more prone to change route in response to traffic information. This is captured by the positive sign of the related binary variable associated with the route change alternative. Information types have been linked in the developed model with route change alternative. These have been coded in the model as zero/one dummy variables. Τhe provision of information increases the propensity of the travelers to switch from their habitual travel pattern, as expected. The travelers' propensity towards a route change is increased when there is information on an incident or a road closure on the habitual route. The presence of the risk aversion latent variable is significant. The higher the risk aversion, the higher the likelihood of not switching route in response to information. The structural equation of the latent variable model suggests that males are more likely to avoid taking risks than females, which is a finding that needs more research since most literature so far suggest that men are more risky than women. Age was also found a significant factor, with travelers of age 35–55 being less prone towards risk than younger travelers. The coefficients of the three indicators of the measurement equation (1) I worry a lot about being late; (2) I don’t like to have a plan ahead; and (3) I prefer to find my own way rather than ask for information, were found significant as expected. The model estimation results can be summarized in the following statements: • There is an inertia associated with the no change alternative • The occurrence of unexpected congestion on the usual route at least twice a week increase the propensity of the drivers to change their habitual route • Access to travel information (pre-trip), increase the propensity of the travellers’ to change departure time • Individuals who don’t like taking risks are less likely to switch route • The internet as information medium, as well as the content of the provided information (incident on the habitual route) and the reason for consulting information has a positive effect in the propensity towards route change. The significance of the content of the information provided on travellers’ switching behaviour is consistent with the findings of Dia (2002) and Mahmassani et al. (2003). In their studies they have identified that, when the content of information involves travel time delays, the probability of (small or major) route changes increases. Kyoung-Sik (2003) also found that the content of pre-trip information motivates people to change departure time. 143 Table 6. Model Estimation Results MNL with Latent Variable MNL CHOICE MODEL Explanatory Variables Coef. (t-stat) Coef. (t-stat) Constant (specific to Change Departure Time) -2.55 (-8.94) -2.56 (-9.19) Constant (specific to Change Route) -4.87 (-6.30) -4.30 (-8.13) Information Source: Internet (specific to No Change) -1.01 (-2.23) -0.93 (-2.07) Consult Travel Information Prior to Departure (specific to Change Departure Time) 1.12 (2.76) 1.14 (2.88) Reason for consulting information: I wanted to be sure I would arrive on time (specific to Change Departure Time) 2.44 (4.92) 0.67 (1.68) Usual Travel Pattern: At least twice a week there is an unexpected delay on my route (specific to Change Route) 1.36 (2.97) 1.15 (3.09) Information provided: There was an incident on my route such as a car accident or overturned truck (specific to Change Route) 0.60 (1.45) 2.48 (5.41) Information provided: Some part of my route was closed or out of service for repairs or construction (specific to Change Route) 1.86 (2.54) 1.78 (3.18) Primary benefit for seeking information: Reduced trip time (specific to Change Route) 3.14 (6.19) 2.71 (7.41) Travel Mode: Car (specific to Change Route) 1.66 (3.24) 1.54 (3.71) Z1 * Risk Aversion (specific to Change Route) -0.91 (-2.77) Initial Log-Likelihood -448.23 -448.23 Final Log-Likelihood -259.93 -261.56 ρ2 0.42 0.41 LATENT VARIABLE MODEL Structural Model for Z1 * Risk Aversion Gender (male dummy) 0.77 (3.47) Age: 35-55 years -0.52 (-3.04) Age: > 56 0.24 (0.94) ρ2 of Structural Equation 0.17 Measurement Model λ1 I worry a lot about being late 1.34 (4.13) λ2 I don’t like to have a plan ahead 0.61 (3.17) λ3 I prefer to find my own way rather than ask for information -0.72 (-3.28) σ1 I worry a lot about being late 2.28 (11.43) σ2 I don’t like to have a plan ahead 2.68 (20.25) σ3 I prefer to find my own way rather than ask for information 2.77 (18.48) A discussion on the meaning of the terms risk averse and risk seeking is useful in the understanding of these results, as they may have more than one interpretation in the literature (e.g., Cohen et al. 1987; Lopes 1984). According to Katsikopoulos et al. (2002), at the behavioral level most interpretations refer to the pattern of choices that a participant makes when presented with an option having one certain outcome and an option or options having an equal (or almost equal) expected value but more than one possible outcome. Usually, a decision maker who chooses the certain option more often is termed risk averse, and a decision maker is risk seeking if he or she chooses the certain