Tuesday, August 20, 2019

My first article for Scientific American

The September issue of Scientific American magazine is titled ‘Truth, Lies & Uncertainty: Searching for Reality in Unreal Times’. It contains articles about how deception works (and not only among humans!), how dishonesty spreads, why we trust lies, and how we make decisions when having incomplete information.

My favorite article is by Jessica Hullman. Perhaps not surprisingly, she writes about visualizing uncertainty. This article alone is worth the price of the magazine, so buy a copy if you can.

I wrote the article on the last page. It's titled ‘Does Obesity Shorten Lives? Misreading data visualizations can reinforce biased perceptions’. It explains the ecological fallacy, amalgamation paradoxes and, above all, focuses on how easy it is to misunderstand a chart if we describe its content sloppily, either to ourselves (mentally) or to others (verbally or textually). The example I used is similar to one that I borrowed from Heather Krause (who also fact-checked the SciAm article,) and that I showcase in How Charts Lie.

As I wrote in the article, it's tempting to describe that first chart as “the more obese we are, the longer we live”, when it doesn't really show that. It just shows that, nation by nation, obesity rates and life expectancy are positively associated.

How do I know it's easy to come up with sloppy verbal descriptions of charts like that? Because I heard them repeatedly in newsrooms when I worked as a visual journalist. I even inadvertently made this mistake in one of my books! On pages 123-128 of The Functional Art there are several graphics about the strong negative association between obesity rates and educational attainment in the U.S. at the state level.

I carelessly described these charts as showing “that better educated people are less likely to be obese”(1). That statement is true—there's no amalgamation paradox here: I checked at the time, and the negative association between education and obesity is obviously weaker person by person, but still pretty substantial. But my maps and graphs don't really corroborate that on their own, something I should have warned readers about; we'd need data about individuals, not states, for that (2). The graphics themselves aren't the problem, though (there may be reasons to design a chart to see whether obesity and education are associated state by state or country by country); the problem is whether we describe what they show accurately.

(1) There are also conspicuous non-weighted means in those graphics; what was I thinking?

(2) Thanks to Hicham Bou Habib for pointing out some embarrassing lapses in his thorough critique of The Functional Art.