Revisiting Categorical Color Perception in Scatterplots: Sequential, Diverging, and Categorical Palettes

Overview

• Current guidelines for selecting categorical colors are based on intuition rather than empirical studies
• Recent research suggests factors beyond just hue differences, like lightness, may be important for effective categorical color palettes
• This study explored how different color palette families (single-hue sequential, multi-hue sequential, perceptually-uniform multi-hue sequential, diverging, and multi-hue categorical) impact people's ability to estimate relative mean positions in multi-class scatterplots

Plain English Explanation

Choosing the right colors to represent different categories in data visualizations is an important design challenge. Existing guidelines for selecting categorical color palettes are often based on rules-of-thumb or intuition, rather than rigorous studies of how people actually perceive and interpret the colors.

While conventional wisdom suggests that maximizing the difference in hue (the technical term for color) between categories is the key, more recent research has found that other factors, like the relative lightness of the colors, may also play an important role in making categorical color palettes effective.

To explore this further, the researchers conducted an experiment where they asked people to estimate the relative mean positions of data points in multi-class scatterplots, using five different families of color palettes. These palette families varied in how they manipulated the hue and lightness of the colors.

The results confirmed that hue-based categorical color palettes tend to be the most effective. However, the findings also suggest that creating scalable, multi-category color encodings requires considering more than just differences in hue. Other visual factors, like lightness, appear to be important as well.

Technical Explanation

The researchers conducted a crowdsourced experiment to assess people's ability to estimate relative mean positions in multi-class scatterplots using different color palette families. They tested five palette families:

1. Single-hue sequential
2. Multi-hue sequential
3. Perceptually-uniform multi-hue sequential
4. Diverging
5. Multi-hue categorical

These families vary in how they manipulate the hue and lightness of the colors to represent different categories.

Participants were shown scatterplots with 2 to 10 categories and asked to estimate the relative mean positions of the data points. They evaluated 20 different colormaps within these palette families.

The results confirmed the conventional wisdom that hue-based categorical palettes, like the multi-hue categorical family, are the most effective for this task. However, the data also provided additional evidence that other factors beyond just hue variance, such as lightness, play an important role in creating scalable, multi-category color encodings.

This suggests that effective dimension reduction and class feature visualization may require considering a broader set of perceptual factors when designing categorical color palettes.

Critical Analysis

The study provides valuable empirical insights to inform the design of categorical color palettes. However, as the authors note, there are several limitations to consider:

• The experiment used synthetic scatterplot stimuli, which may not fully capture the complexities of real-world data visualizations.
• The study focused on evaluating relative mean position estimation, but other tasks (e.g., identifying individual data points) may have different perceptual requirements.
• The palette families tested represent a limited set of the many possible ways to manipulate hue, lightness, and other visual variables.

Additionally, the study did not explore how individual differences in visual perception or color blindness may impact the effectiveness of different categorical color encodings. Accounting for such factors could be an important area for future research.

Conclusion

This study provides empirical evidence that the effectiveness of categorical color palettes depends on more than just maximizing hue differences. Factors like lightness also play a role in how people perceive and interpret multi-class data visualizations.

The findings suggest that designing scalable, multi-category color encodings requires a more holistic consideration of perceptual factors, beyond the traditional heuristic-based guidelines. Continued research in this area can help data visualization practitioners create more effective and inclusive representations of complex, multi-dimensional data.