Wednesday, December 23, 2015

Can you change society by changing your preferences?

A map of New York City, where residential areas are colored with dots according to the race of its inhabitants. In finer resolution, each inhabitant is represented by one dot. Non-residential spaces remain white. Image Copyright: 2013, Weldon Cooper Center for Public Service, Rector and Visitors of the University of Virginia (Dustin A. Cable, creator).

Have you ever heard somebody claim to have no negative feelings towards immigrants while living in a neighborhood with hardly any foreigners? What were your initial thoughts? Chances are, they went along the lines of "well, maybe you think you are not racist". Well, maybe you were wrong. 

The person whose work will assist me in making this point is Thomas Schelling. He created a model of residential segregation, based on an ordinary checkerboard, that is simple enough a child can play it, but it comes from a mind brilliant enough for a Nobel prize
I'll show you an example involving Claire, a Brussels resident. It starts out with the idea that every person, including Claire, prefers to live in a neighborhood where at least part of the neighbors are like her. This initial bias might not be large, she might simply prefer to have about 30% neighbors who are Belgian like her. This does not look like much, considering that the Brussels' population does not contain 70% of foreigners, but rather 33.9% (as of January 1st 2015, according to Statistics Belgium). But what happens if this desire is translated to reality, for all Brussels residents?
Let us assume that we start out with a situation where the population in a residential area consists of 50% Belgians and 50% Non-Belgians, and housing has been randomly assigned. Claire does not mind the 50% of foreigners in her general neighborhood, but would still like to have at least 30% of her immediate neighbors to be Belgians like her. You can see this initial situation below, in figure 1. Although on average there are about half of the neighbors the same, a number of the residents are still unhappy because their immediate neighbors are less than 30% like them. Here, these unhappy people are marked with a cross. 

Figure 1. Initial state of the segregation model. On average, every square has 50% "same" neighbors, and every individual square would like to have 30% neighbors who are like them. Those marked with a cross have less and are thus "unhappy".

The aim of the game is now to move the unhappy residents randomly to free units, or units occupied by other unhappy residents, until all are happy with their living situation. The only rule: once residents are happy in their position, they won't move unless their neighborhood makes them unhappy again. You can see the end-result of this 30% "same" preference in figure 2. 

Figure 2. End state of the segregation model with an initial 30% same preference.

I can repeat this as often as I want (I actually repeated it over 30 times), with different initial random distributions of "Belgians" and "immigrants", always in a 50:50 ratio, but I will always get roughly the same output: highly segregated neighborhoods with around 75% similar neighbors by the time all residents are happy. 
The Schelling model of segregation, or its idea that a small, personal bias can have a large impact on society, has been explored in  domains other than residential segregation. One such example is opinion polarization, the idea that different opinions on the same topic emerge in different groups of society. Michael Mäs and Andreas Flache from the University of Groningen created a computer model as well as an experiment with real participants in 2013. They successfully demonstrated that there is no need for negative influence in order to create two separate "opinion peaks" regarding a certain topic. In the domain of stereotypes, Mark Agars (2004) shows, purely based on calculations, how a very small preference for hiring men over women can considerably change the percentage of women in the upper ranks of companies (or universities, for that matter). These examples (and there are many more) are not necessarily applications of the Schelling model, but the crucial point remains: none of the individuals of a society needs to be racist, misogynist or have extreme opinions; a small initial bias is sufficient. This is not to say that there are no racists out there, only that not having extreme positions is not sufficient to eliminate potentially harmful biases. 
But back to Schelling: what can we do to avoid residential segregation? Let us assume that Claire and the residents in our initial example are serious about wanting to live in the same neighborhood with immigrants. What would be a reasonable initial "same" preference that could lead to a non-segregated community, where half of the residents are immigrants? It turns out, Claire would have to lower her expectations to have neighbors who are like her to around 12%! If she would want to live in a housing block with 100 apartment units, she would have to consider to choose one where only 12 units are occupied with people who are like her. I don't think this is a particularly realistic scenario, especially since the whole point is that the initial, "real" preference happens to be higher, and she does not want to belong to such a small minority. 
But what if they didn't have to adapt their needs to such a degree? Let's try to simply loosen up the conditions a little bit: instead of a fixed level of 30% who are like them, the individuals in the modeled neighborhood could be happy if they live with at least 20% immigrants and a minimum of 20% neighbors who are like them. Here, on average, half of the direct neighbors will be the same, not unsimilar to the initial random state, except that it could be achieved out of the position of an already segregated neighborhood (figure 3).

Figure 3. One possible end state with a minimum of 20% same and 20% different neighbors.

In short, if everybody would chose to have only two out of then close neighbors who are different from themselves, neighborhoods could be less segregated while accommodating a larger variety of personal preferences. 
The caveat of this solution: the residents would have to actively seek out a change in their environment. To get back to Claire, it means that she would have to decide to move out of her apartment, but likely also out of her comfort zone. 

If you found this post interesting, you might like to spend a little time on Vihart and Nicky Case's very lovingly done interactive version of the Schelling model, with variations beyond what I wrote about her and in several different languages
Another possibility is Netlogo. There is an online-version of NetLogo as well as a more powerful one you can install on your computer. In the "model library" section you will find the segregation model (under "Social Sciences") as well as many other models to play around with. This is what I used to create the figures above (for specific references see below)

Julia Eberlen, who wrote this post, is a Ph.D. student in the center for social and cultural psychology. Her work bears on the diffusion of group stereotypes in social networks. 

Article references: 

Agars, M. D. (2004). Reconsidering the Impact of Gender Stereotypes on the Advancement of Women in Organizations. Psychology of Women Quarterly, 28(2), 103-111. doi 10.1111/j.1471-6402.2004.00127.x

Mäs, M., & Flache, A. (2013). Differentiation without Distancing. Explaining Bi-Polarization of Opinions without Negative Influence. PLoS ONE8(11), e74516.

Schelling, T.C. (1971). Dynamic Models of Segregation. The Journal of Mathematical Society 1(2), 143-186. doi: 10.1080/0022250X.1971.9989794

"Thomas C. Schelling - Facts". Nobel Media AB 2014. Web. 22 Dec 2015. <>

Statistics Belgium (2015).  Population par sexe et nationalité pour la Belgique et les régions, 2005 et 2015. retrieved from:

Model, Software and Image References: 

Top image of the City of New York:

Image Copyright, 2013, Weldon Cooper Center for Public Service, Rector and Visitors of the University of Virginia (Dustin A. Cable, creator). retrieved from: 3: 

Figure 1 and 2:  

Wilensky, U. (1997). NetLogo Segregation model. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL.       

Wilensky, U. (1999). NetLogo. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL.   

Figure 3: 

A personal modification of Wilensky's NetLogo Segregation model (cited for Figure 1 and 2), using NetLogo (see above). 


1 comment: