Scale Dependency in Agent-Based Modeling: How Many Time Steps? How Many Simulations? How Many Agents?

Abstract

As with other parameters a modeler might sweep and document in a sensitivity analysis, it is important to explore the scale dependency of a model. Decisions about how long to run a simulation to generate patterns of interest, or how many simulations are necessary to capture the range of patterns generated by a stochastic model, are an important part of the design and testing process. In archaeological agent-based modeling (ABM), though, researchers have only recently begun to approach these issues systematically. More often, pragmatic concerns related to the time required to run simulations have determined scaling rather than a quantitative assessment of the often diminishing marginal returns of adding one more agent to the simulation or one more simulation to the analysis. Documenting the scale sensitivity of a model can help researchers better manage their time and resources. My ABM project on the organization of the Hohokam economy in central Arizona (AD 200–1450) involved a program of systematically exploring the sensitivity of simulation models to scale-dependent parameters. The research has contributed new insights into the Hohokam pottery distribution system, particularly related to the emergence and organization of a nascent market-based economy.