Veri Görselleştirmede Renk: Grafikler ve Haritalar

Color in data visualization is not decoration. It is encoding — a channel that carries information just as position, length, and size do. When color is used well, viewers grasp patterns instantly. When it is used carelessly, charts mislead, confuse, or exclude entire segments of the audience.

This guide covers the three canonical color scale types, how color blindness reshapes dataviz decisions, and the most common mistakes that undermine otherwise solid work.

Why Color Is a Powerful — and Dangerous — Encoding

Among the pre-attentive visual attributes studied in perception research, color stands out for raw speed. Viewers detect a distinctly colored point in a scatter plot of thousands within milliseconds, without conscious search. This power is real, but it comes with constraints.

Color encodes information effectively only when:

1. The viewer can distinguish the encoded values
2. The encoding is intuitive (not arbitrary)
3. The color choices survive different viewing conditions (print, projector, colorblind vision)

Violate any of these, and color stops being a helpful signal and starts being noise. The scale types below are structured responses to each of these constraints.

Sequential Color Scales: Light to Dark

Sequential scales represent data that moves in one direction — typically from low values to high values, or from sparse to dense. The defining characteristic is a single hue that varies in lightness and sometimes saturation.

How Sequential Scales Work

A sequential scale anchors the lightest tint to the lowest data value and the darkest shade to the highest. Every intermediate value receives a proportionally placed color along that continuum.

A practical single-hue sequential scale for geographic population density:

This encoding is intuitive: more blue means more. Viewers with any color perception can read the scale because it operates on luminance, not hue differentiation.

Practical Guidance

• Avoid rainbow (spectral) gradients in sequential contexts. Spectral gradients — red to orange to yellow to green to blue — create false edges at hue boundaries. A sudden shift from yellow to green around the midpoint signals a categorical break that does not exist in the data. Perceptually uniform sequential palettes (like those derived from OKLCH or CIELAB) avoid this artifact.
• Use a perceptually uniform progression. Uniform means equal steps in data produce equal-looking steps in color. Non-uniform progressions over-emphasize certain value ranges.
• Saturated endpoint, desaturated origin. A lightly saturated near-white origin and a fully saturated dark endpoint gives the widest perceptual range.

You can build single-hue sequential scales by starting from a base hue in the palette generator and combining it with the shade generator to ensure smooth steps.

Diverging Color Scales: Two Hue Endpoints

Diverging scales represent data that has a meaningful midpoint. When values above and below that midpoint are conceptually different — not just quantitatively different — a diverging scale makes both halves visible simultaneously.

When to Use a Diverging Scale

Classic use cases include:

• Election results: red for Republican advantage, blue for Democratic advantage, neutral purple or white for a 50/50 split
• Financial performance: red for losses, green for gains, neutral for breakeven
• Temperature anomalies: blue for below-average, red for above-average, white for the historical norm
• Sentiment scores: negative sentiment to positive sentiment, neutral at center

Designing a Diverging Scale

A diverging scale requires three anchors: the two hue endpoints and the neutral center.

Example for a survey satisfaction scale (very dissatisfied → neutral → very satisfied):

The neutral center should be nearly achromatic — a light gray or off-white — so that values on either side of the midpoint read as distinct categories.

Pitfalls

• Unequal perceptual range on each side. Red and green do not have equal visual intensity at equivalent saturation levels. Use a tool to check that your two extreme endpoints are perceptually equidistant from neutral, not just numerically equidistant.
• Choosing hue pairs that collapse for color-blind viewers. Red-green diverging scales are the most common culprit. This is covered in detail in the color blindness section below.

Categorical (Qualitative) Color Scales: Distinct Hues

Categorical scales encode nominal data — categories that have no inherent order. Country A is not more or less than Country B; it is simply different. This calls for colors that are as perceptually distinct from each other as possible, with no implied order or hierarchy.

Principles of Good Categorical Palettes

Maximum perceptual distance. Colors should be spread around the hue wheel, not clustered. A palette that uses three blues, two greens, and one purple will not work because viewers will confuse the similar hues.

Roughly equal lightness. If some categories are light and others are dark, the dark ones will dominate visually even if they are not more important. Aim for similar luminance across all categorical colors.

No more than 7–10 categories, ideally fewer. Human short-term memory holds roughly 7 chunks. Asking viewers to track more than 7 distinct color-to-category associations imposes cognitive strain that undermines the chart. If you have 15 countries, reconsider whether color is the right encoding.

A balanced 6-color categorical palette:

Blue, red, green, amber, purple, and orange are spread widely around the hue wheel and have similar lightness values.

Testing Categorical Palettes

Before finalizing, verify that your palette works when:

• Printed in black-and-white (check luminance similarity does not eliminate distinctions)
• Viewed at small size in a legend
• Viewed by someone with color blindness (see below)

The color blindness simulator handles the third test.

