I think this might be in part due to Edward Tufte's philosophy, as especially described in his book The Visual Display of Quantitative Information [1]. He uses

I think this might be in part due to Edward Tufte's philosophy, as especially described in his book The Visual Display of Quantitative Information [1]. He uses "ink to data ratio" as a measure of the effectiveness of numerical charts. This superficially makes sense: why add more visual noise if it adds no more information? As an example, he suggesting shortening the axes on scatter plots to reflect the range of data in each dimension, which is a win-win: it adds more information and reduces ink. Boxes around objects and lines separating things are specifically called out as especially evil, since they add ink without contributing any information. I think this idea may have leaked from quantitative displays to general design. The flaw of course is that more ink does not necessarily mean more visual noise. With the axes example, disconnecting them means that you now have two high-level visual objects rather than one, and the whole chart starts to looks like a mess of individual objects on the page. If you have several graphs next to each other then it's much easier to focus one and exclude the others if they're in boxes rather than having component objects floating indistinctly near each other. But Tufte only sees a count of black blobs of ink or black pixels, as if each one has its own mental cost. As an aside, the book is still fantastic because it has loads of examples of unconventional charts and figures, and you can just ignore the text (or, better, read it skeptically). I found out about it from xkcd originally [2] and I suspect that many others did too. [1] https://www.edwardtufte.com/tufte/books_vdqi https://www.edwardtufte.com/tufte/books_vdqi [2] https://xkcd.com/124/ https://xkcd.com/124/