Geologic Illustrations (Part 1?)
Something hit GG (well, figuratively; GG can still dodge the physical brick-bats from unhappy students): we have made an effort to teach students about writing, but we haven’t made a similar move in terms of teaching them how to make figures.
This struck home in realizing that the discussion in a writing class centered on what was in an illustration in a paper. And GG realized that the same thing was what typically happened in reading seminars: it was far more frequent to discuss what was shown in a figure than to struggle through some part of the text. And of course most scientific presentations now center on a PowerPoint (or equivalent) presentation. What we show can outweigh what we say.
We need to instruct students on how to make figures that work.
So as a public service, we’ll start looking at problems in figures and alternatives as inspiration strikes. Now there are some nice books out there on scientific illustration; many of the points in such books apply to geoscience and so its likely we’ll repeat some. But the elements of time and space in geoscience add an extra degree of difficulty.
Today consider the simple choice of point size for a scatter plot of some kind.
We’ll not identify the authors simply because this problem is all over the literature and nobody really needs to be singled out (this particular example is quite old). So, here we have a map of values of something, larger symbols for bigger values (in this case, positive is black, negative is gray). Seems clear, right? So what is the problem?
The problem is that areas with no data look identical to areas with a well-determined value near zero. Your eye assigns more significance to the large points compared with the smaller points. What is more, you have no idea which points are significantly different from zero and which are not. You might guess the big points are more significantly different than zero (but in parts of the image, you’d be wrong).
Now in this case, arguably, it is the stuff away from zero that matters for the authors’ point. But there are plots out there where the zero points are every bit as significant as the non-zero points, and yet points are sized by their value (a common plot where this was employed for quite awhile was the plot of travel time anomalies by backazimuth and distance; zero does matter in these cases). Something similar is here:
At first blush you might think there were three points–a plus towards the bottom and two circles toward the top. Closer examination reveals a lot of small points scattered about–and then just what is inside that dashed circle? Are those microscopic dots data points or lakes or reproduction errors? In this plot, arguably, the zeroes, which are nearly invisible, are every bit as significant as the big points–in fact there is a decent chance they are more significant.
So when you think you need to use the size of symbols to convey some value beyond the points’ x-y position in the plot, be sure that smaller = less significant. This might work well if plotting a resolution matrix, where smaller numbers are indeed less significant. But if all the points are equally significant–or their significance is unrelated to the size of the symbols you want to use–rethink what you are plotting.
Of course this points us at the general problem of 3-d plotting (plots where you want x-y-z all on the same plot). We’ll have to consider that, but maybe next time we might discuss the great enemy of good plots: Excel.