option less often. In the context of route choice in the presence of traffic information, a driver could be termed as risk averse if, among travel time distributions that have equal expectations, he or she 144 more often chooses the route with the smaller variability. Conversely, a driver is termed risk seeking if, among travel time distributions that have equal expectations, he or she more often chooses the route with the larger variability. Katsikopoulos et al. (2000) found that risk attitude in route choice is influenced by whether the route choice scenario is classified as belonging to the domain of gains or to the domain of losses. Katsikopoulos et al. (2000) proposed a simple model that described risk attitude reversals quite well. This model represents a realistic break from the tradition of choice models that assume humans have the capacity for complex transformations of all probabilities and values involved (e.g. Tversky and Kahneman 1981, 1992). In an extension of the work of Katsikopoulos et al. (2000), Katsikopoulos et al. (2002) assumed that drivers exhibit bounded rationality (Simon 1957). That is, a simple heuristic is used to process probabilistic travel time information (see also Gigerenzer and Goldstein 1996). Specifically, a driver is assumed to heuristically estimate a probabilistic travel time as simply a point inside the given range. Katsikopoulos et al. (2002) found that diversion frequency among alternative routes with fixed expectation e smaller than the reference travel time c was decreasing in the range of the alternative route r. Also, diversion frequency was increasing in r when expectation e was larger than the reference travel time c. This finding is consistent with risk attitude reversals for other choices (Tversky and Kahneman 1981). More recently, Casas and Kwan (2007) investigated the final choice of people's decision-making process when faced with unexpected events during the commute trip in the presence of real-time information collected using a travel simulator. Results show that people are willing to experiment with other alternatives if provided the information to support their choice, i.e. they are not inherently risk-averse. The above conclusions of the presented model in this paper are consistent with this finding. 6 Conclusions In the presented research, RP data from a 2-day diary survey have been used. In the survey, individuals provided information about the traveller information sources consulted on each trip, as well as how the obtained information influenced their usual travel behaviour. Since multiple information sources are actually available in the Puget Sound area, the results obtained demonstrate the actual choices of individuals in an information-rich environment. A joint choice and latent variable model that explicitly captures the attitude of travellers’ towards travel pattern switching has been developed and may be used to realistically predict travellers’ switching patterns with regards to departure time change and route change. As expected, travelers in general tend to maintain their habitual travel pattern. However, specific travel information - such as incident or road closure – influences behavioural switches such as departure time change and route change. In terms of better-informed travel decisions and more efficient use of transportation infrastructure, Advanced Traveler Information Systems (ATIS) offer an appealing alternative. The information provided by ATIS in a constantly changing environment helps individuals to readjust their travel decisions, make more informed and conscious travel decisions and reduce travel time and stress. The majority of the respondents, who changed their trip in response to travel information, reported that the main reason for making that change was to reduce travel time (60.4%). As the results indicate, the biggest impact of traffic information is reflected in route changes, reducing the peak and most likely causing a re-distribution of travel demand, thus helping reduce air pollution and congestion. 145 While the PSRC data set may well be the richest available to transportation researchers, it still has limitations. For example, the descriptive statistics of the analysed diaries indicate that respondents used some form of information system (internet, television, radio, or other) only for 4.1% of their total trips, while the use of traveller information systems is mostly made during the trip (46.1%). It is believed that with a larger data set, i.e. a higher number of observations receiving traffic information, richer models could be estimated. For example, additional parameters might enter the model specification, or some nesting structures might emerge. The estimated model should therefore be seen in the context of the available data. Generally speaking, the value of the ρ2 of the structural equation (0.17) is rather low. Further research should aim at extending the presented methodological framework with additional variables and structures that would capture the underlying processes more accurately and would provide better fit. An interesting