The use of cannabis for medical reasons in the UK: Prescriptions, sources, products, and high-risk use.

Social learning is a central component of human adaptability (–). But social learning is not only about how much we copy; it is also about who we copy. We shall adopt the word “target” to indicate an individual being copied by a social learner. In large populations, learners encounter potential targets under conditions of demographic mixing, migration, and social stratification, and they routinely use perceptible cues of identity to decide whose information to trust. Formal models for understanding the evolution of social learning have typically ignored identity, focusing on when copying is favored in the face of learning costs and environmental uncertainty (–). Social-learning models that accounted for similarity have focused on a narrow range of benefits, such as learning coordination norms () or improving the efficacy of conformity (). Meanwhile, a parallel empirical and theoretical literature has documented that learners regularly filter potential targets using social categories and markers (–). What remains underdeveloped is a clear account of how demographic composition and tag informativeness jointly determine the extent to which individuals rely on social learning and the ways they use identity markers when they do. Previous evolutionary simulations demonstrate that similarity-biased social learning is adaptive when individual learning is unreliable, indiscriminate copying is degraded by diversity, and identity signals are informative (). This paper extends that work with an analytical model yielding closed-form predictions about evolutionary outcomes. This formalization clarifies the mechanism, generates general comparative statics, and derives results regarding antisimilarity bias and frequency-dependent selection. The core idea is that identity markers matter for learning insofar as they predict transferability of information. Copying someone who is successful in their own circumstances is not necessarily helpful if the learner’s local payoffs are governed by a different behavioral optimum. This problem appears in spatially structured models when migration erodes the correlation between what worked for some individuals and what will work for those who learn from them (–).