Skip to content

Sometimes a graph really is just ugly

I’ve been talking a lot about how different graphical presentations serve different goals, and how we should avoid being so judgmental about graphs. Instead of saying that a particular data visualization is bad, we should think about what goal it serves.

That’s all well and good, but sometimes a graph really is bad.

Let me draw an analogy to the popular media. Books and videogames serve different goals. I’m a reader and writer of books and have very little interest in videogames, but it would be silly for me to criticize a videogame on the grounds that it’s a bad book (or, for that matter, to criticize a book because it doesn’t yield a satisfying game-playing experience). But . . . there are bad books and there are bad videogames.

To restrict our scope to books for a moment: you could argue that Mickey Spillane’s books are terrible or you could argue that, given that they sold tens of millions of copies, they must have had something going for them. I wouldn’t want to characterize such a book as simply bad. But . . . think of all the crappy attempted Spillanes that were produced in the 1950s, all the ones that neither sold well nor had interesting content. Some of them must have been low-quality under any measure. Or we could talk about food. McDonald’s is fine but somewhere there’s a greasy spoon whose burgers are so barfable that even the locals don’t go there. (I remember a place we went to in grad school once at 2 in the morning, a disgusting local restaurant that was open from about 11pm-5am every day and was always empty. The rumor was that its sole function was as a mob hangout. So, sure, it had a function, but food had nothing to do with it. Similarly, some of those attempted Spillanes probably had the intended function of selling books but, at that they failed miserably.)

And now we return to statistical graphics, in particular this graph of trends in military spending:

Screen shot 2011-07-10 at 5.02.55 PM.png

I learned about this beauty from Justin Logan and Charles Zakaib, who wrote:

We have found the charts in this Council on Foreign Relations report [by Neil Bouhan and Paul Swartz] confusing and would appreciate any thoughts you have on the authors’ choice of graphical representation. The charts at the bottom of pages 4 and 5 are particularly vexing. It seems to us that the message could have been conveyed in a much clearer way.

I too was vexed. I’ve been blogging long enough that I think faithful readers could easily reel off ten things I don’t like about the graph. I’m too lazy to list them here (but feel free to do so in the comments!); instead, I just want to make the point that, indeed, this graph could’ve been much better and, no, I don’t see that its ugly clutteredness serves any clear non-statistical goals either. Just as a lot of writing is done by people without good command of the tools of the written language, so are many graphs made by people who can only clumsily handle the tools of graphics. The problem is made worse, I believe, because I don’t think the creators of the graph thought hard about what their goals were.

That said, I applaud their decision to make the presentations graphically. I think the best way to read the report is to read the text and then glance at the graphs to see that they provide evidence for the points made in the text. And you’ll have to decide for yourself whether it’s actually bad news (as claimed on page 5 of the report) that “the United StatesÂ’’ and its alliesÂ’’ share of world military spending . . . is projected to fall further, to 66 percent, by 2015.” I mean, sure, it would be great if other countries didn’t waste any money on expensive tanks, fighter jets, military retirement plans, etc., but can you really blame them for wanting to be like us?

P.S. No, the 30 days are not up, and yes, this entry does have statistical content. It’s part of our ongoing exploration of criteria for understanding and evaluating statistical graphics and, by implication, statistical communication more generally.


  1. Matt says:

    Off topic. The 30 days plan had been awesome. Maybe don't stop after a month?

  2. Andrew Gelman says:


    I'll still blog about statistics; I'll just blog about other things too. You can feel free to skip the non-statistical entries.

  3. Robin Morris says:

    "I think the best way to read the report is to read the text and then glance at the graphs to see that they provide evidence for the points made in the text"

    Isn't the usual way that we read a report to first look at the graphs? If they look interesting, that motivates you to read the text. The advice I've always been given is to make the figures and captions self-contained, as that's all a lot of readers will ever look at.

  4. Andrew Gelman says:


    I agree. But in this case the graphs are pretty bad and so I think the only useful role they can play here is supporting material for the text.

    Still, "supporting material for the text" is a contribution. I think the report is much better with the graphs than it would be without.

  5. Josh Sher says:

    What was the name of the late night restuarant on Highland Avenue near Davis Square? (I assume this is the one you are talking about). I actually was told they had good eggs, but I never went. I always thought it was just a money laundering operation…

  6. Andrew Gelman says:

    Kay and Chips, I believe it was.

  7. […] Statistical Modeling, Causal Inference, and Social Science found a graph that is just ugly, and Observational Epidemiology has another misleading example. […]

Where can you find the best CBD products? CBD gummies made with vegan ingredients and CBD oils that are lab tested and 100% organic? Click here.