Special Issue Article

Social Psychological and

Personality Science
Agent-Based Modeling: 2017, Vol. 8(4) 387-395
ª The Author(s) 2017
Reprints and permission:
A Guide for Social Psychologists sagepub.com/journalsPermissions.nav
DOI: 10.1177/1948550617691100
journals.sagepub.com/home/spp

Joshua Conrad Jackson1, David Rand2,3,4, Kevin Lewis5,

Michael I. Norton6, and Kurt Gray1

Abstract
Agent-based modeling is a long-standing but underused method that allows researchers to simulate artificial worlds for hypothesis
testing and theory building. Agent-based models (ABMs) offer unprecedented control and statistical power by allowing
researchers to precisely specify the behavior of any number of agents and observe their interactions over time. ABMs are
especially useful when investigating group behavior or evolutionary processes and can uniquely reveal nonlinear dynamics and
emergence—the process whereby local interactions aggregate into often-surprising collective phenomena such as spatial segre-
gation and relational homophily. We review several illustrative ABMs, describe the strengths and limitations of this method, and
address two misconceptions about ABMs: reductionism and “you get out what you put in.” We also offer maxims for good
and bad ABMs, give practical tips for beginner modelers, and include a list of resources and other models. We conclude with a
seven-step guide to creating your own model.

Keywords
agent-based modeling, computational social science, social dynamics, social psychologies

From Detroit to El Paso, New York to Los Angeles, urban if low-tech—example of the power of agent-based modeling
environments are divided by race and ethnicity. The pernicious (ABM)1 in understanding emergent social behavior.
consequences of segregation lead us to infer pernicious causes:
People must live in homogenous neighborhoods because they
are racist. This explanation for segregation seems plausible,
as prejudiced individuals do avoid people of other races—but
Agent-Based Modeling
it assumes that the collective behavior of neighborhoods can Agent-based models (ABMs) are computational simulations in
be explained similarly to the behavior of individuals. Almost which artificial entities interact over time within customized
40 years ago, Thomas Schelling (1971) challenged this environments. These entities (agents) are programmed to repre-
assumption, asking whether segregated neighborhoods would sent humans who behave in precisely specified ways. As sum-
form even when individuals had no prejudice, and only wanted marized by Macy and Flache (2009, p. 247), agents are
a few neighbors similar to themselves. adaptive in that they respond to their environment through
Schelling placed red and green pennies on a chessboard to learning and evolution and are autonomous in that they control
represent people in neighborhoods. People were happy—and their own goals, states, and behaviors. They are also intention-
remained in their square—if they were surrounded by at least ally simplified, usually following only one or two basic rules
30% of their “color”; if this number dropped below 30%, how-
ever, people became unhappy and moved to a new square.
1
Schelling played out this model by moving pennies one by one Department of Psychology and Neuroscience, University of North Carolina,
until each person on the board was happy, by which time the Chapel Hill, NC, USA
2
Department of Psychology, Yale University, New Haven, CT, USA
board was highly color segregated. At higher (75%) or lower 3
Department of Economics, Yale University, New Haven, CT, USA
(15%) thresholds of similarity, segregation was more or less 4
School of Management, Yale University, New Haven, CT, USA
5
pronounced (see Figure 1 for an illustration of these effects), Department of Sociology, University of California, San Diego, CA, USA
6
but the key is that even individuals who embraced high diver- Harvard Business School, Cambridge, MA, USA
sity could still end up segregated.
Corresponding Author:
Schelling’s work is an elegant testament to how simple and Joshua Conrad Jackson, Department of Psychology and Neuroscience, Uni-
innocent individual preferences can produce surprising societal versity of North Carolina, Chapel Hill, NC 27599, USA.
outcomes over time. His model also serves as a prototypical— Email: [email protected]
388 Social Psychological and Personality Science 8(4)

Figure 2. Visualization of Gray and colleagues’ (2014) model

displayed at Round 1 (left panel) and Round 300 (right panel).
Retrieved from online simulation at http://www.mpmlab.org/groups/

