Stata Tip 78: Going Gray Gracefully: Highlighting Subsets and Downplaying Substrates

Abstract

In graphics, as in life, going gray is often forced upon us, yet it is also occasionally a deliberate choice. Journals may enforce publication of your graphs in black and white whenever full-blown color is prohibited, or else prohibitively expensive. Even when allowed, color may prove problematic for various and quite different reasons, ranging from physiology and psychology to sociology and aesthetics. For example, many people are red-green color-blind, while the spectral or rainbow sequence from red to violet is not in fact perceived as a monotonic scale. Wilkinson (2005) and Ware (2008) give good introductions to the use of color in visualization. Fortner and Meyer (1997) give a more detailed discussion. Brewer (2005) gives many specific suggestions on color schemes, which are as appropriate for statistical graphics as they are for cartography. Choosing differing shades of gray is also worth consideration for positive reasons. Expressing qualitative contrasts just with gray can be both effective and attractive. A previous Stata tip (Cox 2005) showed how distinct values on an ordered scale could be shown separately on scatterplots by markers of different gray-scale colors. This tip expands the theme with further examples. Naturally, black and white themselves qualify as extreme gray shades and may work very well. Here I will emphasize the use of intermediate shades. Let me underline that the examples here use the sj scheme. See (G) schemes intro for more information. Consider the highlighting of subsets. You may want to show the distribution of a sub- set with the distribution of the complete set as context. Unwin, Theus, and Hofmann (2006); Chen, Hardle, and Unwin (2008); and Myatt and Johnson (2009) include sev- eral examples of this device for various kinds of graphs. On a histogram with a frequency scale, this can be done by laying down the dis- tribution of the complete set first and plotting the distribution of the subset on top. Display of the subset can never occlude the display of the complete set, because at most all the observations in any bin belong to the subset.