The maths does not favour a "side" here. The maths demonstrates the consequences of different distribution of attitudes. Dependent on your views, different parameters will lead to good or bad results.

What the model does demonstrate is that the set and distribution of viewpoints that will lead to desegrated neighbourhoods is far smaller than people tend to think. It is much easier to end up with segregation than most people would expect, even with predominantly mostly benign views in most of the population.

>The maths demonstrates the consequences of different distribution of attitudes.

It demonstrates the consequences of a modeling different distributions of attitudes in a particular way.

>What the model does demonstrate is that the set and distribution of viewpoints that will lead to desegrated neighbourhoods is far smaller than people tend to think. It is much easier to end up with segregation than most people would expect, even with predominantly mostly benign views in most of the population.

It does not demonstrate this, because the parameters and details of the model have no relation to reality. In order for your claim to be true, there would have to be some mapping from real world conditions, to the parameters of the model.

All this model demonstrates is that for some parameters, individual preference for similarity of neighbors dominates over preference for diversity of the community. Which is (to me) obvious anyway.

> It demonstrates the consequences of a modeling different distributions of attitudes in a particular way.

The model is defined in terms of the attitudes.

If the model does not match the stated definition of the attitudes, then that is a bug in the code, not being creative about how to model those attitudes.

> It does not demonstrate this, because the parameters and details of the model have no relation to reality.

You say that, yet you go on to contradict yourself:

> All this model demonstrates is that for some parameters, individual preference for similarity of neighbors dominates over preference for diversity of the community. Which is (to me) obvious anyway.

The model shows that you can assemble a wide range of attitudes that most people wouldn't dream of considering racist, yet that still contributes to make segregation worse.

That's certainly not been the prevailing attitude. A lot of people that are strongly in favour of desegregation have been assuming that relatively small steps (e.g. getting people to be fina about moving into mixed areas) would be sufficient to over time lead to desegregation.

Yet the model blows that idea out of the water.

That there are other factors is largely irrelevant to addressing this attitude.

Are you arguing for the invalidity of most, if not all agent-based simulations?

Yes I am. I don't think they are a useful way to think about the world.

One model in isolation can appear compelling, as some other posts in this thread indicate. But you can't view them in isolation, because all the model shows is that a simulation with certain properties exists. The problem is that:

1. There are a huge number of ways to model the same process. There are many details of the model that could be changed or tweaked, each giving different results. This is a bigger problem for agent based simulations than traditional game-theoretic/economic models (although it's a pretty big problem in both cases).

2. Even if you know that your model is the "right" one for the process you're interesting, in the real world all kinds of processes are interlinked. E.g. racial segregation is highly linked with Southern vs Northern, rural vs urban and wealthy vs poor. It's possible that there things swamp the process that the simulation is modeling. Even if they don't they can make the data so messy that it's impossible to test the model empirically (not that there was any attempt in this case).

Did you play with it at all? It's somewhat flawed--for example, boosting the "prejudice" level to > 85% often leads to diverse, if not stable, communities. But I think it demonstrates the concept well. If a community starts out segregated, it takes being actively displeased with the lack of diversity to get it to change, because people, without a reason to move, won't move.

No I didn't. My point was that when you expand or change the model, you open up a whole new range of possibilities. I'm not particularly interested in how the model is parametrized or what parameters lead to what result.

Again, the point was not the model, but the story.

If the point is the story, what purpose does the model/simulation serve? In this case, the best that the model does is prove that the story is internally consistent. It doesn't prove that the story has any relationship to reality, because there are hundreds of other stories about segregation that you could tell, each with their own model.

It serves the point of illustrating that a common belief about how segregation and level of racism in the population does not hold together: More segregation can occur even with beliefs we would not consider racist; all else being equal, reducing, or even preventing the increase of, segregation takes far more than absence of racism.