Data Gathering to Build and Validate Small-Scale Social Models for Simulation

Data Gathering to Build and Validate Small-Scale Social Models for Simulation

J. Rouchier (GREQAM, CNRS, France)
DOI: 10.4018/978-1-59140-984-7.ch015
OnDemand PDF Download:
No Current Special Offers


This chapter discusses two different approaches that gather empirical data and link them to modeling and simulations with agent-based systems: experimental economics which built reproducible settings and quantitatively defined indicators, and companion modeling which accompanies observed social groups when they negotiate over renewable resource issues. Both developed techniques have different epistemological posture which lead them to promote diverse data comparison and model validation. They both have small limitation. The chapter wishes to put forward that, although both approaches have different goals, some evolutions in research protocol could enhance qualities of both. Some of these evolutions have already started to be used by researchers.

Key Terms in this Chapter

Experiment: Experiments in economics were created imitating psychological experiments. Individuals are gathered in a place where they are made to interact through computers so to eliminate any non-controlled information circulation. They are given a task for which economic theory predicts an optimal behavior. Actual behaviors are then observed, and the divergence between theory and actions is analyzed.

Rationality: An individual’s rationality is the process of decision that organizes its actions. Rationality has to be separated from perfect rationality , which is an economics concept defined by the ability to choose the best action in a situation. Rationality can be sub-optimal (especially in the case of bounded-rationality as defined by Simon). For an artificial agent, rationality is defined by an algorithm which associates information gathered by the agent to a choice of action.

Emergence: A phenomena is said to be emerging if it is the result of the parallel action of individuals of agents who did not have this phenomena as an objective. For example: individuals wish to go from A to B. A traffic jam appears, it was not wanted, nor predictable by the agent before it happened; it emerges from the presence of a number of individuals at the same time, going to the same place. Such an emergence can be detected by the people who are in it or not. Global warming is an emerging phenomenon that we can witness.

Stakeholder: In a decision-making process, a stakeholder is one of those involved in the situation as it is analyzed by organizers, and who have an interest which is at that moment in conflict with others. Along the mediation process, the definition of situation can change and some new actors can be involved. Sometimes people are identified as stakeholders, but do not recognize themselves as such and do not wish to participate in common decision making.

Role-playing Game: Role-playing games are games that are organized around a scenario where each player takes a role that partly defines its abilities and motivation, and where a story is commonly built from the original script. When the aim is just recreational, the master of the game usually makes up in his mind an environmental accident to stimulate the imagination of players. When the role-playing game is used to sustain discussions in a group, events are either: logically deduced from the definition of the environment dynamics, or from scenarios that are proposed by players or have been observed in the represented setting.

Setting: A setting is made up of the whole system of communication, role, and timing in which the agents or individuals are immerged. It is the artificial organization (in a simulation, a game, or an experiment) that represents the institution that is studied.

Protocols: A protocol is the organization of the capture of data in empirical research. In the case of experiments, it will include the way individuals are selected, how the instructions are transmitted, as well as the task that the participants have to achieve in the experiment. In the case of experiments, it will include the type of study that is led before organizing meetings, the number of meetings and participants, the choice of the stakeholders, the tools used for communication, and the number of modifications of these tools.

Parameters: Agent-based models are made of equations that are based on values—numbers or symbols. When a simulation is run, one set of such parameters is used. For example the number of agents in the system, the size of the grid on which the agents evolve, the rules that define how the resource is renewed, the time that each agent has for communication in a time-step, and so forth. Since a complex system cannot be analyzed analytically, what is studied is its sensitivity to each parameter. If the system displays similar behavior when one parameter varies within a certain range, one can say that this behavior is robust regarding this parameter. If the system’s behavior is highly correlated to the change of a parameter, then the model is considered dependent on this parameter.

Validation: Facing any kind of simulation, there is a necessity to define a way to assess the quality of the model in its representation of quality. In the case of simulation models, it is important to first check that the implemented model corresponds to the model description ( internal validation or verification ). Then one has to see if the model has some structural equivalence with the observed real world, to make sure that it is interesting to use the simulation model to understand, predict, or exemplify this reality. It is common to consider that depending on the aim of the modeling process, the model cannot be validated in the same manner.

Complete Chapter List

Search this Book: