Uncertainty Sketch

In his experience of making British Geological Society products, Ben Marchant (BGS) has seen how difficult it is to create measures of uncertainty which are useful to decision-makers, especially when looking at spatial information. This motivated the selection of the paper for this month’s journal club - Decisions, uncertainty and spatial information by Lark et al (2022).  

Themes covered in the paper include: 

  • Generalised and decision-focused measures of uncertainty 

  • Misinterpretation of uncertainty by map users 

  • The need for both verbal descriptions (e.g. likely, unlikely) and numerical values (e.g. 90%) to reduce the chance of bias in interpreting uncertainty estimates

  • Understanding the decision-making process and tailoring expressions of uncertainty to how the map will be used  

Some of the points raised in discussion were: 

  • Separating out measures of risk from measures of uncertainty 

  • Additional challenges that occur when a map has multiple uses in different decisions 

  • How ideas around elicitation might be useful in thinking about how people understand uncertainty   

  • Presenting neutral datasets and interactive tools – allowing users to decide what uncertainty means to them, but with a risk of the data being misused

Ben also covered two papers comparing methods for presenting uncertain information. 

  • The first covers differences in decisions when information is framed positively (probability of exceeding a threshold) and negatively (probability of falling below the threshold) 

  • The second evaluates the effectiveness of different methods for communicating uncertainty by eliciting stakeholder opinions.

For a further look into these topics, the works of these authors were mentioned: 

  • Anthony O’Hagan – expert elicitation 

  • David Spiegelhalter – communicating risk and uncertainty  

Conversations towards the end of the meeting also sparked new ideas for future sessions: 

  • Interacting with policymakers  

  • Software options and approaches for experimental design