Agent-Based Model Output Analysis – A Comprehensive Statistical Framework

Agent-based models (ABMs) better represent complex systems such as social dynamics. However, there are significant obstacles to the adoption of ABMs in decision-making due to the lack of well-defined methods for analyzing the outputs. We are developing a simulation model agnostic extensible framework for the statistical analysis of the outputs. Our framework includes methods to determine the number of runs needed for each experiment; discover simulation configuration groups from several simulation runs; analyze the stability of the simulation experiment and discover the factors affecting output. We demonstrate "ABMInsights" using a rational ABM disease modeling case study to model masking behavior in a pandemic. Our results can discover high, low, and medium masking behavior patterns and explain them in terms of mask efficiency and disease dynamics. "ABMInsights" could find its applications in the larger community to discover insights such as hierarchical dependencies, clusters, and stability of the simulation model results.