Andreas Flache
I use both computational modelling and empirical research to study cooperation and social integration in human societies, and how this relates to social networks. See http://www.gmw.rug.nl/~flache.
Location I am professor of Sociology at the University of Groningen, The Netherlands.
Activity
-
I guess it is fair to say that this is a littlebit more than just an assumption. There are literally hundreds of scientific studies on the effects of intergroup contact on prejudice. The upshot is that by and large it reduces prejudice, but we have a also learned a lot about the conditions under which it does or does not work. Some referenences: The classic is...
-
Indeed, as one good model is. Once we have a good simple model that makes us understand something, we have a useful basis to move on and make the model more complex in meaningful ways. An important question will then always be whether and under what condition a result of interest (e.g. opinion clustering) will hold up. Good suggestions have been made here in...
-
I can only agree with the comments above, the outcome of segregation in this example is not necessarily what people have been choosing for. They may prefer other outcomes , but the "institution" of free and uncoordinated individual decision making entails here segregation as unintended and - at least in this example - by most people undesired result. Thinking...
-
Your last question is very interesting to think about. When seeing an ABM people often say "I could have known this without the model". But such a model is just the start. Once we got it to do "the obvious" we have a tool to ask in an intellectually disciplined way what the consequences would be of different conditions and complications. For example, based on...
-
Quite right that the spatial structure of CAs is rather limited when compared to real social networks. CAs help to understand the basic properties of many dynamics and then a next step is to explore what happens if we assume more complex network structures. Which is what actually quite a few researchers do, for example Centola & Macy in their research on the...
-
Vanessa and Paula, I couldn't agree more. Self-organization may be part of the explanation of clustering but there are of course other possible exogenous factors like those you mention. In the end we also need empirical research to assess how much each possible factor contributes to explaining actual clustering patterns, and people conduct such studies for...
-
As hopefully will become clearer in what follows, CA models can be seen as a subclass of ABMs. CA models impose a particular spatial interaction structure and focus on influences from the local neighborhood on agents' behavior, which in turn is expressed in terms of changes of the state of cells or the spatial location of agents (moving). If you are...
-
Reference for research on intergroup contact: The classic is Allport (Allport, Gordon W. 1954. The Nature of Prejudice. Cambridge, UK: Addison-Wesley.), but there is very much follow-up research, see e.g. this overview paper here: Pettigrew & Tropp (2006), "A Meta-Analytic Test of Intergroup Contact Theory", Journal of Personality and Social Psychology, 2006,...
-
To be sure, no one here is saying that this is the "natural state of affairs". As you will have seen the article ends with "in the next article, we will look a bit more into what the model tells us about the conditions when segregation would arise. And we will relate Schelling’s model to real data about ethnic preferences". Have a look into Clark & Fossets...
-
For this model it seems to not matter whether you use NetLogo 4 or 5. You can run it with NetLogo 5.1.0. Just enter "Continue" when the warning message appears that this model was created under NetLogo 4. After that the models runs without problems as far as I could see. The author of the model is Ian Weaver, a contact address can be found on the site to which...
-
Good point! While indeed much interaction is not bounded locally in geographical space, important structural characteristics of 'real' social networks are similar to the structures we have in a CA. See also my reply to Diloram's comment below for more details and a reference to CA-based work exploring the effects of network structures on social influence...
-
Good point! There are some agent-based models that take into account that some features are more important than others and thus can shape the further process of opinion formation. For example, Baldassarri and Bearman proposed a model of polarization in the American Society in which they elaborated the idea that discussions in interaction focus on those issues...
-
Thanks for checking this! The link embedded in the text brings you to the same page "If you want to try it out yourself, you can use a NetLogo implementation of the >>Axelrod<< model." (http://ccl.northwestern.edu/netlogo/models/community/Dissemination%20of%20Culture)
-
Good point, this particular model always moves into a stable equilibrium. But to be sure, this is not inherent to the CA framework as such. Once you add for example some random noise (agents changing their opinions spontaneously with a small probability) or random movement (agents relocating with a small probability) or random interaction (agents sometimes...