intelligence), we focus on models of social processes. We hope

to provide an in-depth but accessible introduction to ABM for
Figure 1. Visualization of Thomas Schelling’s (1971) segregation
model at its commencement (top panel) and conclusion (bottom social psychologists.
panels). When agents have a 15% threshold for similarity (left panel),
only minimal segregation occurs. However, 30% (middle panel)
and 75% (right panel) thresholds produce striking segregation. Social Psychological Questions Addressed
Retrieved from http://nifty.stanford.edu/2014/mccown-schelling- by ABMs
model-segregation/
Schelling’s (1971) model of segregation addresses one of
social psychology’s core questions: Why do individuals segre-
(representing habits, norms, or preferences) throughout the
gate based on race? ABMs also address other important ques-
simulation.
tions: What is the basis of group formation? What is the best
The outcomes of ABMs, however, are anything but simple.
strategy for maintaining cooperation? Why do couples pair off
A well-programmed model offers insight into how local inter-
in terms of attractiveness? These questions are well suited to
actions between agents can lead to complex group- and system-
ABM because they involve individual behaviors interacting
level phenomena. Consider how a single bird’s tendency to
to produce surprising collective phenomena.
align and remain close (but not too close) to her peers can cre-
ate a swirling flock that appears to be moving with a collective
mind (Reynolds, 1987) or how predator–prey interdependence What Is the Basis of Group Formation?
can explain animal species’ resurgence following near extinc- Social identity is the dominant framework for understanding
tion (Borschev & Filippov, 2004). ABMs are uniquely why people split into “us” versus “them”: People with similar
equipped to shed light on such phenomena and countless other race, religion, or culture form groups, which then square off
applications involving interacting individuals. against each other (Tajfel, 1982). However, these social identi-
Perhaps because of their ability to simulate large-scale ties can only emerge once people separate into groups. This
dynamics with bottom-up processes, ABMs are popular in eco- logic creates a regress in which groups require identity but
nomics (e.g., Tesfatsion & Judd, 2006), sociology (Bruch & identity requires groups. To escape this chicken–egg dilemma,
Atwell, 2015; Macy & Willer, 2002), political science (Ceder- Gray and colleagues (2014) examined whether groups could
man, 2005; Johnson, 1999), and some applied sciences (e.g., form in a completely homogeneous population without any
artificial intelligence; Beer, 1995; Gasser, Braganza, & identities. The authors programmed agents with only two sim-
Herman, 1987; Wooldridge, 2003). In psychology, however, ple characteristics: reciprocity (the tendency to cooperate with
ABMs continue to exist at the field’s margins (see Goldstone those who have previously cooperated with you) and transitiv-
& Janssen, 2005; Smith & Conrey, 2007) perhaps because ity (the tendency to share your network’s social preferences)—
psychologists view them as difficult to implement and see their each of which was a well-established social tendency (Holland
results as only reflecting the assumptions of their programmers & Leinhardt, 1971; Levine, 1998). The model’s results
(you get out what you put in). revealed robust group formation even though agents had no
This article aims to address these concerns and to pique sense of us or them, suggesting that groups can form even with-
social psychologists’ interest in ABMs. We provide examples out identity (see Figure 2).
of classic and recent ABMs that illuminate social behavior,
compare modeling to other methods in social psychology, and
give concrete advice to social psychologists wishing to What Is the Best Strategy for Maintaining Cooperation?
implement their own ABMs. Although there are ABMs that Real-world questions of cooperation are captured by the
simulate nonsocial events (e.g., weather patterns or artificial “prisoner’s dilemma,” in which two people each have the
Jackson et al. 389

