NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Committee on the Assessment of Agent-Based Models to Inform Tobacco Product Regulation; Board on Population Health and Public Health Practice; Institute of Medicine; Wallace R, Geller A, Ogawa VA, editors. Assessing the Use of Agent-Based Models for Tobacco Regulation. Washington (DC): National Academies Press (US); 2015 Jul 17.

Cover of Assessing the Use of Agent-Based Models for Tobacco Regulation

Assessing the Use of Agent-Based Models for Tobacco Regulation.

Show details

Appendix AConsiderations and Best Practices in Agent-Based Modeling to Inform Policy



Agent-based modeling (ABM) is a powerful tool that is being used to inform policy or decisions in many fields of practical importance. Recent examples include land-use and agricultural policy (Berger et al., 2007; Berger and Troost, 2014; Brady et al., 2012; Guzy et al., 2008; Happe et al., 2008; Happe et al., 2006; Heckbert, 2011), ecosystem and natural-resource management (Heckbert et al., 2010; Schlüter and Pahl-Wostl, 2007), control of communicable disease outbreaks (Burke et al., 2006; Epstein, 2004; Epstein, 2009; Eubank et al., 2004; Ferguson et al., 2006; Germann et al., 2006; Lee et al., 2010; Longini et al., 2005; Longini Jr et al., 2007; Yang et al., 2009), marketing (Garcia and Jager, 2011; Rand and Rust, 2011) and private-sector logistics and strategy (Frederick, 2013; North et al., 2010; Rand and Rust, 2011), economic policy (Dawid and Fagiolo, 2008; Frederick, 2013; LeBaron and Winkler, 2008), electoral design (Bendor et al., 2003; Laver, 2005), and education (Maroulis, 2014).

In this paper, I discuss some of the features of ABM that make it compelling for such purposes (especially in the context of public health), lay out the process and challenges involved in using ABM, and offer some important best practices for rigorous and effective use. This is not a textbook or a how-to manual; it is intended as an overview of the major topics and considerations involved in the use of ABM for policy. It just scratches the surface in most cases but provides some references for further reading. I will argue that the use of ABM to inform policy making or decision making can be both promising and practical but is often challenging and requires great care in practice.

1.1. What Is Agent-Based Modeling

Agent-based computational modeling (ABM) is an approach to modeling complex social dynamics that has developed in recent decades, facilitated by increased computational power. In an ABM, actors in a system are represented as autonomous individuals in a computer program. They are given rules that govern their behavior, including adaptation and interaction with each other and with their environment through time, and a starting configuration. The ABM then simulates1 both individual trajectories and population-level patterns or outcomes, which are generated from the bottom up by the decentralized actions and interactions of the agents. Such a model provides mechanistic mapping from individual-level assumptions to coevolving population-level dynamics. Assumptions can be informed by data or theory, and outcomes at both the individual and population levels can be compared with data statistically. ABM allows enormous flexibility in assumptions, and agents can be modeled at any level (or multiple levels) of scale.

1.2. Why Agent-Based Modeling

Like other modeling methods, this technique has both advantages and important limitations. The particular advantages of ABM come from its flexibility, which can help model designers and users to manage three particular challenges that complexity poses for researchers and policy makers alike: heterogeneity, spatial structure, and adaptation.


Real-world complex systems are often characterized by substantial heterogeneity among individuals. Among individuals of a particular type, this might include biological diversity (e.g., in genes, microbiome, sensitivity to reward), behavioral diversity (e.g., in decision-making, psychology, personality), demographic diversity (e.g., in socioeconomic status, race, sex, and age) or diversity in context or prior experiences (see “Spatial Structure” on next page). There may also be substantial heterogeneity in types of actors that are important in a system’s behavior; for example, the outcome of childhood obesity is driven partly by such diverse actors as parents, community stakeholders, school employees, health professionals, food companies, and the children themselves. Types of actors may differ substantially in information sets, goals, incentive structure, constraints, and so on.

By explicitly modeling every individual actor (within the model boundary), ABM allows rich representation of heterogeneity. No aggregation (such as “representative agents,” compartments, or mean-field approximations) is required in an ABM, although aggregation can be accommodated if useful. Taking heterogeneity into account can be critical in the design of successful interventions into complex systems (IOM, 2012; Mabry et al., 2010; Sterman, 2006).

Spatial Structure

An important advantage of ABM is the ability to include structurally rich, dynamic, and heterogeneous representations of social or environmental exposures and influences. For example, ABM can incorporate explicit representations of geography from GIS data (Axtell et al., 2002; Brown et al., 2005a; Brown et al., 2005b; Magliocca et al., 2014; Page, 1999; Sun et al., 2014) or detailed social network structures (Hammond and Ornstein, 2014; Zhang et al., 2014)—representations that may be difficult in standard analytical approaches (Axelrod et al., 2004; Eubank et al., 2004; Page, 1999), which tend to rely on mean-field or other approximations. By directly incorporating sophisticated spatial elements, ABM can effectively model dynamics that result from exposures across space and time (such as advertising or air pollution exposure), patterns of contact between individuals (central to epidemic spread or social influence through networks), the impact of context on decision making, and geographic constraints on choice set (such as the distribution of retailers with heterogeneous characteristics).

ABM not only allows incorporation of spatial elements that affect agents and their interaction with one another, but it also allows modeling of the coevolution of environment and individual behavior, on potentially divergent time scales: for example, the coevolution of retail geography and consumer purchases or of individual choices and social norms (see below).

Adaptation and Coevolution (Potentially Across Scales)

The ABM technique is particularly adept at modeling interaction and adaptation. By modeling at the individual level, ABM allows consideration of multiple interdependent factors that influence an outcome (such as health status). Because ABMs are dynamic, individual-level adaptation can also be represented, whether it takes the form of biological adaptation (as in an addiction process or physiological changes due to weight gain) or of behavioral adaptation (as in learning). A dynamic, individual-level focus also allows ABM to consider such phenomena as path-dependence (Page, 2006), which is important for life-course models that focus on key development windows or accumulation of exposures.

By modeling populations of individuals, ABM can also capture the interaction of actors with each other and with their coevolving environments. This type of interaction and feedback between individual and social levels of scale is important for the study of such phenomena as interacting social influence and social selection processes in adolescents, strategic coevolution of pro-tobacco and anti-tobacco marketing, and the bidirectional influence of social norms and individual behavior.

ABM is also well positioned to study mechanisms or pathways that cross multiple levels of scale. Agents themselves may be modeled on different levels of scale, for example, “employee” agents and the “corporation” agents for whom they work. In addition, ABM offers the opportunity to embed rich depictions of mechanisms within an agent (e.g., physiology or neurobiology) that take as inputs factors outside the agent (e.g., environmental exposure to food or marketing) and interact with between-agent dynamics (e.g., social norms). This enables ABM to consider topics in public health that cross the “skin barrier,” for example (Glass and McAtee, 2006; Hall et al., 2014; Hammond, 2009; Hammond and Ornstein, 2014; Hammond et al., 2012; Mabry et al., 2010).

Policy Resistance

Heterogeneity, spatial structure, and adaptation all complicate analysis, and many analytical approaches struggle to address one or more of these features. The presence of these characteristics in a system may also contribute to policy resistance (Sterman, 2006). Anticipating adaptive (and potentially diverse) responses of a system to an intervention can be critical in designing effective policies. Behavioral and biological adaptation by individual actors can change the impact of an intervention for better or for worse. Interventions that appear promising on a small scale (or in one part of a system) may also run into “equilibrium dilution” or even produce net negative effects as adaptive individual or organizational responses on larger scales (or elsewhere in the system) come into play. The flexibility of ABM in capturing adaptation and heterogeneity thus makes it a potentially useful tool to inform decision-making in complex systems.

