Human Behaviour and Traffic Networks

In this article an attempt is made to close the gap between network wide traffic simulations and resulting forecasts on the one hand and the driver’s behavior as a reaction on his daily experience on the other hand. The first part reports on laboratory experiments involving 200 times repeated interactive choices between two routes. In one condition feedback on travel time was supplied for the chosen round only and in a second one for both rounds. Mean values are near to game theoretic equilibrium but substantial fluctuations persist up to the end. Fluctuations are smaller and payoffs are greater it feedback is given for both roads. There is a negative correlation between a subject’s number of road changes and payoffs. Two types of behavior are observed. A direct responder is attracted to the road which was less crowded in the previous period whereas a contrarian shows the opposite pattern. The second part deals with a description of the progress achieved in the development of cellular automata models of vehicular traffic. The most recent models are able to reproduce free flow, spontaneous jam formation, synchronized traffic, as well as metastability. Here the implementation of this model to simulate the traffic on the autobahn network in North Rhine-Westphalia is described and some of the challenges of such a huge and topologically complex network are discussed.

We present alternative discrete choice models of route choice behavior in road networks. A route choice model has two components: 1. The generation of a choice set of alternative routes; and 2. The choice of route among the alternatives in the choice set. The route choice models used in traffic equilibrium tend to be very simple with implicit generation of very large path choice sets. The advent of Intelligent Transportation Systems has renewed the interest in modeling the effects of traffic information systems on route choice behavior. Developments in discrete choice methods have also led to the more sophisticated route choice models. We present results of a small sample route choice survey of MIT employees showing that only a small percentage of drivers select the minimum travel time, minimum distance or minimum generalized cost paths as calculated from the model of the road network. These results demonstrate that the usual deterministic route choice model can be rejected. We then focus on the formulation of probabilistic route choice models. We consider the problem of overlapping paths and the modified Logit solutions that have been developed: C-Logit and Path-Size Logit. We also consider applications to route choice of more general discrete choice models including Cross-Nested Logit, Probit and, ultimately, the Logit Kernel model, which is a flexible hybrid of Logit and Probit. The properties of the different models are examined using simple network examples. Finally, we present estimation results for the different choice models. The results show that the Logit Kernel model with a Path-Size term gives the best fit. The application of Probit and Logit Kernel models requires simulation. Among the closed-form models, the Cross-Nested Logit with a Path-Size term provides a slightly better fit then the Path-Size Logit. The Path-Size formulation appears to have significant explanatory power of the way travelers perceive the alternative paths in a road network.

In this contribution, dynamical models for decision making with and without temporal constraints are developed and applied to opinion formation, migration, game theory, the self-organization of behavioral conventions, etc. These models take into account the non-transitive and probabilistic aspects of decisions, i.e. they reflect the observation that individuals do not always take the decision with the highest utility or payoff. We will also discuss issues like the freedom of decision making, the red-bus-blue-bus problem, and effects of pair interactions such as the transition from individual to mass behavior.

In the second part, the theory is compared with recent results of experimental games relevant to the route choice behavior of drivers. The adaptivity (“group intelligence”) with respect to changing environmental conditions and unreliable information is very astonishing. Nevertheless, we find an intermittent dynamical reaction to aggregate information similar to volatility clustering in stock market data, which leads to considerable losses in the average payoffs. It turns out that the decision behavior is not just driven by the potential gains in payoffs. To understand these findings, one has to consider reinforcement learning, which can also explain the empirically observed emergence of individual response patterns. Our results are highly significant for predicting decision behavior and reaching the optimal distribution of behaviors by means of decision support systems. These results are practically relevant for any information service provider.

An overview of three experiments to study day-to-day departure time and route adjustment behavior of commuters, conducted in the mid 1980’s, is first given, focusing on behavioral mechanisms used by commuters to adjust these decisions in response to experienced congestion and variability in system performance. Experiments conducted in the past five years using a real-time interactive simulator to study users’ responses to real-time information are discussed. Focus is on experiments where quality as well as content of supplied information is varied. The effect of real-time information on the decisions and behavioral processes followed by commuters in online route switching as well as in departure time adjustment from day to day is addressed.

Congested roads during rush-hours create a common-pool problem in which individual rationality results in inefficienCcies. They are not a mere nuisance for commuters but impose social costs that can be reduced by implementing a system of user fees: road pricing. Road pricing promises substantial efficiency gains if used as an instrument to achieve an efficient allocation of fixed road capacity. The road-pricing model of Arnott, R., De Palma, A., and Lindsey (1993) is tested in an experiment. Although the Nash equilibrium of the model is not unique and Nash behavior is therefore unlikely, the road-pricing experiment supports the predictions of the model as long as each player plays with one vehicle only. Allowing players to play with more than one vehicle makes the outcome more efficient. Players appear to internalize part of the externality.

This paper deals with traffic models designed for the calculation of traffic demand: traffic frequency, traffic distribution and the allocation of traffic to different means of transportation. It engrosses the traffic participation of persons in work traffic. Human behaviour and traffic participation represent a system, which offer a wide area to support planning power. The today’s status of nearest neighbourhood to the “Traffic”-system shall be replaced by making that system an integrated component of it, creating the “Time-Use — Traffic” — System. This paper deals with this aspect. Subjects of detailed presentation are results of mobility research in greater German cities (Düsseldorf 1998, Duisburg 2000 and Essen 2001), the data transfer to the Traffic-Demand-Model and the need to prepare the model according to the new title.

Route choice simulators provide a means of collecting data on travellers’ choice of routes. They offer greater opportunity for experimental control and detailed observation than is possible by monitoring real-life choices and, we will argue, can provide more reliable data than can be obtained via stated preference questionnaires or other simplified exercises. Several route choice simulators were developed in the 1990s. Most were intended to help understand and predict drivers’ responses to information and guidance from the Advanced Traveller Information Systems (ATIS) then under development. Others were developed with a broader remit — namely to understand and predict drivers’ route choice in response to a wide range of stimuli and to shed light on issues such as network learning and dynamics. The paper provides examples to illustrate the range of simulators that have been developed and the uses to which they have been put. Key issues in the design of simulators, which are likely to affect the validity of any resulting data, are discussed. The findings from studies designed to compare data from a simulator with that real life are reported. Simulators are compared with alternative sources of behavioural data and conclusions are drawn on their particular strengths and weaknesses. Finally the paper examines the case for the further development and use of route choice simulators.

With this paper on the irrationality of decision making we try to contribute some new aspects to the discussion about human car driving behaviour. From the biologists point of view we argue that human behaviour in modern societies is still deeply interwoven with behavioural patterns which have been developed in the course of human evolution. We illustrate our thoughts on the example of aggression, describing its biological function and its role and changed values in modern societies. Finally a study on driving behaviour is presented, which put forward biological theories to investigate aggressive driving behaviour.

This paper argues the case, that self-selection effects are pervasive in spatial and transport-related decision making. This argument is not new, but the recent interest in appropriate models of change and of daily behaviour under constraints brings it to forefront of interest again. Using descriptive examples and two choice models (set of mobility tools of a household and housing location choice) the paper demonstrates these effects empirically. While the statistically proper solution is the development of self-selection models, which make the decisions about the constraints endogenous to the system, this strategy is difficult in this context due to the long time-horizons/long histories of the relevant decisions. In this paper the constraints were described with variables reflecting those previous decisions. In the model of the mobility tool choice: housing location (type of location, distance to nearest public transport stop; housing costs; distances to work and shopping); in the model of residential choice: distance to previous residential location, type of previous location, distances to work and education). All are significant and have the expected signs.

This chapter attempts to specify realistic behavioral assumptions in models of linked choices of activity sequence/duration, destination, mode, departure time, and route. It is argued that linked choices or activity scheduling may be feasible because people employ heuristic decision rules that circumvent their limits on information-processing capacity. In addition, learning is argued to play an important role. The outcome of learning is either that aggregated information about linked choices (scripts) is retrieved from memory and executed without deliberation, or that deliberate choices are simplified because they are made among sets of scripts.

A review of an application of the models ASDA and FOTO for reconstruction, tracing and forecasting, of spatial-temporal congested patterns on highways based on local traffic measurements proposed in 1996–1999 is presented. Some non-linear features of spatial-temporal congested patterns which are linked with the individual driver behaviour are considered. The model ASDA (Automatische Staudynamikanalyse: Automatic Tracing of Moving Traffic Jams) is devoted to the tracing and prediction of the propagation of moving traffic jams. The model FOTO (Forecasting of Traffic Objects) is devoted to the identification of traffic phases “synchronized flow” and “wide moving jam” and to the tracing and prediction of the patterns of “synchronized flow”. A short introduction to the three-phase traffic theory by Kerner as the basis of the models ASDA and FOTO is made. It is stressed that the models ASDA and FOTO perform without any validation of model parameters in different environmental and traffic conditions. First results of the application of ASDA and FOTO for the online automatic tracing of traffic flow patterns at the TCC (Traffic Control Center) Rödelheim near Frankfurt (Germany) are discussed.

In modern societies, drivers increasingly experience traffic jams that not only cause pollution but also increase the cost of commuting. Hence, one challenge to researchers dealing with traffic management is to find efficient ways to model and predict traffic flow. Due to the social nature of traffic, most of the decisions are not independent, thus drivers have to somehow co-ordinate their actions. Although there are already systems designed to assist drivers in these tasks (broadcast, internet, etc.), such systems do not consider or even have a model of the way drivers decide. We aim at simulating various forms of driver decisions. For validation we want to compare our simulation results with empirical data from real experiments whose results are reported elsewhere. Here we present a naive model for the route choice adaptation of learning commuters with heuristics based behaviour. In simulation experiments they compete in a given scenario. Our results show that the heuristics used, lead to a situation similar to that obtained in the real experiments, especially concerning the route commitment and adaptation to equilibrium states.

Transportation simulation packages need to generate the routes along which vehicles move through the network, and these routes need to be sensitive to congestion. The traditional solution to this problem is static assignment. Unfortunately, static assignment does not work when confronted with more realistic dynamics such as spatially extended and/or time-varying queues. This paper looks at issues which come up when moving away from the static, flow-based representation toward a dynamic, agent-based representation. The important difference is that learning and adaptation are moved away from the system and toward the agents. There is however rather a smooth crossover than a sharp transition between the two views, and intermediate methods are possible, with trade-offs between fast relaxation vs. realistic modeling of human behavior.