choice to cooperate or defect. The group payoff is maximized Emergence

when both people cooperate, but each player is made better off
Agents in the previous examples were not programmed to seg-
individually by defecting—capturing the essential tension of
regate, to form social groups, to maintain stable cooperation, or
social dilemmas. Political scientist Robert Axelrod asked peo-
to find partners of a similar attractiveness. Instead, these group
ple to program agents with different strategies for repeated
phenomena arose via emergence—when the aggregation of
prisoner dilemma games (e.g., always cooperate, always
small-scale individual behavior yields qualitatively different
defect, copy your partner’s past behavior) and then paired these
collective behavior. Emergence lies at the heart of almost any
agents with each other in a round-robin design (Axelrod, 1980;
complex phenomenon, from traffic jams, to the wetness of
Axelrod & Hamilton, 1981). As long as the agents engaged in
water, to the neural basis of consciousness (Bassett & Gazza-
repeated interactions, the winner was a very simple agent—
niga, 2011; Tononi, Sporns, & Edelman, 1994). For example,
“tit-for-tat”—which began with cooperation and then copied its
while no individual neuron is conscious, their collective inter-
partner’s previous decision. Axelrod’s ABM was important
actions yield human consciousness. Likewise, Schelling’s
because it revealed a simple route for the emergence of coop-
model revealed that segregation could arise from the innocent
eration, even in complex societies.
decisions of relatively egalitarian individuals.
More recently, Bear and Rand (2016) developed an ABM to
Historically, the impact of ABMs has been proportional to
explore the psychological basis of cooperation. Agents played
the amount of emergence they reveal—the apparent disconnect
either one-shot or repeated prisoner’s dilemmas. They could
between individual and collective behavior. For example, the
engage in two different kinds of cognition: a low-cost general-
models from the previous section feature large-scale phenom-
ized intuitive response or a higher cost calculated response that
ena that are difficult to predict from individuals’ behavior.
could tailor its choice to whether the game was one-shot or
Importantly, in explaining complex group-level phenomena
repeated. The results showed that—given a high likelihood of
with simple individual-level rules (see Smaldino, 2014), good
repeated interaction—the best strategy was to intuitively coop-
ABMs typically reduce complexity—leading to these two
erate and deliberatively defect when the game was one-shot.
complementary maxims for research with ABM:
This ABM therefore offered an evolutionary explanation for
why people sometimes cooperate when they can get away with Maxim for good ABMs: Reduce complexity by revealing
defection. how higher level phenomena emerge from the repeated
interaction of simple rules.
Maxim for bad ABMs: Introduce complexity by taking a
Do Couples Seek Out Similarly Attractive Partners? simple phenomenon and inventing complicated rules to
Members of a romantic couple tend to be similarly attractive, explain it.
but it is not immediately clear why. Although some believed These maxims serve as useful criteria in evaluating whether
that people intentionally search for their attractiveness ABMs add to or detract from a paper. The very best ABMs are
“match” (Huston, 1973; White, 1980), Kalick and Hamilton explainable in plain prose and should reveal the emergence of
(1986) used an ABM to test whether matching could occur complex or surprising phenomena using simple principles.
even if all people preferred maximally attractive partners. Conversely, bad ABMs take a straightforward, intuitive phe-
Heterosexual male and female agents were assigned an attrac- nomenon and complicate it with unjustified assumptions and
tiveness score from 1 to 10 and were repeatedly paired up. abstruse mathematics. These maxims also help to address two
Pairs asked each other on “dates,” and if both agreed, they left traditional criticisms of ABMs.
the pool, otherwise they were paired up with new agents.
Kalick and Hamilton ran two variations of the model: one
in which people wanted maximally attractive partners (moti-
vated for supermodels) and another where people wanted Reductionism
similarly attractive partners (motivated for matching). In the ABMs are often seen to be reductionist, destroying the special-
“motivated for matching” condition, agents’ attractiveness ness of psychological processes by explaining them with sim-
was very highly correlated (r ¼ .85) with their partners’—sig- ple agent behaviors. For example, claims of reductionism
nificantly higher than what actually occurs in real life. In have been leveled against research linking love to hor-
contrast, agents who were “motivated for supermodels” had mones—if hormones are involved in love, is love “just” hor-
their attractiveness moderately correlated (r ¼ .5) with their mones? But fears of reductionism ignore the possibility of
partners’—nearly the same correlation as in real life (Critelli emergence, and the fact that all phenomena are embedded in
& Waid, 1980). This moderate matching occurred because a chain of lower and higher level events. Even if love can be
when everyone preferred the prettiest people, the prettiest “reduced” to hormones, there is still an undeniably powerful
ended up together first, and the less pretty were left to pair feeling of love, a higher level emergent experience that moti-
up afterward. As with many ABMs, people’s individual vates people to write sonnets and run through the airport at the
preferences (for attractive partners) led to unexpected collec- last minute. Emergence also provides a defense against claims
tive patterns (attractiveness matching). of reductionism in ABMs. Even if reciprocity and transitivity
390 Social Psychological and Personality Science 8(4)

Table 1. Comparing Agent-Based Modeling to Other Methods.

Research Aspect Field Studies Lab Experiment Archival Studies ABMs

Control and realism Low control; high realism Medium control; medium realism Low control; medium realism High control; low realism
Scale Medium to high scale Low to medium scale High scale High scale
Nonlinear dynamics Medium visibility Low visibility Medium visibility High visibility
Mechanism Medium clarity High clarity Low clarity High clarity
Note. ABMs ¼ agent-based models.

