Valdis Krebs (who is also one of the writers at the Network Weaving blog) has examined the book co-purchasing data from Amazon:

The new thing in Valdis’ regular analyses is that there is a tripartite structure in the book purchasing data: we have purple quantitative/science/liberal (pointy-headed?) books centered on “Freakonomics”, red pro-interventionist religion-partisan books centered on “America Alone”, “Politically Correct Guide to Islam” and “Godless”, and blue anti-interventionist cross-religion books centered on “The Looming Tower” and “Fiasco: The American Military Adventure in Iraq”. The is considerable distance from purples and reds. While Krebs marks “The God Delusion” as purple, I see it more as a separate anti-religion cluster shared both by blue and by purple.

I have written about the recommender technology, an application area of data mining. In particular, with recommender technology, books get recommended to readers based on the history of their purchases. What this means in practice, however, is that people will be recommended more of the same. Such technology will have a clustering effect on the people in the sense that once the machine has established that someone is, say, a Democrat, only those books that are in alignment with the Democrat perspective will be recommended to him. Contrast this with the current recommendation “technology” based on bookstore aisles where one finds books on politics in a single section – with a mix of perspectives.

Internet allows us to transcend geography. When it becomes faster to talk to a colleague on a different floor using email and instant messaging than it is to meet in person, there is not much to geography beyond time zones. But there is only a limited number of associations we can develop with other people. For that matter, when geographic similarity matters less, other aspects that determine whether a bond will develop between two people begin to matter more.

When one allows people to settle where similar people live, the geography is going to mirror these clusters, and political opinions will be increasingly geographically clustered. In particular, observe how blue are the urban centers. While this picture might be of the USA, the same pattern also appears in Europe.

For example, I often wonder if New York is really a diverse environment. It is diverse in the sense that it hosts a vast number of diverse circles, but these circles tend to be disconnected from one another. I would even say that people within each circle are a lot more similar ideologically, intellectually and educationally than any “non-diverse” environment of a small town, where different people have to learn how to live together. In a big city, one can pick who you’re going to spend time with, so one doesn’t have to adapt. Without this, people grow apart, perhaps even stop understanding or feeling for the “other” group of people. With sufficient geographic clustering over the span of several generations, one again ends up with borders, civil wars and similar. And with the internet, one can grow apart without even moving, meaning that we can see this clustering at a faster and faster rate and the consequent conflict even sooner. Some things never change.