Tractability assumptions and derivational robustness, a match made in heaven?

Abstract

The epistemology of modeling in science is a topic that has received no shortage of attention from both philosophers and scientists. Models, with their ubiquitous unrealistic and false assumptions, raise a myriad of interesting epistemic questions. Many have picked up a concern expressed by biologist Richard Levins, which is that, for many models, it is unclear if any particular result is a product of the realistic assumptions, or if the result, is in some critical way, dependent on an unrealistic assumption. His suggestion of a solution to this problem, the use of robustness analysis, has similarly been picked up by many as a plausible solution to this concern.
Since Levins’ time, taxonomies of modeling assumptions and types of robustness analyses have been developed. One common connection made is that of tractability assumption, or assumptions introduced for the purpose of mathematical tractability, and derivational robustness analysis, or a robustness analysis that tests the influence of assumptions on the derivation of some result. Derivational robustness analysis is often singled out as being particularly well-suited for resolving the epistemic concerns introduced by tractability assumptions.
In this paper, I argue that, despite how it is commonly approached in the literature, tractability assumptions do not present a consistent set of epistemic concerns. Given this, derivational robustness analysis cannot be used to resolve the epistemic concerns raised by all tractability assumptions. In use this to motivate some concerns about how well-suited the taxonomy that includes tractability assumptions is to discussions about the epistemic concerns of unrealistic modeling assumptions.