Results from aggregate model runs for The Evolution of Similarity-Biased Social Learning by Paul E. Smaldino and Alejandro Pérez Velilla. Code repository at https://github.com/datadreamscorp/SimilarityBiasedLearning Abstract: Humans often learn preferentially from ingroup members who share a social identity affiliation, while ignoring or rejecting information when it comes from someone perceived to be from an outgroup. This sort of bias has well-known negative consequences---exacerbating cultural divides, polarization, and conflict---while reducing the information available to learners. Why does it persist? Using evolutionary simulations, we demonstrate that similarity-biased social learning (also called parochial social learning) is adaptive when (1) individual learning is error-prone and (2) sufficient diversity inhibits the efficacy of social learning that ignores identity signals, as long as (3) those signals are sufficiently reliable indicators of adaptive behavior. We further show that our results are robust to considerations of other social learning strategies, focusing on conformist and payoff-biased transmission. We conclude by discussing the consequences of our analyses for understanding diversity in the modern world.