Skip to main content
Search form
  • 00:00


  • 00:12

    WILLIAM RAND: Hi, my name is Bill Rand.I'm an Associate Professor of Marketingat North Carolina State University in Raleigh, NorthCarolina.So today I'm going to talk a little bitabout agent-based modeling.Agent-based modeling is a method by whichwe place a computational representationof the agents of interest directly into a computer model.So we're trying to simulate every little detail of what'sgoing on in that model, of what's going on in that systemthat we're looking at.

  • 00:39

    WILLIAM RAND [continued]: Agent-based modeling is about takingrules of agents, rules for the environment those agentsoperate in, and then using that to explaina phenomenon we see around us.A lot of time, we're talking about interacting agents.So we're talking about consumers talking to each other.We're talking about people making purchase decisionsin a store environment.

  • 00:60

    WILLIAM RAND [continued]: We're talking about how the effects of those agentsbubble up to make emergent patterns of phenomenonthat we see around us.So for instance, we go into a store,we make a decision about whether to buy a NintendoSwitch or a PlayStation 4.As many and many of us make that decision,overall, a dominant market share comes about.

  • 01:20

    WILLIAM RAND [continued]: There is a pattern of how many of each of thoseproducts is being bought.And that pattern, that affects future decisions.So the next person walks into the storesays, well, there's more Switches thathave been bought recently.I know a lot of friends who have Switches.So I'm going to purchase a Switch instead of purchasingthe PlayStation 4.And that's really what agent-based modeling is.It's a way of understanding the emergent behavior of consumersand how we see those patterns around us every day.

  • 01:52

    WILLIAM RAND [continued]: So my interest in agent-based modelinggoes all the way back to my undergraduate days.I was very interested in artificial intelligencewhen I was in undergraduate doing some research.And I became interested in how wecould use computers to understandhuman and social systems.And agent-based modeling is a powerful tool for that.So one of the first models I builtwas trying to understand how residents lookingfor a house in the suburbs made their decision about whathouse to purchase.

  • 02:17

    WILLIAM RAND [continued]: And we wound up building a huge modelof suburban sprawl in southeastern Michiganas part of that project.And it was very interesting to meto understand how we could simulate those processesand then see how that played out over time.As I got more and more into that,we decided to spend some time as a postdoc workingat Northwestern University, furtherstudying how agent-based modeling could reallybe used to understand more and different systems.

  • 02:43

    WILLIAM RAND [continued]: And one particular system I became interested inwas diffusion of information.How do people find things out?How do they process that information once they have it?And how do they pass it on to others?Well, it turns out that that actuallyis directly related to marketing in many ways,as you might expect.So I made a transition from a computer science PhD,which is where my background was in, to a Marketing Professorrole at the University of Maryland, at that time.

  • 03:07

    WILLIAM RAND [continued]: And since then, I've really enjoyed it.I feel like agent-based modeling is a great tool for marketing.Marketing is one of the few fields wherewe have theories of how individuals make decisions, howthey decide what to do, how they decide what kinds of actionsto take and how to interact with the people around them--a lot of times, through consumer behavior research.But then we also, at the other end,have quantitative knowledge about the patterns of behaviorthat we're seeing from those decisions.

  • 03:32

    WILLIAM RAND [continued]: And agent-based modeling can kind ofbe a glue between those two things.It can tie the theories of human behaviorto the large-scale patterns of human behaviorthat we see at the other end.One area of research that we've spent a lot of time inis word-of-mouth marketing.

  • 03:53

    WILLIAM RAND [continued]: So obviously, agent-based modeling,one of the aspects it has that's very powerfulis the fact that we can model how individualsare talking to each other.How they're interacting with each other on social media,in a normal circumstance where they meet each other in a storeshopping line, for instance.And we can look to see how those interactions influence purchasedecisions, essentially.

  • 04:15

    WILLIAM RAND [continued]: And then we can, once we have a model of that,once we've put together that model,we've situated the agents and their social networks,and they're interacting, and they'retalking about products and brands,or whatever they're discussing.We can then have, as a marketer, come forwardand start to think about what would be the bestway to alter that conversation or to changethat conversation in some way.

  • 04:36

    WILLIAM RAND [continued]: To bring to light the fact that there is a brand that they'renot thinking of, a product they're not considering,and see if they might be interested in purchasingthat product or purchasing that brand.So we can address questions like price and promotionand placement and all these kind of standard marketing questionsthrough their effects on word of mouth.How does that alter word of mouth?And that's kind of a classic example.

  • 04:57

    WILLIAM RAND [continued]: We've used it in a number of other situations,but that's a good example.So one of the things that agent-based modeling reallyneeds is a basic theory of how the individual behaves.We're representing consumers as individuals,not as groups of consumers.And so we need to understand how that individual behaves.

  • 05:19

    WILLIAM RAND [continued]: So one thing that's very helpful to haveis very low-level individual data.So if we know that a particular individual makesthese kind of purchases on these kind of daysand another individual makes a different set of purchases,and we understand a little bit more about their background,whether they have families, those kind of things,that can be very easy to instantiatewith an agent-based model, right?

  • 05:40

    WILLIAM RAND [continued]: Now, that data is not always available.Sometimes we have more aggregated-level data.And that's fine in many cases.When we have that aggregated-level data,as long as we have a theory about the human decisionprocess--and especially if we have two competingtheories about the human decision process,we can build agents at those competing theories.

  • 06:00

    WILLIAM RAND [continued]: We can build one that represents theory A, one that representstheory B, simulate out what happens,and see if the patterns of behavior actually match out.We actually did some work with this, on this kind of problem,with Michael Trusov and Yogesh Joshi when I wasat the University of Maryland.And there, we built a model to simulate Facebook app adoptionover time.

