This short article in Significance by Allan Reese contains much wisdom. He takes a moderately well known graph and improves it, emphasising the need to choose a message and own it, curate it and take responsibility for its understanding among your audience. He also objects to slavishly following rules. I couldn’t agree more and these messages will, I hope, emerge in my soon-to-be bestseller on the subject out next year. Until then though, I give you my current Figure 9.9, which was kindly provided by Andrew Gelman and comes from the magisterial book Bayesian Data Analysis (BDA3). It might be considered hubristic in statistics to print “B D A” down the spine of your book, for reasons that you either get or you don’t, but this is such a whopper it gets away with it. And this chart is on the front cover.
This is not the same chart that Allan critiqued, being American rather than British data, but is on the same topic. What this one does is to decompose the time series into several additive components. They did so with Gaussian processes, a really flexible smooth modelling method that you can run in Stan and get those non-parametric curves that are like silk, milk, a bedspread or a quilt, icing on a cake or a serene translucent lake.
I like the fact that this is simple but gives you so many different stories, all clear to the viewer without need of any narration. It’s so hard to find data that work like this as exemplar analyses.