Using Statistical Learning to Navigate Signed Social Networks

Knowing the structure of our signed social networks—comprised of both friendly and antagonistic relationships—is critical for successfully navigating human social life. Yet, it is well-known that people have systematically biased mental representations of social network structure. The dominant account explaining these biases is Balance Theory, which posits that people have a desire for cognitive consistency or balance. Prior work, however, shows that people’s biases are not faithful to Balance Theory’s predictions. We propose an alternative account: that statistical learning of natural occurring patterns in the real social world shapes the inductive biases people rely on to navigate signed social networks. Using computational models, we formalize how these biases can be acquired and represented to facilitate strategic social behaviors (e.g., link prediction). Analysis of nine open-access datasets and one self-collected longitudinal, signed social network (which together span over 150,000 nodes and 1.6 million ties) demonstrates that individuals’ biases reflect the real-world statistics of social networks. When we directly pit Balance Theory against statistical learning, we find that individuals can (slowly) adapt to entirely novel network structures that contradict established real-world statistics, revealing that statistical learning is the mechanistic source of these biases. Our findings position statistical learning as a fundamental framework underlying how humans learn about relational patterns that make up their social networks, providing a new lens for understanding social navigation.