IBM did research back in the 90s on perceptually-based colormaps and how to best represent various types of data within the color dimensions of luminescence, saturation and hue [1]. For example, they found that,
(1) Hue was not a good dimension for encoding magnitude information, i.e. rainbow color maps are bad.
(2) The mechanisms in human vision responsible for high spatial frequency information processing are luminance channels. If the data to be represented have high spatial frequency, use a color map which has a strong luminance variation across the data range.
(3) For interval and ratio data, both luminance- and saturation-varying color maps should produce the effect of having equal steps in data value correspond to equal perceptual steps, but the first will be most effective for high spatial frequency data variations and the second will be most effective for low spatial frequency variations.
===
[1] the original link got removed from IBMs website. Back in the day it was under
My favorite “effective” dataviz as of late (in this instance, effective as in conducive to education) is the mlu-explain articles from Amazon, visually explaining machine learning:
With similar objectives, Edward Tufte achieved contemporary fame in data viz with his approach to presentation that can be stated simply: less is more.
Tufte taught and demonstrated that in charts, anything other than the barest of axis and label - any ornamentation beyond basic data - detracts from the presentation through distraction and ambiguity.
Tufte can be quite dogmatic and not all of his principles should be taken as gospel at all times (then again, nothing should, probably). But one thing that stood out to me is that some of his examples remind you not to get too smart — just put the labels directly with what they label, annotate directly on the graph, etc., the kind of stuff you can easily forgot to do when the default is to just use chart templates in whatever software your data lives.
Mirror: https://www.gwern.net/docs/design/2021-franconeri.pdf