1.3. A Brief Overview of Agent-Based Modeling in Various Fields

Application of ABM first proliferated in biology and in social science and initially focused on important uses of the technique for theory and hypothesis development. Early examples in social science included work on cooperation (Axelrod, 1997b), electoral and bureaucratic dynamics (Bendor et al., 2003; Bendor and Moe, 1985; Kollman et al., 1992, 1997; Laver, 2005), conflict (Bhavnani and Miodownik, 2009; Epstein, 2002), and segregation (Bruch and Mare, 2006; Schelling, 1971; Xie and Zhou, 2012). Early work using ABM in evolutionary biology (Axelrod et al., 2004; Hammond and Axelrod, 2006a; Holland, 1992; Nowak, 2006; Ohtsuki et al., 2006) and ecology (DeAngelis and Mooij, 2005; Heckbert et al., 2010) also contributed in important ways to theory development. Many of these efforts leveraged the ability of ABM to capture heterogeneity, spatial structure, and adaptation to generate important new insights. With growing computing power, application of ABM expanded to such fields as education (Maroulis et al., 2014), anthropology (Axtell et al., 2002), economics and finance (Dawid and Fagiolo, 2008; Dawid and Neugart, 2011; Farmer, 2000; Farmer and Foley, 2009; LeBaron and Winkler, 2008; Tesfatsion and Judd, 2006), marketing (North et al., 2010; Rand and Rust, 2011), and land use (Berger et al., 2007; Berger and Troost, 2014; Brady et al., 2012; Brown et al., 2005a; Brown et al., 2005b; Guzy et al., 2008; Happe et al., 2008; Happe et al., 2006; Heckbert, 2011; Magliocca et al., 2014; Sun et al., 2014). ABM also began to be applied in a broader set of ways, including models engaged with large data sets (Axtell et al., 2002; Bruch and Mare, 2006; Farmer and Foley, 2009) and models designed to engage with or inform policy and to address policy resistance (Berger et al., 2007; Brown et al., 2005a; Brown et al., 2005b; Dawid and Fagiolo, 2008; Farmer, 2000; Guzy et al., 2008; Happe et al., 2008; Happe et al., 2006; Heckbert, 2011; LeBaron and Winkler, 2008; Magliocca et al., 2014; Schlüter and Pahl-Wostl, 2007; Sun et al., 2014).

A very recent, but rapidly growing, application area for ABM is in public health. Initial applications of ABM to public health focused on the epidemiology and control of communicable disease (Burke et al., 2006; Epstein, 2004; Epstein, 2009; Eubank et al., 2004; Ferguson et al., 2006; Germann et al., 2006; Lee et al., 2010; Longini et al., 2005; Longini Jr et al., 2007; Yang et al., 2009). A large network of modelers (MIDAS2) funded by the National Institutes of Health (NIH) has had substantial scientific and policy impact using ABM among other modeling approaches (see section 3.1 below). The last 5 years have seen growing recognition of the potential for ABM to yield new insights on a wide array of topics in public health, particularly in light of the importance of heterogeneity, spatial structure, and adaptation that have been informed by other fields (Brown et al., 2005a; Brown et al., 2005b; Magliocca et al., 2014; Sun et al., 2014), and the approach was highlighted in three recent Institute of Medicine reports (IOM, 2010, 2012, 2013). This broader awareness has led to the recent proliferation of work, including the creation of two additional NIH-funded modeling networks that use ABM: one focused on obesity (NCCOR Envision3) and one on health disparities (NICH4). Initial ABM studies in these areas include (COSSA, 2014; Hall et al., 2014; Hammond, 2009; Hammond and Ornstein, 2014; Hammond et al., 2012; Zhang et al., 2014).


The overview above highlighted the growing array of topics to which ABM is being applied, but also began to draw out several distinct ways in which the technique can be used. Models in general (and ABM in particular) can be used for a variety of specific purposes as part of a research, education, or decision-support agenda (Epstein, 2008). Four especially common uses of ABM are (1) formulating or testing explanatory hypotheses about (potentially unobservable) mechanisms driving observed patterns in the real world, (2) bridging individual-level assumptions and population-level dynamics, (3) guiding data collection or empirical analysis by pinpointing especially important gaps or by discovering new questions, and (4) informing the design or evaluation of interventions (including policy choices).

In each of these uses, ABM can yield compelling insights to complement existing approaches—although the particular perspective that it provides is not always well suited for every topic or question (see Heckbert et al., 2010, and others for “litmus tests” of suitability). In the rest of this paper, I will focus on the specific use of ABM as a decision-support tool to inform policy or intervention design and evaluation.

2.1. Policy as a Specific Use for Agent-Based Modeling

Computational or mathematical models (including ABM) offer a number of potential advantages to a decision maker. By making explicit the assumptions, key pathways, and uncertainties involved (along with the mapping of all three of these onto potential outcomes), models can help decision-makers to revisit and discuss implicit mental models that may be driving the decision process. Explicit models are more easily tested, both for internal consistency and for external fidelity. Models can also be especially useful tools when fielding real-world experiments to inform policy choice is difficult, overly expensive, time-consuming, unethical, or impractical. An additional advantage offered by models such as ABM lies in their ability to uncover potentially unanticipated adaptive system responses that a policy or intervention might trigger (see section 1.2). ABM can also help a decision maker understand the implications heterogeneity (across individuals, contexts, or time) may have for the impact of a policy in the longer term or in contexts differing from those for which empirical evidence is available. Finally, models can sometimes be of particular use when processes of policy implementation or even policy making itself are the focus.

2.2. Three Specific “Modalities” for Informing Policy with Agent-Based Modeling

The use of ABM to inform policy or decision making comes with its own particular set of considerations. ABMs that inform policy fall into three distinct categories: prospective policy models, retrospective policy models, and indirect policy models.

Prospective policy models (also sometimes called ex ante models) help to inform the design of policies or interventions by elucidating their potential effects. Such models contain representations of key dynamic mechanisms in a system, along with explicit representations of one or more policy choices, and they allow comparison of policy options within the simulated system. This process can aid in the design of policies or interventions by:

  • Identifying leverage points where small shifts induced by targeted policies can generate large shifts in systemic outcomes or dynamics (such as “tipping points”). This may help to identify previously unnoticed opportunities or strategies for intervention.
  • Elucidating potential linkages (trade-offs or synergy) between multiple policies or intervention elements in a complex system. This may help to facilitate coordination across “silos” in government or society, as needed for “systems” interventions (Huang et al., 2011; Nader et al., 2012).
  • Allowing experimentation “in silico” to understand full potential consequences (intended or unintended) of interventions, which may include counterintuitive or unexpected impacts. This is of particular use when “in vivo” or “in vitro” experimentation (for example, through a randomized clinical trial) is not practical.
  • Anticipating a variety of possible future scenarios that may unfold, incorporating both uncertainty and policy choices, or helping to elucidate how an intervention design might “scale,” translate to a novel context, or play out in the long term.

Models (including ABM) are most effective as one input into a multifaceted decision-making process; they generally cannot eliminate uncertainty or the need for judgment in weighing difficult trade-offs. They can, however, be of substantial help to decision makers in managing both complexity and uncertainty.

Retrospective policy models help to understand the underlying reasons for (retrospectively observed) success or failure of a policy or intervention that is already in place. To do this, they leverage the ability of ABM to provide insight into complex and dynamic mechanisms that are at work in a system which may not be directly observable (see section 1.2 above). In some settings, data may not exist (or might even be impossible to collect) to disentangle multiple simultaneously occurring mechanisms. ABM can help with causal inference in such circumstances. In the context of an intervention evaluation, this type of model can help evaluators to understand why and how elements of the intervention may have succeeded or failed. By facilitating consideration of heterogeneity (see section 1.2), ABM can also help to understand differential success of a policy or intervention across sub-populations or contexts. This may be critical for consideration of scaling and translation of successful interventions. In practice, retrospective modeling may often be combined with subsequent prospective modeling that leverages lessons learned from existing data to design improved interventions.

Indirect influence on policy or decision making may also come from models that are not explicitly aimed at consideration of policy choices. ABM offers extensive capabilities for understanding etiology, bidirectional relationships between system structure and individual behavior over time, and the operation of pathways that cross levels of scale. This type of model generally does not contain any explicit representation of policies or interventions and thus does not directly simulate the potential impacts of policy choices. Nonetheless, discoveries derived from this type of model may have important implications for policy—including identification of key leverage points, mechanisms, or windows of opportunity for intervention. Application of such insights within a policy-making process must be done with care and may require further simulation modeling that explicitly contains representation of the policy choices under consideration.


In this section, I provide brief descriptions of models that illustrate how policy may be informed by ABM in each of the three ways described above. The examples chosen are focused on public health where possible but also include a sampling of work from the social sciences; the set of examples here is by no means comprehensive.

3.1. Examples of “Prospective” Agent-Based Modeling to Inform Policy Design

Models of Infectious Disease

One of the earliest applications of ABM in public health has been in the modeling of communicable disease, and much of this work has had an explicit prospective focus on policy or intervention design. In 2003, the NIH National Institute of General Medical Sciences formed a collaborative network of scientists who were using modeling to understand infectious disease dynamics (MIDAS). The network, which now includes almost 100 scientists, has helped to pioneer the use of computational models (including ABM) to inform policies aimed at preparation for or response to epidemics. MIDAS has generated numerous scientific advances (Burke et al., 2006; Epstein, 2004; Epstein, 2009; Eubank et al., 2004; Ferguson et al., 2006; Germann et al., 2006; Lee et al., 2010; Longini et al., 2005; Longini Jr et al., 2007; Yang et al., 2009) and received the Distinguished Service Award from the U.S. Department of Health and Human Services for contributions to policy. The use of ABM in MIDAS has included both small-scale or exploratory models (Epstein, 2004; Epstein et al., 2008) and large-scale ones (Epstein, 2009; Eubank et al., 2004), with complex models built up in layers through iteration with exploratory and empirical work over a number of years (see section 4.3, BP3 below). These models leverage ABM’s ability (see section 1.2) to include more realistic mixing patterns (explicit geography and/or networks), extensive heterogeneity (demographic, immunological, or behavioral), and adaptive behavior change by individuals in response to epidemics or to intervention elements (for example, protective self-isolation or decisions about care-seeking or vaccine acceptance). Some models cross many levels of scale from biological (disease progression and host-response within an individual person or virus evolution) to global (air travel or vaccine production). The experience of MIDAS has also helped to elucidate best practices for communicating models to policy makers (see section 4.3) and has underlined the value of multiple methods and multiple models in increasing confidence in policy-oriented findings (also see section 4.3).

