Social Science Microsimulation: A Challenge for Computer Science

Motivation

Microsimulation in the social sciences, i.e. the simulation of dynamic feedback (in both directions) between individual states and states of the population as a whole or certain groups within a population, as well as dynamic feedback between the individuals and the emergence of new phenomena on the group or population level is still a task which suffers from at least three deficiencies:

• Either a model is straightforwardly described in a high level simulation language (like DYNAMO for the Systems Dynamics Tradition, or MIMOSE for younger concept based microsimulation approaches), then large scale models which a great number of interacting individuals cannot be run efficiently (in the case of MIMOSE and also in SmallTalk based multi agent models) or are impossible or extremely difficult (DYNAMO)
• Or a model can be run efficiently with a large number of individuals, then it must have been written down in a general purpose language and is only difficult to communicate in its details, as is the case with most data driven microsimulation models (as, e.g. the models of Sonderforschungsbereich 3 or the Darmstadt Micro Macro Simulator).
• In the second case there are at least two different traditions (of course, very short traditions): data based dynamical microsimulation with no or little interaction between the individuals, and the individuals regarded as black boxes behaving stochastically, and concept driven microsimulation models based on the distributed artificial intelligence approach, with the individuals modeled as agents with memory, goals, and rules, and acting in an environment. Both approaches have evolved in almost total ignorance of each other and a synthesis might be valuable.

Our plan is to discuss which solutions can be found (or developed) by computer science to the problems that arise from social science microsimulation in order that such models can be run efficiently from a user/modeler friendly surface by a modeler who wants to describe his/her model in a problem oriented language (like MIMOSE, e.g.), and not with the help of a general purpose language which is not communicable among social scientists.

The five days from May 1st to 5, 1995, were devoted to the following subjects:

• Social Science Microsimulation / Microanalytic Simulation Models
• Social Science Multilevel Simulation
• Cellular Automata
• Game and Decision Theory
• Distributed Artificial Intelligence

Participants included economists and social scientists applying microsimulation techniques of various different kinds as well as computer scientists interested in helping the former to solve their problems more elegantly and efficiently. Thus representatives of at least three different scientific communities gathered in order to discuss their problems and to help each other find solutions.