Numerical Simulations

Numerical simulations were conducted in order to check the face validity of the suggested model. The aim of those simulations was threefold: to check the behaviour of the model under a basic case, to monitor the behaviour of the model under specific scenarios and to illustrate the effect of every parameter on the overall behaviour of the model. In the current chapter, a few basic case results are going to be presented in order to help the reader to get a clear image of how the model works. However, the focus of this chapter is on the effect of the γ and λ2 memory-activation parameters on the model behaviour and especially on the process of habit formation, as well as on the effect of the α1 and a2 emotion-related parameters on the decision-making process.

Simulation Settings

Due to simplification reasons the conducted simulation focused on three activity types (work, shopping and leisure), while there are five agents acting. When an agent becomes 'awake', it stops acting. The time needed for processing one agent is 1.29 seconds. On simulation day t = 600, the simulation run is stopped by the user.

The spatial setting of the simulation is based on the Eindhoven area. A residential and a work location, as well as a supermarket close to their neighbourhood, a social club and a friends' home location correspond to each agent. The schedule of the three activities is predefined and same for all the agents. There are four possible context-conditions c (weekday/rush-hour, weekday/non-rush-hour, weekend/rush- hour, weekend/non-rush-hour). A Monte Carlo simulation, based on predefined probabilities for each context c, determines which c is experienced by an agent, on simulation day t.

The input universal choice set, including the possible activity profiles, is similar for the five agents. Travel time and cost for every activity profile were calculated via maps.google.com. Due to simplification reasons, travel cost is regarded as static; while travel time as dynamic (three levels of delay are considered). The model starts with the agents not being aware of the simulation setting. Two tables are considered, one representing the agents' beliefs regarding the probability of delay for every context and one representing the system probabilities. Based on the first one (which is updated after every experience), agents' beliefs are formulated, while based on the second one, a Monte Carlo simulation generates the actual delay that is experienced by each agent, on simulation day t. The surprise term for experienced utility is generated based on a normal distribution with mean 0 and standard deviation 0.25.

When an activity needs to be conducted, each agent selects an activity profile in order to reach the location of that activity. Based on this experience, the emotional value, the awareness and the activation level of the experienced activity profile are updated. On the other hand, the same indicators of the non-selected profiles decrease, based on a forgetting process. The basic settings for these mechanisms are (1) awareness threshold, (2) awareness retention rate, (3) parameter for updating activation levels γ=1.5, (4) activation level retention rate , (5) parameters for the probability of lowering aspirations , and , (6) stress threshold for short-term horizonand for long-term horizon and (7) uncertainty parameter for exploration