Bill Gates’s favorite graph of the year

Under the subject line “Blog bait!”, Brendan Nyhan points me to this post at the Washington Post blog:

For 2013, we asked some of the year’s most interesting, important and influential thinkers to name their favorite graph of the year — and why they chose it. Here’s Bill Gates’s.

image001-800x742
Infographic by Thomas Porostocky for WIRED.

“I love this graph because it shows that while the number of people dying from communicable diseases is still far too high, those numbers continue to come down. . . .”

As Brendan is aware, this is not my favorite sort of graph, it’s a bit of a puzzle to read and figure out where all the pieces fit in, also weird stuff going on like 3-D effects and the big space taken up by those yellow and green borders, as well as tricky things like understanding what some of those little blocks are, and perhaps the biggest question, what is the definition of an “untimely death.” But, as often is the case, the defects of the graph form a statistical perspective can make it attractive to readers: the 3-D design is grabby, and all the puzzles give a Chris Rock effect. So overall I think the graph’s a winner. And its use of the three colors is excellent—a simple but effective way of conveying the three groups (just don’t ask me to guess the relative sizes of the yellow and red parallelograms!).

The click-through solution

But what the really needs—and this is something I’ve said before too—is for the reader to be able to click through to a more standard statistical graphic (for example, a dotplot showing causes of deaths in order, with three columns corresponding to the three colors on this graph, with smaller, replica dotplots showing results just for the U.S., China, India, Indonesia, and other large countries, also maybe some continents and other groupings such as the E.U.), then click through again to a spreadsheet with the numbers.

Also some hypertext, i.e., a caption for the graph. There seems to be a style thing where designers like to use the minimum of words. But remember the BD principle: a picture plus a thousand words is better than 2 pictures or 2000 words. The hypertext could give lots of information, including their definition of “untimely death.”

This graph does not say what you think it says

Most interesting to me, though, was Gates’s claim that the graph shows that “while the number of people dying from communicable diseases is still far too high, those numbers continue to come down.” I guess I’ll buy that the graph shows that those yellow numbers are “far too high,” not really from the graphic itself but because we can see that big fat rectangle for Diarrhea. It just doesn’t seem like so many people should be dying of that. But I don’t think the graph does such a great job of showing the trend (“those numbers continue to come down). I mean, sure, after seeing Gates’s words, I can go back and say, yeah, lots of bright yellow there. But I didn’t catch that at all on first viewing. Also, that bright yellow thing is a bit of a cheat. For the red and green sections of the graph, the sharpest declines are indicated by a very pale green and a very very pale pink. But in the yellow box, the sharpest declines are screaming yellow.

That’s all fine, of course. The point of a display like this is to get people’s attention. It’s the role of the follow-up graphs in the click-through to provide some perspective and allow some comparison.

Here’s the quick rule:

Infographic: Grabs your attention.

Statistical graphic: Allows you to make comparisons.

Both these goals are important, and there’s no reason whatsoever to expect that they can be most effectively achieved by a single display. Also don’t forget captions. If there’s a key message you want to convey in your graph (for example, infectious diseases kill too many people but are in decline), I recommend putting that message front and center, in words, on your display.

P.S. Funny that they call Bill Gates a “thinker.” He seems like more of a doer!

25 thoughts on “Bill Gates’s favorite graph of the year

  1. Thanks for linking back to those supporting graphs. Looks like wapo hyperlinking the picture to that url would satisfy most of Andrew’s concerns.

    As for “Untimely Deaths”, looks like this infrographic is based on the graph of “YLLs (Years of Life Lost)”. Basing this view on the infographic’s caption: “War casualties account for just 0.05 percent of total life-years lost annually.”

  2. Fleshing it out the distinction btw “thinker” and “doer”: Thinker, dilettante, tinkerer, doer. I think they fit along an unclosed circle.

    • Maybe I’ve been ruined by too much topology, but “unclosed circle” translates to “line segment” in my mind…that said, I’ve never tried to drink from a doughnut or take a bite out of a coffee cup, so I guess I’m not all they way gone yet.

  3. “Diarrhea. It just doesn’t seem like so many people should be dying of that.”

    First – that quote should be a billboard, possibly in DC somewhere, with rotating pictures of random politicians looking confused.

    Second – it gives me a chance to mention the best/worst acronym in international development: International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b). And yes, you will in fact hear long strings of Bangla spoken, in the middle of which you hear “i-c-d-d-r-b”. And yes, it is lower case, and yes, it has a comma in the acronym. Took me about a week to get it to roll off the tongue. I joke about their name, but they have probably saved hundreds of thousands (or more) of lives by pushing oral rehydration packets into rural areas (as well as working on vaccines for cholera, running a massive hospital, etc.)

    Of substance (or, On Humanity): Suicide costs more life-years than dying in fires, floods, and falls all put together, despite the growth of natural disaster related deaths. And it looks like about twice as many life-years are lost to self-killing compared to other-person-killing-you. Part of that may be that suicides are a little younger than soldiers or victims, but probably not much.

