In effect, what I’ve done here is choose to break a different rule from Duncan. Instead of putting two scales on the same axis, I have made one axis discontinuous between panels, skipping values in order to compress the horizontal size. Hence the reminder at the top of each panel that you’re shifting up an order of magnitude each time. Despite the rulebreaking, there’s still some principle at work because instead of just putting a discontinuity right at the end (to incorporate the largest value) the panels are split consistently by powers of ten, and it makes sense to think of WPA expenditures as falling into groups like “stuff they spent billions on” versus “stuff they spent tens of millions on” or “stuff they only spent a few million dollars on” and so on.
I like this, too. I can foresee an entire lecture on different ways of presenting information about the WPA…. Students will be so happy.