One early MIDAS model that provides clear illustration of the prospective use of ABM to inform policy design can be found in work on smallpox preparedness (Burke et al., 2006; Epstein, 2004; Longini Jr et al., 2007). This model began with a stylized representation of individual movement across key social contexts identified in previous epidemiological work—agents in the model move between and spend time in households, workplaces or schools, and hospitals—and drew on appropriate demographic data. An “index case” of smallpox was introduced into this artificial population, and the spread of the pathogen (with characteristics drawn from empirical evidence on natural history) was simulated. The model was then used as a “virtual laboratory” to allow experimentation with varying policies for containment of the epidemic through vaccination. The model allowed comparison of potential impacts of policies already under consideration, but it also made use of the individual-level dynamic data created by the simulations (which provided a detailed account of how smallpox spread through a community) to identify novel policy options that focus on particularly high-leverage intervention targets to allow maximum effectiveness with minimal vaccine use.

The use of this type of model is of particular importance for policy discussions surrounding potential responses to bioterror—a circumstance that does not lend itself to real-world experimentation but would demand well-articulated and rapid policy response. By allowing prospective consideration of options in silico, making use of the best available data and capturing the inherent uncertainties (e.g., inexact pathogen parameters, timing and location of early cases) the models can be a key input into planning and decision-making. ABM on a much larger scale also proved useful in assisting policy response to the emerging H1N1 influenza epidemic of 2009–2010 (Epstein, 2009; Eubank et al., 2004; Ferguson et al., 2006; Germann et al., 2006; Lee et al., 2010; Longini et al., 2005; Yang et al., 2009). In that case, models had to address more extensive variation both in geographic context (from emergence of the virus in Asia to its spread around the globe to the United States) and in potential policy options (from antiviral prophylaxis to school closure to quarantine), but benefited from ongoing surveillance as the early epidemic unfolded. Earlier work in MIDAS helped to make possible the development and deployment of sophisticated models that were needed to inform policy response at both national and regional levels during the crisis. As in the case of smallpox, a primary use of ABM was for prospective consideration of varying mixtures of policy options across various contexts, with a clearly defined objective of effectively containing the epidemic.

Other Exploratory Work in Public Health

The use of ABM as a tool for prospective consideration of policy options in a public health context has begun to spread outside of infectious disease, including work in disaster preparedness (Epstein et al., 2011). In tobacco control, early development work for this type of ABM is under way. One example is the Tobacco Town project (Luke et al., 2014), which leverages the flexibility of ABM in representing detailed geography (see section 1.2) to consider tobacco control policies that are inherently spatial in nature (such as point-of-sale policies). This effort draws on demographic data, travel data from ecological momentary assessment, and retail exposure and purchase data to simulate representative communities in which point-of-sale policies might be deployed (FDA, 2013). The models include consideration of adaptive individual responses to environmental changes. The goal is to provide an in silico policy laboratory to understand the potential effects (intended or unintended) of retailer-based policy options such as zoning, licensing, and type-specific retailer density reduction across a variety of contexts and over both the short term and the longer term.

Outside of Public Health

ABM has been used extensively outside of public health as a tool for prospectively informing policy or interventions, including work on retirement policy (Axtell and Epstein, 1999), anticorruption interventions (Hammond, 2008), and agricultural production policies (Berger and Troost, 2014). This type of application for ABM has also appeared in the private sector (e.g., for consideration of changes to logistics, marketing, or strategy) (Frederick, 2013; North et al., 2010; Rand and Rust, 2011). Agents in these models represent (inter alia) current or potential retirees, bureaucrats, farmers, landowners, consumers, and employees.

3.2. Examples of “Retrospective” Agent-Based Modeling

Retrospective use of ABM to understand differential success of policies and interventions in public health has only recently begun to emerge,5 but examples of this use are more widespread in social science. One illustrative example comes from political science consideration of real-world electoral systems and their implications for party competition and bureaucratic politics (Laver, 2005; Laver and Sergenti, 2011). In this work, agent-based models of multiparty competition (in which political party leaders and voters are types of agents) are applied to understand the historical trajectories of party policies and vote shares in 10 European countries. Another example is the recent use of ABM to study the economics of systemic risk in the housing market (Geanakoplos et al., 2012). This work looks retrospectively at policies that were in place during the housing boom and bust of 1997–2009 and develops an ABM of the underlying mechanisms (individual-level incentives and behavioral adaptations to the policies) that produced the observed outcome.

3.3. Examples of “Indirect” Policy Implications from Agent-Based Modeling Focused on Mechanisms

Early applications of ABM in obesity research have focused on elucidating complex etiology. Obesity results from a multiscale system, with behaviors and outcomes driven by interacting mechanisms that sometimes cross levels of scale. ABM has the potential to offer new insights into mechanisms and to connect research focused on “below the skin” with that focused on “above the skin” (Hammond, 2009). Some of these models are beginning to offer insights that may have important indirect policy implications. For example, recent work focused on understanding preference formation builds on existing evidence in neuroscience on key brain systems that are involved in controlling eating behavior (Hall et al., 2014), making use of ABMs’ ability to cross the “skin barrier” by embedding dynamic reward-learning processes within agents while exposing agents to different external sequences of environmental food exposures. The resulting model (Hammond et al., 2012) illustrates how early food exposures can strongly shape food preferences in ways that have substantial inertia in the face of subsequent changes in food opportunities; the model also shows how preference formation can be path dependent in the sequence of experience exposures (see section 1.2). Although this model does not directly consider any specific real-world policy, it offers potential implications for both targeting and timing of interventions to prevent obesity by encouraging the formation of preferences for healthy food. Another group of recent ABM papers focuses on the role of social networks and social influence in obesity, elucidating potential dynamic mechanisms through which social influence occurs (and which may potentially be harnessed for interventions) (Bahr et al., 2009; Hammond and Ornstein, 2014; Zhang et al., 2014). Some of these models include empirically grounded biological processes (Hall, 2010) that interact with the social level (Hammond and Ornstein, 2014); others begin to explore modalities for interventions to harness social forces while stopping short of prospectively modeling any specific real-world policy in detail (Bahr et al., 2009; Zhang et al., 2014).

In social science, examples of ABM with important indirect implications for policy include canonical work on the drivers of segregation (Bruch and Mare, 2006; Xie and Zhou, 2012) and work on the underlying mechanisms that may explain the ubiquity of ethnocentrism (Hammond and Axelrod, 2006b). In both cases, the models do not directly simulate specific policy or intervention choices—but they elucidate powerful pathways (and sometimes specific levers in the form of key variables) that could be harnessed for policy or intervention purposes (see, for example, Axelrod, 2004, which discusses potential application of insights from the ethnocentrism model to security issues of central Asia).


As described in the previous sections of this paper, ABM can provide a powerful, flexible tool with high potential to offer meaningful insights for both scientific research and policy design. The very power and flexibility that make ABM appealing can also make it challenging to use appropriately, however. section 1.2 (and 4.1 below) make clear that ABM involves many distinct choices about implementation, as well as many assumptions in translation from the real world into the computational world. Like any modeling technique, results from ABM flow directly from the inputs; thus the conclusions reached are only as strong as the inputs on which they are based. The specificity of an ABM (its ability to elucidate very specific operationalizations of mechanisms and actors) is part of its power; but it involves similar challenges, with great care needed in generalizing from the conclusions of a particular model. Because ABM is also a relatively new technique, opportunities for formal training and available reference materials, such as textbooks, are still limited. For all of these reasons, attention to emergent best practices in ABM is of particular importance.

This section of the paper begins by laying out the key elements and steps that go into constructing and using an ABM and then lays out a number of best practices for each of the steps in the process.

4.1. Key Building Blocks of an Agent-Based Model

Although ABMs are quite diverse as a group, reflecting diversity in topic (see section 1.3) and goal (see section 4.2)—and understanding the number of design choices implicit in these building blocks helps to motivate some of the best practices discussed below (see section 4.3).