-
Good point, Farid! I do of course agree that before we can use ABM for serious policy advice, these models need to be much more sophisticated, based on empirical input data and tested against empirical output data. In some areas, like traffic modelling or models of disease spreading, ABM modellers are quite successful in this. In other areas, like opinion...
-
Good points! Further on in this module we discuss Axelrod's model of "cultural dissemination" in which individuals have many different features. This model shows that social influence in combination with homophily (birds of a feather flock together) can produce clustering of like-minded people and alignment of their opinions across many dimensions (e.g....
-
Very good points. Further on in this module we discuss Axelrod's model of social influence. This models adds the assumption that you are more likely to interact with others who are more similar to yourself. This further promotes cluster formation and - indeed - the emergence of a small world in which people mostly interact with those who are similar to them...
-
Yes, there are CA models of what has been called "fads" or "cultural drift", based on Axelrod's model of opinion dynamics (see later in this module). For example, modelling work by physicists (Klemm, San Miguel) showed that if you add to Axelrod's original model a small chance of "random perturbation" of opinions, clusters can break down but also new "fads"...
-
Good point! CA can help to understand well what happens in a clustered world with local interaction. And while it is true that much interaction nowadays is not bounded by geographical space, we also know that the networks within which people interact are often strongly clustered. That is: typically people interact with others who are also connected to each...
-
Indeed, social psychologists point to this mechanisms, sometimes called "striving for uniqueness". Agent-based modellers have explored the consequences this may have for social influence dynamics. Broadly, if there is sufficient anti-conformism in combination with social influence, this can generate permanent diversity and clustering in society. See for...
-
This is obviously a deviation from reality. But not every deviation from reality is a problem in modelling and some are actually quite useful. In this case, we would first have to answer the question whether assuming a torus makes a big difference. Well, for many results of CA models it does not matter (you can try this out with many of the models in the...
-
Good to see you are already discussing the fundamental assumptions! A general definition of interaction would be any form of encounter in which an 'agent' (or, actor) responds to perceived actions of other agents in his local environment such that these other agents - in turn - respond to the actions of that focal agent. In that sense actions like moving into...
-
Andreas Flache replied to Dirk Koenig
Axelrod's model has as a possible outcome that distinct and stable cultural groups arise that are so different from each other that no further influence between the groups will take place. This is very similar to the situation that you describe. But the reason that these groups emerge in this model is social influence. Social influence has in such a case led...
-
I fully agree that social-self organization does not imply that policy makers should not try to influence social dynamics in directions that can be seen as socially desirable (where it is of course a matter of democratic debate what that exactly means). But social self-organization also teaches us that social behavior is not easy to influence, can develop in...
-
Very interesting comment: "then as people travel more, interact more then gradually their views will coincide and everyone will co-exist peacefully (not much sign at the moment)! ". This is indeed an implication of Axelrod's model, but as so often it depends a lot on the assumptions you make. Elsewhere I have argued that actually the opposite is also possible,...
-
About the statement "none of the agents is dissatisfied in the final outcome": in modelling, it is very important to distinguish between purely analytical statements describing the behavior of a theoretical model, and statements about the real world. The statement above has a clear meaning within the context of the model. We have a definition of what an agent...
-
I know of at least one model by my my colleague Michael Maes and his co-author Lukas Bischofberger. See here: http://www.maes-sociology.eu/news/bestpaperaward
One conclusion is that when people tend to link mainly to those they agree with, this may lead to opinion polarization. -
Interesting idea to analyze the follow network from FutureLearn to see whether there is clustering! There are studies of the links between political blogs in the U.S., or of twitter networks showing clear evidence of clustering along partisan lines. See for example here for some instructive visualizations and background:...
-
I definitely agree that housing prices are also an important factor. But as it seems, so are ethnic preferences and various other factors (e.g perceptions of crime rates). The challenge for modellers is trying to understand how all these factors operate together. For example, whether differences in housing prices and income may amplify segregation dynamics, or...
-
There are models that include differences in wealth and housing prices. Some of this work can be found in the paper by Clark and Fosset to which I also refer in one of the articles about the Schelling model. Fosset also has a simulation model (simseg) that allows to play around with variation in things like income levels, housing prices, social status of...
