Redundant is Not Redundant: Automating Efficient Categorical Palettes Design Unifying Color & Shape Encodings with CatPAW

Paper Title

Redundant is Not Redundant: Automating Efficient Categorical Palettes Design Unifying Color & Shape Encodings with CatPAW

Publication Info

• Topic area: Data visualization and categorical palette design
• Keywords: Redundant encoding, categorical visualization, color-shape pairing, data-driven design, scatterplots, graphical perception, palette optimization, CatPAW, crowdsourced experiments, visualization tools

Background and Problem

• Problem / challenge: Existing guidelines for designing redundant encodings (color and shape combinations) are limited, and prior evidence on their effectiveness is conflicting. Current tools pair color and shape palettes without considering their interactions, leading to suboptimal designs.
• Significance: Effective redundant encodings can improve the accuracy of categorical data visualizations, especially in tasks like correlation estimation, and enhance accessibility for users with color vision deficiencies.
• Motivation and related work: Previous research has explored color and shape encodings individually but lacks comprehensive guidance on combining them. Studies show mixed results on the benefits of redundancy, with some highlighting perceptual interactions between channels. This paper aims to address these gaps by systematically studying redundant encodings and providing actionable design tools.

Solution

• Proposed approach: A data-driven framework for designing effective redundant categorical palettes, implemented in the web-based tool CatPAW (Categorical Palette Automation Wizard).
• Novelty:
  1. Empirical evaluation of redundant color–shape encodings across varying category numbers.
  2. Identification of interaction effects between color and shape in redundant encodings.
  3. Development of a statistical model to generate optimized palettes based on empirical data.
  4. Implementation of CatPAW, a tool for customizable and effective palette design.
• Procedure and key techniques:
  1. Conducted four crowdsourced experiments to evaluate redundant encodings and build pairwise accuracy matrices for colors and shapes.
  2. Developed a statistical model using experimental data to predict palette performance.
  3. Integrated the model into CatPAW, enabling users to generate and refine palettes based on category numbers and user-defined constraints.

Results

• Concrete findings:
  • Redundant encodings improved accuracy in correlation estimation tasks, with the highest benefits observed for 5–8 categories.
  • Redundant encodings achieved an average accuracy of 76.9%, outperforming color-only (72.9%) and shape-only (67%) encodings.
  • Interaction effects between color and shape were significant, with certain pairings (e.g., unfilled shapes with low-lightness colors) performing poorly.
  • Lightness and chroma magnitudes were critical for effective color differentiation.
• Advantage over baselines:
  • CatPAW-generated palettes outperformed designer palettes (77.5% vs. 68.4% predicted accuracy) and user-selected palettes (66.2%).
  • The tool effectively ranked redundant palettes, with a strong correlation (r = 0.97) between predicted and empirical performance.
• Experiments / evaluation:
  • Four experiments with 455 participants evaluated redundant encodings, color-shape pairings, and pairwise accuracy matrices for 39 colors and shapes.
  • Tasks included correlation estimation in multiclass scatterplots with varying category numbers (2–10).
  • Statistical analyses (ANOVA, Welch’s t-tests) confirmed significant effects of redundancy, category numbers, and color-shape interactions.
• Limitations and future work:
  • Limited to specific colors, shapes, and white backgrounds.
  • Focused on correlation tasks in scatterplots; generalizability to other tasks and visualization types is untested.
  • Future work should explore broader palettes, additional tasks, and user preferences, as well as refine CatPAW’s model and extend its capabilities.

Summary

This paper investigates the effectiveness of redundant encodings (color and shape combinations) in categorical data visualization, demonstrating their benefits for correlation estimation tasks, particularly for 5–8 categories. It highlights the importance of carefully pairing colors and shapes to maximize performance and introduces CatPAW, a web-based tool for generating optimized categorical palettes based on empirical data. The study provides actionable guidelines for redundant encoding design and lays the groundwork for future research on categorical visualization and automated palette generation.