The elements of an ABM may be organized according to the “PARTE” framework: Properties, Actions, Rules, Time, and Environment. The first three elements (Properties, Actions, Rules) define the agents, while the next two (Time and Environment) define the context (see Figure A-1).

FIGURE A-1. PARTE framework.


PARTE framework.

Properties are characteristics of individual agents (such as sex, age, disease state, wealth, and body mass index). An agent property can be:

  • Mutable or immutable over time (within the simulation). This is a design choice: a model focusing on a single school year might treat age as immutable, whereas allowing age to change within the simulation might be central to models considering the lifecourse or overlapping generations (Axtell et al., 2002).
  • Observable, partially observable, or unobservable to other agents. A simulation of farm behavior might treat size in acres as observable to other farms, but income as only partially observable (Brown et al., 2005a; Brown et al., 2005b; Happe et al., 2006; Magliocca et al., 2014; Sun et al., 2014). In a model of tobacco use, agents may be able to observe numerous cues that suggest whether another agent is a current smoker or not, but previous smoking history is likely harder to observe.
  • Stored in a variety of data structures, from simple booleans to complex lists or arrays. A dichotomous variable can be easily stored in a simple data structure. Storing the mapping between food types and associated reward values (Hammond et al., 2012) or agents’ preferences for other types of choices (Kollman et al., 1992, 1997; Maroulis et al., 2014) might require a more complex data structure, such as a hash table or vector.

Not all types of agents represented in a model need to have all properties; for example, the property “market capitalization” is relevant for agents that represent firms but not for agents that represent employees of the firms. All properties must have well-defined conditions for initialization and for change through time. Initialization can involve draws from predefined distributions or from data and may be conditional on values assigned to other properties. By representing each individual actor as a separate software object, ABM allows enormous flexibility to capture heterogeneity across agents in their properties (see section 1.2).

Actions define the repertoire of specific behaviors that agents can perform within the simulation, such as moving around the environment, eating food, smoking tobacco, communicating information to a neighbor, forming a friendship tie, or buying a product. Agent actions can:

  • Change an agent’s own Properties. For example, taking the action “eat” may affect the property “body mass index” over time (Hammond and Ornstein, 2014); taking the action “buy cigarettes” will immediately affect the property “inventory of cigarettes” (Luke et al., 2014).
  • Change the Properties of other agents. For example, models of cooperation and reciprocity often contain an action “donate (to agent x)” which increases a wealth or wellbeing property of agent x while decreasing the same property for the actor (Axelrod et al., 2004; Hammond and Axelrod, 2006a; Hammond and Axelrod, 2006b; Nowak, 2006).
  • Change the Environment. In models of land use and commons, for example, exploitation of a resource by one agent may reduce the amount of resource available at that location in the environment (temporarily or permanently) to other agents.
  • Change an agent’s own Rules, for example through learning.

For every action included for any agent in an ABM, the modeler must define conditions under which the action is triggered or may be performed. Each action must also have defined consequences (which may include one or more of the changes above); actions that have no consequences do not affect simulation dynamics and do not belong in the model.

Rules are the central drivers of model dynamics, defining how agents choose an action, update properties, and interact with each other and their environment. Rules in an ABM can:

  • Take as an input the current or past value of Properties (an agent’s own, those of others, or those of the environment); for example, “if age > 18, purchase tobacco.”
  • Be dependent in some way on Time, and may involve learning or adaptation—for example, image scoring in reputation models, or the process of preference formation (Hammond et al., 2012; Nowak, 2006).
  • Vary enormously in complexity from simple heuristics (for example, “when reaching any four-way intersection, always turn left”) to detailed internal models or calculations (for example, agents who collect data about the simulated world and optimize over some objective function).
  • Cause not only modification of agents but creation or removal of agents, for example, creation of offspring in a demographic model or killing of other agents in a model of genocide (Bhavnani and Miodownik, 2009; Epstein, 2002).
  • Involve stochastic probability, for example, “when reaching any four-way intersection, turn left with probability 50 percent.”

Time is central to a dynamic simulation model. Agent-based models (and other related simulation models) generally have a single, lowest-level fundamental unit of time that represents one pass by the computer through the set of instructions that embody the simulation. This is sometimes referred to as an “iteration,” a “tick,” or a “round.” Time in an ABM:

  • Can remain abstract or can be calibrated to real-world time with additional design work. Such calibration may be easier for some models than for others, depending in part on whether behaviors in the model occur on known time scales. A model in which an agent’s age in years changes every 12 iterations or in which agents travel back and forth between home and office every other iteration is relatively easy to calibrate to calendar time (Luke et al., 2014). A model of chronic disease incidence driven by smoking or a model of opinion-change dynamics may require more work to calibrate (Garcia and Jager, 2011).
  • Can involve multiple distinct “speeds” at which change occurs, for example, the speed with which a virus spreads through social contact versus the speed with which the virus itself evolves.
  • Is the unit in which rules, action, and changes in agent properties or environment are defined.
  • May also shape the simulation results through decisions about the order in which instructions are followed by the computer. For example, some types of diffusion models start with a single agent that deviates from the population and calculate the likelihood that the deviation will spread through the population. The choice between two implementations of diffusion—one in which the early adopter influences others before being influenced by its own neighbors and one in which the two directions of influence are evaluated in the reverse order—can result in very different outcomes, such as whether the deviation persists or dies out. A deterministic agent activation order could mistakenly lock the model into one of those outcomes (see section 4.3).

Environment provides the context for agents and their interactions in the model. The flexibility to represent many different types of environment effectively is a major strength of ABM (see section 1.2). The Environment in an ABM can:

  • Range from simple and relatively abstract geometries (e.g., a lattice, ring, or torus) to highly complex ones (often empirically informed) such as a GIS shape file or a network structure.
  • Contain “agent types” itself, with their own properties, actions, and rules. For example, a model of subsistence agriculture may contain rules for crop regeneration at any particular environmental location that are dependent on farming and harvest practices as well as intrinsic soil and water conditions (see Axtell et al., 2002).
  • Can change over time, endogenously (as a result of such agent actions as crop rotation) or exogenously (for example, as a result of a policy change or external shock).

The PARTE framework describes fundamental building blocks that are present in every ABM, but as the illustrative examples above show, enormous variation is possible in the form that each element (P, A, R, T, and E) takes from model to model. This flexibility is part of the power and potential of ABM, but also underlines the importance of following best practices in making the numerous design choices required (see section 4 below).

4.2. Key Steps in Agent-Based Modeling

Just as ABMs share key building blocks in common (while exhibiting extensive heterogeneity in instantiation), the process of constructing and using an ABM generally follows six key steps. These steps are not unique to ABM—they are shared with many other forms of computational modeling—but several steps raise particular considerations for ABM. Table A-1 briefly describes the key steps.

TABLE A-1. Key Steps in Model Development.


Key Steps in Model Development.

With ABM in particular, progress from step to step may not be linear, but instead may involve iterative cycles or back-and-forth—especially as models are built up from simple to complex in stages (see section 4.3 BP3 below).

4.3. Considerations and Best Practices for Model Developers

This section outlines best practices that have emerged to guide design and use of ABMs. Some are specific to policy-aimed modeling; others are general best practices for good ABM (or even for modeling in general) but may have special relevance or importance when the aim is to inform policy. The best practices are organized according to the six steps of modeling (see section 4.2), and discussed in the subsections below.

1. Definition of Question or Goal

Models can be put to many different uses and can help to achieve a number of distinct goals (see section 2.1). The full utility that a model will ultimately have cannot always be foreseen, of course, but models tend to be most useful when they are focused and tailored for a specific purpose. This is because different questions or goals are likely to lead to very different decisions about model structure, different design choices, and different data needs. ABM in particular, because of the specificity that it requires, involves many specific design and implementation choices early in the process (see sections 4.1 and 4.2). Thus, a key best practice is

Best Practice (BP) 1. Start with a clear question or goal, and let this drive early modeling decisions

Having the question or goal clearly in mind guides the initial steps of model development, including inventory of relevant existing literature, available data, and needed team expertise. Agreement on question and goal between the model design team and potential end users of results can be especially important for policy-oriented modeling. A clear statement of question also helps model designers to ensure that the method chosen is appropriate and well suited—ABM may not always be the best choice (for guidance on when to choose ABM, see section 1.2 above and also Axelrod, 1997a; Axelrod, 2004, 2006a; Axelrod and Tesfatsion, 2006; Heckbert et al., 2010).

