Understanding a complex social system requires an accurate model of the behavior of the micro-entities, as even seemingly innocent differences between micro models can generate markedly different macro dynamics. While the theoretical understanding of complex social systems has increased tremendously in the past decade, the field lacks empirical research testing competing micro-models and their macro implications. As a consequence, theoretical models are based on various competing micro-models - such as myopic best-response, reinforcement learning, and imitation - that have not been validated.

Here, we report results of a pilot study informing a bigger laboratory experiment on a simple coordination problem in a network of 15 to 20 nodes. We experimentally manipulated which information was provided to subjects in order to allow them to or prevent them from using alternative decision rules. We tested which micro-model best described observed decisions depending on the provided information, using both the micro-data and the macro-patterns that we observe during the experiment.