Skip to content

Why is this graph actually ok? It’s the journey, not just the destination.

Josh Miller was in my office and started flipping through Kieran Healy’s book on data visualization, a book that I like a lot—I even use it in my class, replacing Cleveland’s Elements of Graphing Data which is wonderful but things have changed in 35 years so time for a new book.

Josh noticed Figure 8.17 (see below left) and his reaction was that the series of distributions of the changing age distribution was pretty but that it would be more evocative to show a series of images showing graying hair.

My response was that a creative graph with graying heads could be great, but for a textbook you want something generalizable, something that students can do themselves for their own data. Cool graphs are cool, but they don’t represent a direct way forward.

In teaching, it’s important to give students a path for them to do things themselves.


  1. Anoneuoid says:

    It would be better if the y-axis gridlines were on top of the densities.

  2. Michael Nelson says:

    More evocative but less rich. In its current form, the graph vividly shows trends in fatness of tails, modality, spread, central tendency, and skew–and the change in relationships among them. I can’t think of a way of adding artistic detail without losing some of that information, or making it less vivid. Of course, audience and author intent matter: maybe Miller was thinking of a poster that has to grab conference attendee’s fleeting attention, or in an article for the general public in USA Today.

    Besides, you don’t want your students learning more slowly just because the graph provokes thoughts of oldness, do you???

  3. Klaas van Dijk says:

    Some of the years are lacking. This is not listed in the caption of this figure it is as well not indicated in the distances between the graphs. Anyone any idea why these years are lacking?

  4. Adede says:

    People go gray at different ages, so I don’t know what you’d get out of that except “older by an indefinite amount.” And how would you eve make that visualization? Two stock photos put together?

  5. Kaiser says:

    This graph is definitely popular, and usually well received. It brings back the usual entertainment v. information debate. It’s not easy to understand the age trends from it; for one thing you can’t put y-axis on them as they overlap. A more boring version would plot the time trend at discrete ages – one line for each chosen age. That would show the trends more clearly but it won’t invoke gray hair.

  6. Seems like a good example of how making something slightly more difficult to read can grab attention. Something about the overlap makes me tense, like I just want to stretch it out. But not sure I’d stare at it as long if that weren’t the case.

    • Andrew says:


      Good point about the overlap. I assume Kieran just did it that way so the graphs would fit, but the overlap conveys all sorts of things, including some tension or irritation as you say, but also an implication of efficiency and importance, kind of visually implying that this display is particularly effective by making full use of the available space.

      All of this is somewhat reminiscent of what they say about the Go board, how the grid is rectangular, not square, with not quite enough room for the stones to fit, so that the position can be perfect only in the players’ minds, not on the board itself.

      Also there’s something disturbing about how the years are unequally spaced on the display. I assume that was not on purpose but was just something that happened in the editing process?

      • jim says:

        ” the years are unequally spaced ”

        it’s not the years it’s the text labels. If you look at the last three or four text labels they’re in different positions relative to the x axis. Looking at the X axes they look evenly spaced to me.

      • jim says:

        yes looking at it top to bottom many text labels are off relative to the x axis, and presumably while these were being jacked around the 1988 one was duplicated and replaced 1989.

  7. Kieran Healy says:

    Thanks for the discussion and the comments! As a few people have suggested, the issues with the label positioning (and the 1988/89 label error) were introduced at the proof stage, as the typesetters endeavored to make the look of some of the graphs better. I caught some of these during the production process but not all of them, unfortunately. This sort of change of course kind of defeated one of the purposes of the book, which was to have a 1:1 correspondence between the code shown and the images on the page. The press did a really fantastic job with the book production on the whole, but this was one area where I didn’t anticipate what was going to happen. I’ll know better next time.

    More substantively, it might be worth noting that this plot appears as part of a more general discussion of how and whether to make bespoke plots, so it’s not as if I recommend this (or indeed any particular) plot type as something people should go for unquestioningly.

Leave a Reply