Jaffery Young (1998) discusses the advent of computer models in social sciences in a very story-telling fashinon in his article. While the technique is not new, its use is expanding in several arenas in recent times from understanding history to impact of economic reforms. One of the early projects is Anasazi project which simulated the environmental conditions from 400 AD to 1400 AD to understand why the Anasazi population moved out from Arizona's Long House Valley. The group of anthropologists, archeologists and computer scientist got together for simulating the virtual "agents" representing the population of Anasazi of that time. They simulated how an individual agent reacted based on environmental conditions weather to move or to have children and explained why the Anasazi moved out. The scientists can change the rule of behaviour to simulate various reaction of the societies. Basically, the experiments with social phenomena can be done like it is done in a physics laboratory.
Simulation is a young science. Axelrod (2003) provides a clear way forward for the simulation to be a field in it's own right. He also considers it important to build community of social scientists working in this area to promote simulation as a science. This is especially important since the simulation in social science is done in diverse fields. He suggests three major foundations for developing this science: developing the methodology, standardizing and institution building.
He provides the value of simulation by providing several purposes for which simulation can be useful. The various purposes for which simulation can be used are: 1. Prediction 2. Performance 3. Training 4. Entertainment 5. Education 6. Proof and 7. discovery. Amongst these, prediction, proof and discovery are the most valuable purposes for which simulation can be used as a scientific method.
According to Axelrod, simulation is third way of doing science apart from induction and deduction. Simulation starts as deduction (with explicit assumptions and set rules) and ends as induction (based on data generated from the model rather than based on data collected from the real world).
Rational choice assumption was popular because it allowed deduction. Adaptive behaviour is difficult to model for deduction and that's where the simulation helps. Simulation allows to model both adaptive and rational behaviour simultaneously.
He outlines the three important tasks for conducting simulation research: programming the simulation model, analyzing the data and sharing the results with others. While programming, three goals should be kept in mind: (internal) validity, usability and extendability. While analyzing the data it should be noted that due to stochastic events and randomness built in most social science models, the results differ in each run and often path dependent (based on history). Presentation of history of several runs would prove useful. Simulation also provides an opportunity to change the parameters and see the effects (which allows what-if kind of analysis). While presenting the results, there are four problems that social scientists face: 1. Model needs to be described which increases the length of the article 2. Narrative history of simulation runs requires a great deal of space for describing them 3. For making them easy to understand for audiance from various disciplines, the use of commonly understood terminology in a discipline is restricted and there is a need to explain several concepts at length 4. Computer simulations are relatively new and hence requires a long description of methodology even if shared with a single discipline. Hence, publishing in a social science journal becomes difficult. Alternatives that Axtell suggests are: to provide a CD or hard copy to those who ask for it, archive the data on websites and mention the URL in the published article in a journal.
Axtell also discusses the general lack of replication of simulation so far. He presents one of the projects that he did for replicating his own model as well as of some others and the difficulties he faced in replicating simulations to provide pointers for future researchers. He also provides the basis for replication of simulation and test the validity.
Carley (1995) provides the review of literature demonstrating use of computational and mathematical models in organization theory. He suggests how the organization theory has benefited from the advent of computational and mathematical models. There are four major areas that benefited from the mathematical and computational modelling in organization theory: organization design, organizational learning, organization & information technology, organizational evolution and change.
Mathematical models were "weberian" in the sense that they assumed rational behaviour, limited number of agents and less complexity of processes. Computational models could accomodate more agents and complexity but complete sensitivity analysis on all parameters were not possible. These models could not be scaled up for bigger and complex organizations but provided the opportunity to think how the individual agents changed the collective behaviour. Another genre of models is organizational engineering models that precisely modelled and were used for specific purposes. The extent to which these models can be used for other purposes they were not designed for provides the adequacy of these models.
Another set of models were based on social networks. How individuals and organizations are connected with other individuals and organizations is modelled to understand power, turnover, information flow, diffusion, innovation etc. Logic models are the most limited in their scope.
Models working on organization design tried to find design that produce optimal decisions and building concensus. However, these normative models were soon challenged and gave way satisfactory decisions and distributed decisions. These models provide trade-offs between different designs and consequences on performance etc. However, there is a way to go as culture needs to incorporated as traditional parameters such as task, cognition and structure fail to explain several behaviour.
Organizational learning was another area where extensive work was done in terms of mathematical and computational modelling which contributed to several findings of significance in organization theory. Two challenges remain to be incorporated in further research: linking models of individual agent's learning with models of organization's learning (or adaption) and second, linking diffusion and organization structure. Diffusion doesn't occur simultaneously irrespective of structure.
Models linking information technology and organizations are used both to alter the organization design and improve informatino systems. Information technology is either modelled based on its characteristics or as an artificial agent.
Change and evolution are major themes studied in organization theory. Modelling has been done to study the dynamic processes of change not only for organization design but for cooperation, coordination, planning etc. Methodological issues such as how to represent organization which is changing and non-applicability of statistical techniques to parameters such as organization design.
Points to ponder:
- Is it a breakthrough in social sciences or a fad?
- Can human behaviour be modelled or there is something special about humans that makes human behaviour too complex to model in a computer simulation?
- If the modelling became popular, would it draw researchers away from the field?
- Is the use limited to situations where people look to others for appropriate behaviour?
- What is in simulation that reveals much more from the same information that the analysts couldn't see in several years e.g. in history?
- How would one trust he premitive models that can produce only "rough estimates"? If there is no practical direct use of this kind of research, who would patron them?
- Rather than top-down, agent-based modelling is bottom-up.
- Assumptions should be kept simple for other researchers to replicate. But accuracy is lost when simplicity is preferred. On what basis you strike the balance between the accuracy and the simplicity for your research?
- Would it develop if kept within traditional boundaries of several disciplines or to develop it as a seperate discipline? How would traditional disciplines be facilitated to interact when all of them are using simulation and advancing it? Is it a inter-disciplinary science?
Important Links:
http://www.soc.surrey.ac.uk/JASSS/
Important People:
Robert Axtell, Brookings Institute
John Casti, Santa Fe Institute
