Anticipated Travel Time , Information Acquisition and Actual Experience : The Case of Hanshin Expressway Route Closure

Travel time is one of the most fundamental and important determinants of travel behavior. However, the travel time on which travel decision is based is subjective one, i.e., anticipated travel time. We propose a conceptual model of the formation of an anticipated travel time through information acquisition and initial driving experience. Fitting this model to day-to-day data of anticipated travel times collected during a Hanshin Expressway route closure, we test the Information Dominance Hypothesis (i.e. as a driver acquires more information on travel time, he can predict travel time more precisely and refers less to anticipated travel times he has had in the past in order to anticipate travel times) and the Experience Dominance Hypothesis (i.e. the influences of information which is not from driving experience on anticipated travel time is weaker with actual driving experience than without actual experience). Although words-of-mouth information does not have such impacts as are consistent with these two hypotheses, the results with the other types of information support these two hypotheses. Fujii and Kitamura Paper No. 00-0712 2 INTRODUCTION Travel time has been regarded as one of the most fundamental determinants of both longterm and short-term travel decisions including residential location choice, travel mode choice, destination choice and route choice. Frequently, objective travel time measurements obtained from transportation network models have been used as explanatory variables in travel behavior models. However, the travel time on which travel decision is based is subjective one, i.e., anticipated travel time. Objective travel time may serve as a proxy variable for anticipated travel time if and only if the difference between them is small and non-systematic. In cases of auto trips, however, the difference between an anticipated travel time and the objective travel time would not necessarily be trivial because it is difficult for a driver to predict uncertain travel time precisely when travel time fluctuates from time to time depending on traffic condition, which in turn depends on a number of factors, not all of which are well understood. In the field of travel behavior research, departure time choice models, mode choice models and route choice models have been developed, incorporating the notion of anticipated travel time (1, 2, 3, 4, 5), and dynamic traffic network simulation models that account for the process of learning how to predict travel times (6, 7) have been proposed. One crucial element in these endeavors is understanding how drivers anticipate travel time. Several studies have in fact examined anticipated travel time per se empirically and theoretically (1, 8, 9, 10, 11, 12, 13, 14). Using experimental data, these studies have indicated that the information on travel time gained from past experience influences the anticipated travel time a driver may have for the future. In addition, it has been reported that the rational expectation hypothesis that the expected value of the distribution of Fujii and Kitamura Paper No. 00-0712 3 anticipated travel times would equal that of actual travel times, was supported by experimental data. Although findings based on experimental data would be useful for travel demand analysis, they alone would not be sufficient (15). It is the anticipated travel time a driver has on an actual transportation network, not on a hypothetical network, that is important for travel demand analysis. In particular, for a better understanding of the cognitive process of perception and prediction of travel times, it is important to understand how a driver forms an anticipated travel time when he faces a traffic condition he has never experienced in the past. This is because the anticipated travel time of a route would not change substantially once the driver establishes it well by repeated driving experience, while the anticipated travel time under unfamiliar traffic condition can be expected to be influenced substantially by information acquired by the driver, which in turn depends to an extent on the prevailing information dissemination policy and practice. In addition, the initial driving experience may substantially influence the anticipated travel time formed under unfamiliar traffic condition. The traffic simulation analysis conducted by Nakayama et al. (6), indicated the possibility that drivers' beliefs on travel time formed by initial driving experiences may be "deluded", i.e., the travel time of a route, on which a driver experienced an excessive travel time, and which therefore is not used any longer by the driver, may be believed by the driver to be longer than it actually is. Furthermore, the simulation analysis has shown that this delusion may result in "deluded equilibrium", which is different from user equilibrium and may represent much less efficient traffic condition. Thus, understanding the formation of an anticipated travel time through initial driving experience and Fujii and Kitamura Paper No. 00-0712 4 information acquisition would be crucial for understanding drivers’ learning in route choice, and also for linking microscopic drivers’ route choice behavior and macroscopic dynamics in network traffic flow. The data used in this study were collected in a survey in the Osaka metropolitan area to investigate commuting drivers’ adaptive behaviors during the closure of a segment of the freeway system in the area for maintenance purposes. A freeway closure forces its regular users to behave differently from their travel habits, and also affects the traffic condition throughout the area. Consequently drivers must acquire information on the traffic condition during the highways closure (16, 17) in order to be able to anticipate their travel times. The highway closure in Osaka generated such drivers in the road network of the area who did not know the traffic condition during the closure, yet presumably intended to anticipate travel times as precisely as possible for their commuting trips. It can be expected that surveying such drivers would provide information that can be utilized to understand the process of anticipated travel time formation under the absence of information on traffic condition obtained from driving experiences. The survey therefore collected information on commuting travel mode, route, actual travel time, anticipated travel time, and acquisition of information each day during the highway closure. Using this data set, this study analyzes the impact of acquired information about traffic condition on anticipated travel time. The following media are considered as the source of information: mass-media, word-of-mouth, telephone-based traffic information services, and actual initial driving experience during the closure. Fujii and Kitamura Paper No. 00-0712 5 A CONCEPTUAL FRAMEWORK FOR UNDERSTANDING DRIVERS' ANTICIPATION OF TRAVEL TIME State Dependence and Information Effect It has been demonstrated repeatedly that a behavioral or