Abstract

Ever since Richard Levins’s influential article “The Strategy of Model Building in Population Biology” there has been a growing discussion surrounding the types of tradeoffs that confront scientific model builders. However, almost all the discussion of Levins’s modeling tradeoffs has focused on what Michael Weisberg calls the ‘representational ideals’ of scientific models. In this paper I argue that this emphasis on representational aims misses many of the pragmatic tradeoffs that scientific modelers confront due to limited experimental data, measurement tools, modeling frameworks, and other modeling resources. In response, this paper aims to investigate the pragmatic modeling tradeoff between (1) having a model be constructable from, and testable against, the available experimental data and (2) building models that are able to generalize across a wide range of contexts of application. I argue that this experiment-applicability tradeoff is a relationship of attenuation rather than a strict or necessary tradeoff between model properties. I then use three case studies to show that, rather than a strict theoretical limit, how this tradeoff is best navigated is highly context sensitive. I then explore the philosophical implications of this tradeoff for how we ought to think about theories as collections of models, how models connect with experiments, and how modelers balance various modeling aims.