-
If it is seriously flawed it should produce some central outcomes that can be empirically refuted. ABM is a tool that can help us to find such refutable predictions from a model.
-
I agree that for many social phenomena it is not so difficult to invent some model that can produce the phenomenon. That is why the assumptions we make in these models should be rooted in theoretical and empirical knowledge about human behavior. The assumptions of social influence and homophily do not just come from somewhere but have been tested in much...
-
Very interesting that you raise the point about Axelrod assuming a nominal opinion space. I can only agree that this is often an unrealistic assumption. Actually in some other paper with Michael W. Macy we showed how diversity can become much less likely once you assume continuous features on which agents can move towards each others' opinion in gradual steps ...
-
Andreas Flache replied to Dirk Koenig
It certainly makes sense to assume that some features are less easily influenced than others. This could surely be built into this kind of model, but it is not yet in the program in NetLogo. An interesting implication may be that if there is consensus on such a hard-to-change feature, this may prevent clustering because agents always have something in common.
-
Andreas Flache replied to John Hendry
This is possible in the model. You may first pick A as focal agent who then interacts with B and adopts a trait of B on some feature. Then at some time later you may pick B and B interacts with A, then adopting a trait from A on some other feature. They more similar they become (as consequence of such interactions), the more likely they are to interact again.
-
Clearly in online networks people are not restricted to interacting with others within a close "distance" from them. But interestingly "clustering" can (and does) also occur quite a bit. In online networks you can choose people (or, say, blogs) based on their similarity to your opinions. This is also called homophily. If everyone does that, people end up in...
-
That could be a good justification for this assumption. It actually draws on an early version of Schelling's model, where the idea is that people have a perception about how a neighborhood (in the sense of a couple of hundred or maybe even thousands of households) is composed ethnically and then make their decision whether to stay or move to another...
-
I (obviously) agree with the argument in our paper that Axelrod's model produces (some) unrealistic results. But I think one should not conclude that therefore all models based on this approach must be unrealistic. To take up the example quoted above that Axelrod's model predicts diversity only for small populations: we show in our paper that a relatively...
-
Your understanding of the model is exactly right. I am also not sure how realistic it is... But then there are many examples of models that aim to introduce more realism in this type of modelling. One example is our more recent paper referred to in the text (Flache & Macy, 2011), but there are many others (you can e.g. find references in our paper to models...
-
Very important discussion. To be sure, models like this one do in no way imply a value statement about whether diversity is good or bad. But they may help to understand the conditions under which diversity emerges or is stable. Much research (e.g. in organizations, school classes, neighborhoods) shows that homophily is a fact of life, a very robust social...
-
Very interesting point. There are actually models of self-organization of hierarchies in animals that take into account differences in strength and 'personality', but also show how hierarchies can become self-reinforcing at some point. Interesting work in this area is done for example by Charlotte Hemelrijk and her group...
-
Andreas Flache replied to Megan S
I can only agree that things are different when agents have a wider range of interaction than just their small local neighborhood. Later in this week you will also encounter Axelrod's model of opinion clustering. Based on this model it has indeed been argued that 'globalization' (in this sense) may foster consensus or monoculture. But this in turn depends on...
-
I agree this is a very valid point. Actually, there are models that add to the basic CA framework the assumption that with a small probability interactions can also occur across large spatial distances. This can profoundly change model behavior, for example accelerate the diffusion of information or opinions. See for example work by Damon Centola and Michael...
-
Andreas Flache replied to [Learner left FutureLearn]
Admittedly, the rules here are way simpler than real life. But that has the advantage that we can better understand what goes on. Agent based modellers are also working on building models with more realistic assumptions and more realistic behavior. For example, in work with Michael Mäs and Dirk Helbing I proposed a model that can generate "dynamic clusters"...
-
Good point. There are also models that take into account that people are different in the extent to which they can influence others or are open to being influenced. In such models, a minority can also be very influential and eventually change the majority's opinions. See e.g. work by Deffuant and others about the emergence of extremism. But that goes beyond...
