Our new paper (with Klaus Keller) has been published in Environmental Modeling & Software.

Using a perfect model experiment, we demonstrate how small increases in agent-based model complexity can result in large differences in the data required to calibrate the model, as well as identifying the correct (in this case, the data-generating) model structure. This result emphasizes the importance of treating model structural uncertainty as a deep uncertainty, rather than committing to a particular parameterization. It also demonstrates the importance of leveraging multiple independent sources of information to construct sound and informative prior distributions.

The paper is open-access, and there is a preprint available as well.