Nonlinearity in Social Service Evaluation: A Primer on Agent-Based Modeling
Israel, Nathaniel, Wolf-Branigin, Michael, Social Work Research
Measurement of nonlinearity in social service research and evaluation relies primarily on spatial analysis and, to a lesser extent, social network analysis. Recent advances in geographic methods and computing power, however, allow for the greater use of simulation methods. These advances now enable evaluators and researchers to simulate complex adaptive systems (CASs) by applying agent-based modeling (ABM). CASs reflect the interactions of competitive and cooperative tendencies found in agents. ABM simulations create and test generated observable patterns using the fewest number of plausible decision rules and agents. This primer presents essential concepts for understanding ABM as social service applications of complexity theory shift from a metaphorical perspective to a formalized evaluation method. Further developments in ABM methods need to focus on concepts emanating from the study of complexity science, including the concepts of the wisdom of groups, strengths found in diverse perspectives, robustness, interconnectedness, sustainability, and conflict and cooperation. Appropriate software programs for developing and testing agent-based models are provided.
KEY WORDS: agent-based modeling (ABM); complexity theory; nonlinearity; social network analysis; spatial analysis
As social service evaluators increasingly perceive their services as complex adaptive systems (CASs), the methods for evaluating such systems remain primarily qualitative and metaphorical (Agar, 1999; Stevens & Cox, 2008; Wolf-Branigin, 2009).Taking the next step involves application of agent-based modeling (ABM), a method focusing on building models (computer programs) that simulate the resources and decision rules of agents (clients) to represent social realities (Gilbert, 2008). ABM computationally simulates the interactions of autonomous agents to assess their effect on whole systems. For social program planners, possible uses include identifying resources, defining emerging needs, and building and sustaining capacity.
Complexity arises from the interactions of competitive and cooperative tendencies found in agents. An ABM simulation's test of fit is riot "variance explained" but, rather, a generated observable pattern using the fewest number of plausible decision rules and agents (Gilbert, 2008). It is not hypothesis testing in the traditional sense of attempting to demonstrate that a null hypothesis is false. ABM identifies emergent social behaviors and structures occurring when decision rules are changed. CASs occur in a continual state of dynamic equilibrium, navigating being between rigid order and chaos. While operating under a set of simple rules, patterns emerge from these simple interactions (Gell-Mann, 1994). As a tool, ABM moves social program evaluation beyond nonlinear methods such as spatial analysis and social network analysis (SNA) and seeks to account for more than the simple cause-and-effect explanations (Halmi, 2003). With a CASs view of iterative and adaptive processes, evaluators determine how agents organize into a collective behavior, given changing environmental conditions (Hudson, 2000;Warren, Franklin, & Streeter, 1998).
In contrast to a reductionist, classical research approach that imposes a top-down model of individual and aggregate client behavior, ABM provides a bottoms-up approach that accounts for individual (agent) behavior and investigates the collective effects of individual responses to social interventions (Mitchell, 2009). Developed from economic game theory, ABM emphasizes the importance of unintended consequences arising from the interactions of heterogeneous agents by applying simple rules to their behavior (Axelrod, 1997). Examples of social issues modeled and generated through ABM include racial segregation (Schelling, 1978) and drug epidemics (Agar, 1999).
Software programs like NetLogo (2009) and MASON (Luke et al., 2009) provide the primary tools with which to model and investigate critical issues relating to human behavior and environmental interactions. …