Statistical Learning Shapes Cognitive Maps of Signed Social Networks

Building cognitive maps of signed social networks—those containing both friendly and antagonistic ties—is essential for navigating social life. These cognitive maps are often not faithful reproductions of social networks but instead exhibit systematic biases. While dominant accounts attribute these biases to schemas, such as ones derived from Balance Theory, they leave open the question of where such biases come from in the first place. We propose that these biases emerge from statistical learning of real-world relational patterns. Across four studies that combine behavioral experiments, computational modeling, and analyses of ten longitudinal signed networks (over 150,000 nodes and 1.6 million ties), we find that people’s biases simply reflect statistical structures of their social environments. Biases align more closely with naturally occurring motifs than the patterns predicted by Balance Theory. When exposed to artificial networks that violate real-world statistics, individuals gradually, yet flexibly alter their biases, implicating statistical learning as the underlying mechanism. These findings position statistical learning as a foundational process shaping how humans encode and reason about complex signed social networks, providing a new lens for understanding socio-cognitive map building.