Three essays on models of social conventions

Humans once lived in a state of relative equality. At some point in the last 24,000 years, stratification began to emerge. There is a research program in philosophy that employs the tools of evolutionary game theory and cultural evolution to illuminate how this transition might have occurred. The main idea is that humans developed social categories to help solve coordination problems. When we need teams of people to perform complementary tasks, we also need a system of norms and expectations to delegate who will take on what roles. As a by-product, these social categories introduced mechanisms for power to be accumulated by one group over another. Small differences in initial power can shift the bargaining relationship between groups and generate a system of runaway inequality. This dissertation explores two questions about this story. First, the formal models involved assume people engage in similarity-biased social learning - people in one category learn primarily from other members of that category. Without this assumption it is difficult to maintain distinctive group behaviours that persist across generations. Similarity-biased social learning has not received an adequate evolutionary explanation in the literature. I develop several models to show the conditions under which the learning bias is adaptive. It is unclear how we should evaluate the accuracy of these models. They are highly simplified representations of ancient societies. The second component of this dissertation is on the philosophy of modeling in the cultural evolution community. There is one set of norms that treats simple theoretical models as tools for generating ideas. Those ideas are, in turn, tested using standard statistical models in the linear regression family. There is another set of norms that treats simple theoretical models as potential statistical tools. The models are taken to represent the assumed data-generating process. Empirical work requires enriching the theoretical model with context-specific assumptions to make it a more credible depiction of actual historical societies. I argue this second approach is necessary to make substantial progress in understanding cultural dynamics.