Likes Attract, Opposites Repel?
Homophily and the Glass Ceiling Effect in Social Networks (click on the PDF link followed ‘Full Text:’)
It’s common knowledge that people tend to associate with others whom they have a great deal in common with. This idea is formalized in the article Homophily and the Glass Ceiling Effect in Social Networks with a concept known as homophily, which is the tendency for people to build network connections with those in which they find similarities with themselves. Such similarities can be formed on the basis of genetics, religion, race, culture, wealth level, age, gender, behavior, and many other factors. When constructing graphs representing social networks between people, often times it is clear that the nodes in the graph form groups (connected by a single local bridge perhaps) and even components (groups not connected by any local bridges) as a result of homophily-induced interconnections. We’ve discussed these properties of graphs in class, and this article provides some reasoning as to why these patterns arise in graph theory.
If we treat the people in a homophilic network as nodes in a graph, then we’d notice that the path length between two arbitrary nodes tends to decrease as the number of similarities increases. This makes sense, since if there exist more similarities, then there’s a higher chance of two nodes having a direct edge linking them. This implies that homophily could have an effect on a graph’s diameter, or the longest shortest path. It could also have an impact on whether or not two nodes have an edge connecting them. People who find similarities with each other are more likely to have a connection, while people who are relatively dissimilar are less likely to share a connection.
In this graph, nodes are color coded according to level of similarity. Only the nodes sharing the same color have connections. Homophily causes connections, but it can also cause the lack thereof, and ultimately increase the number of components of a graph.
Not only can components form as a result of this phenomenon, but groups may form as well. The article provides numerous examples of types of groups that can form within graphs under the influence of homophily, one of which being race and ethnicity. This is arguably one of the stronger influences on social network connections. While people belonging to the same ethnic group tend to form positive connections with each other, strife between races may cause negative connections (represented by positively and negatively signed edges between nodes, graphically). There are many other categories of groups that can be formed, similar to the example we worked out in class about the social network of the average individual, consisting on group categories such as coworkers, family, Cornell friends, other friends, etc.
A group can also be formed based on gender. The authors of this article mention an interesting phenomenon—that gender “influences the formation of cliques and larger evolving network structures.” (p. 42). Homophily and gender differences often cause young boys and girls to gravitate towards different social circles. In the case of adults, it was observed that men still tended to have more homophilic networks than women do. The evolution of such a network was modeled with a graph in a similar way that we model networks in class. Men generally choose nodes similar to themselves, and then choose nodes that were already connected to the people that they’re directly connected with to form their network (p. 42). It’s evident the homophily can have a large impact on graphical models of social networks.
Image from the article depicting the difference between a graph not under the influence of homophily, and a graph completely under the influence. The group distinctions in the graph on the right is clear.