Talk (Work in Progress): Leon Assad (MCMP)

Title:

Is model robustness a confirmatory virtue?

Abstract:

Is model robustness a confirmatory virtue? Model robustness denotes the convergence of many models to the same results. Often this is taken as a confirmation of these models and the hypotheses they embody. Model robustness is, for instance, often invoked by climate scientists. But is robustness really confirmatory? Some philosophers of science believe that it is not.

In my paper, I formulate a formal argument in favor of model robustness as confirmatory. In doing so, I draw on non-formal positions optimistic on this issue (in particular on Elisabeth Lloyd’s 2015 paper concerning climate science) and use a Bayesian framework (developed in Bovens & Hartmann, 2003). My analysis suggests that when an agent is confronted with a set of related models that converge to the same empirically confirmed result, they have reason to strengthen their belief that the models indeed represent the cause of the real target phenomenon. This belief may be strengthened further as the number of converging models increases. However, this holds only under certain conditions; it remains to be discussed whether these conditions are satisfied in each specific case and, more importantly, whether the assumptions underlying my Bayesian analysis are justified.