2. Model Scope and Conceptual Design

This step begins with decisions about model scope and how to best represent key conceptual ideas within the model. Appropriate representation of concepts is an important part of the skill of modeling, is aided by clear questions or goals as guideposts, and works differently depending on the modeling method used. For ABM, the best model designs tend to take as a point of departure consideration of key actors in the system rather than an emphasis on variables or factors (see Macy and Willer, 2002), as is more usual in, for example public health. An ABM-specific best practice is

BP2. Take an “agent” perspective in initial design, identifying key actors in the system that will be the focus of the model

Initial model design also involves choices about scope and model boundaries. Here, a common tension occurs in the balance between parsimony and breadth. Models often yield the clearest insights when they remain relatively simple—a principle sometimes referred to as Occam’s Razor or the KISS principle (Axelrod, 2006a). Parsimony allows effective tracing from inputs via specific mechanisms to outputs of interest (giving clear answers to “why” and “how” key results obtain; see BP11). Keeping the model simple also helps in managing more pragmatic challenges, such as computational speed or tractability. However, especially with complex-system models, there is often pressure to include as much realism as possible; indeed, part of the motivation for selecting an ABM approach is the increased flexibility that it offers to capture realism and interaction.

Managing this tension is a key part of initial model design. Decisions about what to include in the model and what to leave out are guided in part by clarity in the question statement (see BP1), but a key best practice for ABM is

BP3. Start with (relatively) simple models and build up complexity iteratively, one step at a time

A common experience with ABMs is that simple models are rich enough to generate complex dynamics, counterintuitive surprises, and important insights. And even simple models involve many distinct design and implementation choices (see section 4.1), which require careful sensitivity analysis and testing (see BP9). Of course, what simple means may be contextual and depend in part on the starting point in existing studies and the ultimate goal. Even when a more complex model is envisioned, however, starting simple is usually the right choice. By building up complex models from simple ones, one moving piece at a time, the modeler can maintain clarity about how each piece affects results (see BP11) and can greatly facilitate interpretation (see BP12).

When the goal of the modeling effort is prospective or retrospective policy assessment, an additional design consideration is appropriate representation of policies themselves. Engagement with stakeholders can be an important input into this type of model to ensure, for example, that policies considered in the model are “realistic” ones (e.g., of interest in the real world). At the same time, the modeler must maintain enough independence from stakeholder concerns to avoid building foregone conclusions into the model. (For more on working with stakeholders, see IOM and NRC, 2015.)

3. Model Specification

This step involves operationalizing the model “ingredients” in an implementation-ready way, moving from a conceptual design to a specific and explicit sketch of the model. For an ABM, this involves fully specifying P, A, R, T, and E (see section 4.1).

Part of the power of computational and mathematical models comes from a clear and explicit statement of the assumptions that drive results (see section 2.1), and models are only as good as their assumptions. The ability of a model to provide clear and convincing insights thus depends critically on supporting assumptions. An important best practice in model specification is

BP4. Each assumption should be well grounded and have a strong motivation

Assumptions can be grounded in data, grounded in theory (or external face validity), and (sometimes) included for the specific purpose of considering the sensitivity of model results to their formulation (e.g., as in hypothetical policy scenarios). Regardless of their origin and grounding, full sensitivity analysis is needed for all assumptions (see BP9).

Recourse to a sufficiently interdisciplinary group of content experts can be critical in developing a well-grounded specification and a model that can meet its goals. Pragmatic consideration of opportunities for testing or calibration (including data availability) may also be important at this stage, depending on the stated goals for the model. If the goal includes empirical explanatory power or forecasting, for example, inclusion of model outcomes for which no data are available may be problematic; if the goal is to develop theory or design experiments, this may be desirable instead.

4. Model Implementation

This step involves translation of the specific model into an operational form to conduct simulations. For an ABM, this involves writing computer code—either from scratch, or using one of a number of packages that provide some functionality for routine tasks (see Axelrod and Tesfatsion, 2006; Axelrod, 1997b; Rand and Rust, 2011, for discussion of these). Translation from prose descriptions, schematics, or “pseudocode” to formal mathematical and computational instructions is challenging and requires close attention for a number of reasons.

BP5. Use care in translation of an ABM design into computational code

Computers require very specific instructions and cannot “fill in” any gaps—initial attempts to implement a model often lead to the discovery that the specification (step 3) is insufficiently detailed and requires further thought. Specific choices of functional form or algorithm, required for computational implementation, can affect the results of an ABM and may require additional consideration at this stage (see also BP9). If the person doing the computer coding is not the same person as the model designer in steps 2 and 3, there is a danger of miscommunication or “loss in translation” (see Axelrod, 2006a; Axelrod, 2006b). Additional considerations may arise from the choice of computer language or modeling package. Because software for developing ABM is not standardized and is often open-source and continually evolving, it is important to check whether any design choices or algorithms are “hard-coded” by default. The implementation stage can also lead to tension between the conceptual design and goals of the model (steps 1–3) and the capabilities of the software platform. Although pragmatic considerations concerning feasibility and effort can be real constraints, it is important to avoid letting the available software tools drive the model design away from the imperatives of goal and question.

Given the importance of an implementation that accurately reflects design, and the number of specific choices involved in implementing an ABM, it is an important best practice to conduct several rounds of error-checking and “partial model testing” once code has been written.

BP6. Conduct error-checking and partial testing as models are implemented

There are many approaches to this type of testing (see Axelrod, 1997a; Axtell et al., 1996; Miller and Page, 2007; Rand and Rust, 2011; for discussion), which has two fundamental goals. The first is to ensure accurate translation from conceptual to computational, and to catch any errors in coding—this often involves both review of the computational code line by line, and the design and application of simple tests of functionality that match actual computational outputs from small pieces of the code with those expected. The second goal of partial testing is to ensure that the model specification itself (assuming proper implementation in code) represents concepts and meets design goals appropriately. Boundary-adequacy tests and extreme-event tests can help to uncover flaws in the model specification that result in dynamics that, for example, violate face validity or clash with conceptual design and require revisiting step 3 in the process. The ability to conduct this type of testing effectively is another important reason to build model complexity slowly (see BP3).

For ABM in particular, there are many details of implementation that can strongly shape dynamics. These may include interaction topology, agent activation regime, randomization of lists, and handling of pseudorandom generation (see Axtell, 2000); each of these topics deserves consideration in the implementation step but may also require sensitivity testing (see BP9). Of particular importance for ABM are decisions about initialization and halting conditions. Every property included in agents will require a starting (initialization) value in the computer, and generating results from a simulation requires instructions to the computer about when (in Time) to stop the simulation and calculate the outputs. Both decisions can affect results, and they require special care and consideration but sometimes do not arise until the implementation phase. Like other assumptions, these decisions should be grounded and have a strong motivation (see BP4).

During implementation, documentation of all the specific decisions made becomes a key best practice. Models can go through several iterations of design, specification, and implementation, and maintaining alignment between the actual computer code and the description (in prose or mathematics) is critical, as is version numbering to ensure a match between any particular set of results and the exact code that generated them.

BP7. Fully document model specification and implementation, and maintain up-to-date documentation throughout the process

Documentation should cover not only the key ingredients in the model specification (P, A, R, T, and E) but also specific choices in implementation, such as those described above. Given widely varying standards across journals and fields about source-code availability, and the variation in packages for ABM, the documentation should aim where possible to be precise enough to allow replicability on its own. Use of an open-source programming language or package, and provision of programming code for published ABMs are also important best practices.

5. Analysis

Once the model has been fully implemented computationally, it can be used to conduct analysis. Depending on the goals and questions (see BP1), the analysis may take a variety of different forms. For most of these, an important early step will be testing and/or calibrating the model. There are many approaches to testing (see Epstein, 2012; Heckbert et al., 2010; Manson and Evans, 2007; Miller and Page, 2007; Rand and Rust, 2011), which may involve “stylized facts” from published literature, primary data collection, or use of secondary data, such as those from surveys and experiments, GIS data, and surveillance data. It is important that testing and calibration procedures be consistent with the goal of the model and the question being considered, so it is often important to consider testing from the very outset of design.

BP8. Undertake carefully considered testing and calibration of the model, consistent with the goal or question

Testing of a model often focuses on comparing outputs with reference data but may also involve comparison or manipulation of inputs (see Rand and Rust, 2011). All procedures and datasets used in testing or calibration should be part of the documentation for the project.

For almost any question or goal, a key part of analysis using an ABM is sensitivity analysis. This process involves testing the dependency of model outputs to variation in each of the inputs (assumptions and parameters) and sometimes specific implementation choices.

BP9. Conduct thorough and appropriate sensitivity analysis

A good sensitivity analysis will usually go beyond testing inputs one by one, instead co-varying inputs over wide ranges to understand sensitivity to differing combinations of parameters. Special attention may be needed both to halting conditions and to initialization (see BP6). Although increasing computational power makes conducting thorough sensitivity analysis easier, the importance of this step can act as a practical limit on model complexity and helps to motivate BP3. The design of (and results from) sensitivity analysis should be well documented and should serve the central goals of internal consistency and increased confidence in the robustness of results being reported.