are sufficient conditions for group genesis (Gray et al., 2014), (Shariff, Willard, Anderson, & Norenzayan, 2016), or even
groups themselves prompt powerful feelings of solidarity and their transient mood (Forgas, 1995). Despite the flexibility of
important behaviors—from war to religious movements— experiments, they are also limited by questions of ethics and
which cannot be reduced to these lower level processes. feasibility—there is only so much that participants can do (or
be asked to do) in the lab. In contrast, ABMs offer exceptional
You Get Out What You Put In control: Agents in computational models can be instructed to
perform almost any initial behaviors and will follow their
Critics of ABM have also claimed that the results of ABMs are instructions with complete uniformity. This control also
closely tied to researchers’ decisions in setting their models’ remains high over indefinitely large samples and infinitely long
parameters. In some sense, this is a strength of ABMs: Unlike simulations.
in the laboratory or the field, the behavior of agents can be iso- The trade-off to ABMs’ high control is a low degree of
lated and specified with precision—which forces researchers to external validity. For example, the agents in Schelling’s model
explicitly formulate their theories. ABM-derived hypotheses moved neighborhoods without incurring the financial or social
are therefore decidedly falsifiable, with no ambiguity about costs inherent in relocation. Kalick and Hamilton’s date choice
what a model should predict. Of course, this level of experi- model similarly assumed that individuals who accept dates per-
menter control has the potential to make the final outcome manently leave the dating pool, which seldom occurs in real
seem obvious—but again, this criticism holds primarily with life. Because of these shortcomings, ABMs are most effective
models that fail to show emergence. In Schelling (1971), there when used in conjunction with laboratory or field experiments,
is nothing obvious about a slight preference for similarity caus- which can use human subjects to validate an ABM’s
ing rampant segregation, and in Gray and colleagues (2014), parameters (as in Luhmann & Rajaram, 2015) or its causal
there is nothing obvious about two simple rules of interac- pathways (see Bear & Rand, 2016; Kalick & Hamilton, 1986).
tion—reciprocity and transitivity—leading to stable grouping
within homogenous populations.
Scale
Comparing ABM to Other Methods One clear advantage of ABMs over other methods is statistical
power. Obtaining sufficient N can prove difficult, as research-
In addition to the theoretical framework of emergence, ABMs
ers struggle against a subject pool deadline or limited funding
offer several methodological advantages that complement
for participants. Even in field studies, researchers may obtain
other methods. In comparison to laboratory experiments, field
large sample sizes, but these samples may be incomplete or fea-
studies, or archival investigations (including “big data” analy-
ture troublesome attrition. In ABMs, sample size is simply a
sis), ABMs offer a unique combination of experimental control
parameter specified in the model. ABMs can also operate over
and massive scale, along with the ability to capture nonlineari-
any amount of time and sample at any rate. Of course, large N,
ties and underlying mechanisms. However, like any tool in a
long-term and high sampling-rate ABMs may take longer to
social psychologist’s toolbox, ABMs come with limitations,
run, but this typically means extras days and not years (and
of which external validity is most notable. This drawback is
computing superclusters can substantially reduce this time).
mitigated by supplementing ABMs with other tools—such as
The critical point is that by analyzing large samples over an
laboratory or field experiments—in multimethod investiga-
extended time, ABMs can reveal large-scale societal emer-
tions. Table 1 shows a comparison of the relative advantages
gence (e.g., segregation and homophily), which is often impos-
and disadvantages of ABMs compared to other methods.
sible to observe with more traditional paradigms (and even
with “big data” analyses; Lewis, 2015).
Control and Realism
In psychology, maximum control is often ascribed to experi-
mental lab paradigms featuring random assignment, but even
Nonlinear Dynamics
experiments have their limits. Participants may respond differ- Most social psychology paradigms often only assess the beha-
ently to experimental manipulations based on their cultural vior of one group at one specific time point, but social pro-
background (Hong et al., 2003), their religious upbringing cesses unfold dynamically across time and individuals.
Jackson et al. 391

Considering conformity, People generally follow behaviors National Academy of Sciences (Vol. 99, Supplement 3,
more as they become more common (Asch, 1956; Boyd & 2002). Finally, websites like “OpenABM” (www.openab
Richerson, 1985; Henrich & McElreath, 2003), except for non- m.org) and “Agent-Based Models” (www.agent-based-mod
conformists who follow the behavior less (Efferson, Lalive, els.com) provide courses, videos, and code libraries of pre-
Richerson, McElreath, & Lubell, 2008). As a result, conformity vious models from which researchers can adapt code.
follows an oscillating pattern of increases, decreases, and Aspiring ABMers must develop some level of computer
stability, which is difficult to fully capture with static experi- programming. Python, MATLAB, R, and C have often been
ments (Jarman et al., 2015). The spread of social attitudes previously used to program ABMs. However, there are also
(Nowak, Szamrej, & Latane, 1990) and stereotypes (Kashima, more accessible tools available for those who do not have time
2000) and the process of group formation (Halberstadt et al., to master a traditional programming language. The software
2016; Jackson, Halberstadt, Jong, & Feldman, 2015) also fol- package Netlogo (http://ccl.northwestern.edu/netlogo/; Tisue
low nonlinear patterns. In fact, there are few social phenomena & Wilensky, 2004) is free and relatively simple and provides
that behave truly linearly over time, given the dynamic nature the code and explanation behind several of the models in this
of social–cultural interactions and the unpredictable impacts of article, such as the predator–prey model, the flocking model,
initial conditions (Vallacher & Nowak, 1999). ABMs are an and Schelling’s segregation model. Netlogo also comes with
ideal method for modeling these nonlinear processes, as they an extensive manual for researchers to learn the programming
can include millions of time points and multiple runs (Abbott, language as well as practical tips for building an ABM. Other
1988). tools that offer ABM training include “Swarm” (Minar, Burkhart,
Langton, & Askenazi, 1996), which requires some program-
Mechanism ming ability (C or Java) but comes with a tutorial and example
code to get new users started, and “Flexible Large-Scale Agent
With their high controllability, ABMs are often able to isolate Modeling Environment,” which is a more accessible computa-
and directly manipulate the discrete psychological processes tional environment, since models are specified in XML.
underlying complex social phenomena. Of course, psychologi- “Cellular Automaton Explorer” offers a manageable interface
cal mechanisms can take many forms and can exist on many to program simple ABMs and is particularly well suited for
levels of analysis. ABMs are best suited to study how manifes- demonstration purposes (see, e.g., a popular Wolfram demon-
tations of individual (or dyadic) behavior influence larger scale stration: http://demonstrations.wolfram.com/CellularAutomato
group-level phenomena, such as when a slight individual desire nExplorer/).
for similarity catalyzes neighborhood segregation (Schelling, To augment these resources, we provide a seven-step con-
1971). One question is whether the mechanism provided by ceptual ABM algorithm, with each step illustrated by Schel-
ABM is the same in real life: Just because a mechanism suffi- ling’s (1971) segregation model and Gray and colleagues’
ciently generates some outcome does not mean this mechanism (2014) grouping model. For more examples of the seven steps,
necessarily or always generates the outcome. However, reveal- we also provide a substantial (though not exhaustive) supple-
ing even likely mechanisms is valuable for both basic research mental table with 35 additional ABMs on social–psychology
and policy decisions. topics ranging from the dynamics of online chatting to deci-
sions about expressing pain. This collection offers insight into
how other researchers have translated their research question
Building an ABM into simulations.
After being inspired by ABM’s rich history and unique metho- Some of these steps do not apply to all models or all research
dology, readers might want to try their hand at model building. questions, and so researchers should feel free to adapt them to
While training in ABMs is absent from most PhD programs in their own needs. Nevertheless, the steps provide a useful guide
social psychology, many articles have linked ABMs to specific for exploring social processes and for creating simulated
research questions (e.g., Axelrod, 1997; Carley, 2002; Schel- worlds with the potential for collective emergence.
ling, 1971) with others providing more detailed, technical
guides (e.g., Smith & Conrey, 2007). An edited volume by Tes- Step 1: What are your world’s dimensions? Is your world flat or
fatsion and Judd (2006) includes chapters on ABM’s history multidimensional? Schelling’s segregation model is two
and its applications in economics as well as an introductory dimensional (2-D)—like land—but group formation models
appendix with extensive practical tips for newcomers. Gilbert are often multidimensional to represent complex social spheres
and Troitzsch (2005) provide a broader overview of ABM in (although these models often still involve 2-D visualizations to
the social sciences. Epstein (2008) includes a discussion of present data). In choosing the dimensionality, researchers must
ABM’s benefits over other methodologies, and Nowak consider if the actions of one agent necessarily constrain the
(2004) gives an in-depth overview of emergence in ABM and behavior of other agents—the more the mutual constraints, the
the utility of simple models for simulating complex pro- lower the degrees of freedom and the lower the dimensionality
cesses. Journal issues focusing on ABMs have included (e.g., if I move across town from you, I not only move further
American Behavioral Scientist (Vol. 42, August 1999), Sci- from you but also your neighbor). Note that dimensions only
ence (Vol. 284, April 1999), and the Proceedings of the apply to models where interactions between agents are
392 Social Psychological and Personality Science 8(4)

