Abstract
Robustness analysis faces a confirmatory dilemma. Since all of the models in a robust set are idealized, and therefore false, the set provides no confirmation. However, if a model is de-idealized, there is no confirmatory role for robustness analysis. Against this dilemma, I draw an analogy between robustness analysis and experimental replication. Idealizations, though false, can play the role of controlled experimental conditions. Robustness, like replication, can be used to show that some means of control is not having an undue influence. I conclude by considering some concerns about this analogy regarding the ontological difference between models and experiments.