Once a model has been implemented (and often after testing), it can be put to use. Many models are designed to yield specific insights or answer specific questions. Serving this goal requires designing clear experiments to conduct in the artificial world of the model.

BP10. Design clear experiments to yield clear insights

The accessibility and flexibility of ABM can lead to a temptation to “explore” the model’s behavior in an undirected way (Macy and Willer, 2002), but this rarely yields clear insights and can quickly become overwhelming. Thinking carefully about the questions of interest (see BP1) and how to design appropriate experiments to generate clear answers within the model is an important best practice for using ABM effectively and efficiently. For policy-oriented models, this may involve consideration of how to represent a “policy” or “intervention” in the model appropriately. Just as with model implementation, documentation of the experiments conducted with the model (specific parameterizations, code version used, and so on) is critical.

Results from model analysis (whether in testing, sensitivity analysis, or experimentation) can sometimes be surprising or counterintuitive. This can occur even in simple models (see BP3) but especially in more complex ones. A critical best practice is to investigate surprises so that why and how they arise can be understood (Axelrod, 2006a,b).

BP11. Always investigate surprising results, and make sure that you understand how they arise

This may require additional work (and even new elements of code to help track internal states of the model), but is crucial both to ensure that errors are caught and to effectively communicate complex and surprising results by providing intuition to accompany them.

6. Synthesis and Reporting

Once a model has been implemented, tested, and analyzed, the next step is to interpret the findings. Drawing appropriate conclusions from simulation results is not always straightforward, especially when models are stochastic, involve numerous inputs, and include multiple mechanisms. Caution is needed to avoid overclaiming (or underclaiming) and to convey appropriate nuance and uncertainty in findings (see BP14).

BP12. Draw appropriate conclusions from the model analysis

Forthright disclosure of findings, design issues, and sensitivity of results to input assumptions is important. Transparency in (and documentation of) the process used to design, implement, and analyze the model is important. Engagement with subject-matter experts or stakeholders may be needed at the stage of conceptual interpretation.

For almost any goal, an important step for a computational model such as an ABM is to translate the quantitative output of the simulation back into conceptual language that is appropriate for the intended audience. This may involve connecting the model and its results to an existing literature or conversation. ABM in particular can often lend itself to very visual depictions of model dynamics, and designing and executing effective visualization can often be a time-consuming process (and may involve additional computer programming).

BP13. Visualize and translate results into conceptual language

For policy-oriented use, particular care is needed to ensure that visualizations and conceptual descriptions of model findings are designed with the likely audience in mind and are accurately representing the modeling results (including nuance and uncertainty in findings). Visualization and conceptual description may also cover the analysis and testing procedures used.

Tension can arise with complex models between the goal of descriptive accuracy and the goal of clear communication to a nontechnical audience (Happe et al., 2006). Managing this tension is facilitated by starting with simple models and building up complexity in layers with a clear sense of the contribution of each layer to the outcomes (see BP3 and Macy and Willer, 2002; Tesfatsion and Judd, 2006).

A central issue in conveying models and their results, particularly in a policy context, is managing and communicating uncertainty appropriately. This involves first quantifying uncertainty and its origins—often a distinction is made between “aleatory variability” (natural randomness in a process that cannot be removed) and “epistemic uncertainty” (driven by limited data or knowledge) (see Berger and Troost, 2014). Uncertainty must then be effectively communicated along with the model design and results, and expectations about accuracy in forecasting or policy assessment must be managed.

BP14. Manage and communicate uncertainty appropriately

Even the best models almost never remove uncertainty and the need for judgment in interpreting results and applying them to real-world situations. Conveying the degree of uncertainty, and its nature and source, is often a key task for a modeler working in a policy context. Recognition (both by modelers and by model consumers) that models are just one input in the decision-making process is important.

The use of modeling to inform a decision process may go beyond the design, execution, and interpretation of any single model. For complex real-world problems and decisions, the use of multiple models or multiple methods can be particularly helpful (see section 2.1 discussion of modeling networks, such as MIDAS).

BP15. Consider multiple models or methods in the context of a broader decision-making process

Models may be designed independently to answer the same question (giving additional confidence where they agree) or may be designed to complement one another by covering different parts of a topic to preserve parsimony within each individual model while increasing the scope of the overall effort. Models may also sometimes be linked directly (for example, outputs of one model used as inputs in another model), but this requires consideration early in the design process.

4.4. A Few Considerations for Model Consumers

The sections above have described the many elements (section 4.1), steps (section 4.2), and best practices (section 4.3) involved in constructing and using ABM. With these in mind, a few guidelines arise for decision-makers who wish to use modeling as an input into the decision process.

Early engagement with the modeling effort can be helpful in communicating the goal or question of interest to model designers and in ensuring that the fit between the desired use of the model and the method and design of the model is appropriate. Model consumers are not always involved in the design and implementation phases of modeling, but they can be. Engagement in design itself can take many forms but often involves helping to ensure “face validity” and relevance of key design choices. Model consumers should ask the right questions throughout the process to ensure that they understand the choices being made.

Later in the process, model consumers should also ask the right questions to ensure that they fully understand the results and their boundaries and possible interpretations, the sensitivity of results to assumptions, and the role of uncertainty. It can be especially helpful for decision makers to understand the intuition and pathways behind results that seem counterintuitive.

4.5. Common Misperceptions About Agent-Based Modeling

A few misperceptions about the use of ABM commonly arise, especially in its use for policy purposes, and are worth brief discussion and clarification. One common misperception is that ABMs are necessarily “ad hoc” or reliant on poorly grounded inputs and assumptions. ABMs certainly can suffer from this problem (as can models of all types!) but they needn’t—there is nothing inherent in the ABM method that prohibits well-grounded assumptions. As described above (see section 4.3), care is needed in model design to motivate and ground assumptions and to avoid growing models too rapidly in complexity and stretching the ability of the modeling team to defend assumptions and explore sensitivity. The flexibility and individual-level focus of ABM confer great power, but they also require careful attention to and responsibility for assumptions on the part of the modeler.

A second common misperception concerns reuse of models. As described above, good models are usually designed for quite specific purposes with clear questions and boundaries in mind, and modelers make many specific implementation choices that flow from these goals. One implication of this is that “repurposing” models to answer questions or address topics and contexts for which they were not designed must be done with great care. Models can be used effectively in this way, but it requires carefully revisiting assumptions and design choices to ensure that they remain appropriate for the new application.

A third common misconception is in regard to the skill set required for ABM. Although ABMs are computational models, their rigorous design and use require much more than the skill of computer programming or computer science. Navigating the many elements and challenges of design, implementation, and interpretation of ABM requires another skill, the skill of “modeling,” and benefits from extensive experience and familiarity with the best practices outlined above.


This paper has reviewed the many potential uses of ABM to inform policy or decision making, the features of the technique that make it compelling to use for such purposes, and the best practices involved in doing so responsibly and rigorously. The central message of the paper is that the use of ABM as an input into the policy process is promising and practical, but it is also challenging and complex. As the use of ABM in this way continues to become more widespread, I hope that the overview of key considerations given here will contribute to careful and appropriate use of this powerful tool.


