The challenge with any simulation is to identify the relevant and non-relevant factors for the model. As we can imagine from the cases above, showing how we may consider the development of the system, there are always opportunities to complicate models. Meadows (2008, 95-99) speaks about boundaries for the system. All systems must be somehow limited. She continues to observe:
We have to invent boundaries for clarity and sanity; and boundaries can produce problems when we forget that we have artificially created them. When you draw boundaries too narrow, the system surprises you. [- -] System analysts often fall into the opposite trap: making boundaries too large. [- -] This “my model is bigger than your model” game results in enormously complicated analysis, which produces piles of information that may only serve to obscure the answer to the question at hand. [- -] Ideally, we would have the mental flexibility to find the appropriate boundary for thinking about each problem. We rarely are that flexible.Similarly, almost 40 years earlier Forrester (1969, 12-13) states a similar idea:
To develop a complete concept of a system, the boundary must be established within which the system interactions take place that give the system its characteristic behaviour. [- -] To build a computer simulation model of a system, one must first estimate what components are interacting to produce the behaviour being investigated. The choice selects the sets of those components that lie within the dynamic boundary for the particular study and excludes all other potential components as being irrelevant to the study and therefore outside the dynamic boundary.
The long quotes speak about the issues for scoping the modelling activity correctly. Quantitatively oriented social scientists might recognise similar issues in regression analysis. Adding new variables usually improves the results, but the improvements may be marginal considering the overall model performance. They are speaking about a similar issue. If the model is too limited, it does not capture the phenomena under investigation. If it is too complex, results are difficult to deduce from it - and its development becomes complex.
However, the elegance of simulation models emerges from their simplicity. For sociological imagination, simulation models allow thinking and experimenting as the most essential parts for the argument. Think about all the rules not present in the Schelling (1971) model, such as housing price or different preference profiles for groups A and B (for discussion on developments of this model accounting such factors, see e.g. Aydinonat, 2007). However, for the main contribution of his original model, these aspects are not required. He makes a case to analyse the phenomena only accounting for one simple rule and a limited case but shows that, even with such limited perspective, relevant findings emerge from the analysis. This example shows that even blunt oversimplifications can provide a fertile opportunity for social theorising. This is why Macy and Willer (2002) highlight that simplified models may support sociological imagination:
Analysis of very simple and unrealistic models can reveal new theoretical ideas that have broad applicability, beyond the stylised models that produced them. Models should start out simple, and complications should be added one at a time, making sure that the dynamics are fully understood before proceeding.
Beyond possibilities for creative imagination, it is clear that for all simulations we must reduce it. The reality is overly complex for us. (This is true even for scholars who are drawn towards simulation of societies as they are complex and computationally intensive, thus allowing them to take benefit of their advanced skills in this area; Conte et al., 2012) Some aspects of what is considered a sufficient level of simplification emerges from the intended use. For a theoretic analysis we can allow narrower boundaries than for a policymaking tool, where more real-world realities might be demanded for the analysis. One might suggest that boundaries could be developed around a particular theory or hypothesis. Meadows (2008) highlights that boundaries rarely exist neatly in academic disciplines and encourages opening up when considering what to model and how to model phenomena. However, she later acknowledges that academia has the habit of maintaining the disciplinary boundaries (which we return to further in Section 1.6). The strategy to limit to particular theories therefore may be good publication advice.
Therefore, my recommendation is to focus on understanding what limitations emerge from the simplifications - what is not included in the model at this time. These should be clearly stated and discussed, like any good academic work. Furthermore, it may be wise to initially start from a simpler model and complicate it as needed to avoid feature creep, demands and complications that may lead to too much complexity of the model in the long run. Simulation building is often iterative. The model may evolve during the experimenting and exploring process to acknowledge observations and answer potential critique.