Demystifying Tufte's data-ink ratio

I think the idea that a bar chart is usually improved by narrowing its bars while keeping them in the same positions is pretty silly. Yes, the article does say that one can take it too far, but I think that's the wrong framing. It's not that using less ink is good, but sometimes if you take that too far then it competes with visual clarity. Visual clarity is always the goal; sometimes clearer designs have more non-data ink, sometimes less; many designs are too busy so "less non-data ink" is usually good, but not always.

Tufte's "data-ink ratio" is a pretty good heuristic, but e.g. it would probably become a better match for actual visual clarity if what you counted wasn't "amount of ink" but something that measures spatial change in ink. It's the transitions that feel like noise, much more than patches of constant colour. But "data-ink ratio" is easier to say and easier to read than "ratio of integral of norm of gradient of colour".

Also, I think this sort of graph is better with very light gridlines than with no gridlines, though I think Tufte would disagree.

Also also, in Tufte's book, immediately before the section about "data-ink" there is an copy of a graph drawn by Playfair showing a country's balance of trade over time. Tufte gives it as an example of a good informational graphic. You can see it at https://en.wikipedia.org/wiki/William_Playfair#/media/File:P... if you like. This graph uses shading to indicate which direction the balance of trade goes in; that information is completely redundant with the curves by which the shading is bounded; but it would very much not be improved by removing that "redundant" ink even though according to his principles as interpreted by the page linked here that would make a huge improvement in the data-ink ratio.

It makes me think of https://xkcd.com/688/

That’s interesting. However, with charts, reducing the amount of ink can sometimes compromise readability.

That page is a lot of ink for very little data. It's also somewhat wrong - no, you can't claim the axis marks and labels aren't data ink at all just because it doesn't directly reflect the measure. They are still data that's extremely relevant to reading the chart. Yes, even if Tufte says otherwise.

Meanwhile, even Tufte's advice was explicitly to grey them down, but not completely remove them.

(And we can debate Tufte's advice - lots of other comments on this page how he had great aesthetic sensibility, but his advice wasn't grounded in reason or actual readability research, and so is limited to the examples he chose)

This isn't a great example, because they're effectively using a chart to show a handful of data points. The key with the data/ink ratio is to densify the amount of data shown so that the ink that is used can show additional dimensions or statistics -- such as using a 2 dimensional plot, turning the axes into box plot equivalents, or other relationships.

I think I prefer the 10-12% range myself.

What in the slop?

"The data to ink ratio is the ratio of data to ink. The data to ink ratio can be increased by decreasing the amount of ink. Decreasing the ink, while keeping the data constant, will increase the data to ink ratio. Here, we decreased the amount of ink in the diagram, and you can readily see that the data to ink ratio went up. The data to ink ratio should be increased, except when it shouldn't."

I remember getting Tufte's book ("The Visual Display of Quantitative Information") and being really excited because I'd heard all these great things about it, and then being disappointed because it had no real justification for its claims. The "data ink" thing is a typical example. I'm willing to accept the general concept that many graphs are too cluttered. But there's no particular reason to suppose that quantifying the proportion of "data ink" and trying to optimize it is the best way to handle that. A lot of the book is just taking subjective aesthetic judgements and trying to frame them as some kind of objective truth.

If we really wanted reliable guidelines on how to format graphs, we would want to ground them in psychological research that actually demonstrates that, for instance, some versions result in better understanding or recall of certain information. But as far as I can see Tufte's book makes no attempt to do this. (Perhaps he or those who built on his works did it later on?)

To be fair, if someone wanted to present that exact dataset efficiently, they would have probably proposed something much different, like an horizontal chart with thin lines, eg

North [------------] 7

South [--------] 5

East [--------] 5

West [-----] 3

or something, which would have saved a lot of data/ink ratio

Viewing in dark mode, so most of the ‘ink’ is used to render the empty background. Every suggested change here decreases the ink/data ratio.