Graph Color

Link Between Data And Color

Match data types to color palettes/schemes

• Sequential ^1dc7c8
  • suitable for ordinal variables
  • easy to tell the level of a color
  • e.g. Heatmap
• Diverging
  • suitable for Likert Data
  • able to divide colors into groups
• Qualitative
  • suitable for Categorical Data
  • There should not be some colors that stand out more than other colors
    • Unless you are stressing certain values

Perceptually Uniform Color Spaces

We want to perceive the difference between data by observing colors.

Distinctions In Data

• The rate of color change should be consistent
  • This is also required by perceptually uniformness
  • Color scheme rainbow is not perceptually uniform
• Large range
  • the color range should be large enough to help distinguish differences
• Sharp break at important thresholds
  • 

With ggplot2

• Continuous data
  • scheme function: +scale_color_viridis_c() (c for continuous)
  • palette option: +scale_color_distiller(palette = "PuBu")
  • own sequential: +scale_color_gradient(low="white", hight="red")
  • own diverging: +scale_color_gradient2(low="blue", mid="white", hight="red")
• Discrete data
  • scheme function: +scale_color_viridis_d() (d for discrete)
  • palette option: +scale_color_brewer(palette = "PuBu")
  • own: +scale_color_manual(values=c("red", "yellow", "#FFFFFF")
• The color in the above scales can be fill for scaling the fill variable
• continuous vs discrete: c vs v, distiller vs brewer, gradient vs manual

With Other Packages

library(RColorBrewer)
colors <- brew.pal(4, "Reds") # get the color codes
barplot(1:4,  col = colors)

Color Vision Deficiency

To make color vision deficiency (CVD) friendly graphs,

• Use palettes that have already been tested
  • viridis, scale_color_colorblind() in ggthemes
• Use a color vision deficiency simulator
  • Color Oracle
• Use high contrast

General Tips

• Be consistent with colors
  • the color of the same object should be consistent among different graphs
  • manually set up the color if needed
• Legend order matches graph order
  • If your legend is on the right, it better matches the right ends of the graph
  • try label instead of legend
  • for some plots, like Bar Plot, legends at the bottom may be better
• Sequential scheme for Continuous Data; qualitative scheme for Categorical Data
• Do not use diverging scheme for non-likert data
• Colors should be a separate dimension
  • if a variable is already presented along the x-axis, there is no need to color the variable