-
Interesting idea. Actually, one example I can think of is Fosset's model of ethnic residential segregation, based on Schelling's model (see later this week). Fosset assumes that people respond to the ethnic composition of their neighborhood without necessarily having direct contact with everyone in that neighborhood.
-
The most important difference is the specific spatial structure that you impose with a CA model. That agents change their states in response to input from the environment indeed is a feature of all ABM.
-
You point to a very important aspect of social self-organization: it always happens within a given social context and under given constraints. The context and the constraints often can not fully explain or determine the outcome, but they surely have an influence in addition to the processes of self-organization. For example, while low-price neighborhoods...
-
I surely agree that humans can evaluate their behavior much more in moral terms than animals do. But very often it is not clear to us what the consequences of our actions are, so in order to morally evaluate our actions, we need a good theoretical understanding of how the interplay of the actions of many people can produce results that no one wanted nor...
-
Dear Jennifer,
I totally agree. In our paper on modelling school segregation we went for a somewhat simpler model through a similar exercise to identify conditions under which an integrated distribution can be stable. See here: http://jasss.soc.surrey.ac.uk/17/1/5.html, paragraphs 3.10 to 3.14. -
To Stephen's remark: very important point. There are a number of extensions of these models that take into account factors like wealth and housing prizes. With these models people study how income and ethnicity can come to be correlated locally by segregation dynamics. See some of the references I gave in other comments: e.g. work by Zhang or further work by...
-
Jennifer and Michael,
although I pointed to possible complications above, I think the basic story is simpler. If the data represent real preferences, full integration is not sustainable. No matter how you describe the data. Suppose we would randomly mix these four groups based on the actual composition in LA (see e.g. data of US Census Bureau:... -
Dear Jennifer,
I think this is an excellent idea! I agree someone else might have done this before, but I wouldn't know of an example right away. Perhaps some work by Van de Rijt and Macy. I also agree that even if we model the preferences closer to the empirical pattern they find in their study, the outcome of the model may still be segregation. Worth to... -
Andreas Flache replied to Marc Bond
Tom,
quite right, but how do we know how much each of these things matters for the outcomes of opinion formation dynamics? One way to get more insight in this is ... further modelling. People do this on basis of models like the one we discuss here, add in the role of media influence, of opinion leaders, of geography, and so on. And then they use their models... -
Dear Tom,
you are absolutely right that besides ethnic composition also income composition (and crime rates, housing prices ...) play a role in the residential decisions people make. I think also your intuition is right that this may lead to segregation not only by ethnicity but also by income and that sometimes the latter may be stronger than the first. In... -
Dear Rashid,
I couldn't agree more with you! Scrutinizing and pulling the models apart is very important. I hope in this course we convey that these models most of all should stimulate critical thinking about social processes before we rush to conclusions. In the end, hopefully, the models, the critical thinking and research building on these models will... -
Dear Rashid,
It seems what you describe fits with James Coleman's insight "if a social policy does not actively employ the interests of those on whom it has an impact, it will find those interests employed in directions that defeat its goals". According to Schelling's model, segregation is an outcome that can occur even if it is undesirable for the... -
Tom, there is research analysing on basis of census data why people move in and out of neighborhoods. A number of papers are by Robert Mare and co-authors. Here is one quote from one of their papers analysing about 1700 moving decisions in Los Angeles county: "The probabilities that either blacks or Hispanics remain in their neighborhoods vary inversely with...
