Assortativity of suicide-related posting on Twitter

Abstract

Networks in which similar individuals are more likely to associate with one another than their dissimilar counterparts are called assortative. Such patterns are a hallmark of human social networks, in which numerous phenomena (e.g., mood, weight) cluster among like-type members. The clustering of suicides in time and space implies such fatalities likely also have socially assortative features, and suggests other forms of suicide-related behavior may as well. This investigation evaluated the assortativity of suicide-related verbalizations (SRV) by machine coding 64 million posts from 17 million unique users of the Twitter social media platform, collected over two distinct 28-day periods. These data were used to assemble a network, in which users were defined as socially linked if they mutually replied to each other at least once. Bootstrapping revealed that SRV was significantly more assortative than chance, up to at least six degrees of separation (i.e., people six links apart were still more similar on SRV than chance). When user mood was controlled, SRV assortativity still remained significantly higher than chance through two degrees of separation, indicating this effect was not just an artifact of mood. Discussion demonstrates how exploiting assortative patterns can improve the efficiency of suicide risk detection.