Color Scales for Data Visualization: Sequential, Diverging, Categorical, and When Each Applies

Color in data visualization is an encoding channel, not a decorative layer. When you assign color to a data dimension, you are making a claim about how readers should interpret the visual differences between color values. The three primary scale architectures — sequential, diverging, and categorical — make different claims, and using the wrong one for a given data type systematically misleads readers. Sequential color scales encode a single ordered dimension: more of something is represented by more of a visual property. In a well-designed sequential scale, darker or more saturated means more. The most legible sequential scales move from a near-neutral light value to a single saturated hue at the high end. Single-hue sequential scales are interpretable by color-blind readers and work well in print. Multi-hue sequential scales (yellow to green to blue) can encode a wider value range with more perceptual steps, but require careful construction to maintain perceptual ordering. The OKLCH color space is better than sRGB for constructing sequential scales with consistent perceptual lightness steps. Diverging scales encode data that has a meaningful midpoint: a zero value, a neutral value, or a target value. A temperature anomaly map should be diverging. A sentiment score from -10 to +10 should be diverging. A range from 0 to 100 with no meaningful midpoint should not be diverging — using a diverging scale on data without a meaningful midpoint introduces false emphasis on the middle of the range. The canonical diverging scale architecture uses two distinct hues at the extremes with a near-white or neutral midpoint. Categorical scales encode discrete classes with no ordinal relationship. The requirement is maximum distinctiveness between classes rather than ordered progression. Categorical scales fail when they use similar hues for different categories, implying that the categories are ordered along some dimension they are not. The maximum number of distinguishable categorical colors for typical data visualization is eight to twelve; beyond that, patterns, shapes, or text labels become necessary. Functional accessibility in data visualization requires testing with simulated color vision deficiency, not just WCAG contrast ratios. Deuteranopia (the most common form of red-green deficiency) collapses green and red into the same perceived range, making standard traffic-light status systems unreadable. The fix is to add luminance contrast between the green and red levels so they are distinguishable by value even when hue is lost, or to use blue-orange as the primary two-color system instead of red-green.