evolgen

I’m working on a few graphs for a presentation. In a previous incarnation, I distinguished two partitions of my data using the colors red and green. This made sense intuitively (the red ones had something broken, and the green ones were a-ok), but I realized that people with red-green colorblindness would not be able to distinguish the different graphs. I switched the color scheme to white and gray, which should enable everyone to distinguish the two groups.

i-d84a62bfb1ffc7f1a3f9f28bc5e2d69d-blue_yellow_graph.gif

I also have a second way of partitioning the data, and I don’t want to use the same color scheme for both. I initially used blue and yellow for these two groups, but that may be problematic. I can’t use gray and white for this partition because then the two different partitions would look the same. Does anyone have any suggestions for another color scheme for this data set? I’d like to find two color schemes that are both acceptable for the colorblind and unique enough that the graphs showing different partitions won’t be confused.

Comments

#1 tbell
May 17, 2007

I have some kind of red/green color defective vision, and I find that color is less important than contrast for me. So, use highly contrasting schemes (dark, light). I couldn’t care less about the color then. There are many shades of gray…And at about 25% increments they are pretty easily distinguishable. You can have it both ways, color *and* contrast. Most people will pick up on the color distinction and the rest of us will get the contrast (or texture even) distinctions.
#2 red
May 17, 2007

Try vischeck: http://www.vischeck.com/ — I found it from Cindy Brewer’s colorbrewer site: http://www.personal.psu.edu/cab38/ColorBrewer/ColorBrewer_intro.html which was the first thing I thought of.

Edward Tufte likes lighter colors with minimal differences — Maybe various shades of gray would be good — it reproduces well and does not overload the graphic with saturated colors.
#3 Donaco Smyth
May 17, 2007

Instead of blocks of color, can you try using patterns?

Maybe diagonal stripes or dots or plaids or something?
#4 red
May 17, 2007

Maybe just use good labels. If you get tricky with the color, then people have to decode it. Tufte’s “The Visual Display of Quantitative Information” talks of boxplots on page 123-125. What would the color add to the communication that labels like “Able and not-Able” versus “Broken and not-Broken” would?
#5 Roy
May 17, 2007

In your example, color dominates. Is the color actually important? If I were watching your presentation on a black-and-white monitor, I could tell the one on the left from the one on the right. If the distinction is Treatment A versus Treatment B, an ‘A’ and ‘B’ would be more useful than color keying.

I have seen acres of PowerPoint presentations with saturated colors there for no good reason. I can still tell left from right without having neon colors clamoring for attention.

Even Excel graphs are full of ‘color’ even when run through black-and-white printers — with gray super-dominating the images yet signifying nothing.

Serious science publications (Science, Nature, etc) prefer not to be entertained by graphs, but rather informed.

Take your advice from Tufte.
#6 Roy, again
May 17, 2007

I thought of a simple example. I have two guest for you to meet. Which strategy should I use?

1. Give them nametags saying ‘Frick’ and ‘Frack’.

2. Give them colored squares, and then put up a color key on a poster, so you’ll know the one with the red square is Frick and the one with the yellow square is Frack?

This, to me, captures the silliness of using color just because the department has a color printer.

(I’m old enough to have done my undergraduate graphs in India ink.)
#7 Chris
May 17, 2007

Why not use a color and a pattern? Go ahead and use red and green, but also use diagonal stripes for one and a cross-hatch for the other.
#8 J Daley
May 17, 2007

Hi- I’m taking a genetics course this summer, and I’d like to know if you can reccomend any good supplemental texts that can help. I can’t find anything other than The Cartoon Guide to Genetics and a Schaum’s Outlines. Any suggestions?
#9 RPM
May 17, 2007

J Daley, if the course is good you shouldn’t need a supplemental text. sorry, but i can’t be of much help.
#10 Andrew Gelman
May 17, 2007

Do a scatterplot–I prefer this to a boxplot anyway. If you want to show medians and other quantiles, you can plot these as lines.
#11 Tex
May 17, 2007

What is wrong with black and white for your two colors? If you need only two colors, it is hard to beat these two classics. Show the two different types of partitioning on two separate graphs with labels to clearly indicate how they differ.

If you look at movies and photographs made before 1938, you will see that the entire world existed in black and white back then, and we got along just fine. I know your computer has 254 million other colors, but just ignore them.
#12 Keith Robison
May 17, 2007

I also suffer from deficient, but not non-existent, color vision.

One thing to keep in mind, and ideally test, is that many of the projecting monitors have much worse color contrast than your CRT or LCD monitor which you use directly. Red vs. blue bargraphs look fine to me on a monitor; I’m constantly baffled by them when projected. Blue vs. green disambiguation is sometimes tricky for me even on a monitor. On the other hand, I have found blue vs. yellow personally very useful for heat maps where I want to see gradations.

Junk Charts is a good blog on graph design.

Tufte should be required reading, especially for graphic artists (the ones at my last company routinely came up with badly junked-up, color careless charts). He isn’t perfect — there is one plot in the first book he praises that has a wicked moire effect for me — but he is very good. I also can recommend his course — he is just as engaging in person as he is in the books. William S. Cleveland is also very good.

On the other hand, if Excel gives it to you as a default, it is guaranteed to be awful. I definitely would stay away from most of the cross-hatching patterns.
#13 greg
May 18, 2007

There’s been a study in Japan. Refer to this site:
http://jfly.iam.u-tokyo.ac.jp/color/
#14 Daniel
May 18, 2007

I’m red-blind and totally agree to tbell’s answer. It’s not that important which colors you use but you have to watch out for a high contrast. Your yellow/blue chart looks good to me, but for somebody with a tritan defect (commonly known as blue-yellow color blindness) it might cause some problems.

You can look at your webpage through the eyes of a colorblind with this colorblind webpage filter.
#15 RPM
May 21, 2007

Thanks for the comments, everyone. I’ve been checking out those websites for pointers on designing figures.

Regarding patterns: I have tried that on another figure. The one issue I have, though, is that it interferes with reading the boxplot. Of course, I could abandon the box plot entirely, but there are two discrete categories. A scatter plot would still be possible, but I don’t like that aesthetic.

The one thing I have to keep in mind is that these data will be presented in two different ways (by splitting them up differently). The graph shown above is for one of those splits. For the other split, I went with white and gray. For this split, I need a different color/pattern scheme (that’s why black and white wouldn’t work). I am including a legend on the graph. But I’m trying to keep the colors/symbols consistent throughout the presentation, so I’m, once again, limited in my options.
#16 Krista
May 24, 2007

Perhaps you could use green and yellow. Both blue-yellow and red-green colourblind folks should be able to see a difference between green and yellow. But if not, there should be enough of a contrast that even if the slide was viewed in black and white the columns should appear different (using a dark shade of green and a bright yellow).