Beyond their support for policymaking and understanding complex interactions, simulations force scholars to explicate their assumptions (Ylikoski, 2014). This can benefit scholars to be even clearer about what they think is going on. Furthermore, social scientists can use simulation models - especially those working on a more granular level (agent-based simulations and microsimulations) to gap between micro- (individual) and macro- (society) level analysis in a fruitful manner. As researchers have full control of the assumptions, environment and rules of the model, they provide a flexible approach to conduct studies. By defining rules at the individual level (such as the likelihood of moving if the neighbourhood is too different; Schelling, 1971), they may observe that wide societal-scale implications emerge (such as segregation). Thus, they give an idea of how individual thoughts and actions (micro-level) transform into outcomes at a society level (macro-level).
However, the disconnect between `the reality' and simulation models limits the usefulness of the simulation models for many more empirically orientated social scientists. Some scholars discuss simulations as artificial societies, where the aim is to `grow a phenomena' observed in society and use this to explain it (Epstein and Axtell, 1996). For example, when examining inequality, Pluchino et al. (2018) were able to grow the power law distribution of wealth in an artificial society by dividing lucky and unlucky events in a specific manner in the simulation model. This does not yet explicate that this is the only way the power law distribution of wealth could emerge in the society. Maybe in reality there are factors beyond talent (which was distributed based on normal distribution) and luck, which explains oneâs position in society. Thus, the simulation model presents one possible explanation for the observed phenomena, but it is not fully conclusive. Similarly, what can be simulated sets limits to their utility. Simulation models assume that actions follow rational and described rules. However, that is rarely the case (Bonabeau, 2002). There are many ways to address these challenges.
The first strategy is to ignore these challenges and explicate simulation models as tools for exploration and thinking and cut the connection to `realityâ. The second strategy is to make simulation models more related to empirical knowledge by ensuring these factors are accounted for in the simulation activity. Microsimulations do this in the area of parameters, but they are limited as interactions between agents are limited. In system dynamics models and agent-based simulations, the choice of parameters is often less connected to reality. Similarly, there are empirically calibrated models, where the aim is to ensure models correspond to reality in their behaviour. These observations may emerge from previous literature or come out from smaller human-behaviour studies (Squazzoni et al., 2013). Naturally, these are much thicker models than models used for thought experiments. Therefore, like in all methods, simulations require careful consideration when in use.