Section: New Results

Supporting Judgment and Decision Making with Visualizations

Participants : Pierre Dragicevic [correspondant] , Luana Micallef, Jean-Daniel Fekete.

People have difficulty understanding statistical information and are unaware of their wrong judgments. Cognitive biases abound, particularly in Bayesian reasoning (see http://youtu.be/D8VZqxcu0I0 for a classic example). Psychology studies suggest that the way Bayesian problems are represented can impact comprehension, but few visual designs have been evaluated and only populations with a specific background have been involved. We conducted a study where a textual and six visual representations for three classic problems were compared using a diverse subject pool through crowdsourcing [22] . Visualizations included area-proportional Euler diagrams, glyph representations, and hybrid diagrams combining both. Our findings were inconsistent with previous studies in that subjects’ accuracy was remarkably low and did not significantly improve when a visualization was provided with the text. A follow-up experiment confirmed that simply adding a visualization to a textual Bayesian problem is of little help for crowdsource workers. It however revealed that communicating statistical information with a diagram, giving no numbers and using text to merely set the scene significantly reduces probability estimation errors. Thus, novel representations that holistically combine text and visualizations and that promote the use of estimation rather than calculation need to be investigated. We also argued for the need to carry out more studies in settings that better capture real-life rapid decision making than laboratories. We proposed the use of crowdsourcing to partly address this concern, as crowdsourcing captures a more diverse and less intensely focused population than university students. Doing so, we hope that appropriate representations that facilitate reasoning for both laymen and professionals, independent of their background, knowledge, abilities and age will be identified. By effectively communicating statistical and probabilistic information, physicians will interpret diagnostic results more adequately, patients will take more informed decisions when choosing medical treatments, and juries will convict criminals and acquit innocent defendants more reliably.

For more information, see http://www.aviz.fr/bayes .