By joining this group you will also get the opportunity to subscribe to its activity. Join this group
PURE Research Programme

Articles

« go back

PURE Research Blog: Designing to communicate uncertainty, by Professor Alison Black, Reading University

Designing to communicate uncertainty.

In the latter stages of the RACER stream of the PURE project, the Centre for Information Design Research at the University of Reading will be working with Reading’s Meteorology and Psychology Departments and industry partners, to examine methods of visualising uncertainty in disaster forecasting.

There are, currently, no standards in how uncertainty is displayed visually although, conventionally, a line of forecast is shown in the context of a spread of prediction. Broad et al. (2007) have described how this type of visualisation was misinterpreted by members of the public in the hurricane season of 2004, with underestimation of hurricanes’ impact on areas not on the line of forecast and confusion of probability of impact with intensity (Figure 1).

Figure 1: Cone of uncertainty visualisations used by US National Hurricane Center in 2004.
(Reproduced from http://www.nc-climate.ncsu.edu/edu/k12/.tropicalcyclones)

On a more positive note Stephens et al. (2012) found that participants in an on-line weather game (possibly a self-selecting user group) made better game decisions when presented with probabilistic forecasts (fan charts) indicating the spread of prediction. However research by Pappenberger et al. (2013, p.140) suggests that even experienced weather scientists may tend to pick a preferred model out of an ensemble, rather than consider the full range of an ensemble prediction. And although they recognise the role of uncertainty in improving the skill of forecasts they cannot agree on the range of uncertainty to indicate on a forecast visualisation (Pappenberger et al op cit).

These kinds of studies provide background for considering the communication of uncertainty in contexts such as the insurance or construction industry. Here analysts are working with both the risk of extreme weather events and the potential losses associated with them. They need to visualise possible scenarios, not only for their own decision-making but also for communication with clients who may not share their fluency with probabilistic data. Our team will need to understand the detail of the communications between these analysts and their clients, their different needs for information and levels of understanding before attempting to design for them.

Solutions may lie both in how data are represented and how they can be layered, with detail included, excluded or given different levels of emphasis in order to facilitate decision-making. It may be that solutions lie not only in data presentation but also in options for animating or annotating representations so that, for example, a client can replay a case that has been presented by an insurer in order to absorb it in detail. Readers will, no doubt, be aware of the wonderful visualisations of world data developed by Hans Rosling’s Gapminder team. It’s worth noting that these are not presented as stand-alone communications but fronted by Rosling who provides a commentary on what we are seeing. Considering a similar, albeit less dramatic, approach may be appropriate in our work.

Figure 2: Hans Rosling provides a personal commentary on his data visualisations in order to explain their significance to his viewers.
(Image from gapminder.org.)

We may find that, although there are no standards for presentation, there are conventions that professionals have grown accustomed to and that we may either challenge or work around. The popular design press still reports a salutary 2007 case study of (unsolicited) proposed re-designs of the interface for the highly complex Bloomberg terminal (Figure 3a). Simplified, calmer approaches (for example Figure 2b) were rejected by Bloomberg’s then Chief Executive because they diverged from the terminal’s existing look and feel. Some commentators suggested that Bloomberg had an emotional investment in the complexity of their terminals that they were loath to lose. But there is substantial evidence (from fields as diverse as chess-playing, computer programming and medical diagnosis) that expertise leads to qualitative changes in use of complex visual data which could, at least initially, be vulnerable to significant changes in data visualisation. Note that Bloomberg have since revised their terminal design, probably influenced both by competition and by user expectations based on other everyday and business software.

a    

Figure 3: (a) Current Bloomberg terminal and (b) one of the proposed re-designs by consultancy, IDEO.

A final point, that I hope will be clear from the preceding discussion, is that information design is not simply about choosing the colours or typeface for displays (although these are important for clarity and legibility); it is about ensuring fit between data visualisation and the analysis and presentation tasks their users need to carry out. Just to reinforce, though, that usability is enhanced by a combination of fit to task and appropriate, detailed design I am closing with an illustration (Figure 4) of work carried out by Linda Reynolds, who, over 20 years ago, was commissioned by UK’s Air Traffic Services (NATS) to design the colour palettes for their first colour screen control systems. In Reynolds’ (1995) description of the work she explains her use of luminance contrast, hue and saturation to ensure fit between the screen representation and conceptual ‘layers’ of information that were important for controllers’ tasks and to ensure that users were able focus on critical information without distraction.

Figure 4: Illustration of colour design work carried out by Linda Reynolds for UK air traffic control systems’ first colour displays.
(Reproduced from Reynolds 1995.)

We hope our involvement in the RACER programme will enable us to make similar kinds of recommendations for the detailed visualisation of uncertainty data for expert and non-expert users.

 

References

Broad, K., Leiserowitz, A., Weinkle, J. and Steketee, M. (2007) Misinterpretations of the “Cone of Uncertainty” in Florida during the 2004 Hurricane Season. Bulletin of the American Meteorological Society, 8/5, 651–667.

Pappenberger, F., Stephens, E., Thielen, J., Salamon, P., Demeritt, D., Andel, S. J., Wetterhall, F. & Alfieri, L. (2013). Visualizing probabilistic flood forecast information: expert preferences and perceptions of best practice in uncertainty communication. Hydrological Processes27(1), 132-146.

Reynolds, L. (1995) The functional use of colour on visual display units. Information Design Journal, 8/2, 109-24.

Stephens, E., Spiegelhalter, D., Mylne, K. and Harrison, M. (2011) Using an online game to evaluate effective methods of communicating ensemble model output to different audiences. Poster presentation at American Geophysical Union (AGU) Fall Meeting.

Comments

Comments

1 person has had something to say so far

If I may comment on my own post: readers may be interested in this animation, by Oli Hawkins, of the uncertainty underlying published UK immigration figures http://olihawkins.com/visualisation/1
Posted on 03/10/13 12:41.

Most read articles