Rework data:ink ratio chapter

[clauswilke]

The chapter on data:ink ratio should be reworked to take into account the feedback by @steveharoz. See this Twitter thread: https://twitter.com/sharoz/status/1005868631023112192

Transcribed:

Why perpetuate the myth of the importance of a data-to-ink ratio? It's based entirely on Tufte's opinion books rather than empirical evidence. Debunked many times.

Bateman et al. CHI 2010
Borgo et al. TVCG 2012
Borkin et al. TVCG 2013
Haroz et al. CHI 2015
Skau et al. CGF 2015

Collectively, these articles refute the notion that “ink” or non-minimal graphical elements is predictive of performance: 1. Bateman et al 2010 and Haroz et al 2015 showed that some embellishments improve performance. 2. Bateman et al 2010, Haroz et al 2015, and Skau et al 2015 failed to find a measureable performance hit for some embellishments. 3. Haroz et al 2015 and Skau 2015 showed that some embellishments harm performance. So non-minimal ink can either improve, reduce, or not affect performance. It’s an irrelevant dimension. Of course, not every form of ink (e.g. grids and backgrounds) was tested. That could become a bit of a no true scotsman issue. But they do show that ink quantity fails to predict much of anything. So why use the term at all? And what evidence is there that it’s worth the effort to minimize contrast of grids or outlines? It's fine to like and advocate the style. But no need for the psuedosciency term. As for Borkin et al 2013, it showed an improvement in recognizability, which I completely agree is not the same as memorability.

Plan for revisions:

• Rename the chapter to "Balance the data:ink ratio".

• I consider the data:ink ratio useful to think about extreme cases: all the way to one end or all the way to the other end figures become ugly. In the middle, though, there is a large range of options that can work well.

• Cite some of the relevant research literature.

• Add a version of Figure 18.2 with a frame around the plot panel, as proposed by @hadley.

• Make it clearer that many of the recommendations in this chapter are design choices that are guided to some extent by personal taste. Different people may make different choices, and that's fine.

[hadley]

Just for completeness — an oldie, but a goodie, is this exploration of Tufte's mid-gap plot vs the boxplot, which finds his more minimal version gives a less accurate estimate of whisker length:

@article{stock:1991,
	Author = {Stock, William A. and Behrens, John T.},
	Journal = {Journal of Educational Statistics},
	Number = {1},
	Pages = {1-20},
	Title = {Box, Line, and Midgap Plots: Effects of Display Characteristics on the Accuracy and Bias of Estimates of Whisker Length},
	Volume = {16},
	Year = {1991}}

[clauswilke]

Thanks!