Mixed Uncertainty in One View: Co-Visualizing Statistical Variability and Qualitative Confidence

This research investigates how to effectively communicate two distinct types of uncertainty in time-series forecasts: quantitative uncertainty (statistical variability) and qualitative confidence (expert judgment regarding model reliability). While prior work has established that qualitative confidence influences decision-making, there is little guidance on how to visually integrate this information into standard line charts. The authors conducted three preregistered human-subjects studies (n=923) using a mock Public Safety Power Shutoff (PSPS) task. Participants acted as grid operators, deciding whether to shut off power based on wind speed forecasts that varied in both statistical spread and expert confidence levels.

The researchers compared traditional text-based labels (e.g., "Low," "Medium," "High" confidence) against various visual encodings. These included glyph-based icons positioned below the chart and integrated encodings applied directly to the confidence intervals (CIs) using color, transparency, and blurred strokes. The study used a well-defined incentive structure to ensure participants faced a clear trade-off between the costs of unnecessary power shutoffs and the risks of wildfire ignition, allowing the researchers to measure how different visual designs shifted decision thresholds.

The study successfully replicated prior findings that viewers spontaneously incorporate qualitative confidence into their decisions, even in the more complex context of time-series data. Crucially, the researchers identified several non-textual encoding techniques—such as specific color and transparency variations—that perform comparably to text labels. These findings suggest that designers can integrate qualitative confidence directly into forecast visualizations without sacrificing decision quality, providing a more compact and potentially more intuitive way to present multidimensional uncertainty.