Improving the representation of smallholder farmers' adaptive behaviour in agent-based models: Learning-by-doing and social learning

Computational models have been used to investigate farmers' decision outcomes, yet classical economics assumptions prevail, while learning processes and adaptive behaviour are overlooked. This paper advances the conceptualisation, modelling and understanding of learning-by-doing and social learning, two key processes in adaptive (co-)management literature. We expand a pre-existing agent-based model (ABM) of an agricultural social-ecological system, RAGE (Dressler et al., 2018). We endow human agents with learning-by-doing and social learning capabilities, and we study the impact of their learning strategies on economic, ecological and social outcomes. Methodologically, we contribute to an under-explored area of modelling farmers' behaviour. Results show that agents who employ learning better match their decisions to the ecological conditions than those who do not. Imitating the learning type of successful agents further improves outcomes. Different learning processes are suited to different goals. We report on conditions under which learning-by-doing becomes dominant in a population with mixed learning approaches.