Calls to communicate uncertainty using mixed, verbal-numerical formats ('unlikely [0-33%]') have stemmed from research comparing mixed with solely verbal communications. Research using the new 'which outcome' approach to investigate understanding of verbal probability expressions suggests, however, that mixed formats might convey disadvantages compared with purely numerical communications. When asked to indicate an outcome that is 'unlikely', participants have been shown to often indicate outcomes with a value exceeding the maximum value shown, equivalent to a 0% probability -an 'extremity effect'. Recognising the potential consequences of communication recipients expecting an 'unlikely' event to never occur, we extend the 'which outcome' work across four experiments, using verbal, numerical, and verbal-numerical communication formats, as well as a previously unconsidered numerical-verbal format. We examine how robust the effect is in the context of consequential outcomes and over non-normal distributions. We also investigate whether participants are aware of the inconsistency in their responses from a traditional 'how likely' and 'which outcome' task. We replicate and extend previous findings, with preference for extreme outcomes (including above maximum values) observed in both verbal and verbal-numerical formats. Our results suggest caution in blanket usage of recently recommended verbal-numerical formats for the communication of uncertainty.Additional Supporting Information may be found online in the supporting information tab for this article. key words verbal probability expressions; numerical probabilities; mixed-format; risk communication; geological hazards; extremity effect tion formats: verbal probability expressions (VPEs) (e.g. 'unlikely'), numerical expressions (e.g. '20%'), and mixed expressions in two orders (e.g. 'unlikely [20% likelihood]' and '20% likelihood [unlikely]'). 1 Budescu and Wallsten (1995) proposed that the choice of format for communicating likelihood information should be governed by the congruence principle: the precision of the communication should be consistent with the degree of certainty that can reasonably be expected for estimates about the event described. In the domain of geological hazards, estimations of events, such as the probability of a large earthquake, might not be precisely quantifiable. In such instances, a specific numerical expression of the probability of this event might be perceived as overly precise. Using a VPE would seem to better represent the uncertainty and underlying imprecision associated with the probability estimate. VPEs are also thought to be easier to understand and 1 In line with standard, dictionary definitions, we use 'likelihood' as a synonym for 'probability' in the present paper, although note that, mathematically, each has a unique and specific definition.
Communication formats