  • 06:21

    WILLIAM RAND [continued]: And so what we did was, we said, wedon't know what the underlying network structure of Facebookis, but we have some theories about whatthat structure might look like.Let's build an agent-based model wherewe simulate that structure.And so we try and play around with howpeople might be talking about apps, how they mightbe discussing apps on Facebook, how they might actually justbe sending those, "hey, sign up for thisbecause your friend signed up for it", links on Facebook.

  • 06:45

    WILLIAM RAND [continued]: And then we simulated how the app adoption patternswould look like under those different structures, howpeople were interacting.And we found, over time, that there was one particular setof app adoption structures that really made sense, that reallymatched the empirical data.So we looked at the empirical dataof how quickly those apps took offand we saw that this one structure really showed up.

  • 07:06

    WILLIAM RAND [continued]: And it was a particular structurethat had a kind of preferential attachment natureto it, where there were a few people whowere really influential in the networkand there were a lot of people who aren't as much influential.And that's, interestingly enough,a story that comes up again and again in marketing, that wehave these influentials who really make the decisionsabout a lot of things going on.And agent-based modeling is a powerful toolfor that, for understanding, whatif we have a small group of users who are really powerful,who can really affect an overall system quite dramatically?

  • 07:42

    WILLIAM RAND [continued]: In some recent work, we kind of went the other way.We started with a lab experiment.This is work I did with Ned Smithat Northwestern University.We had an experiment that Ned had run where he had actuallygone into a lab and he had asked people, "what wouldyou do if you were to lose your job today?How would you best decide how to find another job?"And what he found was very interesting.

  • 08:06

    WILLIAM RAND [continued]: He found that there is a sharp socioeconomic statusdifference between individuals who are more wealthyand individuals who weren't and howthey addressed this problem.Individuals who are wealthy wouldgrow their social networks.They would reach out to more people.They would try to expand who they were connected to.Individuals who are less wealthy wouldcollapse their social networks.

  • 08:28

    WILLIAM RAND [continued]: They would focus on their family and friendsand really talk to them about the job opportunitiesthat they might be facing.We were curious.And at the end of this, Ned speculatedthat maybe this particular lab experimentcould show why we see wealth divides in many industrializednations.Maybe it's because of the way people approachthe job question that the rich get richer and the poor staypoor.

  • 08:52

    WILLIAM RAND [continued]: And so, sure enough, we built an agent-based model that kindof simulated those dynamics.And what we found was that the wealth gap showed upquite easily in that system.Moreover, we found that not only did the wealth gap show upif we gave the agents the rule-- we just literally told them,hey, if you're under the median wealth,then collapse your network.If you're over the median wealth, expand your network.

  • 09:14

    WILLIAM RAND [continued]: But also, that gap showed up even if we allowed the agentsto learn that rule over time.So we started all the agents out with the same ruleand then we made another rule that said,if you keep getting rejected from the jobsyou're looking for, if you keep tryingto approach jobs that are maybe higher-status jobsand you keep not receiving those jobs,you keep being told you're not going to get that job,then maybe you'll change a rule.

  • 09:39

    WILLIAM RAND [continued]: Maybe you'll start to collapse your social network ratherthan expanding it.And what we found is that, even in thiswhat I'll call adaptive modeling sense,agents were able to reproduce the same kind of wealthinequality patterns that we see in many of the industrializedworld today.And that's one area that agent-based modelingis really good at.Agent-based modeling is excellentwhen you have rules of behavior that are actuallygoing to change over time, where the agents themselvesaren't just following the same rule all the time,but they're changing their behaviorbased upon past experience.

  • 10:10

    WILLIAM RAND [continued]: One of the few modeling methods that can reallytake into account that kind of behavioral changeis an agent-based model.And so I find it very powerful to usein these kinds of contexts.For students of marketing, probably the best placeto start is a paper I wrote with Roland Rust called "Agent-BasedModeling in Marketing--Guidelines for Rigor."And it's a great, just, how to get startedin agent-based modeling.

  • 10:38

    WILLIAM RAND [continued]: It tells you how to think about problemsthat agent-based modeling might be appropriate forand how to build those models in a way that makes surethat they're both verified and validated,and that they're rigorous enough to publishin academic journals, and that youcan use them for making actual business decisions, as well.In addition to that paper, I also have a textbook outthat I've published that really kind of digsinto that a little bit deeper.

  • 11:03

    WILLIAM RAND [continued]: So if you're a marketing student who'skind of got a little more interest in this area,or you want to look into a little bit more,that might be a good place to go.And I teach an online course that complements that, as well.[MUSIC PLAYING]

Video Info

Publisher: SAGE Publications Ltd.

Publication Year: 2020

Video Type:Tutorial

Methods: Agent-based simulation, Artificial intelligence, Quantitative data analysis, Marketing research

Keywords: agent-based models; artificial intelligence; consumer behavior; decision making; marketing research; pattern recognition; Social media; Social network analysis ... Show More

Segment Info

Segment Num.: 1

Persons Discussed:

Events Discussed:



Bill Rand, PhD, Associate Professor of Marketing at North Carolina State University, discusses his research using agent-based modeling for understanding consumer behavior.

Looks like you do not have access to this content.

An Introduction to Agent-Based Modeling for Consumer Behavior

Bill Rand, PhD, Associate Professor of Marketing at North Carolina State University, discusses his research using agent-based modeling for understanding consumer behavior.

Copy and paste the following HTML into your website