Key Takeaways — Chapter 3: Color

1. Color Is Three Variables, Not One

Color is not a single visual channel. It is three distinct perceptual dimensions: hue (the wavelength — red, blue, green), luminance (the lightness — dark to light), and saturation (the vividness — vivid to muted). Each dimension is suited to a different data type. Hue encodes categories. Luminance encodes ordered magnitude. Saturation provides emphasis but is unreliable for precise encoding. Treating these three dimensions as interchangeable is the root cause of most color mistakes in visualization.

2. Three Palette Types for Three Data Types

Sequential palettes (light-to-dark in a single hue) encode ordered data going one direction — revenue, population, concentration. Diverging palettes (two hue anchors with a neutral midpoint) encode data that deviates from a meaningful center — profit/loss, temperature anomalies, sentiment. Categorical palettes (distinct hues at similar luminance) encode nominal categories — product lines, countries, experimental groups. Choosing the wrong palette type for your data is a fundamental encoding mismatch, equivalent to using a pie chart for time-series data.

3. The Rainbow Colormap Is Broken

The rainbow (jet) colormap fails on every criterion that matters: it is perceptually non-uniform (creating false boundaries in smooth data), it has no monotonic luminance gradient (making it unreadable in grayscale), and it is hostile to colorblind viewers (collapsing to ambiguous brown-yellow bands for the most common forms of color vision deficiency). There is no legitimate use case for rainbow on continuous data. Use viridis, inferno, cividis, or any perceptually uniform alternative.

4. Luminance First, Always

The luminance-first principle is the single most important rule for accessible color design. If your chart's essential information can be read from the luminance channel alone — if the chart still makes sense in grayscale — then it will survive colorblindness, black-and-white printing, and low-quality displays. Design luminance first, then layer on hue for additional discrimination. Never rely on hue alone to carry critical information.

5. Design for Colorblindness as a Baseline Requirement

Approximately 8% of men have some form of color vision deficiency, overwhelmingly red-green confusion (deuteranopia and protanopia). This is not a rare edge case — it is a predictable portion of any audience. Avoid red-green as a critical distinguishing pair. Use redundant encoding (shape, pattern, or labels in addition to color). Test every chart with a colorblind simulation tool before publishing. The tools exist and are free; failing to use them is a professional lapse.

6. Limit Categorical Colors to Seven or Eight

Viewers can reliably distinguish and remember about seven to eight distinct hues. Beyond that threshold, the legend becomes a lookup table, pre-attentive color discrimination breaks down, and the viewer must constantly shift gaze between chart and legend. If your data has more than eight categories, the solution is not a bigger palette — it is a different design: grouping, faceting (small multiples), interactive filtering, or the gray-plus-highlight strategy.

7. Every Color Must Mean Something

Color without meaning is noise. If a bar chart uses twelve different colors for twelve months but all bars represent the same variable, the colors add no information and imply false categorical distinctions. Before adding color to any chart element, ask: "What data does this color encode?" If the answer is nothing, use a single neutral color. Color is information, not decoration.

8. Semantic Color Is Powerful but Not Universal

Leveraging shared associations (red for danger, blue for cold, green for growth) makes charts intuitive for the right audience. But semantic associations are culturally specific: red means good fortune in many East Asian contexts, and the red-green "traffic light" convention fails for colorblind viewers. Use semantic color deliberately, document your color-meaning mappings across a report or dashboard, and provide redundant cues (labels, icons, shapes) so that no viewer depends solely on a cultural association to interpret your chart.