Data-ink ratio

Tufte defined data-ink as the non-erasable core of a graphic, the ink whose removal would take a number with it, and the data-ink ratio as that quantity divided by the total ink the chart spends. Two rules follow, and they are the whole method: maximise the ratio, within reason, and then revise and edit. In practice you do it by erasure. Take a finished chart, delete one thing, and ask what the reader lost. A plot fill, a box frame, a legend for a single series, and every other gridline usually cost nothing. The line and the dots cost everything.

The word sits next to its complement. The data-ink ratio measures how much of a chart’s ink carries data; chartjunk names the ink that carries none. One is a number you can watch move as you delete, the other is a name for what you are deleting, and neither is much use without the other.

Treat the ratio as a heuristic, not a law, because it can be gamed and it can be pushed past the point of use. It counts ink, not comprehension, so a chart can score well and still be hard to read: erase the axis labels and the ratio improves while the reader loses the scale, and strip the gridlines from a plot people are meant to read exact values off and you have optimised the measure instead of the chart. Critics have made this point since the 1980s, and controlled studies of “minimal” versus embellished charts have repeatedly failed to show the clean advantage the rule implies. The defensible version is narrower and still worth having: given two charts that answer the reader’s question equally well, prefer the one that spends less ink saying it.

The form Tufte pushed the idea furthest with is the sparkline, a chart reduced to almost nothing but its marks and set inline with the sentence it belongs to. It has no axes, no gridlines, no legend and no frame, and it works because it is not being asked for precision: the exact figure is printed beside it in the text, and the picture is carrying only the shape of the trend. That is the honest reading of a high ratio. It is not that decoration is immoral, it is that every mark should be answering something the reader actually asked.