Common errors are choosing colors that have inappropriate meaning in the context, or are loud enough to be distracting. Also, color is a poor encoding for ranked or ordered information*. Our brains don’t naturally order color, the way we do with brightness, length, and size. Associating an order with colors a learned and inconsistent encoding, and should not be relied upon.

* Heat maps typically use color in an ordered way, and we’re used to it because it is familiar, not because it is intuitive.

Another example is the redundant use of color and placement to encode the three main topics in my speech diagram. Using just color, the nodes would look like colored beads all mixed up on a chronological necklace. Using just placement, the boundaries of the three main topics would be less easy to discern. As it is, the three topics are clearly differentiate.

Here’s the entirety of section 4.1.4 of my thesis. For more examples, see the few sections that follow.

4.1.4 The four components of diagrams
The content of a graphic can be classified into four groups. The relevant, useful content of a graphic is either message or redundancy, and the balance is either decoration or noise (Horton, 1991, p. 27). The message is the information content that is being conveyed to the reader. If part of the message is encoded in multiple ways, such as by shape and by color, it is redundantly encoded. Redundant encoding provides the reader with multiple ways to extract the same knowledge, and has been shown to improve performance (Mayer, 1989).

Visual content that does not contribute to readers’ understanding is either decoration or noise. Horton differentiates the two by defining decoration as visual elements that increase the appeal of the image without interfering with the actual message. Noise is anything that does not contribute to the message or the appeal of the image, and therefore interferes with the message (Horton, 1991, p. 28).

Those are insightful comments.

One alternative to the current arrow heads is to print them inside the bars, so they don’t affect the perception of length as much. Obviously for some short bars (Seattle) the heads would be tiny, but if the entire image used them, the meaning of the directionality would be clear from the markets with more change.

In the case of the red/green of the bars, yes, the colors could be lost. However, the position of the bars above or below the segment redundantly encode the same information as color (and the arrow heads). Without the color it might take a moment longer to understand what the bars mean, but I think that an interested party could figure it out pretty quickly.

In my example with the text in the segment, some of the contrast is very poor, and would probably be lost in a grayscale or photocopied version of the image. There are few enough segments that there isn’t really a risk of confusion if they all had the same pale fill color, which would allow good contrast with the digits.

Personally, I like the arrowhead–it makes it quick to understand that the item represents change. I don’t get that from the headless bars.

Question for you: what are your thoughts on color coding? I’ve worked with just enough people who don’t see red and green as distinct that I’ve started thinking of color coding as a trap for a diagram designer. I look at the diagram and the information looks really clear, but I can’t trust that any particular viewer is going to see the information in that channel. Between colorblindness and monochrome printers/photocopiers, I’m not happy with a diagram unless it works in grayscale.

(Yeah, I know, the chromosome comic is dependent upon color-coding, but I wasn’t putting in the make-a-good-diagram effort on that, and it shows in more ways than just color.)