FIGURE 1Fire Model Running Using NetLogo

524 DecemberAcademy of Management Learning & Education

Agent-Based Modeling Toolkits NetLogo,RePast, and Swarm

Reviewed by Duncan A. Robertson, University of War-wick, UK.

Management educators are always seeking newand exciting ways of teaching key concepts to theirstudents. Agent-based models are one way of com-municating management ideas using a methodthat is inherently visual—one that can capture theattention of students and executives alike.

Agent-based models are gradually making theirway into management research. Although this re-view does not consider the use of agent-basedmodels in management research and does not givean in-depth insight into agent-based modelingconcepts (readers may wish to consult Robertson,forthcoming, for these purposes), what is intendedis to expose readers to a small selection of agent-based modeling packages to see whether and howthey can incorporate them into their teaching.

“Agents” can include firms, people within orga-nizations, or entire industries: Agent-based model-ing allows the user to define the interactions be-

tween these agents and then use these to generatemodels. Agent-based models define the interac-tions between these agents, whether they be forexample interactions between firms or the interac-tions between people within a firm. What makesthis technique exciting for a taught audience isthat the models are inherently visual, as can beseen from the screenshots (Figs. 1, 2). Students cansee the dynamics of the interaction betweenagents and the subsequent effect on other agents.This generally sparks a discussion between stu-dents, and can therefore, be used as a discussiontool in order to initiate debate, debate that maywell be augmented with the use of other teachingmethods such as case studies. While the use ofagent-based models alone is often not sufficient toexplain all the intricacies of a chosen manage-ment concept, they can supplement traditionaltechniques. Agent-based models are particularlysuitable for modeling complex systems, wheremany agents interact; there is therefore scope forusing these toolkits to illustrate some of the morerecent concepts within management, particularlythose where the micro level behavior of individu-als affects the global, macro properties of a sys-tem.

As with any new method, knowing where to startcan be a significant obstacle in introducing thetechnique into a course. I review three agent-based modeling toolkits in order to point readers to

FIGURE 2SugarScape Model Running Using RePast

2005 525Book & Resource Reviews

an appropriate place to start investigating agent-based models, to decide whether there is a placefor them in their courses: NetLogo, RePast, andSwarm.

The first is NetLogo, a system that is very easy toinstall and to operate: Users can explore modelswithout any technical knowledge; an extensive“models library” exists so that users can see thetypes of models that have been made using thetoolkit. These are generally well documented, thedocumentation being easily accessible throughthe “information” tab within the program. Param-eters of the model are easily changed using graph-ical “sliders.” For those who want to change thedetails in a model, clicking on the “procedures” tabbrings up the entire model code, which can easilybe changed in order to extend the sample models.NetLogo provides a very accessible introduction toagent-based modeling. This is a very useful firststep after which users can decide whether thetechnique is one that warrants further investiga-tion (Figure 1).

For those more adventurous readers, or thosewith some experience in programming, RePast isthe obvious next step. It provides an extremely

wide set of “libraries” for developing models. Aswith NetLogo, several sample models are avail-able that are easily downloaded and run. However,to extend these models requires a reasonable ef-fort, for example in installing a facility such asJBuilder for editing Java programs. One of the keyfeatures of RePast is that it was set up and ismaintained by a group of social scientists, whichmeans that social networks, genetic algorithms,systems dynamics, and geographical informationsystems (GIS) are all supported. However, the timeto develop a new model will be significant unlessyou have relatively good Java programming skills.But there is a good community of model builders,all keen to support new members of the RePastcommunity (Figure 2).

For completeness, I also mention Swarm, as thismay often be considered as a viable option bypeople new to agent-based modeling. Swarm isnow looking relatively geriatric at the early age ofabout 10 years. Although still very usable by theexpert, it is not designed specifically for socialscience applications, and is not well supported. Itis also extremely tricky to install or run success-fully.

TABLE 1Selected Agent-Based Modeling Packages

NetLogo RePast Swarm

Developed by Northwestern University(Center for ConnectedLearning and Computer-Based Modeling)

Repast Organization forArchitecture andDevelopment (originallyUniversity of Chicago SocialScience ResearchComputing)

Swarm Development Group(originally Santa Fe Institute)

Advantages Very easy to set up and runmodels.

Many users, good supportfrom the RePast community

The precursor to RePast and to alesser extent NetLogo

Example models Many Many Only a few are readilyavailable.

Disadvantages Very complicated models areoutside the capability ofNetLogo, mainly for abstractmodels.

Need to be able to understanda programming language(e.g., Java)

Extremely difficult to install andrun on a Windows computer.

Time required to start buildingmodels for programmingnovice

A few hours A few days A few weeks

Limitations Models created cannot beeasily extended.

Very versatile system. Modelscan be extended very easilysubject to at least someknowledge of e.g., Java.

Difficulty in installation, notwell supported, especially bythe social science community

Support from academiccommunity

High High—especially from socialsciences academiccommunity

Relatively low

Active maintenance ofsoftware

High High Low

Additional features 3D visualization of models.Models can be embedded inweb pages.

GIS implementation Complicated to install

526 DecemberAcademy of Management Learning & Education

So which toolkit to use? In summary, NetLogo (forbeginners wanting to make models work), or Re-Past (for people wanting to extend models or forthose with good programming skills). Swarm is thegrandfather of them all and is not recommended,at least initially for the nonexpert.

All the toolkits ultimately provide a similar endresult: the ability to model, to manipulate, andinspect agents. NetLogo has a simple interface,models can be set up and run with only the push ofone or two buttons; to produce the model such asthe one illustrated takes just 50 lines of simplecode. Compare this to the simulation in Figure 2,using RePast, which requires 950 lines of code,much of it quite complicated. However, this reflectsthe potential complexity of each of these toolkits:NetLogo is good for setting up simple simulationsvery quickly, while RePast allows simulations, per-haps trialed using NetLogo, to be expanded andmade more sophisticated (Table 1).

As more educators develop their own models,

management-specific models will become avail-able to our community. For the moment, we have tostart with models from other disciplines (such associal science or natural science) that can beadapted to shed light on management problems.This is a call to arms for educators to get theirhands dirty—to download NetLogo to see whatmodels exist, to use tools such as NetLogo to de-velop their own models, then (if still enthusiastic)to develop their own models using RePast. Andfinally, to report their findings back to AMLE sothat we may share in their rich experiences.

REFERENCES

NetLogo, http://ccl.northwestern.edu/netlogo/

Repast, http://repast.sourceforge.net/

Robertson, D. A. In press. Agent-based models in managementresearch. Organizational Research Methods.

Swarm, http://www.swarm.org/

2005 527Book & Resource Reviews