Errors in social network knowledge predict how ties evolve in the future

A striking feature of people’s social knowledge is that it is riddled with errors. These errors—e.g., making mistakes about who are actually friends in a social network—has typically been attributed to cognitive limitations or heuristics that prioritize knowledge of general social patterns over details about specific ties. Here, we demonstrate that such errors instead reflect an adaptive link prediction process that anticipates how the network will evolve far into the future. By longitudinally following a large, real-world social network (N=196) and tracking the formation and dissolution of social ties over the course of a year, we show that judgments that are wrong in the moment accurately forecast which relationships will emerge or dissolve up to six months later. On average, participants’ errors predicted tie formation nearly 3x, and tie loss more than 1.2x above chance levels. Beyond the use of simple heuristics like triadic closure and homophily, computational modeling reveals that successful link prediction relies on mental representations of the social network that integrates information about both direct and indirect, long-range ties between individuals. Having a social network representation that is future-oriented yields distinct social advantages: those who are better able to predict changes in the network become more centrally positioned over time, contingent on their initial position.