Describing likelihood and impact in impact-based forecasts: What’s likely? What’s severe?

Impact based forecasts and warnings (IBFW) combined information about the potential consequences (impacts) of weather events with the estimated probability of them occurring. In representing IBFW ordered verbal categories such as low, medium and high may be used to classify both probability and impact severity. However, there has to date been limited exploration of how these categories are interpreted in the context of IBFW. While work in the social and behavioural sciences suggests that there can be variability in interpretation of verbal representations of probability, such as those used in IPCC reports, statements about impact severity – where numeric estimates may not be as easily attached to descriptions – remain under-investigated. In regionally representative survey with n > 1500 members of the public in England, UK, we asked participants to indicate on sliders how they would interpret a series of descriptors of likelihood (0=will definitely not happen,100=will definitely happen) and impact severity (0=not serious at all, 100=very serious) related to severe weather. Examining mean, median and distribution of responses for each verbal likelihood and impact descriptor we find that while descriptors indicating a middle point (i.e. medium, moderate) tended to elicit judgements around the central point of the scale for both impact and likelihood statements, there was overall less consistency around phrasing intended to denote low and high categories. For impact statements, we observe overlap between ‘very low’ (mean=33, median=26) and ‘low’ (mean=34, median=28) as descriptors. Catastrophic and extreme were interpreted to denote the highest level of impact, though in both cases mean and median ratings of impact seriousness were below 75 with indication of wide variability in interpretation. For likelihood, we find that only the descriptor ‘high’ elicited an estimated likelihood above 60. Surprisingly phrases that may be intended to denote very high likelihoods (‘expected’) or certainty/near-certainty (‘observed’) received mean interpretations close to the centre of the scale. Taken together our findings suggest that while terms typically associated with mid-points may be understood as intended, gaps between communicator and recipient may occur when discussion events with very low and very high likelihood. We discuss potential approaches to addressing this, including providing numeric values with likelihood descriptors and avoiding terminology where the greatest gaps between intention and interpretation may occur.