  • Axelrod R. Resources for agent-based modeling. Princeton, NJ: Princeton University Press; The complexity of cooperation: Agent-based models of competition and collaboration. 1997a:206–222.
  • Axelrod R. The complexity of cooperation: Agent-based models of competition and collaboration. Princeton, NJ: Princeton University Press; 1997b.
  • Axelrod R. Comparing modeling methodologies. Ann Arbor, MI: Gerald R. Ford School of Public Policy; 2004.
  • Axelrod R. Advancing the art of simulation in the social sciences. Rennard JP, editor. Hershey, PA: Idea Group Reference; Handbook of research on nature-inspired computing for economics and management. 2006a:90–110.
  • Axelrod R. Agent-based modeling as a bridge between disciplines. Tesfatsion L, Judd KL, editors. Amsterdam, Netherlands: North Holland Publishing Co.; Handbook of computational economics. 2006b:1565–1584.
  • Axelrod R, Tesfatsion L. Guide for newcomers to agent-based modeling in the social sciences. Tesfatsion L, Judd KL, editors. Amsterdam, Netherlands: North Holland Publishing Co.; Handbook of computational economics. 2006:1647–1659.
  • Axelrod R, Hammond RA, Grafen A. Altruism via kin-selection strategies that rely on arbitrary tags with which they coevolve. Evolution. 2004;58(8):1833–1838. [PubMed: 15446434]
  • Axtell R. Effects of interaction topology and activation regime in several multi-agent systems. Washington, DC: The Brookings Institution; 2000.
  • Axtell RL, Epstein JM. Coordination in transient social networks: An agent-based computational model of the timing of retirement. Washington, DC: The Brookings Institution; 1999.
  • Axtell R, Axelrod R, Epstein JM, Cohen MD. Aligning simulation models: A case study and results. Computational & Mathematical Organization Theory. 1996;1(2):123–141.
  • Axtell RL, Epstein JM, Dean JS, Gumerman GJ, Swedlund AC, Harburger J, Chakravarty S, Hammond R, Parker J, Parker M. Population growth and collapse in a multiagent model of the Kayenta Anasazi in Long House Valley. Proceedings of the National Academy of Sciences of the United States of America. 2002;99(Suppl 3):7275–7279. [PMC free article: PMC128597] [PubMed: 12011406]
  • Bahr DB, Browning RC, Wyatt HR, Hill JO. Exploiting social networks to mitigate the obesity epidemic. Obesity (Silver Spring). 2009;17(4):723–728. [PubMed: 19148124]
  • Bendor J, Moe TM. An adaptive model of bureaucratic politics. The American Political Science Review. 1985;79(3):755–774.
  • Bendor J, Diermeier D, Ting M. A behavioral model of turnout. American Political Science Review. 2003;97(2):261–280.
  • Berger T, Troost C. Agent-based modelling of climate adaptation and mitigation options in agriculture. Journal of Agricultural Economics. 2014;65(2):323–348.
  • Berger T, Birner R, McCarthy N, Díaz J, Wittmer H. Capturing the complexity of water uses and water users within a multi-agent framework. Water Resources Management. 2007;21(1):129–148.
  • Bhavnani R, Miodownik D. Ethnic polarization, ethnic salience, and civil war. Journal of Conflict Resolution. 2009;53(1):30–49.
  • Brady M, Sahrbacher C, Kellermann K, Happe K. An agent-based approach to modeling impacts of agricultural policy on land use, biodiversity and ecosystem services. Landscape Ecology. 2012;27(9):1363–1381.
  • Brown DG, Page S, Riolo R, Zellner M, Rand W. Path dependence and the validation of agent-based spatial models of land use. International Journal of Geographical Information Science. 2005a;19(2):153–174.
  • Brown DG, Riolo R, Robinson DT, North M, Rand W. Spatial process and data models: Toward integration of agent-based models and GIS. Journal of Geographical Systems. 2005b;7(1):25–47.
  • Bruch EE, Mare RD. Neighborhood choice and neighborhood change. American Journal of Sociology. 2006;112(3):667–709.
  • Burke DS, Epstein JM, Cummings DA, Parker JI, Cline KC, Singa RM, Chakravarty S. Individual-based computational modeling of smallpox epidemic control strategies. Academic Emergency Medicine. 2006;13(11):1142–1149. [PubMed: 17085740]
  • COSSA (Consortium of Social Science Associations). OBSSR conference held on complex systems, health disparities & population health: Building bridges. 2014. [December 18, 2014]. http://www​​/volume33/OBSSRsystems.pdf .
  • Dawid H, Fagiolo G, editors. Agent-based models for economic policy design [Special issue] Journal of Economic Behavior & Organization. 2008;67(2)
  • Dawid H, Neugart M. Agent-based models for economic policy design. Eastern Economic Journal. 2011;37(1):44–50.
  • DeAngelis DL, Mooij WM. Individual-based modeling of ecological and evolutionary processes. Annual Review of Ecology, Evolution, and Systematics. 2005;36:147–168.
  • Epstein JM. Modeling civil violence: An agent-based computational approach. Proceedings of the National Academy of Sciences of the United States of America. 2002;99(Suppl 3):7243–7250. [PMC free article: PMC128592] [PubMed: 11997450]
  • Epstein JM. Toward a containment strategy for smallpox bioterror: An individual-based computational approach. Washington, DC: Brookings Institution Press; 2004.
  • Epstein JM. Why model? Journal of Artificial Societies & Social Simulation. 2008;11(4):12.
  • Epstein JM. Modelling to contain pandemics. Nature. 2009;460(7256):687. [PMC free article: PMC3785367] [PubMed: 19661897]
  • Epstein JM. Generative social science: Studies in agent-based computational modeling. Princeton, NJ: Princeton University Press; 2012.
  • Epstein JM, Pankajakshan R, Hammond RA. Combining computational fluid dynamics and agent-based modeling: A new approach to evacuation planning. PLoS ONE. 2011;6(5):e20139. [PMC free article: PMC3105018] [PubMed: 21687788]
  • Epstein JM, Parker J, Cummings D, Hammond RA. Coupled contagion dynamics of fear and disease: Mathematical and computational explorations. PLoS ONE. 2008;3(12):e3955. [PMC free article: PMC2596968] [PubMed: 19079607]
  • Eubank S, Guclu H, Kumar VA, Marathe MV, Srinivasan A, Toroczkai Z, Wang N. Modelling disease outbreaks in realistic urban social networks. Nature. 2004;429(6988):180–184. [PubMed: 15141212]
  • Farmer JD. A simple model for the nonequilibrium dynamics and evolution of a financial market. International Journal of Theoretical and Applied Finance. 2000;3(3):425–441.
  • Farmer JD, Foley D. The economy needs agent-based modelling. Nature. 2009;460(7256):685–686. [PubMed: 19661896]
  • FDA (U.S. Food and Drug Administration). Modeling and statistical methods for the regulatory assessment of tobacco products—a public workshop. Rockville, MD: Dec 6, 2013. 2013.
  • Ferguson NM, Cummings DAT, Fraser C, Cajka JC, Cooley PC, Burke DS. Strategies for mitigating an influenza pandemic. Nature. 2006;442(7101):448–452. [PubMed: 16642006]
  • Frederick R. Agents of influence. Proceedings of the National Academy of Sciences of the United States of America. 2013;110(10):3703–3705. [PMC free article: PMC3593873] [PubMed: 23471958]
  • Garcia R, Jager W. From the special issue editors: Agent-based modeling of innovation diffusion. Journal of Product Innovation Management. 2011;28(2):148–151.
  • Geanakoplos J, Axtell R, Farmer J, Howitt P, Conlee B, Goldstein J, Hendrey M, Palmer N, Yang C-Y. Getting at systemic risk via an agent-based model of the housing market. American Economic Review. 2012;102(3):53–58.
  • Germann TC, Kadau K, Longini IM, Macken CA. Mitigation strategies for pandemic influenza in the United States. Proceedings of the National Academy of Sciences of the United States of America. 2006;103(15):5935–5940. [PMC free article: PMC1458676] [PubMed: 16585506]
  • Glass TA, McAtee MJ. Behavioral science at the crossroads in public health: Extending horizons, envisioning the future. Social Science and Medicine. 2006;62(7):1650–1671. [PubMed: 16198467]
  • Guzy MR, Smith CL, Bolte JP, Hulse DW, Gregory SV. Policy research using agent-based modeling to assess future impacts of urban expansion into farmlands and forests. Ecology and Society. 2008;13(1):37.
  • Hall KD. Predicting metabolic adaptation, body weight change, and energy intake in humans. American Journal of Physiology-Endocrinology and Metabolism. 2010;298(3):E449–E466. [PMC free article: PMC2838532] [PubMed: 19934407]
  • Hall KD, Hammond RA, Rahmandad H. Dynamic interplay among homeostatic, hedonic, and cognitive feedback circuits regulating body weight. American Journal of Public Health. 2014;104(7):1169–1175. [PMC free article: PMC4056226] [PubMed: 24832422]
  • Hammond R. Endogenous transition dynamics in corruption: An agent-based computer model. Center on Social and Economic Dynamics; Washington, DC: Brookings Insitution; 2008. (Working Paper No. 19, revised).
  • Hammond R. Complex systems modeling for obesity research. Preventing Chronic Disease. 2009;6(3):A97. [PMC free article: PMC2722404] [PubMed: 19527598]
  • Hammond RA, Axelrod R. Evolution of contingent altruism when cooperation is expensive. Theoretical Population Biology. 2006a;69(3):333–338. [PubMed: 16427106]
