Communicating uncertainty: design patterns for framing model results in scientific publications

Abstract

Uncertainty is a prominent issue in modelling. We learn early in our studies that “all models are wrong, but some are useful.” We also learn accompanying techniques for quantifying performance, and methods for addressing uncertainty within our analyses. When it comes to publishing our results, communicating uncertainty appears to be part of the craft side of modelling, one that we learn best by experience. Sooner or later, we discover that reviewers (and the reader) are willing to accept limitations of our modelling if we use certain key phrases (e.g. “left to future work”) or subtly change our wording (e.g. “seems to indicate” vs. “proves”). Our writing effectively frames the model results, implicitly conveying the author’s judgement about model uncertainty, confidence about results and shaping the reader’s expectations of how the model may be wrong and how it is still useful. While it does not appear to have been broached in the literature on uncertainty in modelling, the framing of model results appears to be one of the primary means by which modellers have addressed uncertainty, and specifically communication of uncertainty, within scientific publications. It is one of the core practices that new modellers need to learn to ensure that their model-based analyses are considered to be credible and useful. Unfortunately, this practice cannot be easily distilled into an algorithm, method or recipe. As with other aspects of the ‘art’ of modelling, there does however appear to be some knowledge that should ideally be transferable