psychological state is influenced by the past state (18, 19, 20). It is assumed in this study also that an anticipated travel time affects anticipated travel times held in the future. This is labeled state dependence. As state dependence implies that a driver anticipates the travel time for a future trip while referring, consciously or unconsciously, to the anticipated travel times he had in the past, we can determine if the anticipation held in the past affects the anticipation held for the future by statistically testing the presence of state dependence. As the predictive ability of a statistical model increases, predictions match actual values more precisely (21). This implies that anticipated travel time becomes correlated with actual travel time when predictive ability is high. We label this emergence of correlation as information effect, since predictive ability might be influenced by the amount and quality of information an individual has acquired (see Figure 1). Because the ability to predict travel times is bounded (22, 15, 6), however, the information effect may not always be appreciable. How extensive a driver’s ability to predict travel time is, can be inferred by statistically testing the presence of information effect. State dependence implies that the anticipated travel time held at a point of time, t, is influenced by that held at a previous time point, t-1. In order to test the existence of information effect, then, it is necessary to estimate the correlation between the actual Fujii and Kitamura Paper No. 00-0712 6 travel time at t and the anticipated travel time at t after eliminating the state dependence effect (i.e. the effect of the anticipated travel time at t-1 on the anticipated travel time at t). Such a correlation can be obtained as an coefficients estimate of a multiple regulation model whose dependent variable is the anticipated travel time at t and whose independent variables include the anticipated travel time at t-1 and actual travel time at t, as shown in Figure 1. Note that this model is considered as a devise to test the significance of information effect, and is not intended to represent causal relationships. In particular, it is not at all the intent here to assert that the anticipated travel time at t, held prior to trip making, affects the actual travel time at t. The Effects of Information Acquisition and Actual Driving Experience If a driver acquires information on the travel time of a route from, e.g., telephone information services, radio traffic reports, or actual driving experiences on the route, the driver should be able to predict travel time more precisely than could a driver without such information. Consequently, the information effect would be intensified with the amount of information that has been acquired. Furthermore, if the driver is capable of anticipating travel times well based on the information he has acquired, it would not always be necessary to refer to anticipated travel times from the past. With these conjectures, we hypothesize: Information Dominance Hypothesis: Information effects become larger as the driver acquires more information on travel time, and dominate state-dependence effects. Fujii and Kitamura Paper No. 00-0712 7 The information provided by the media, e.g. TV, radio or telephone services, is, under the current technology, generic and not specific to each driver or each trip, whereas actual experience provides information specific to the driver, i.e. actual door-to-door travel time for his trip (the former type of information shall be called “generic travel time information”). Therefore, driving experience, especially the initial one, is expected to affect substantially the anticipated travel time and tend to negate the influences of generic information. So, we hypothesize as follows: Experience Dominance Hypothesis: The influences of generic information on the anticipated travel time are weaker with actual driving experience than without it. Shown in Figure 1 is the proposed model of the formation of anticipated travel time. In this model, the information effect and state-dependence effect are both posited, and the influences of information acquisition and driving experience on these effects are also postulated, representing the two hypotheses presented above. This model is formulated as a structural equations model (SEM) and estimated using the maximumlikelihood procedure available in the LISREL8 software (23). Special Features of the Conceptual Framework The proposed conceptual framework for understanding drivers' anticipation of travel time is different in three ways from those in previous studies that have investigated anticipated Fujii and Kitamura Paper No. 00-0712 8 travel time (1, 8, 9, 10, 11, 12, 13, 14). First, the present study is different from the previous studies in that it investigated the preciseness of prediction or predictability (21) of travel time which might be based on the driver’s cognitive skill for prediction (24) and which can be represented by the information effect in the model proposed in this study. Although a line of studies hypothesizing rationality in expectation (9, 10) investigated the predictability or preciseness of prediction, it did not offer knowledge why and when such rationality which leads to high predictability might emerge. This study, on the other hand, proposes the two hypotheses with respect to theoretical relations among the predictability, information acquisition and actual experience, and tests them. In other word, this conceptual framework may be used to enhance our knowledge about predictability of travel time. Second, the present study can be distinguished from the previous studies in that it investigates drivers' anticipation under unfamiliar traffic conditions. The previous studies investigated the anticipation of travel time when drivers uses one or few routes repeatedly (1, 8, 9, 10, 11, 12, 13, 14). These studies would be useful to understand anticipation of travel time, for example, of drivers who commute everyday using one or few routes. However, actual drivers whose cognitive ability is substantially bounded (22) and may have developed the habits of choosing particular routes (16, 17), must be unfamiliar with majority of the routes and the traffic conditions in actual large-scale traffic networks. Accordingly, research into the anticipation of travel time under unfamiliarity is indispensable for understanding drivers' decision making and behaviors in actual largescale networks. Finally, some studies investigated the anticipation of travel time when travel time Fujii and Kitamura Paper No. 00-0712 9 information is provided by roadside variable message signs or advanced traveler information systems (e.g. 8, 12, 13, 14), but they have not investigated the anticipation when information is acquired through other media (e.g. mass media, telephone services, or words-of-mouth) as well. The framework proposed in the present study is expected to be able to provide useful fundamental information to transportation planners and aid them in developing effective information provision schemes.