It also makes plots easier to talk about. Most people would be able to quickly ascertain that there are five distinct categories in the plot on the left and, if asked to characterize the “blue” points, would be able to do so. With the plot on the right, where the points are all blue but vary in their luminance and saturation, it’s harder to say how many unique categories are present. And how would we talk about a particular category? “The fairly-but-not-too-blue points?” What’s more, the gray dots seem to fade into the background, de-emphasizing them relative to the more intense blue dots. If the categories are equally important, this is a poor representation. So as a general rule, use hue variation to represent categories. With that said, here are few notes of caution. If you have more than a handful of colors in your plot, it can become difficult to keep in mind what each one means, unless there are pre-existing associations between the categories and the colors used to represent them. This makes your plot harder to interpret: rather than focusing on the data, a viewer will have to continually refer to the legend to make sense of what is shown. So you should strive not to make plots that are too complex. And be mindful that not everyone sees colors the same way.
Varying both shape (or some other attribute) and color can help people with anomalous color vision understand your plots, and it can keep them (somewhat) interpretable if they are printed to black-and-white.
On the other hand, hue variations are not well suited to representing numeric data. Consider this example, where we need colors to represent the counts in a bivariate histogram. On the right, we use a palette that uses brighter colors to represent bins with larger counts: On the left, we use a circular colormap, where gradual changes in the number of observation within each bin correspond to gradual changes in hue. With the hue-based palette, it’s quite difficult to ascertain the shape of the bivariate distribution. In contrast, the luminance palette makes it much more clear that there are two prominent peaks.
0 kommentar(er)
