How to map some data into a factor for visualisation is one of the key questions to answer when designing a new graphic. Sometimes, the variables available are fixed by some external consideration, such as a requirement for a particular type of chart, and sometimes the only constraints are those inherent in the message to convey and the data available.

The table below from a paper by J.D. Mackinlay in ’86 signals which variables are the most effective for which types of data. This isn’t anything new or groundbreaking, it’s just sufficiently important to aspects of my work that I wanted to reproduce it here.

Quantitative Ordinal Nominal
Position Position Position
Length Density Hue
Angle Saturation Texture
Slope Hue Connection
Area Texture Containment
Volume Connection Density
Density Containment Saturation
Saturation Length Shape
Hue Angle Length
Texture Slope Angle
Connection Area Slope
Containment Volume Area
Shape Shape Volume

Which of course also implies that position should map to your most important metric, or if possible, your two most important metrics. For a bar chart, the choice of X axis can sometimes fall between category or time, with the Y axis often representing amount. This choice can be informed by these relative strengths.

Hue decreases accessibility, though, so for nominal items I would still favour texture over hue when possible.