In lots of contexts, one finds humans more or less sorted into groups of similar people. Most readily noticed when the sorting occurs in terms of racial and ethnic features, there are all sorts of other ways that people also cluster in terms of similar characteristics. It is of course unlikely that there is a simple single cause for the sorting in any particular case, much less that there is the same single simple cause for all of them. At the same time, its curious that a similar phenomenon occurs in lots of different contexts, and perhaps productive to wonder whether there might be something similar that is at least contributing to all of them.
An obvious possibility is that people have a preference for associating with people like themselves. One might well suspect that this would by itself lead to clustering, but the intuition is too vague to do much more with it. In situations like this, an explicit model can be helpful, not only to verify the intuition but to provide a way of addressing some further questions raised by the intuition. Is there, for example, more clustering if peoples' preference for being with similar people is greater? What is the relation between the strength of preference and and the amount of clustering? How dependent are the answers to these questions on population density? What happens if people prefer to be around people different from instead of similar to themselves? Robert Schelling, an economist, was interested in the 1960's in patterns of racial segregation and integration in American cities, and came up with an interesting model along the lines just suggested. Schelling's model, published in 1971, led to some conclusions that surprised people at the time, and for this reason became one exemplar for of the potential usefulness of modeling as a way of exploration. At the time, Schelling had to make do with paper, pencil, and objects to work out the outcomes of his assumptions (as had to Conway when he created the Game of Life). Since then, it has become possible to work out these consequences much more rapildy, using computers. In this exhibit we provide a computer implementation of Schelling's model and some modifications of it. The model is quite simple and straightforward. Locations in a two dimensional grid can be occupied by two differing sorts of entities, one green and the other red. Each location has eight neighbors and each type of entity has a controllable preference to be surrounded by neighbors like (or different from) themselves. The preference is expressed by a wish to have some particular percentage of the eight neighbors of any entity to be like (or different) from the entity itself. At each time step (itertion) of the model entities check the relevant percentage of their eight neighbors. If the preference is greater than the preference value, the entities remain where they are. If it is less than the preference value, they move randomly to any open location. The model continues running until no one moves on successive iterations. That's all you need to know to discover for yourself some of the basic (and suprising) conclusions that Schelling, and others working with the model since, were able to reach. The basic model will let you start with either an integrated (random) or a segregated distribution of reds and greens, and control the population size (and hence density), the strength of preference, and the direction of preference (similar or different). See what you discover about the kinds of questions raised above. And if you get interested, there's a more sophisticated version of the model that will allow you to explore some other variations. We've also provided a page that lists some of the important conclusions that Schelling and others reached. But don't look at that until you've done some exploring yourself, and reached your own conclusions. Go to it ... and enjoy. |
(Partial Spoiler: Play Before You Look, and then Play Some More) This exhibit was created by Doug Blank, Ann Dixon, Paul Grobstein, and Ted Wong, and is a project of the Emergent Systems Working Group of the Center for Science in Society at Bryn Mawr Colllege. The models provided here are modifications of Wilensky, U. (1998). NetLogo Segregation model. http://ccl.northwestern.edu/netlogo/models/Segregation. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL and were translated into java applets using NetLogo software made available by the Center for Connected Learning. |