Handbook of Applied System Science

Zachary Neal

• Publisher: Taylor & Francis
• Year: 2016
• Print ISBN:
  9780415843324
• Edition: 1
• eBook ISBN:
  9781317607076
• Topic: Social Sciences
• Subtopic: Sociology
• Index: Social Sciences

The Handbook of Applied System Science is organized around both methodological approaches in systems science, and the substantive topic to which these approaches have been applied. The volume begins with an essay that introduces three system science methods: agent-based modeling, system dynamics, and network analysis. The remainder of the volume is organized around three broad topics: (1) health and human development, (2) environment and sustainability, and (3) communities and social change. Each part begins with a brief introductory essay, and includes nine chapters that demonstrate the application of system science methods to address research questions in these areas. This handbook will be useful for work in Public Health, Sociology, Criminal Justice, Social Work, Political Science, Environmental Studies, Urban Studies, and Psychology.

Chapter 14 of this book is freely available as a downloadable Open Access PDF under a Creative Commons Attribution-Non Commercial-No Derivatives 4.0 license available at http://www.taylorfrancis.com/books/e/9781315748771.

1
WHAT IS SYSTEM SCIENCE?

Zachary P. Neal

DOI: 10.4324/9781315748771-1

A system is a collection of separate parts that, because they are linked to and affect one another, are interdependent. This interdependence means that understanding how the system works requires understanding not only how each of the parts works individually, but also how they interact with each other. A mechanical clock offers a nice example. Each gear and spring is a separate part, but each of these separate parts interacts with other parts to make the clock run. To understand how the clock runs, it would be useless to study each gear and spring separately. The parts must be studied together, with a view toward understanding how each part interacts with and affects the other parts, and how in combination they produce the outcome that we call “telling time.”

Even very complex mechanical clocks with hundreds of parts represent relatively simple, determinist systems. Because each gear interacts with the others in the same way every time, their behavior is highly predictable. This is, after all, what makes clocks useful. But, most of the systems we encounter on a daily basis are much more complicated and their behaviors cannot be so easily predicted. Consider your own neighborhood community. The community is composed of multiple, separate parts – you, your neighbors, the buildings and roads, the businesses and institutions, and so on – that interact with and affect one another. To really understand how the community works, it would be useless to study each part separately. Instead, we need to understand, for example, how you interact with each of your neighbors, or how the roads provide access to different businesses. Still further, we might need to understand how your interaction with your neighbor gave you information about a shortcut road to a business you frequently visit. By understanding how each of these separate parts interacts with the others, we can start to understand how in combination they produce the outcome that we call “the community.”

System science methods are a broad collection of methods designed to help understand systems. Although there is no definitive list of such methods, this volume focuses on three that adopt different perspectives on understanding systems and that the Office of Behavioral and Social Sciences Research at the U.S. National Institutes of Health has identified as dominant. Agent-based models adopt a bottom-up perspective, seeking to understand systems by examining each interaction among each individual part of the system (e.g. each person’s decision about where to live in the neighborhood). System dynamics adopts a top–down perspective, seeking to understand systems by examining the relationships among key variables that describe the system (e.g. how the number of people and number of businesses depend on each other). Finally, network analysis focuses on understanding a system’s behavior as a function of the pattern of relationships that exist in-between the system’s parts (e.g. which pairs of neighbors are friends and which are not). The common thread that unites these and other system science methods, and distinguishes them from other more conventional research methods, is their focus on understanding systems by looking at all their parts not individually but in combination as interdependent parts of a larger whole. The goal of this chapter is to briefly introduce the logic of these methods; more detailed technical introductions to using these methods can be found in the references below.

Bottom–Up: Agent-Based Models

Agent-based models are a type of computer-based simulation methodology in which agents interact with each other following simple behavioral rules. Because these types of models require the use of computers, and in some cases can be computationally intensive, they have a relatively short history, starting in the early 1970s with Schelling’s model of segregation and Conway’s “Game of Life” (Gardner 1970). They are often rooted in an epistemological stance known as methodological individualism, which views system-level outcomes as arising or emerging from the micro-level interactions of individual agents. For methodological individualists, a truly complete explanation of a system’s outcomes must be framed in terms of the individual agents whose behaviors led to the outcome. This is a kind of reductionist epistemology, but one that innocently asks, if outcomes aren’t caused by the actions and interactions of agents and their environments, where else could we possibly look for an explanation? Accordingly, the goal of many agent-based models is to understand the kinds of micro-level interactions that could be responsible for generating a known system-level outcome, or what Epstein (1999) called the generativist’s question. To answer this question, Epstein proposed that researchers conduct what he called the generativist’s experiment: “Situate an initial population of autonomous heterogeneous agents in a relevant spatial environment; allow them to interact according to simple local rules, and thereby generate–or ‘grow’–the macroscopic regularity from the bottom up.” This experiment provides a kind of recipe for building and testing agent-based models.

Agents

All agent-based models are composed, at a minimum, of agents and the rules that guide their behavior. These models are highly flexible and can be used to explore systems in many different contexts. In an agent-based model designed to understand a neighborhood community, the agents might be individual people, and the behavioral rules might specify how these people decide with whom to talk or form a friendship. Alternatively, in an agent-based model designed to understand an ecosystem’s population dynamics, the agents might be individual animals of various species, and the behavioral rules might specify who eats who, under what conditions, and how often. Thus, what kind of entity the agents represent and what kind of rules they follow depend on the context of the system under investigation.