Figure 3. In the lattice network (A), agents only interact with their neighbors (applicable to residential models). In the small-world network (B),
cross-network connections compliment neighboring connections, so that any two agents are connected by only a few degrees of separation
(applicable to almost any social network). In the scale-free network (C), densely connected agents are more likely to generate new connections
compared to sparsely connected agents (applicable to the Internet and citation networks).

governed by space. In network models, for example, there are & Szolnoski, 2010). For example, prisoner’s dilemmas can rep-
no dimensions. resent people’s decisions to either act selfishly or coopera-
tively. In any ABM, researchers should ensure that agents’
Application of Step 1. In Schelling’s model, agents were
behavior approximates the type of social behavior of interest,
paired in a 2-D space (as illustrated in Figure 1), while in Gray
which often involves programming in a degree of randomness
and colleagues’ grouping model, agents interacted in a multidi-
for variability.
mensional space where one agent’s position did not impede
other agents’ movement. Application of Step 3. In Schelling’s model, agents decided
whether to stay in their neighborhood or to move to another
vacant space on the grid. Gray and colleagues’ agents played
Step 2: How do agents meet? Behavior in ABMs is usually
a prisoner’s dilemma game.
divided into rounds, and on each round, some number of agents
interact with each other. One question is how to select which
agents interact. Do they interact only with their neighbors, or Step 4: What is the payoff? Payoffs correspond to what agents
can they be paired up with any other agent in the simulation? get out of an interaction and can represent money, happiness,
These choices stem in part from the dimensionality (see Step or social bonds. In some ABMs, there is no payoff system, but
1), but there are other choices within each of these sets. In some in many ABMs that feature interactive decision-making, pay-
models, agents can avoid interactions entirely—perhaps offs are determined by considering an agent’s decisions and
because they are “unpopular”—while in others, agents can those of that agent’s partner(s). In a prisoner’s dilemma, for
interact with more than one agent. In the latter case, what rules example, an agent’s decision to cooperate yields a different
will govern interaction order? And will agents prioritize some payoff depending on whether their partner chooses to also
interaction partners over others? Will interactions be governed cooperate or to defect.
randomly or according to a rule (or a bit of both)? The answers Application of Step 4. Schelling’s agents received no payoff,
to these questions (along with your world’s dimensionality) since there was no interactive decision-making. Gray and col-
will determine the network you choose for your model. Three leagues’ agents, however, received a payoff that depended on
popular networks are displayed in Figure 3. their prisoner’s dilemma decisions.
Application of Step 2. Since Schelling’s segregation model
focused on neighborhood dynamics, he programmed agents Step 5: How do agents change? Agents can change in a number
to only interact with their next-door neighbors. In contrast, of ways throughout the simulation. In many economic models,
Gray and colleagues’ agents could interact with any other agent agents “remember” the way their counterpart treated them and
in the model, though they were more likely to interact with adjust their behavior in future rounds. In evolutionary models,
“friends” than with “enemies”—and they only interacted with each round will end with some agents dying (often if they have
one partner per round. received a low payoff) or reproducing (often if they have
received a high payoff). In mating models, agents can pair up
(or break up). In models where agents form groups, agents can
Step 3: How do agents behave? When agents meet, what do they
become closer to some agents and move further from others.
do? Do they ask other agents on dates (Kalick & Hamilton,
1986)? Do they share food (Jahanbazi, Frantz, Purvis, Purvis, Application of Step 5. Both Schelling’s and Gray and col-
& Nowostawski, 2014)? In many social science ABMs, agents leagues’ agents changed via movement, moving to a randomly
repeatedly play economic games, which allows for experimen- selected grid space (Schelling) or closer to those who treated
ters to mathematically approximate real social behavior (Perc them nicely (i.e., their friends; Gray and colleagues).
Jackson et al. 393

