Color is one of the most popular channels used to represent data. However, color is difficult to use well, and its misuse can lead to ineffective or even misleading visualizations. Visualization research offers a suite of heuristic rules and guidelines for using color use. However, most of these guidelines are based on intuition and results from psychology. We lack a holistic empirical understanding of color use grounded directly in visualization use. This project sought to measure color perception and use in the context of visualization to build guidelines, models, and techniques for effectively using color in visualization. The project achieved the following objectives: 

1. Model how people perceive colors mapped to data in visualizations

We modeled how the differences we see in data visualized using color change based on the datapoint?s shape and size, measured how semantic color attributes affect our confidence in and perceptions of that data, quantified how cognitive attributes of color affect people?s abilities to reason about data, and modeled how colors can highlight important data.

2. Capture and assess designer practices in color use in visualization

We collected 222 hand-crafted designer colormaps reflecting known best practices. We developed an algorithm to identify key patterns in the perceptual and mathematical structures of these colormaps and assessed how they apply common design rules.

3. Develop tools for more effective color use in visualization

We created an algorithm for generating color encodings for visualizations using single guiding colors and for selecting colors for interactions. We integrated these approaches into web-based design tools that allow rapid encoding generation, design, and editing. 

Intellectual Merit: Our results challenge fundamental conventions for visualizing data using color and provide experimentally-grounded models, metrics, algorithms, and tools for using color more effectively.  We show how the ways we reason about color in visualizations are unique to the visualization?s design and goals in ways that long-used models from psychology may miss. Our results offer new methods and models quantifying the differences people see in a visualization. These models help visualizations more honestly and accurately communicate differences and patterns in data. 

We also provide evidence that color may affect not only perception (what we see) but elements of decision making and higher-level cognition (how we reason about what we see). For example, coloring missing values red leads to higher confidence in the data than making them transparent. Rainbow colormaps, while long discouraged due to known perceptual flaws, may help people infer whether differences between datasets matter. Visualization designers can use these observations and measures to consider how the ways that they represent data best align with the nature of the data, including its patterns, quality, and context, to more effectively and responsibly visualize data. 

We also reduce fundamental barriers to creating effective color encodings. We developed tools allowing any user to create effective and aesthetically-pleasing color encodings comparable to designer encodings using only a single guiding color. Our tools allow users to edit those encodings using simple actions like rotating and scaling. Our results show how coupling perception, design, and automation can empower developers to generate more expressive and creative visualizations with minimal labor. Our tools empower users to craft the overall appearance, usability, and effect of their visualizations without relying on pre-generated encoding choices, enhancing rather than reducing agency through automation. 

Broader Impacts: People are using data more than ever before, from making predictions in sports to understanding a pandemic. Our results collectively expand our ability to communicate data more responsibly and effectively. For example, modeling how color may affect decision-making, such as potential biases and perceived quality and confidence, data journalists can design visualizations that instill appropriate confidence in the data to draw appropriate conclusions. Our results offer new insight into the ways we perceive and reason about visual phenomena. Our methods inspired a systematic characterization of experimental techniques from vision science that provides primer for rigorous experimental design. These efforts will improve experimental research in visualization, leading to better insight into how people perceive and reason with visualizations.  

The tools we?ve developed also empower people to create novel representations of their own data, giving them agency over communication while making it significantly easier to follow a nuanced and often complex set of heuristics and conventions for making effective color choices. By requiring only a single color and allowing people to edit using well-defined actions, we encourage effective visualization practices while giving people agency over the ways they represent their data. These benefits also make visualization design more accessible: by modeling designer practices using perceptual models, we enable people with color vision deficiencies to more effectively develop novel color encodings. 

The grant supported six graduate students in interdisciplinary research. Three women involved in the project pursued advanced academic positions related to the project. The PI also led a computer graphics workshop for high school women.

Last Modified: 08/07/2020
Modified by: Danielle Szafir