Abstract
What constitutes a validated model? What does the model validation process involve? These questions still have not been completely resolved by the model validation community. This paper argues that a validated model for “best-estimate plus uncertainty” (BEPU) predictions requires and implies something very different from what many contemporary paradigms of model validation are focused on and built around. Starting from what is required of a validated model for BEPU predictions, the author works back to the implications on the model validation process (conceptually, operationally, interpretationally, etc.). Ultimately a shift is required in the conceptualization and articulation of model validation, away from contemporary paradigms. Indeed, the paper introduces and illustrates a “model conditioning” step that is often essential in a model validation process that supports BEPU predictions (i.e., extrapolation or interpolation with the validated model), yet does not appear to have been previously recognized in the validation literature. Another aspect of the paradigm shift advocated by the author is that hypothesis tests for model rejection or acceptance are often arbitrary and not useful for best-estimate predictions with estimated uncertainty. The conceptualization and articulation of model validation in this paper accommodates a cohesive framework that extends from validation experiments, to model conditioning, validation, and accreditation, to optimized model predictions within a BEPU philosophy.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have