10  Programming agent-based model analyses

In this Manual chapter we will mostly ignore the theoretical motivations for doing agent-based modeling to focus on developing the technical skills and techniques required to efficiently develop analyses of agent-based model. This is the development cycle, i.e., the process for developing and analyzing agent-based models.

  1. Sketch it out: Write a brief justification and documentation explaining the model. A simple “box and arrow” diagram or table of model variables can be a great help to stay focused when writing code to implement the model. Specify which learning strategies, social networks, initial conditions, stopping conditions, or anything else that will need to decide before you write the code to implement your model.
  2. First code draft: Program the model in computer code. Start small. Develop a simplified version of the model first if your idea seems complicated. Use a small population size and inspect whether the model dynamics . Prototype the analyses, too, creating time series plots for a range of one (or maybe two) of the hypothized-most-important, explanatory input model parameters or initial conditions, holding other explanatory variables constant to one or a few spot-test values.
  3. Code review and refactoring: As you program you will likely need to revise your code just as you would revise prose, writing multiple drafts to get the meaning right and clarify your ideas as much as possible. Professional software developers call this process refactoring. As you model the process you’re interested in you get new insights about it, which will make you want to change the model from what you started with. This is fine! This is refactoring. At this point it is good to experiment, prototype, and tweak until you are sure the model address the problem of interest. Note that through the practice of modeling, the problem statement also gets sharpened since modeling forces us to specify and thereby clarify our theoretical explanations. Prototype all results by analyzing a relatively small number of trials (e.g., 5-20) for all (or nearly all) the input variable values to be tested in the full analysis.
  4. Write a draft results section with the preliminary results