Leveraging Modularity During Replication of High-Fidelity Models: Lessons from Replicating an Agent-Based Model for HIV Prevention

High-fidelity models are increasingly used to predict, and guide decision making. Prior work has emphasized the importance of replication in ensuring reliable modeling, and has yielded important replication strategies. However, this work is based on relatively simple theory generating models, and its lessons might not translate to high-fidelity models used for decision support. Using NetLogo we replicate a recently published high-fidelity model examining the effects of a HIV biomedical intervention. We use a modular approach to build our model from the ground up, and provide examples of the replication process investigating the replication of two sub-modules as well as the overall simulation experiment. For the first module, we achieved numerical identity during replication, whereas we obtained distributional equivalence in replicating the second module. We achieved relational equivalence among the overall model behaviors, with a 0.98 correlation across the two implementations for our outcome measure even without strictly following the original model in the formation of the sexual network. Our results show that replication of high-fidelity models is feasible when following a set of systematic strategies that leverage the modularity, and highlight the role of replication standards, modular testing, and functional code in facilitating such strategies.