candidate direction for analysis of the travellers’’ behaviour could be found in the field of psychological foundations from behavioural theory (e.g. based on the works by Tversky and Kahneman 1981, 1992) for items such as character development, stress and anxiety, as well as risk behavior and trade-offs for better choices (Kahneman and Tversky (2000), Rabin and Thaler (2001), and Schwartz (2004). One consideration relates to the extent at which the presented model may be useful to practitioners. While currently the software to estimate such models may not be widely available to practitioners, it is very likely that in a few years it will become more accessible. Research in this area could then be used as a basis for the practitioners to adopt this more powerful type of models and integrate them into their arsenal. As has been illustrated from the model estimation results, the presented methodology outperforms the simpler MNL model that is widely used by practitioners today. Acknowledgements The authors would like to thank Ms. J. Lappin and Mr. S. Pierce from the Volpe National Transportation Center at Cambridge for providing us with the relevant data to conduct this research. We would also like to thank Prof. Denis Bolduc for providing us with the software package ICVL that was used for the estimation of the models. References Abdel-Aty, M. and Abdalla, M. F., 2004. Modeling drivers' diversion from normal routes under ATIS using generalized estimating equations and binomial probit link function. Transportation, 31(3), 327-348. Abdel-Aty, M., Kitamura R. and Jovanis P., 1997. Using stated preference data for studying the effect of advanced traffic information on drivers’ route choice. Transportation Research Part C, 5(1), 39-50. Bierlaire, M. and Thémans, M., 2005. Development of Swiss models for transportation demand prediction in response to real-time traffic information. Proceedings of the 5th Swiss Transport Research Conference, Monte Verita/Ascona, Switzerland. Bonsall, P., 2004. Traveller Behavior: Decision-Making in an Unpredictable World, Journal of Intelligent Transportation Systems. 8(1), 45-60. 146 Casas, I., and Kwan, M. P., 2007. The Impact of Real-Time Information on Choices During the Commute Trip: Evidence from a Travel Simulator. Growth and Change, 38(4), 523–543. Choocharukul, K., 2008. Effects of Attitudes and Socioeconomic and Travel Characteristics on Stated Route Diversion: Structural Equation Modeling Approach of Road Users in Bangkok. Transportation Research Record: Journal of the Transportation Research Board, 2048, 35-42. Chorus, G. C., Molin, J. E. E. and Van Wee, B., 2006a. Use and effects of advanced traveller information services (ATIS): A review of the literature. Transport Reviews, 26(2), 127–149. Chorus, C. G., Molin, J. E. E., Arentze, T. A., Hoogendoorn, S. P., Timmermans, H. J. P. and Van Wee, G. P., 2006b. Observing the making of travel choices under uncertainty and information: validation of travel simulator. Paper presented at the 85th Annual Meeting of the Transportation Research Board, Washington, DC. Cohen, M., Jaffray, J.-Y and Said, T., 1987. Experimental comparison of individual behavior under risk and under uncertainty for gains and losses. Organizational Behavior and Human Decision Processes, 39(1), 1-22. Dia, H., 2002. An agent-based approach to modelling driver route choice behaviour under the influence of real-time information. Transportation Research Part C, 10(5), 331-349. Ettema, D. and van der Horst, R., 2005. Use of travel information and effects on mode choice for recreational trips. Proceedings of the 84nd Annual Meeting of the Transport Research Board, Washington DC. Farag S. and Lyons, G., 2008. What Affects Use of Pretrip Public Transport Information? Empirical Results of a Qualitative Study. Transportation Research Record, 2069, 85-92. Gan, H. C., Ye, H. and Gao, W. S., 2008. Drivers’ En-Route Diversion under the Influence of Variable Message Sign Information: An Empirical Analysis. Proceedings of the Transportation Research Board Annual Meeting, Washington DC. Gigerenzer, G., and Goldstein, D. G., 1996. Reasoning the fast and frugal way: Models of bounded rationality. Psychological Review, 103(4), 650-669. Goulias, C., Kilgren N. and Kim, T., 2003. A decade of longitudinal travel behavior observation in the Puget sound region: sample composition, summary statistics, and a selection of first order findings. Proceedings of the 10th International Conference on Travel Behaviour Research, Moving Through Nets: The Physical and Social Dimensions of Travel, Lucerne, August 10-14. Han, Q., Timmermans, H., Dellaert, B. G. C. and van Raaij, F., 2008. Route Choice Under Uncertainty: Effects of Recommendation. Transportation Research Record, 2082, 72-80. Hato, E., Taniguchi, M., Sugie, Y., Kuwahara, M. and Morita, H., 1999. Incorporating an information acquisition process into a route choice model with multiple information sources. Transportation Research Part C, 7(2), 109129. Jou, R., Lam, S., Liu, Y. and Chen, K., 2005. Route switching behaviour on freeways with the provision of different