    I think almost every suggestion Andrew makes on the blog is pretty interesting, but in the case of graphs I sometimes wonder if we’ll never live up to his expectations until we have a full time graphics person for every researcher (or I guess a grad student who actually wants to be a graphic designer). Animated-interactive informational-statistical-graphics? Yep, awesome. Nope, not easy.

    • “…. in the case of graphs I sometimes wonder if we’ll never live up to his expectations”

      Indeed. I used to be awed by Tufte and his graph prescriptions but then I slowly discovered the “graph theory” prescriptions are a jumble of often contradictory & esoteric edicts & opinions.

      I think the problem here is too much subjectivity in deciding what is a “good” graph. What we sorely need is an empirical, data-heavy approach to graph design.

      Let’s actually show people differently plotted graphs and measure comprehension / retention / recall etc. Till this this debate just is a lot of punditry & quackery.

      • Rahul:

        It would be great to have more rigorous work on measuring graphical perceptions. But I do think that systematic discussions of our own reactions and practices can be helpful as well. I’ve published thousands of graphs and I think I’ve learned some lessons along the way. I, and others, also learn from open discussion. For example, it is through years of discussion on this blog and elsewhere that I came to the realization (or, one might say, the opinion) that one thing holding us back in discussions of graphs was the attitude that one graph can do it all. I find it extremely freeing to allow that infoviz and stat graphics serve different goals and operate in different ways, and that there’s room for both styles (and room for improvement in both, as well).

        • Andrew:

          I dunno. To some extent, it’s like doctors trying to decide if Drug-A or Drug-B will work better by discussing the pharmacology & physiology in a conference room. There’s some merit but one always needs an empirical study eventually.

          I think we are at that point. A lot of the “good graphics” discussions are ideological. Having published thousands of graphs helps, but I think we have very sparse feedback on how well each graph did.

          Armchair theorizing about good graphs has reached its utility maximum I feel. I won’t be surprised if quite a lot of what we think makes a good graph is proved wrong by empirical measurement.

        • Rahul:

          I don’t think your term “armchair theorizing” is appropriate. I’m not just theorizing, I’m trying out different things and publishing them for various different audiences. Sure, I agree that it would be better if we could define our outcomes and measure them, but unfortunately we’re not there yet. In the meantime, I think it’s helpful to consider the goals of our different graphics in a qualitative fashion. In that sense, statistical graphics is like most human endeavors in that its aims are not clearly defined and are not clearly measured. Much of medical research is atypical in this regard, as the outcomes often are defined and can be measured well.

        • See. This is what happens when you accidentally post anonymously because you forgot to type in your name before you hit enter. You start a fight about quantitative evaluation of representations of quantitative analyses, with a side of child death by dehydration.

          The good news is, that is the one topic that Andrew apparently did not publish on this year. Journal of Post-Modern Economic Epidemiology, here comes a manuscript titled “Deconstructing Quantitative Evaluations of Representations of Quantitative Evaluations – a Qualitative Assessment of Practitioner Responses to Child Death by Diarrhea Graphics”. I invite Raul and Andrew to coauthor, but only if Entosophy will write a 17 page introduction of which I understand nothing, and Daniel Lakeland is listed as corresponding author.

      • Tufte created a nice little cult of highbrow folks. When I was in marketing research, I had the Napoleon-invading-Russia graph made famous by Tufte on my wall, and maybe 1 person in 100 or 500 who walked by my office recognized it. But I always found that I had a lot in common with the handful of guys who did recognize it, and found them to be interesting conversationalists.

        • Ooh, I hate that graph! I mean, it’s fine for what it is, it’s beautiful and conveys a lot of information effectively, but I hate the implicit message that it’s sent to people, that it’s desirable and possible to tell an entire story in a single graph. I get so frustrated when people try to do this with an infoviz. That’s one reason I preferred Tufte’s second book, because there he focused on small multiples.

    • Andrew wonders “Diarrhea. It just doesn’t seem like so many people should be dying of that.”

      Diarrhea is a major killer of infants in developing countries (bad water+ignorance=dead infant). Since it hits infants, it has an enormous impact on life-years and is responsible for shortened country-wide life-expectancy where prevalent.

      Prevention: water treatment at the household or community level. Treatment: continued hydration, replacing essential minerals lost to diarrhea.

      The Gates Foundation has been a major contributor to such programs. Not so much to statistical graph development.

      • Steve:

        Just in case it wasn’t clear, of course I think that public health improvements are a zillion times more important than statistical graphics development. But I do think that graphs that can help us visualize and understand comparisons can, ultimately, be helpful in real-world research that helps people in their daily lives.

        • Public health improvements have been helped by statistical graphics development going back at least to Florence Nightingale using a very early pie chart to impress upon Parliament that a huge fraction of military deaths in the Crimean War were due to disease, not battle.

  4. What catches my eye are all the small, bright spots (indicating rapidly increasing causes), often too small to even label. (Typhoid, neurological diseases, natural disaster, etc.) Of course I suspect an association with sampling variance (lower mean=higher coefficient of variation), but I’m having trouble convincing myself that’s the whole story. Eyeballing, typhoid is on the order of 1/10 the size of most of the other big yellow blocks, so the coefficient of variance should be about 3x larger?

Comments are closed.