Color Blindness in Data Visualization

Approximately 8% of men and 0.5% of women have some form of inherited color vision deficiency. In a typical workplace audience of 50 people, 3–4 viewers likely experience color differently. In a public-facing visualization, the number is not negligible.

Types Most Relevant to Dataviz

Deuteranopia / Deuteranomaly: Reduced sensitivity to green. The most common form (~5% of men). Red and green appear similar in hue; the distinction collapses to a yellow-brown-gray range.

Protanopia / Protanomaly: Reduced sensitivity to red. Less common (~1% of men). Red appears darker and less saturated; red-green confusion occurs.

Tritanopia: Rare (~0.001%). Blue-yellow confusion. Affects very few but becomes relevant when blue is a primary categorical color.

The Red-Green Disaster

The most common dataviz color blindness failure is the red-green diverging scale — used constantly for financial data, maps, and status indicators. For deuteranopic and protanopic viewers, the two hue endpoints collapse to nearly identical brownish or olive tones. The chart conveys almost no information to roughly 6% of male viewers.

Alternatives:

Replace red-green with:

• Blue-orange (#1D4ED8 to #EA580C): Widely distinguishable across all common color vision types
• Purple-green (#7C3AED to #059669): Also robust across deuteranopia and protanopia
• Red-blue with strong lightness difference: Works for most forms if the lightness contrast is strong enough

Traffic light (red-yellow-green) status indicators have the same problem. Replace with shapes and labels as the primary encoding, and use color as a secondary redundant cue.

Design Strategies for Color-Blind Robustness

1. Use luminance contrast, not just hue contrast. Colors that differ significantly in lightness remain distinguishable under most forms of color blindness because luminance perception is largely preserved.

2. Add redundant encodings. Don't rely on color alone. Use:

3. Different line styles (solid, dashed, dotted) in line charts
4. Different point shapes in scatter plots
5. Direct labels on data series

6. Hatching or patterns in bar charts

7. Test with a simulator. Use the color blindness simulator to view your palette through deuteranopic, protanopic, and tritanopic filters before publishing.

8. Use established accessible palettes. Wong (2011) published an 8-color palette specifically designed for color-blind robustness: black, orange, sky blue, bluish green, yellow, blue, vermillion, and reddish purple.

Best Practices and Common Mistakes

Best Practices

Match scale type to data type. Sequential for ordered continuous data. Diverging for data with a meaningful center. Categorical for unordered nominal data. Mixing these (using a sequential scale for categorical data) creates implied ordering that misleads.

Put the darkest colors on the most important data. Darker values command more attention. If your highest values are the most actionable, anchoring them to the dark end of a sequential scale aligns visual attention with data priority.

Consider the background. Colors appear different on white, gray, or dark backgrounds. A scale designed on a white background may look flat or muddy on a gray dashboard. Test on the actual background.

Use a consistent scale across related charts. If three charts in a dashboard all show sales figures, use the same color scale for the same metric. Inconsistency forces viewers to re-read every legend.

Keep legends close to the data. The further the legend, the more cognitive work viewers must do to decode the colors. Direct labels eliminate legend lookup entirely.

Common Mistakes

Rainbow scales for continuous data. As noted above, spectral gradients introduce false perceptual edges and distort the data. A perceptually uniform single-hue scale is almost always superior.

Using color as the only encoding. If your chart would fail entirely when converted to grayscale, it is over-relying on color. Add shape, position, or direct labels.

Too many categorical colors. More than 7–8 distinct colors in a legend is a signal that the data needs to be restructured, not that you need a larger palette.

Low-contrast colors on map fills. Geographic maps display many colors simultaneously at small polygon sizes. Colors that look distinct in isolation may be very hard to tell apart at map scale. Test at actual display size.

Ignoring the projector effect. Projectors wash out color significantly. A palette that reads beautifully on screen may lose all mid-range distinctions in a conference room presentation. Add contrast buffer.

Encoding with color alone in print. Print may reproduce colors differently than screen, and accessibility for printed materials is equally important. Design with grayscale print in mind.

Key Takeaways

• Use sequential scales (light-to-dark, single hue) for ordered continuous data where one extreme means "more" of something.
• Use diverging scales (two hue endpoints, neutral center) for data with a meaningful midpoint — financial, political, deviation from baseline.
• Use categorical scales (perceptually distinct hues) for nominal data with no implied order; limit to 7–10 categories maximum.
• Avoid rainbow gradients for sequential data — they introduce false edges and mislead perception.
• Red-green encoding fails for approximately 6–8% of men; replace with blue-orange or purple-green diverging pairs.
• Test every palette with the color blindness simulator before publishing.
• Add redundant encodings (shapes, labels, line styles) so the visualization remains readable when color fails.
• Use the palette generator to build and evaluate color palettes for your specific data type.