Calibrating and Evaluating Simulations

This dissertation consists of three papers that examine philosophical questions concerning computer simulations and modeling. In my first chapter, I argue that it is essential to consider the role of simulation calibration when making epistemic comparisons between experiments and simulations. Doing so allows us to be sensitive to differences in their respective methodologies. Simulation calibration, which takes place during the model’s construction, results in changes to the simulation model. As such, in theses epistemic comparisons between simulations and traditional experiments, it matters whether we are considering calibrated or uncalibrated simulations. My second chapter continues the inquiry into the philosophical importance of calibration and argues that simulation calibration is like metrological calibration in substantive ways, namely that they both depend on comparison and have a goal to make what is being calibrated like the object of comparison. Simulation calibration, however, goes beyond metrological calibration in that the parametrization that takes place during simulation calibration can be considered novel information about the target system in cases where the parameters that are calibrated represent physical properties in the target system. In my third and final chapter, I consider implications of the computational nature of simulations on model evaluation. I consider two theories of model evaluation, the adequacy-for-purpose and similarity views. The adequacy-for-purpose theory posits that models are evaluated on their representation adequacy, but also other criteria such as methodology and the user, making it a more holistic approach to model evaluation. I argue that even though the adequacy-for-purpose theory is focused on evaluating the representational adequacy of the model, it can accommodate computational aspects of simulations by including them in the methodological considerations. Similarity accounts of model evaluation focus only on representational similarity, and this approach cannot account for the computational aspect of simulation modeling. Reframing computational aspects in representational terms fails to consider the kinds of questions that raised by new modeling methods. The upshot of these three investigations into simulation modeling is that examinations of simulation modeling need to consider more fully the nuances of simulations, namely the model-building techniques and the computational aspect.