Johannes Nystrom
Stockholm University
When scientists seek further confirmation of the result of a scientific model, they often try to replicate it in a variety of different models that share the same theoretical core. If successful, the result is said to be robust. Traditionally, the confirmation value of robustness has been understood as being based on the probabilistic independence of the different model’s results. Given that they are independent, observing that they give equivalent results is ‘surprising’ and should increase our credence in the given result. However, recent criticism (Schupbach 2018) questions the possibility of achieving such independence in practice. In this paper, I argue that the relevant probabilistic independence must be conditional on the models sharing the same core. This leads to a partial (unconditional) probabilistic independence that is more well-suited to account for robustness analysis. While slightly weaker, it may be substantial depending on scientific context.
Still, Harris (2021) defends another strong concern. Even if a satisfying notion of probabilistic independence can be defended, it is not clear why the correctness of the result is a variable that can explain the ‘surprise’. What does the correctness of the result, in the real world, have to do with how our models behave in model land? In response, I argue that the probabilistic relationship should be represented as a three-variable network. What explains the ‘surprise’ is a ‘new’ propositional variable Y which states that the number of results that can possibly be derived by the relevant class of models is very low. In turn, Y straightforwardly increases the likelihood that the modelling result of interest is correct. This three-variable network offers an accurate representation of the non-empirical structure of robustness analysis, and I show that it amounts to (non-empirical confirmation) under some plausible assumptions. Throughout, I operate within the framework of Bayesian epistemology.
Ref.
Harris, M (2021) The epistemic value of independent lies: false analogies and equivocations, Synthese, 199, (5-6): 14577-14597
Schupbach, J (2018) Robustness analysis as explanatory reasoning, British Journal for Philosophy of Science, 69, 275–300
Chair: Josef Kohlmaier
Time: September 8th, 10:00-10:30
Location: SR 1.007
SOPhiA 2026 