Why do opinion clusters form?
We have discussed Axelrod’s model of opinion dynamics and you have seen how the model produces clustering. Let us now see why this happens.
The process is very similar to the one that we know from the much simpler “voting model” that we introduced on step 4.3. At the outset, in some regions and by random chance, we see a bit more similarity of neighbours. In these regions, agents interact more with each other and less with the rest of the world. Thus, here they become more homogeneous over time. This leads also to more dissimilarity from neighbouring regions, because there is less interaction with them. After a while, there is no more influence between different local regions and stable diversity emerges as the final result.
The agents did not plan the opinion clusters that emerged nor could they be predicted. Every time we run the model again, different clusters will arise. We can use this model to study conditions that lead to more or less diversity and clustering. Many papers, including Axelrod’s original research, have been written about effects of conditions such as the number of features or traits, the size of the neighbourhoods, the size of the cellular world, etc. on cultural diversity in this model. Next to Axelrod’s original paper (Axelrod, 1997) you can find more on this for example in work by Flache & Macy (2011). Let us just have a look at one of those examples.
Suppose we increase the complexity of the opinion space in a society. That is: let us add more features on which agents can have different opinions. Think a moment about what will happen then, or use the NetLogo model to find this out yourself and try to explain it. Below you see the results of a simulation that uses exactly the same assumptions as the one we showed above in the previous article. But this time we assume there are 10 different features, rather than 5. That is: every agent has now an opinion on 10 different things, not 5.
Axelrod NetLogo model
As you see, we have now much less diversity in the end. To be precise, there is no cultural diversity at all. There is in the end only one cultural region left. All agents agree on all 10 features. That is: there are no different clusters surviving and all agents adopt the same opinion on all different features. The result does not always have to be that extreme, but it is clear that diversity is much lower now than with only 5 different features. This may not be what you expected: more features means a more complex opinion space, thus more possibilities for diversity. So why does the opposite happen: why do we see less diversity?
A very important parameter in Axelrod’s model is the probability that two neighbouring agents have at least some probability to interact with each other. Once they can interact, they will likely become more similar, interact more in the future and eventually merge into one cultural region. The more features there are, the more possibilities there are that in the random initial situation two neighbouring agents will agree on at least one of these features. That is, the more features there are, the higher the probability that two neighbouring agents can interact from a random start. This is the reason why we see less cultural diversity if we increase the number of features in the opinion space. Every feature gives a possibility to agree on something by random chance. The more features, the more likely neighbours can interact in the random starting situation. What do you think will happen if we increase the number of cultural traits? (You will get the opportunity to try this with the NetLogo model and then try to understand what happens. This will be covered later in this week’s material and is optional.)
Let us now summarise what you have learned in this article about Axelrod’s model of cultural dissemination:
You have learned how cellular automata models can be used to understand how clustering of opinions in society could emerge from social self-organisation. This is possible when many individuals interact driven by two simple but powerful social principles: social influence and homophily.
You have also seen that social influence and homophily do not always lead to this result. For example, if the number of features in the opinion space of a society is high, these models predict much less diversity. Cellular automata models help us to understand which conditions could lead to more or less diversity and when such conditions may have unexpected and counter-intuitive effects.
Of course, we also have to ask whether such models are sufficiently realistic to study real human societies. But before that, let us try to work with the model ourselves in Netlogo.
© University of Groningen