Behavioral experiments in computational social science

Behavioral experiments are rarely used as an empirical strategy in computational social science, where empirical studies typically focus on analyzing large-scale digital trace data. We argue that behavioral experiments have a role in computational social science, in particular in combination with agent-based modeling – a key theoretical strategy in computational social science. We highlight three ways in which behavioral experiments can contribute to theory building in computational social science: by testing macro-level predictions from agent-based models, by evaluating behavioral assumptions on which these models are based, and by calibrating agent-based models. We illustrate these points through three examples from our work concerned with the emergence of conventions.