Step 6: How long does your world last? As mentioned earlier, one fail to generate new knowledge. As we suggest, good ABMs
of the major advantages of ABMs is their scale. Researchers harness the power of emergence, in which higher level phe-
can collect data for any specified amount of time, meaning that nomena derive from the simple behavior of agents. As with any
an ABM investigation will almost never be underpowered. method, ABM is imperfect, but it does offer social psycholo-
However, researchers should set a theoretically meaningful gists a powerful way to implement precise hypotheses and to
length to their model. In some cases, models should run until explore emergence. Not only can researchers build whole
they have reached some form of equilibrium. In other cases, worlds to examine social processes, they also can sample from
models should run for a length that approximates some phe- these worlds over thousands of generations to yield unprece-
nomenon of interest (e.g., Luhmann & Rajaram’s, 2015, model dented insight into collective behavior. Whether studying rela-
of collective memory) but still allows the researcher to conduct tionships, stereotypes, culture, attitudes, emotions, religion, or
analyses with adequate reliability. In either case, decisions are the self, social psychologists should consider adding ABM to
limited only by (practically unlimited) computer storage space their methodological toolbox.
and CPU speed.
Declaration of Conflicting Interests
Application of Step 6. Both Schelling’s and Gray and col-
leagues’ models ran until a point of equilibrium. In Schelling’s The author(s) declared no potential conflicts of interest with respect to
model, this equilibrium was the point at which agents were no the research, authorship, and/or publication of this article.
longer moving across neighborhoods. For Gray and colleagues,
equilibrium represented the point at which agents had all Funding
formed groups or group formation was impossible. The author(s) disclosed receipt of the following financial support for
the research, authorship, and/or publication of this article: Kurt Gray
Step 7: What do you want to learn from your world? At the end of received funding from the Templeton Funded Imagination Institute
the day, ABM is a theory testing and development paradigm (RFP-15-11), the National Science Foundation (SES-1534083), the
(Smith & Conrey, 2007) with independent and dependent vari- Russell Sage Foundation (93-16-08), and the Charles Koch
ables. In the case of ABMs, independent variables (or Foundation.
“parameters”) are customized by the experimenter, while
dependent variables are measured throughout the model or at Note
the model’s conclusion. If experimental hypotheses are con- 1. While we use the term “agent-based modeling” in this article, the
firmed, researchers should consider adding other independent terminology around agent-based model is diverse and potentially
variables into the model as moderators. Using new variables confusing. Alternative terms include “multiagent systems,”
or situations to test the generalizability of a phenomenon is “agent-based simulation,” “agent-based computing,” and
often called a “robustness analysis,” and it can reveal surprising “individual-based modeling.”
new effects or nonlinearities.
Application of Step 7. Schelling’s central parameter was References
agents’ desired similarity, while his dependent measures were Abbott, A. (1988). Transcending general linear reality. Sociological
agents’ positions at the conclusion of the simulation. His find- Theory, 6, 169–186.
ing was that a relatively low rate of similarity seeking (*30%) Asch, S. E. (1956). Studies of independence and conformity: I. A
could produce relatively homophilous agent distributions at the minority of one against a unanimous majority. Psychological
conclusion of the simulation. Monographs: General and Applied, 70, 1–70.
In Gray and colleagues’ model of “us and them,” the central Axelrod, R. (1997). The complexity of cooperation: Agent-based
parameters were agents’ tendency to show reciprocity and tran- models of competition and collaboration. Princeton, NJ: Princeton
sitivity, and the central-dependent variable was group cluster- University Press.
ing. Varying parameters and measuring clustering revealed Axelrod, R. (1980). More effective choice in the prisoner’s dilemma.
how reciprocity and transitivity could produce stable grouping. Journal of Conflict Resolution, 24, 379–403.
Gray and colleagues also examined a moderating role for Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation.
“trust”—the baseline tendency for cooperation or defection. Science, 211, 1390–1396.
Bassett, D. S., & Gazzaniga, M. S. (2011). Understanding complexity
in the human brain. Trends in Cognitive Sciences, 15, 200–209.
Conclusion Bear, A., & Rand, D. G. (2016). Intuition, deliberation, and the evolu-
ABM is not a new technique, but its promise and power are tion of cooperation. Proceedings of the National Academy of
often overlooked by social psychologists. We believe that Sciences, 113, 936–941.
there are two assumptions that have hindered their increased Beer, R. D. (1995). A dynamical systems perspective on agent-
use. The first is that ABMs are difficult to learn or understand. environment interaction. Artificial Intelligence, 72, 173–215.
However, good ABMs should be easy to conceptually under- Borshchev, A., & Filippov, A. (2004, July). From system dynamics
stand, and the resources discussed above should make their and discrete event to practical agent based modeling: Reasons,
implementation easier. The second assumption is that ABMs techniques, tools. Paper presented at the 22nd International
394 Social Psychological and Personality Science 8(4)