types of real-time traffic information. Transportation Research Part A, 39(5), 445-461. Kahneman, D. and Tversky, A., 2000. Choices, Values and Frames. New York, Cambridge University Press. Katsikopoulos, K. V., Duse-Anthony, Y., Fisher, D. L., and Duffy, S. A., 2000. The framing of drivers' route choices when travel time information is 147 provided under varying degrees of cognitive load. Human Factors, 42(3), 470-481. Katsikopoulos, K. V., Duse-Anthony, Y. Fisher, D. L. and Duffy, S. A., 2002. Risk attitude reversals in drivers' route choice when range of travel time information is provided. Human Factors, 44(3), 466-473. Khattak, J. A. and Khattak, A. J., 1998. A comperative analysis of spatial knowledge and en route diversion behavior in Chicago and San Fransisco: Implications for advanced traveler information systems. Transportation Research Record, 1621, 27-35. Khattak, J. A., Yim, Y. and Stalker, L., 1999. Does travel information influence commuter and non-commuter behavior? Results from the San Fransisco bay area TravInfo project. Transportation Research Record, 1694, 48-58. Kyoung-Sik, K., 2003. Role of information sources in driver switching decision. EC Workshop on Behavioural Responses to ITS, Eindhoven, The Netherlands. Lappin, J., 2000. Advanced Traveler Information Service (ATIS): Who are ATIS Customers? Report for Federal Highway Administration, U.S. Department of Transportation. Lappin, J. and Bottom, J., 2003. Understanding and Predicting Traveler Response to Information: A Literature Review. Technical Report No.: FHWA-JPO-04014, US Department of Transportation Lappin, J. and Pierce, S., 2003. Evolving awareness, use, and opinions of Seattle region commuters concerning traveller information: Findings from the Puget Sound Transportation Panel Survey, 1997 and 2000. Proceedings of the 82nd Annual Meeting of the Transport Research Board, Washington DC. Lopes, L. L., 1984. Risk and distributional inequality. Journal of Experimental Psychology: Human Perception and Performance, 10(4), 465-485. Ma, J. and Goulias, K.G., 1997. A dynamic analysis of activity and travel patterns using data from the Puget Sound transportation panel. Transportation, 24(1), 1-23. Mahmassani, H., Huynh, N., Srinivasan, K. and Kraan, M., 2003. Trip maker choice behavior for shopping trips under real-time information: model formulation and results of stated-preference internet-based interactive experiments. Journal of Retailing and Consumer Service, 10(6), 311-321(11). Mahmassani, H. and Liu, Y., 1999. Dynamics of commuting decision behavior under advanced traveler information systems, Transportation Research Part C, 7(2-3), 91-107. Marchal, F. and De Palma, A., 2008. Measurement of Uncertainty Costs with Dynamic Traffic Simulations. Proceedings of the Transportation Research Board Annual Meeting, Washington DC. Polydoropoulou, A. and Ben-Akiva, M., 1999. The effect of advanced traveler information systems (ATIS) on travelers behaviour. In Emmerink R.and Nijkamp, P. (Ed.) Behavioural and Network Impacts of Driver Information Systems. Polydoropoulou, A., Ben-Akiva, M. and Kaysi I., 1994. Revealed preference models of the influence of traffic information on drivers route choice behavior, Transportation Research Record, 1453. Polydoropoulou, A., Ben-Akiva, M., Khattak, A. and Lauprete, G., 1996. Modeling revealed and stated en-route travel response to ATIS. Transportation Research Record, 1537, 38-45. Rabin, M., and Thaler, R. 2001. Anomalies Risk Aversion. Journal of Economic Perspectives, 15(1), Winter 2001. Simon, H. A., 1957. Models of man. New York: Wiley. 148 Schwartz, B., 2004. The Paradox of Choice. Why more is less. Herper Perennial 2004. Toledo, T. and Beinhaker, R., 2006. Evaluation of the Potential Benefits of Advanced Traveler Information Systems. Journal of Intelligent Transportation Systems, 10(4), 173-183. Tsirimpa, A., Polydoropoulou, A., and Antoniou, C., 2007. Development of a Mixed MNL Model to Capture the Impact of Information Systems on Travelers' Switching Behavior. Journal of Intelligent Transportation Systems, 11(2), 1-11. Tversky, A., and Kahneman, D., 1981. The framing of decisions and the psychology of choice. Science, 211(4481), 453-458. Tversky, A., and Kahneman, D., 1992. Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297-323. Van der Horst, R. S. N. and Ettema, D. (2005) Use of travel information and effects on mode choice for recreational trips. Paper presented at the 84th Meeting of the Transportation Research Board, Washington, DC, USA. Walker, J. L., 2001. Extended Discrete Choice Models: Integrated Framework, Flexible Error Structures, and Latent Variables. Ph.D. Dissertation, Department of Civil and Environmental Engineering, Massachusetts Institute of Technology. Xiaohong Pan and Khattak, A. J., 2008. Evaluating Traveler Information Impacts on Commercial and Non-Commercial Users. Transportation Research Record, 2086, 56-63. Zhang, F., Shen, Q. and Clifton, K., 2008. An examination of Traveler Responses to Real-time Bus Arrival Information Using Panel Data. Transportation Research Record, 2086, 56-63, 2082, 107-115.