KAIST study: Visualization channel rankings shift across four perceptual tasks

A new paper from KAIST researchers—Soohyun Lee, Seokhyeon Park, Minsuk Chang, and Jinwook Seo—challenges long-held assumptions about which visual encoding channels work best. "Revisiting Channel Effectiveness: A Multi-Dimensional Evaluation with Primitive Visual Stimuli" (arXiv:2608.04435), accepted to IEEE VIS 2026, argues that classic rankings (like Cleveland & McGill's) focus almost exclusively on magnitude estimation accuracy in full chart contexts. The team instead tested seven core visual channels—position, length, tilt, area, curvature, luminance, and saturation—using primitive visual marks stripped of chart scaffolding. This isolates how each channel performs on four distinct perceptual tasks: accuracy, discriminability, separability, and pop-out.

The results, based on crowdsourced experiments, show that channel effectiveness is fundamentally task-dependent. Spatial channels generally lead, but accuracy hinges on whether a fixed spatial anchor exists. Discriminability varies wildly across channels and value ranges—a pattern the authors formalized into a new Anchored Harmonic Weber model. They also found pairwise channel interactions are often strongly asymmetric, meaning the choice of one channel can unevenly affect another. Most striking is the dissociation between estimation accuracy and preattentive detection: length excels at accuracy but only moderately at pop-out, while area is the best for detection yet poorly supports quantitative estimation. The paper synthesizes these findings into a scenario-driven framework for context-sensitive channel selection, directly helping visualization designers decide which encoding to use based on the user's actual task rather than a one-size-fits-all ranking.