Studying the Separability of Visual Channel Pairs in Symbol Maps

Paper Title

Studying the Separability of Visual Channel Pairs in Symbol Maps

Publication Info

• Topic area: Evaluation of visual channel separability in bivariate symbol maps for multivariate data visualization.
• Keywords: Visual separability, bivariate symbol maps, visual channels, color, shape, size, orientation, graphical perception, multivariate visualization.

Background and Problem

• Problem / challenge: Existing guidelines on the separability of visual channels are largely based on theoretical assumptions and expert heuristics, with limited systematic empirical testing, particularly in map-based contexts.
• Significance: Understanding separability is critical for designing effective multivariate visualizations, as interference between visual channels can compromise the ability to isolate and interpret individual data dimensions.
• Motivation and related work: Prior studies have explored separability in scatterplots and other visualizations but have not systematically tested separability in spatially embedded symbol maps. This study addresses this gap by empirically evaluating four visual channel pairs in bivariate symbol maps.

Solution

• Proposed approach: A preregistered, within-subjects experiment to systematically test the separability of four visual channel pairs—Color × Shape, Size × Color, Size × Shape, and Size × Orientation—in bivariate symbol maps.
• Novelty:
  1. Systematic empirical evaluation of separability in a map-based context.
  2. Analysis of asymmetries in channel assignment (task-relevant vs. task-irrelevant).
  3. Examination of how specific values of the interfering channel affect performance.
  4. Design recommendations for bivariate symbol maps based on empirical findings.
• Procedure and key techniques:
  1. Participants viewed maps of 11 western U.S. states with bivariate symbols encoding two data dimensions (Dimension A: task-relevant, Dimension B: task-irrelevant).
  2. Four visual channel pairs were tested under two flipped conditions (which channel encoded Dimension A).
  3. Participants performed a forced-choice identification task, selecting the state with the highest value of Dimension A.
  4. Accuracy and response times (RTs) were recorded and analyzed using repeated-measures ANOVA and post-hoc comparisons.

Results

• Concrete findings:
  • Color × Shape yielded the highest accuracy (0.882) and fast response times, while Size × Orientation had the lowest accuracy (0.823) and slowest response times.
  • Size × Color and Size × Shape formed a mid-tier cluster, with Size × Color being faster but Size × Shape slightly more accurate.
  • Asymmetries were observed in channel assignment: Shape and Color outperformed Size when encoding the task-relevant variable.
  • Specific values of the interfering channel influenced performance, with high values (e.g., large size, dark color, triangle) generally aiding recognition.
• Advantage over baselines:
  • Empirical evidence confirmed that Color × Shape is the most separable pair, aligning with theoretical expectations, while Size × Orientation showed the weakest separability.
  • Asymmetries and value-level effects highlighted nuances not captured by traditional separability rankings.
• Experiments / evaluation:
  • 200 participants (197 after exclusions) completed 24 trials each, testing four channel pairs under two flips with three target variations.
  • Metrics included accuracy and log-transformed response times for correct trials.
• Limitations and future work:
  • Simplified design (e.g., fixed map layout, three discrete levels per channel) limits generalizability to more complex real-world maps.
  • Did not include single-channel baselines to isolate interference effects.
  • Future work should explore additional channel pairs, redundant encodings, and higher-level tasks like trend detection or correlation assessment.

Summary

This study empirically evaluated the separability of four visual channel pairs in bivariate symbol maps, finding that Color × Shape offered the best performance, while Size × Orientation was the weakest. Asymmetries in channel assignment and the influence of specific interfering channel values were also observed. These findings provide actionable design recommendations for bivariate symbol maps and challenge the assumption that separability rankings derived from isolated channels directly transfer to multivariate contexts. Future work should expand to more complex map designs, additional channel pairs, and higher-level analytical tasks.