Conference of the System Dynamics Society, Oxford, UK. doi:10. communities. Proceedings of the Web Intelligence (WI) and Intel-
1.1.511.9644 ligent Agent Technologies (IAT) 2014 IEEE/WIC/ACM Interna-
Boyd, R., & Richerson, P. J. (1985). Culture and the evolutionary tional Joint Conferences, 3, 64–71. doi:10.1109/WI-IAT.2014.151
process. Chicago, IL: University Chicago Press. Jarman, M., Nowak, A., Borkowski, W., Serfass, D., Wong, A., &
Bruch, E., & Atwell, J. (2015). Agent-based models in empirical Vallacher, R. (2015). The critical few: Anticonformists at the
social research. Sociological Methods & Research, 44, crossroads of minority opinion survival and collapse. Journal of
186–221. Artificial Societies and Social Simulation, 18, 6. doi:10.18564/
Carley, K. M. (2002). Smart agents and organizations of the future. jasss.2663
The Handbook of New Media, 12, 206–220. Johnson, P. E. (1999). Simulation modeling in political science.
Cederman, L. E. (2005). Computational models of social forms: American Behavioral Scientist, 42, 1509–1530.
Advancing generative process theory. American Journal of Sociol- Kalick, S. M., & Hamilton, T. E. (1986). The matching hypothesis
ogy, 110, 864–893. reexamined. Journal of Personality and Social Psychology, 51,
Critelli, J. W., & Waid, L. R. (1980). Physical attractiveness, romantic 673–682.
love, and equity restoration in dating relationships. Journal of Kashima, Y. (2000). Maintaining cultural stereotypes in the serial
Personality Assessment, 44, 624–629. reproduction of narratives. Personality and Social Psychology
Efferson, C., Lalive, R., Richerson, P. J., McElreath, R., & Lubell, M. Bulletin, 26, 594–604.
(2008). Conformists and mavericks: The empirics of frequency- Levine, D. K. (1998). Modeling altruism and spitefulness in experi-
dependent cultural transmission. Evolution and Human Behavior, ments. Review of Economic Dynamics, 1, 593–622.
29, 56–64. Lewis, K. (2015). Three fallacies of digital footprints. Big Data &
Epstein, J. M. (2008). Why model? Journal of Artificial Societies and Society, 2, 1–4.
Social Simulation, 11, 12. Luhmann, C. C., & Rajaram, S. (2015). Memory transmission in small
Forgas, J. P. (1995). Mood and judgment: The affect infusion model groups and large networks an agent-based model. Psychological
(AIM). Psychological Bulletin, 117, 39. Science, 26, 1909–1917.
Gasser, L., Braganza, C., & Herman, N. (1987). MACE: A flexible Macy, M., & Flache, A. (2009). Social dynamics from the bottom up:
testbed for distributed AI research. In M. N. Huhn (Ed.), Distribu- Agent-based models of social interaction. In P. Hedström & P.
ted AI (Vol. 1, pp. 119–152). London, England: Pitman. Bearman (Eds.), The Oxford handbook of analytical sociology
Gilbert, N., & Troitzsch, K. (2005). Simulation for the social scientist. (pp. 245–268). New York, NY: Oxford University Press.
New York, NY: McGraw-Hill Education (UK). Macy, M. W., & Willer, R. (2002). From factors to actors: Computa-
Goldstone, R. L., & Janssen, M. A. (2005). Computational models of tional sociology and agent-based modeling. Annual Review of
collective behavior. Trends in Cognitive Sciences, 9, 424–430. Sociology, 28, 143–166.
Gray, K., Rand, D. G., Ert, E., Lewis, K., Hershman, S., & Norton, M. Minar, N., Burkhart, R., Langton, C., & Askenazi, M. (1996). The
I. (2014). The Emergence of “us and them” in 80 lines of code swarm simulation system: A toolkit for building multi-agent
modeling group genesis in homogeneous populations. Psychologi- simulations. Retrieved from http://www.santafe.edu/media/work
cal Science, 25, 982–990. ingpapers/96-06-042.pdf
Halberstadt, J., Jackson, J. C., Bilkey, D., Jong, J., Whitehouse, H., Nowak, A. (2004). Dynamical minimalism: Why less is more in psy-
McNaughton, C., & Zollmann, S. (2016). Incipient social groups: chology. Personality and Social Psychology Review, 8, 183–192.
An analysis via in-vivo behavioral tracking. PLoS One, 11, Nowak, A., Szamrej, J., & Latané, B. (1990). From private attitude to
e0149880. public opinion: A dynamic theory of social impact. Psychological
Henrich, J., & McElreath, R. (2003). The evolution of cultural evolu- Review, 97, 362–376.
tion. Evolutionary Anthropology: Issues, News, and Reviews, 12, Perc, M., & Szolnoki, A. (2010). Coevolutionary games—A mini
123–135. review. BioSystems, 99, 109–125.
Hong, Z., Bednarek, S. Y., Blumwald, E., Hwang, I., Jurgens, G., Reynolds, C. W. (1987). Flocks, herds and schools: A distributed
Menzel, D., . . . Verma, D. P. S. (2003). A unified nomenclature behavioral model. ACM SIGGRAPH Computer Graphics, 21,
for Arabidopsis dynamin-related large GTPases based on homol- 25–34.
ogy and possible functions. Plant Molecular Biology, 53, Schelling, T. C. (1971). Dynamic models of segregation. Journal of
261–265. Mathematical Sociology, 1, 143–186.
Holland, P. W., & Leinhardt, S. (1971). Transitivity in structural mod- Shariff, A. F., Willard, A. K., Andersen, T., & Norenzayan, A. (2016).
els of small groups. Comparative Group Studies, 2, 107–124. Religious priming a meta-analysis with a focus on prosociality.
Huston, T. L. (1973). Ambiguity of acceptance, social desirability, and Personality and Social Psychology Review, 20, 27–48.
dating choice. Journal of Experimental Social Psychology, 9, Smaldino, P. E. (2014). Group-level traits emerge. Behavioral and
32–42. Brain Sciences, 37, 281–295.
Jackson, J., Halberstadt, J., Jong, J., & Felman, H. (2015). Perceived Smith, E. R., & Conrey, F. R. (2007). Agent-based modeling: A new
openness to experience accounts for religious homogamy. Social approach for theory building in social psychology. Personality and
Psychological and Personality Science, 6, 630–638. Social Psychology Review, 11, 87–104.
Jahanbazi, M., Frantz, C., Purvis, M., Purvis, M., & Nowostawski, M. Tajfel, H. (1982). Social psychology of intergroup relations. Annual
(2014, August). Agent-based modelling of primitive human Review of Psychology, 33, 1–39.
Jackson et al. 395

