I might expand this into a larger article at some point, but for now it’s just something I decided to cobble together for a quick post. Thinking about data visualization was a big part of my job at Scout Labs, and this represents my palette for expressing data in picture form.

Since color consists of three factors (hue, value and saturation), it’s three for the price of one from a data visualization standpoint. Hue can communicate difference, but value and saturation can communicate other dimensions – like degree of difference. Color is tricky though. You have to be careful to accommodate colorblind people and black and white printing.
Size is good for expressing one dimension of difference between things. It suggests something quantitative. If precision matters, then it’s safer to vary size along just one axis (e.g. length). Studies show that people are bad at judging area and angles. They can tell when one line is roughly twice as long as another, but they’re wildly off when they try to guess the exact difference in area between, say, two adjacent circles or two sections of a pie chart.
Shape is a good way of creating very basic distinctions between things – or classes of things. It works well, for example, in scatter diagrams and other visualizations that plot data in two- or three-dimensional space.
Decoration is good when you want to make an item or a small subset of items stand out from a larger set. Decoration can be more or less subtle, so I like to use it to represent variation as opposed to difference.
For position to mean anything, it helps to have stable reference points – like x and y axes (i.e. a grid). Meaning is expressed by the position of objects relative to each other of course, but more importantly it’s expressed in the position of objects relative to the axes.
Motion can be a powerful way to add directional nuance around things like trends, or to wrap in concepts like velocity, but the biggest drawback, obviously, is that motion isn’t possible on paper and needs to be translated into something else.

Obviously these aren’t mutually exclusive. People are capable of grokking a number of concepts from a single visualization, so I usually combine dimensions from the palette. Sometimes I combine things just for efficiency – to get more out of each pixel so to speak. More often, I combine things when I feel like they make sense together.

For example, I might use hue to represent positive or negative sentiment in a product review, saturation or value to represent the intensity of the sentiment, and size to represent the reach of the source.

2 thoughts on “The Data Visualization Palette

Leave a Reply

Your email address will not be published. Required fields are marked *