-
Dear Tom,
yes, many elements of real life are not taken into account in these models. An important question to ask is then: would anything change fundamentally in the conclusions we draw from such a model if we take more things into account? And what would change and why? I believe that ABM is a very powerful tool to search answers to these questions.... -
Dear Tom,
I think that Vincent (below) makes an important point: in order to explain segregation we may not always need to understand the underlying causes of ethnic preferences, as long as people act on them. I also agree with you that we can learn more if we dig deeper. For example, for school segregation ethnic preference dynamics like Schelling's are also... -
Dear André,
I totally agree that assuming the possibility of changing population size is much more realistic. My answer would be that the basic type of tool we have here can relatively easily be extended to explore theoretically how population growth could affect segregation. Arguably the basic result of emergent segregation may not change, as people moving... -
Dear Jacky and Clovis,
good point. The models discussed here indeed strongly draw upon 'backward-looking' or heuristic decision making. But especially for the Schelling model there are also versions that model boundedly rational decision making. For example, Zhang - building on work by Peyton Young - developed versions of the Schelling model based on random... -
Dear Bram,
we agree on that we need a combination of theoretical and empirical research. I don't quite agree that this type of modelling is just a source of inspiration like music, or that it is in contrast with 'cool scientific thinking'. What else is scientific thinking than fully explicating the assumptions that you make (an ABM forces you to do so), then... -
Dear Carla,
I could not agree more with your statement that a simple model can help us understand which variables may play a role. As someone else also wrote: once we have understood this we can move on and add more complicated assumptions (but I wouldn't do it any earlier).
About the internet example: an interesting thing is that although we now can easily... -
Hi Bram,
I agree that models are no substitute for empirical research. But I would add that data are no substitute for theoretical understanding either. We need both and we need to integrate models and empirical research. Models building on Schelling can help us to formulate testable predictions about the conditions under which unintended segregation can... -
Dear Darren and Peter,
these are important questions from a social-philosophical point of view. My answer would be that models like this one can help us to understand how we could prevent outcomes that are undesirable from the perspective of the individuals involved. These models show how segregation can be a likely outcome although people could be equally... -
Hi Bram,
housing prices surely play a role when you think of residential clustering (or, segregation). But so do, for example, preferences for certain types of neighbors. See the research discussed later in this model about ethnic preferences in residential segregation. Another point is that religion, musical taste etc. do play a much more important role when... -
Andreas Flache replied to Marc Bond
Dear Mark and Peter,
very important point! Actually, physicists have taken this model and added to it some noise to model 'cultural perturbation', i..e changes of opinion that occur spontaneously. They found that clustering breaks down for very small amounts of noise and remains only in the form of dynamic, ever changing clusters. But then, Axelrod's... -
Why modelling if there is so much data out there? Some people even claim that 'big data' bring the 'end of theory'. But in my view the big difference is that with modelling we can understand the processes that bring certain patterns about and can find out why and under what conditions we would get the patterns given the model. And we can make predictions for...
-
Andreas Flache replied to Ken Carew
Dear Ken,
that is a very good question to ask about this type of CA modelling in general. I haven't tried it for this exact model, but for a number of other models I could show in earlier work with Rainer Hegselmann that results are rather robust against changing the geometry of the cellular space. Here is the paper: http://jasss.soc.surrey.ac.uk/4/4/6.html.... -
Dear Margarita,
interesting idea! Suppose under the rules we have now the system would move into a stable state (whether it does will be told in the next video). Then an interesting question could be: how many 'deviants' that follow other rules (e.g. random switching, or stubbornly sticking to your opinion) do we need to make that state unstable? Something... -
Dear Tom,
certainly influence by media channels etc. also plays a role. From an CA perspective an important question is how local interaction operates together with influence from central channels. Once you have a good CA model of local interaction you can add to it effects of e.g. 'media channels' that influence all agents at the same time. In fact,... -
Hi,
Great points, see also the comment I just posted in this discussion: it is a very important point that with modern communication technologies much interaction is not local anymore. But at the same time, much interaction also still is local. CA can help to understand what the consequences of that assumption are. They also can help to understand what... -
Andreas Flache made a comment
Hi everyone,
it is a very important point that with modern communication technologies much interaction is not local anymore. But at the same time, much interaction also still is local. CA can help to understand what the consequences of that assumption are. They also can help to understand what happens if you relax it, for example by increasing the range... -
Andreas Flache made a comment
Hi everyone,
the comments made so far here raise very important points. Some reactions: 1) Is there an essential difference between ABM and CA? No, arguably CA are just a subclass of ABM . Typically assumptions about individual actors are more simply in CA, and so are assumptions about social structure (usually regular patterns etc.). But you can make CA...
