Fantastic piece. I love the systematic approach to refine my intuitions. Not just how to reason about statistics, but how to effectively second guess myself. And the visualisations really help to reflect on that.

Side note: I thought the visuals we're too impressive, turns out it's written by Mike Bostock of d3.js and bl.ocks.org/mbostock fame. His article "Visualizing Algorithms"[0] is a must read.

[0]: https://bost.ocks.org/mike/algorithms/

Observable, the notebook platform hosting the piece, is his (along with Jeremy Ashkenas who posted this to HN, and Tom MacWright) new startup, and I think this post is meant in part to demonstrate its capabilities.

And for everyone reading: go play with Observable. It’s fantastic!

And this doesn’t even discuss the effect of changing the color scale on the visual appearance (linear vs. log vs. some binning algorithm etc.), measuring counts per region vs. per capita vs. per square mile, drawing a cartogram, etc.

There are a lot of choices when displaying data on a map using colors.

Mike & Jeremy: one that you might want to toss in here is normalizing so that regions matching the national average increase are colored neutrally (instead of no change being colored neutral). i.e. normalizing relative to the national trend instead of relative to the previous year’s value. Depending on the data, this can help separate national vs. regional trends, though it takes more careful explanation to avoid reader confusion.

If you take one thing away from this article, hopefully it's this:

> No method is better universally, and none of them is “the best” even in the context of the dataset.

Ha, I thought this was going to be an article comparing BeyondCompare, WinMerge, xxdiff, etc..

But it was interesting nonetheless!

Useful for some politicians.

Journalists. I'd say journalists must read this. Maybe every Monday or so.

And statisticians. And readers of infographics.