Agent-based simulation has become a key technique for modeling and simulating dynamic, complicated behaviors in social and behavioral sciences. As these simulations become more complex, they generate an increasingly large amount of data. Lacking the appropriate tools and support, it has become difficult for social scientists to interpret and analyze the results of these simulations. In this paper, we introduce the Aggregate Temporal Graph (ATG), a graph formulation that can be used to capture complex relationships between discrete simulation states in time. Using this formulation, we can assist social scientists in identifying critical simulation states by examining graph substructures. In particular, we define the concept of a Gateway and its inverse, a Terminal, which capture the relationships between pivotal states in the simulation and their inevitable outcomes. We propose two real-time computable algorithms to identify these relationships and provide a proof of correctness, complexity analysis, and empirical run-time analysis. We demonstrate the use of these algorithms on a large-scale social science simulation of political power and violence in present-day Thailand, and discuss broader applications of the ATG and associated algorithms in other domains such as analytic provenance.

Agent-based simulation has become a key technique for modeling and simulating dynamic, complicated behaviors in social and behavioral sciences. Lacking the appropriate tools and support, it is difficult for social scientists to thoroughly analyze the results of these simulations. In this work, we capture the complex relationships between discrete simulation states by visualizing the data as a temporal graph. In collaboration with expert analysts, we identify two graph structures which capture important relationships between pivotal states in the simulation and their inevitable outcomes. Finally, we demonstrate the utility of these structures in the interactive analysis of a large-scale social science simulation of political power in present-day Thailand.

A novel method of constructing graphs of agent-based simulation data enables social scientists to better understand complex systems.Agent-based simulation (ABS) has become an important tool for studying complicated group behaviors in social science. 1 In ABS, large numbers of autonomous entities, or 'agents,' interact with one another and, over time, they influence and are influenced by the agents around them. Given that these simulations are stochastic, that is, they use small random perturbations, they generally have to be run hundreds of times to generate a distribution of sample behavioral patterns. Increased computing power enables scientists to simulate and study increasingly complex systems. For example, ABS is being used to model cooperative behavior, 2 ethnic mobilization and conflict, 3 violence and genocide, 4 and population growth and collapse. 5 Since these systems are being used to simulate elaborate patterns of human behavior, they require thousands of agents each with a large number of variables to direct its actions and interactions with those around it. Such complexity means that these systems often generate gigabytes of raw data for each simulation. Unfortunately, these very large data sets can prove incredibly costly to interpret and analyze. Lacking appropriate tools and support, scientists studying these systems are driven to oversimplify their models or perform purely numerical analyses that, by design, overlook many of the subtle yet important forces driving behaviors.To get around this issue, we reframed this process as a graph exploration problem. In other words, we began by asking questions about which configurations in the simulation can be reached from a given starting position. In this context, a graph is a representation of objects (called vertices) and the relationships between those objects. Vertices are unique configurations of the variables controlling the simulation, and the relationship between two configurations is a transition in time. This reframing enables us to employ concepts from graph theory to describe behavioral patterns in these data sets. Figure 1. Visualization of an aggregate temporal graph generated from 100 runs of an agent-based simulation of political hierarchies. The two yellow vertices represent an interesting feature: a highly stable vertex pair (note the high degree of revisitation).To begin our analysis, we first constructed a graph formulation called an aggregate temporal graph (ATG) from the data. Here, a vertex represents a unique state in the simulation, and a directed edge represents a transition between two states in one time step. To build the graph, we combined a large set of sample sequences from the agent-based simulation using the following rule: if two sequences have identical values for all variables at some point in their execution, then that state is represented as a single vertex that is shared by both sample sequences. Each of these sequences can be thought of as a pathway through the simulation space (the set of reachable configuration...