  • Hammond RA, Axelrod R. The evolution of ethnocentrism. Journal of Conflict Resolution. 2006b;50(6):926–936.
  • Hammond RA, Ornstein JT. A model of social influence on body mass index. Annals of the New York Academy of Sciences. 2014;1331:34–42. [PMC free article: PMC4133329] [PubMed: 24528150]
  • Hammond RA, Ornstein JT, Fellows LK, Dube L, Levitan R, Dagher A. A model of food reward learning with dynamic reward exposure. Frontiers in Computational Neuroscience. 2012;6(82):1–8. [PMC free article: PMC3468814] [PubMed: 23087640]
  • Happe K, Kellermann K, Balmann A. Agent-based analysis of agricultural policies: An illustration of the agricultural policy simulator Agripolis, its adaptation and behavior. Ecology and Society. 2006;11(1):49.
  • Happe K, Balmann A, Kellermann K, Sahrbacher C. Does structure matter? The impact of switching the agricultural policy regime on farm structures. Journal of Economic Behavior & Organization. 2008;67(2):431–444.
  • Heckbert S. Agent-based modelling of emissions trading for coastal landscapes in transition. Journal of Land Use Science. 2011;6(2-3):137–150.
  • Heckbert S, Baynes T, Reeson A. Agent-based modeling in ecological economics. Annals of the New York Academy of Sciences. 2010;1185(1):39–53. [PubMed: 20146761]
  • Holland J. Adaptation in natural and artificial systems: An introductory analysis with applications to biology, control and artificial intelligence. Cambridge, MA: MIT Press; 1992.
  • Huang TTK, Grimm B, Hammond RA. A systems-based typological framework for understanding the sustainability, scalability, and reach of childhood obesity interventions. Children's Health Care. 2011;40(3):253–266.
  • IOM (Institute of Medicine). Bridging the evidence gap in obesity prevention: A framework to inform decision making. Kumanyika SK, Parker L, Sim LJ, editors. Washington, DC: The National Academies Press; 2010.
  • IOM (Institute of Medicine). Accelerating progress in obesity prevention: Solving the weight of the nation. Washington, DC: The National Academies Press; 2012. [PMC free article: PMC3648752] [PubMed: 22983849]
  • IOM (Institute of Medicine). Evaluating obesity prevention efforts: A plan for measuring progress. Washington, DC: The National Academies Press; 2013. [PMC free article: PMC3951801] [PubMed: 24618760]
  • IOM and NRC (National Research Council). A framework for assessing effects of the food system. Nesheim MC, Oria M, Yih PT, editors. Washington, DC: The National Academies Press; 2015.
  • Kollman K, Miller JH, Page SE. Adaptive parties in spatial elections. American Political Science Review. 1992;86(4):929–937.
  • Kollman K, Miller JH, Page SE. Political institutions and sorting in a Tiebout model. The American Economic Review. 1997;87:977–992.
  • Laver M. Policy and the dynamics of political competition. American Political Science Review. 2005;99(2):263–281.
  • Laver M, Sergenti E. Party competition: An agent-based model. Princeton, NJ: Princeton University Press; 2011.
  • LeBaron B, Winkler P, editors. Agent-based models for economic policy advice [Special issue] Journal of Economics and Statistics. 2008;228(2–3)
  • Lee BY, Brown ST, Korch G, Cooley PC, Zimmerman RK, Wheaton WD, Zimmer SM, Grefenstette JJ, Bailey RR, Assi TM, Burke DS. A computer simulation of vaccine prioritization, allocation, and rationing during the 2009 H1N1 influenza pandemic. Vaccine. 2010;28(31):4875–4879. [PMC free article: PMC2906666] [PubMed: 20483192]
  • Longini IM, Nizam A, Xu S, Ungchusak K, Hanshaoworakul W, Cummings DAT, Halloran ME. Containing pandemic influenza at the source. Science. 2005;309(5737):1083–1087. [PubMed: 16079251]
  • Longini IM Jr, Elizabeth Halloran M, Nizam A, Yang Y, Xu S, Burke DS, Cummings DA, Epstein JM. Containing a large bioterrorist smallpox attack: A computer simulation approach. International Journal of Infectious Diseases. 2007;11(2):98–108. [PubMed: 16899385]
  • Luke DA, Sorg AA, Mack-Crane A, Hammond RA, Kasman ME, Ribisl KM, Henriksen L. “Tobacco Town: Modeling the effects of tobacco retail reduction.”. State and Community Tobacco Control Initiative, National Cancer Institute; Chicago, IL: 2014. (Poster presentation at Annual Investigators Meeting).
  • Mabry PL, Marcus SE, Clark PI, Leischow SJ, Méndez D. Systems science: A revolution in public health policy research. American Journal of Public Health. 2010;100(7):1161–1163. [PMC free article: PMC2882409] [PubMed: 20530757]
  • Macy MW, Willer R. From factors to actors: Computational sociology and agent-based modeling. Annual Review of Sociology. 2002;28(1):143–166.
  • Magliocca NR, Brown DG, Ellis EC. Cross-site comparison of land-use decision-making and its consequences across land systems with a generalized agent-based model. PLoS ONE. 2014;9(1):e86179. [PMC free article: PMC3906050] [PubMed: 24489696]
  • Manson SM, Evans T. Agent-based modeling of deforestation in southern Yucatán, Mexico, and reforestation in the midwest United States. Proceedings of the National Academy of Sciences of the United States of America. 2007;104(52):20678–20683. [PMC free article: PMC2413086] [PubMed: 18093928]
  • Maroulis S, Bakshy E, Gomez L, Wilensky U. Modeling the transition to public school choice. Journal of Artificial Societies and Social Simulation. 2014;17(2):3.
  • Miller JH, Page SE. Complex adaptive systems: An introduction to computational models of social life. Princeton, NJ: Princeton University Press; 2007.
  • Nader PR, Huang TTK, Gahagan S, Kumanyika S, Hammond RA, Christoffel KK. Next steps in obesity prevention: Altering early life systems to support healthy parents, infants, and toddlers. Childhood Obesity (Formerly Obesity and Weight Management). 2012;8(3):195–204. [PubMed: 22799545]
  • North MJ, Macal CM, Aubin JS, Thimmapuram P, Bragen M, Hahn J, Karr J, Brigham N, Lacy ME, Hampton D. Multiscale agent-based consumer market modeling. Complexity. 2010;15(5):37–47.
  • Nowak MA. Five rules for the evolution of cooperation. Science. 2006;314(5805):1560–1563. [PMC free article: PMC3279745] [PubMed: 17158317]
  • Ohtsuki H, Hauert C, Lieberman E, Nowak MA. A simple rule for the evolution of cooperation on graphs and social networks. Nature. 2006;441(7092):502–505. [PMC free article: PMC2430087] [PubMed: 16724065]
  • Page SE. Computational models from A to Z. Complexity. 1999;5(1):35–41.
  • Page SE. Path dependence. Quarterly Journal of Political Science. 2006;1(1):87–115.
  • Rand W, Rust RT. Agent-based modeling in marketing: Guidelines for rigor. International Journal of Research in Marketing. 2011;28(3):181–193.
  • Schelling TC. Dynamic models of segregation. Journal of Mathematical Sociology. 1971;1(2):143–186.
  • Schlüter M, Pahl-Wostl C. Mechanisms of resilience in common-pool resource management systems: An agent-based model of water use in a river basin. Ecology and Society. 2007;12(2):4.
  • Sterman JD. Learning from evidence in a complex world. American Journal of Public Health. 2006;96(3):505–514. [PMC free article: PMC1470513] [PubMed: 16449579]
  • Sun S, Parker DC, Huang Q, Filatova T, Robinson DT, Riolo RL, Hutchins M, Brown DG. Market impacts on land-use change: An agent-based experiment. Annals of the Association of American Geographers. 2014;104(3):460–484.
  • Tesfatsion L, Judd KL. Handbook of computational economics: Agent-based computational economics. Vol. 2. Amsterdam, Netherlands: North Holland Publishing Co.; 2006.
  • Xie Y, Zhou X. Modeling individual-level heterogeneity in racial residential segregation. Proceedings of the National Academy of Sciences of the United States of America. 2012;109(29):11646–11651. [PMC free article: PMC3406804] [PubMed: 22753466]
  • Yang Y, Sugimoto JD, Halloran ME, Basta NE, Chao DL, Matrajt L, Potter G, Kenah E, Longini IM. The transmissibility and control of pandemic influenza a (H1N1) virus. Science. 2009;326(5953):729–733. [PMC free article: PMC2880578] [PubMed: 19745114]
  • Zhang J, Tong L, Lamberson P, A Durazo R, Luke A, Shoham DA. Leveraging social influence to address overweight and obesity using agent-based models: The role of adolescent social networks. Social Science and Medicine. 2014;125:203–213. [PMC free article: PMC4306600] [PubMed: 24951404]



ABM simulations typically involve stochastic elements to represent phenomena that may be either inherently unpredictable (for example, everyone with whom an agent will come into contact on a given day) or about which the researcher cannot make precise assumptions (such as the explicit structure of a complex network).


For more information, see http://www​

Copyright 2015 by the National Academy of Sciences. All rights reserved.
Bookshelf ID: NBK305917


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (1.9M)

Related information

  • PMC
    PubMed Central citations
  • PubMed
    Links to PubMed

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...