Yonatan Zunger’s Blog

[Edited: Link fixed]

I’ve been reading this article, and it’s quite interesting: Peter S. Bearman, James Moody and Katherine Stovel, Chains of Affection: The structure of adolescent romantic and sexual networks. The National Longitudinal Study of Adolescent Health, among many other things, did detailed surveys of behavior. This paper takes the data for a single large (~1000 people) high school, in which every student was surveyed and information about their romantic and sexual partnerships was acquired, and analyzes it in depth. The major conclusion is that the network structure of sexual relationships in high school is qualitatively different from the structure of such relationships in the world as a whole.

Specifically, other studies (I wish I had the reference next to me – will update if I find it) have shown that the sexual relationship networks of the general public are preferential-attachment networks, in which a new connection is most likely to form to someone who already has a lot of preexisting connections. (Not a big surprise, really…) This leads to a network of connected hubs, somewhat like American air routes. High school networks, OTOH, have a much more tree-like structure, with long branches and less clumping. The authors of this paper conjecture that this is due to a social norm against cycles of length 4 – i.e., against dating your ex’s current’s ex. They show that adding this assumption to ordinary models produces trees that look a lot like the ones they found experimentally. (Not too surprising, again – in a predominantly heterosexual network, if 4-cycles are excluded the smallest possible cycle is of length 6, which is already getting too big for real clumps to form)

Why is this interesting? Well, first of all there’s our usual prurient interest in who everyone else is shtupping. (cf. also this ScienceBlog entry, Monkeys will pay to look at porn) But the network of sexual connections is also the network along which STDs propagate, and so network interruption theory takes on serious public-policy implications. For example: In a preferential-attachment network, it turns out that the network is disproportionately vulnerable to interruption of its highest-linkage nodes. This means that sexual health programs aimed at the most active people will have a larger than expected effect, while broadly-aimed programs will be ineffective. In a less clumped network like the high school network, though, removing a random node is enough to disconnect major graph chunks, so a broadly aimed education program may be a lot more effective.

There are other interesting features of this, too, and for those I recommend you read this article.