Tesfatsion, L., & Judd, K. L. (Eds.). (2006). Handbook of compu- Author Biographies
tational economics: Agent-based computational economics
Joshua Conrad Jackson is a PhD student at University of North Car-
(Vol. 2). Amsterdam, the Netherlands: Elsevier.
olina. He studies culture, religion, and morality, among other topics.
Tisue, S., & Wilensky, U. (2004, May). Netlogo: A simple environ-
ment for modeling complexity. Paper presented at the Fifth David Rand is an associate professor of psychology, economics, and
Proceedings of the International Conference on Complex management at Yale University. His research combines mathematical
Systems, Boston, MA. and computer modeling with behavioral experiments to understand
Tononi, G., Sporns, O., & Edelman, G. M. (1994). A measure for brain human cooperation.
complexity: Relating functional segregation and integration in the
nervous system. Proceedings of the National Academy of Sciences, Kevin Lewis is an assistant professor at the University of California,
91, 5033–5037. San Diego. He studies the formation and evolution of social networks.
Vallacher, R. R., & Nowak, A. (1999). The dynamics of self- Michael I. Norton is the Harold M. Brierley Professor of Business
regulation. In R. S. Wyer Jr. (Ed.), Perspectives on behavioral Administration at the Harvard Business School, and a member of Har-
self-regulation (pp. 241–259). Mahwah, NJ: Lawrence vard’s Behavioral Insights Group. He is the co-author—with Elizabeth
Erlbaum. Dunn—of the book, Happy money: The science of happier spending.
White, G. L. (1980). Physical attractiveness and courtship
progress. Journal of Personality and Social Psychology, 39, Kurt Gray is an assistant professor of psychology at University of
660–668. North Carolina, Chapel Hill, who studies mind perception and morality.
Wooldridge, J. M. (2003). Cluster-sample methods in applied econo-
Handling Editor: Simine Vazire
metrics. The American Economic Review, 93, 133–138.