The agents in an agent-based model, whether they represent people or animals or molecules, are assumed to be autonomous. Each agent has some degree of agency and is not fully controlled by external forces. They are capable of acting on their own, in accordance with the specified behavioral rules. However, agents are not assumed to be completely independent. Indeed, the very essence of system science is the explicit recognition that parts of a system are interdependent. Thus, as agents interact with one another, these interactions can influence their behaviors. In addition to being autonomous yet interdependent, the agents may also be heterogeneous – that is, the agents may have unique characteristics that distinguish them from one another.

Considering the role of agents in a specific agent-based model helps to make these ideas more concrete. Schelling’s model of segregation was designed to answer a specific generativist question: what kinds of neighborhood preferences, which guide individuals’ decisions about where to live, could be responsible for generating residential segregation? That is, given that we already know residential segregation occurs, what kinds of individual behaviors could be responsible for it? In this model, the agents represent individual people or households. The agents are autonomous because they are free to choose where to live; their decisions are not dictated by an external force such as a law requiring mandatory segregation. But the agents are also interdependent: to the extent that people have preferences about who their neighbors are, one person’s decision about where to live depends in part on other peoples’ decisions about where to live. Finally, the agents are not all the same. In Schelling’s model, people differ from one another on a single, binary characteristic: there are Type A people and Type B people. This characteristic is unspecified and could represent race, ethnicity, religion, or nearly any other observable social characteristic. For Schelling, and indeed for many agent-based models, the focus is on general processes rather than specific variables.

Rules

The agents populate an agent-based model, but the rules tell these agents what to do and how to interact with one another. Although the flexibility of agent-based models means that the rules could specify nearly anything, they nonetheless are expected to be simple, local, and limited. First, the rules that govern agents’ behaviors must be simple because people and other entities with agency are not like supercomputers that can consider all possible courses of action and then select the optimal one. Instead, they tend to follow heuristics and rules of thumb that are easy to remember and implement. Second, the rules must be local. People are not omniscient; they do not know what is happening in distant parts of the world, or what is going on inside another person’s head. Thus, the information they use to make decisions about what to do comes only from their local, observable environment. Finally, the rules must be limited in number. People generally do not have well-formed plans about what they would do in the event of every conceivable situation (e.g. what would I do if I saw a unicorn?), but rather have just a few behavioral strategies intended to cover the most likely situations.

The stark simplicity of the behavioral rules in Schelling’s segregation model offers an excellent example: move to a new house if you are dissatisfied with the composition of your immediately surrounding neighbors. First, it is simple: look at who your neighbors are, and if you don’t like them, move. Second, it is local: only take into account the people living immediately adjacent to your house. Finally, it is limited: there is only one rule. Of course, to be implemented, it is also necessary to define what it means for a person to be satisfied or dissatisfied with the composition of the neighborhood. For Schelling, satisfaction depends on a threshold value, X, that can be manipulated by the researcher. People are satisfied with their neighborhood if at least X% of their neighbors is the same type as themselves. Thus, if the researcher sets X = 75, then a Type A person would be satisfied if at least 75% of his neighbors were also Type A people, and a Type B person would be satisfied if at least 75% of her neighbors were also Type B people.

Modeling Process

The agent-based modeling process consists of several steps and is frequently iterative. The first step involves building the agent-based model itself, but there are multiple approaches to model building. In some cases, models are built from existing and well-established theories, for example, about the factors that influence when neighborhood residents interact with one another (see Chapter 23). In other cases, models are also informed by empirical data about real people and places, which are used as starting values to simulate realistic populations and environments (see Chapter 3). More recently, agent-based modelers have also explored participatory model-building strategies that directly involve research subjects in the development and testing of the model (see Chapter 18). Finally, as in many fields of research that are incremental, new agent-based models are often developed by adapting and extending earlier models (see Chapter 22). These different model building strategies – theory, data, participatory, and adaptive – are not mutually exclusive, but often are used in various combinations.

After a preliminary agent-based model is built, it is run multiple times not only to debug the model and ensure it is working properly, but also to develop an initial understanding of the types of system-level outcomes the model produces. Running an agent-based model typically consists of two stages: an initialization stage, and an interaction stage. In the initialization stage, the population of agents is created and assigned the various characteristics that might distinguish them from one another. In Schelling’s model, the initialization stage is quite simple: a population of N people (the size of N can be set by the researcher) is created; half are Type A and half are Type B. These people are then randomly assigned locations in a grid, where each square in the grid represents a house in the neighborhood. The left box in Figure 1.1 shows a graphic depiction of a neighborhood of 100 people simulated by Schelling’s model after the initialization stage. The two types of people are represented by different colors (here, black and gray). Because the people are assigned random locations, there is no segregation present; the Type A and Type B people are evenly mixed throughout the neighborhood.

Figure 1.1 Agent-based model of segregation

In the interaction stage, the agents each take a turn following the behavioral rules defined by the model. A single model run, called a “tick,” concludes after each agent has taken its turn. Because the focus is often on how system-level outcomes unfold or emerge from agents’ repeated interactions, the interaction stage of frequently repeated many times. In Schelling’s model, during the interaction stage each person surveys their immediately surrounding neighborhood to determine whether they are satisfied with the composition of their neighbors, given the researcher specified value of X. If the person is satisfied, they do nothing, while if they are dissatisfied, they move to a random unoccupied house. After each person has taken a turn applying this behavioral rule, asking “should I stay or should I go,” the interaction stage starts over with each person again taking a turn and applying the same rule. This process repeats until all the people are satisfied with their locations, or until it becomes clear that it is impossible for all peopl...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Contents
• Detailed contents
• 1. What is System Science?
• Part 1 Health and Human Development
• Part 2 Environment and Sustainability
• Part 3 Communities and Social Change
• About the Contributors
• Index