New HCI study: Blur and color encode forecast confidence as well as text

A new paper from UC Merced and Savannah River National Laboratory researchers tackles a common data-viz problem: how to show both statistical variability (confidence intervals) and subjective expert confidence in the same line chart. The team, led by Racquel Fygenson, conducted three preregistered human-subjects studies with a combined 923 participants. Experiment 1 replicated earlier findings that adding qualitative confidence cues changes decision-making. Then Experiments 2 and 3 tested five visual encodings—text labels, icons, color gradients, transparency, and a blurred stroke design—for representing that qualitative layer.

Surprisingly, the non-textual encodings performed just as well as explicit text at getting participants to incorporate expert confidence into their judgments. Blurred confidence intervals in particular proved intuitive, likely because visual fuzziness intuitively maps to vagueness. The authors emphasize that no single approach dominates; instead, designers should match the encoding to the data's narrative and user context. The paper, "Mixed Uncertainty in One View," offers concrete guidelines: use redundant encodings, avoid misleading precision, and test with real users. It's a timely contribution as AI-generated forecasts and dashboards increasingly need to communicate both quantitative and qualitative uncertainty to non-experts.