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

TL;DR

The paper addresses how to design effective redundant color–shape palettes for multiclass scatterplots. It uses four crowdsourced studies to quantify when redundancy helps, how color and shape interact, and how to optimize color–shape pairings, culminating in CatPAW, a web tool that recommends empirical, category-aware palettes. The findings show redundancy yields meaningful accuracy gains most prominently for moderate category counts ($N \approx 5$–$8$), but effectiveness hinges on palette pairing and channel interactions; the authors provide a data-driven framework and a practical design tool to implement these insights. CatPAW’s data-backed approach offers a concrete path toward more reliable categorical visualizations and lays groundwork for future generalization to additional tasks and contexts.

Abstract

Colors and shapes are commonly used to encode categories in multi-class scatterplots. Designers often combine the two channels to create redundant encodings, aiming to enhance class distinctions. However, evidence for the effectiveness of redundancy remains conflicted, and guidelines for constructing effective combinations are limited. This paper presents four crowdsourced experiments evaluating redundant color-shape encodings and identifying high-performing configurations across different category numbers. Results show that redundancy significantly improves accuracy in assessing class-level correlations, with the strongest benefits for 5-8 categories. We also find pronounced interaction effects between colors and shapes, underscoring the need for careful pairing in designing redundant encodings. Drawing on these findings, we introduce a categorical palette design tool that enables designers to construct empirically grounded palettes for effective categorical visualization. Our work advances understanding of categorical perception in data visualization by systematically identifying effective redundant color-shape combinations and embedding these insights into a practical palette design tool.

Figures (14)

• Figure 1: The four color palettes used in Experiment 1 and 2. Each palette has 10 colors, drawn from (1) ColorBrewer/Paired harrower2003colorbrewer, (2) Tableau/Tab10 tableau, (3) Stata/S2 statagraphics19, and (4) Carto/Pastel carto
• Figure 2: Example stimuli from Experiment 1: scatterplots encoding categorical data using color-only, shape-only, and redundant (color-and-shape) encodings. All three visualizations use the same data distribution with 8 categories. Correlation comparison tasks were used in Experiment 1.
• Figure 3: Results from Experiment 1. (a) Line chart showing average accuracy across category sizes (2 to 10), separated by scatterplots using color-only, shape-only, and color-and–shape (redundant) encodings. (b) Bar chart with 95% CI showing the accuracy difference between redundant and non-redundant encodings across category numbers. The green background highlights the category range where redundant encoding provides the largest performance gains. For each category number, we conducted a Welch’s t-test; statistically significant differences are shown in bold green text, indicating conditions where redundant encoding significantly outperforms single-channel encodings.
• Figure 4: The six shape palettes used in Experiment 2. Each palette consists of six shapes and was generated using Shape It Up tseng2024shape.
• Figure 5: Example stimuli from Experiment 2: three scatterplots using different combinations of color and shape palettes, shown with the same data distribution and 6 categories. Correlation comparison tasks were used in Experiment 2.