Importance of the order of the modules in TransMob [Huynh et al., 2015]

Nowadays, a wide range of microsimulations are performed thanks to agent based simulations. This frame-work allows to model the interactions between agents and environment in a flexible way. Such models should be initialised with a given set of agents representing the initial population, and their interactions rules should be set. Very often the initial population of agents is obtained using a synthetic population generator. TransMob is an agent-based model that aims to simulate transport and land use interdependencies for urban planning. TransMob contains two types of modules responsible for: the dynamics of the social structure within the population (ageing, divorces,...); and the travel behaviour of the individuals (assigning diaries, ...). In the current version of TransMob, the processes happening in the agents everyday life are modelled and performed in a specific order: first ageing, then death, divorce, marriage and finally giving birth. This work focus on the impact of the ordering in which the different modules responsible for the update of the social structure are applied. This presentation aims at analysing the impact on the results if the order of the procedures is changed. For instance, how will the results change if the divorces are performed after the marriages? Let us denote the processes age, die, divorce, marriage and birth by 0, 1, 2, 3, 4 respectively. All possible orders are then given by the set of all permutations of the integers from 0 to 4. Thus, 120 different orders could be analysed. However, if birth is applied before age, then in the first iteration, we will add new babies to the babies already in the initial population, resulting in an artificial peak of 1 year old agents in the first simulated year , 2 years old in the second simulated year, etc. For this reason, we only consider orders performing age before birth. This reduces the number of feasible orders to 60. The analysis is performed using clustering and decision trees. The first results show that the place of ageing with respect to the place of death influences strongly the results. For instance, when ageing the agents is done before death, the final population is younger. We aim to continue in this direction to further identify the consequences of choosing any particular order. The goal of this work and future research is to make scholars aware of the impact of a choice and provide them with findings to help them chose the best order for their application.