Legends and Colors
January 21, 2004 | Sean Gerety
11 Comment(s)
I am working on a report for a company that has a large number of products. The client has stated that in one report the data will appear as both a scatter plot and a table. There could be upward of 40 items in the legend for the scatter plot. So I’ve hit the following dilemmas.
1. The number of discernible hues available. (Not to mention that this will appear in both soft and hard copy format.)
2. The report will be created by a software package and “tweaking” will not be an option.
Thanks,
Sean
Don’t envy you this! If the report is to appear in both hard and electronic
versions, one point (which you may already know) is to print out the range of
hues selected by the software to check that the differences on screen (RGB)
really show up differently on paper (CMYK). Blues and greens are a
particular problem. I have the Excel/Word/PowerPoint palette printed out as
an A4 flag on my desk.
Is it possible to allocate the most distinctive hues to the most interesting data points?
Or have the points in different shapes or with different sizes or different shaded textures?
Or print short names next to the data points? Perhaps this would be too squashed if everything had a name by it but suppose you had a column for this short description and left it blank except for the most interesting data points?
You have my complete sympathy. This is the type of task from which
nightmares are made…I’m afraid these links will not give you any solutions but
they should make your dilemna more easily understood….
http://www.easyrgb.com/matching.php
http://www.personal.psu.edu/faculty/c/a/cab38/ColorBrewerBeta.html
Good luck.
Thanks everyone, the “ColorBrewer” site was very interesting and possibly I could use some of the schemes. And in order to get the high number of values that I need the best option that I have come up with is use a base set of colors and then to add a thin border to the colors to double my legend. The thin black border should add enough visual contrast to trigger a mental difference.
One of the handy items on the “ColorBrewer” site is the information about whether or not a certain color scheme is laptop friendly, color printer friendly, etc.
And another note I’d like to add, when we were trying to create a 40 item color scheme we would print it and then cut out all of the color squares and then try to match the color with legend. This exercise alone will make you appreciate Albers “Interaction of Color”. If you don’t have to book, please buy it.
Sean
The Rochester Institute of Technology has a nice
mini-course on
color from Rob Roy Kelly (who recently died). Kelly draws heavily from Albers,
under whom he studied at Yale.
Glad you found the Color Brewer of help…
Would that I could afford the original Albers with all the plates….
Also of interest…
“The Elements of Color” by Johannes Itten ISBN 0-471-2829-9
“Color and Culture” by John Gage ISBN 0-520-22225-3
..and just for fun
“Colors the Story of Dyes and Pigments” Francois Delamare and Bernard
Guineau
and a terrific light read…
“Colour, Travels Through the Paintbox” by Victoria Finlay
It was a sad day when John Gleason left us. His color lecture for lighting
design was not only elegant but razor sharp in its simplicity.
Use overlays. Allow the reader to add/remove overlays. Label the overlays. Do this on both the soft and hard versions. If the client does not wish to spring for this or another more usable solution, then they deserve a scatterplot with 40 colors.
Hello. ET, I love your books. They are superb.
I have a question that might be more appropriate for separate topic: is there an algorithm to generate a color scheme of a certain size where the colors are “maximially different”. (relative to what the human eye can readily and rapidly discern)
That is, I have a stock chart and it is unknown in advance how many symbols will be plotted. It could be upwards of 20-30. The problem is that many stocks move in unison, so when plotting the lines for each stock, they often overlap. When looking at the chart, it is often very difficult to tell the lines apart, even if they are not physically close, or to associate color with a stock: it would be nice if someone saw “orange” and immediately associated it with, say “MSFT”. Which is fine, except that with 20 colors they often end up looking similar and immediately telling the difference between “orange” and “orange red” and “salmon orange” is basically impossible.
Any suggestions? I could generate color schemes by hand but I am wondering if any research has been done on an algorithmic way of doing it. Any sources or individuals who publish in this area?
Thanks.
Samples (with a poor color scheme, and note that the second chart is much harder to tell what is what):
P.S. I changed the color scheme and things are much easier to see:
I imagine I am barking up the wrong tree and should just manually design a color scheme (nevermind all the other changes I could make to this chart to make things more visually informative).
Color Brewer (mentioned earlier on this thread and others, search on “brewer”):
http://www.colorbrewer.org/
Sequential, divergent, and qualitative color schemes, maxing out at 12 classes. Also try
directly labeling each line. It’s harder to associate labels with colors with lines than just
labels with lines. I haven’t figured out if it’s better to color the labels to coincide with each
line, or leave them black.
Thank you.