Visual communication of uncertainty in continuous distributions
This dissertation investigated how accurately and effectively different graphical display formats can communicate quantitative information about an underlying continuous distribution of an uncertain outcome to knowledgeable but not necessarily technical audiences. Across three studies, display formats were evaluated on three metrics: 1) Correct responses on questions with objective answers; 2) projection of a future forecasters' range estimates (a proxy for how well the uncertainty was conveyed); 3) subjective experience using the display. The role of numeracy as a main effect and its interaction with Display was also assessed. Study 1 found that for objective measures the Quantile plot and CDF plot performed better than either a Box or Density plot, but this outperformance only manifested among the high numerate; for projection, the Box plot induced the widest range bounds and the Density plot the smallest; for subjective measures, the Quantile plot was rated highest and its objective / subjective scores were best aligned. Study 2 extended the first study by testing whether communication can be improved by combining multiple plot elements into a single display or by allowing participants to toggle between different plots. For objective measures, the addition of a CDF element boosted performance; for projection, the Toggle condition yielded wider range bounds. In both cases, numeracy moderated the effect, with the differences only manifesting among the high numerate. A Toggle CDF+Box display had the greatest alignment between objective and subjective measures. On common events across studies, however, a Quantile plot alone still yielded the best performance on objective measures and Box plot the best on projection - suggesting an inherent trade-off. Study 3 extended the findings from binary to categorical / ordinal events by testing the effectiveness of four different display types intended to effectively scale as the number potential outcomes increase. The Box+Scatter plot, a Box plot with a 1-dimensional scatterplot of the raw forecasts superposed (using transparency to provide a pseudo-density) performed best on objective measures and its objective and subjective scores were best aligned.
Marcus, James Charles, "Visual communication of uncertainty in continuous distributions" (2014). ETD Collection for Fordham University. AAI3630168.