Color Scale Bias: How Heatmap Colors Distort Data
Heatmap color scales change how readers see your numbers. Learn why rainbow scales mislead, what perceptually ordered palettes fix, and how to test yours.

Color scale bias is what happens when the colors of a heatmap or map make equal differences in the data look unequal. The numbers are fine. The scale that turns them into colors is not, and readers trust what they see. This guide covers why common scales mislead, what the research shows, and seven rules for a scale that tells the truth.
What is color scale bias?
Color scale bias is a distortion that comes from the palette, not the data. A scale that changes hue sharply, or whose lightness does not rise steadily with the value, makes some differences look bigger than others and invents borders where none exist. A scale ordered from light to dark avoids both problems.
Heatmaps are used in epidemiology, finance, climate science and medical imaging because they pack a lot of values into a small space. That is also why a poor scale does damage: it changes how every value in the grid is read.
Why rainbow scales mislead
People read lightness more reliably than hue. A good sequential scale gets steadily lighter or darker as values rise, so order is obvious without a legend. A rainbow scale (blue, green, yellow, red) changes hue at uneven points and its lightness goes up and down, so three things happen:
- A step between two values can look larger or smaller than the same step elsewhere on the scale.
- Sharp hue changes, such as green to yellow, look like category borders even when the data is smooth.
- The extreme colors pull the eye to the highest and lowest cells whether or not they matter.
Crameri, Shephard and Heron make this case in The misuse of colour in science communication (Nature Communications, 2020): the rainbow scale fails to give a smooth, meaningful gradient, and its extreme values dominate the picture. They also note that many common software packages once used it as the default.
There is experimental evidence that it costs accuracy. In Evaluation of Artery Visualizations for Heart Disease Diagnosis (Borkin et al., IEEE TVCG, 2011), viewers made fewer diagnostic mistakes with a perceptually appropriate diverging scale than with a rainbow scale, and the rainbow scale was much slower to use in 3D. The data was the same. Only the colors changed.
What a hue-driven scale looks like
The heatmap below is a PlotSet example of the coronavirus death toll by region at the end of 2020. Its scale runs blue, then yellow-green, then red. Look at how the colors group the columns into bands: Africa is blue, Asia is yellow-green, Europe and North America are red. The borders between bands come from the hue changes in the scale, not from gaps in the numbers.
A sequential scale in one hue family, light for low and dark for high, would show the same order with no false borders. Putting the actual numbers in the cells, as this chart does, also helps, because it lets readers check their impression against the data.
Counts, rates and the map that shows where people live
Scale bias gets worse when the thing being colored is wrong. A map of raw counts mostly shows where the population is, so large cities look intense and small places look quiet whatever their risk. Divide by population, area or another base before you color, and say which one in the legend. A choropleth map of rates, with a light-to-dark scale and labeled breaks, answers the question readers are actually asking.
Defaults, color spaces and the choice nobody makes
Most people never change the palette a tool gives them, so the default becomes the standard. Two further settings matter:
- Interpolation: blending colors in RGB does not give equal visible steps. Scales built in a perceptual space such as CIELAB are designed so that equal steps in the data look like equal steps in color.
- Domain and breaks: clipping the range, binning by quantile or centering a diverging scale in the wrong place all change which cells stand out. Use a diverging scale only when there is a meaningful midpoint, such as zero or a target.
Color vision deficiency
The National Eye Institute says about 1 in 12 men have color vision deficiency. A red and green diverging scale is a poor choice for them, because the two ends can look alike. Blue and orange, or a single hue running light to dark, stay readable. Our guide to designing blind-friendly colors goes deeper, and making data visualizations accessible covers the rest of the checklist.
Seven rules for an honest heatmap
| Problem | Fix |
|---|---|
| A rainbow scale for continuous data | Use a sequential scale that gets steadily lighter or darker |
| A diverging scale with no real midpoint | Use it only around zero, a target or an average |
| Lightness that goes up and down | Check that the scale gets lighter or darker in one direction |
| Raw counts that follow population | Normalize per person, per area or per base unit |
| Hidden scale boundaries | Label the breaks and show the range in the legend |
| Red and green as the two ends | Test in a color vision deficiency simulator, or switch hues |
| Red used to signal loss by habit | Use red when it means something, and say what |
Build one in PlotSet
PlotSet's heatmap charts let you pick the palette yourself, and the table template's heatmap columns come with single-hue palettes such as blues, teal and grays, plus a diverging option for data with a midpoint. For a first build, follow the heatmap how-to, browse the heatmap templates, or read how to choose chart colors that do the legend's job.
FAQ
Why is the rainbow color scale bad for heatmaps?
Its hue changes at uneven points and its lightness is not ordered, so equal steps in the data do not look equal and false borders appear. Research on medical visualizations found more diagnostic mistakes with a rainbow scale than with a perceptually appropriate one.
What is the best color scale for a heatmap?
For values that run from low to high, a sequential scale in one hue or a perceptually ordered multi-hue scale that gets steadily lighter or darker. For data with a midpoint, a diverging scale centered on it.
Can the colors of a chart really change a decision?
Yes. Color changes which cells draw attention and how large the differences look, so readers can reach a different conclusion from the same numbers. That is why the scale deserves the same care as the data.
How do I make a heatmap that works for color blind readers?
Avoid red and green as the two ends, prefer a single hue running light to dark or a blue and orange diverging pair, and put the values in the cells or the tooltip so color is not the only carrier of information.


