Tuesday, December 10, 2013

Discussing an infographic by Reuters

There's been some discussion on Twitter today about the Reuters infographic on the top left. Jonathan Keller redesigned it (top right) to include a full Y-scale. Christopher Ingraham has summarized the conversation.

Which one of the graphics is right? I don't really know. Keller's does a much better job at showing the actual proportions, and at revealing the tiny amount of women at the top levels of the corporate world. But it bothers me because it obscures the slight progress made in the last couple of decades. That was my original objection. Can we get the best of both worlds? Perhaps. Here's a quick and rough sketch of what I'd try to do (IMPORTANT: See discussion and updates below):

UPDATE1 : Francis Gagnon has created a new version of the line chart in which the Y-axis goes up to 50%, instead of 100%. What do you think?

UPDATE 2: Chris Moore's design (see comments.)

UPDATE 3: Another idea from a reader (see comments.)

UPDATE 4: Three new excellent contributions by Robert Kosara, Kaiser Fung, and Krisztina Szab√≥


  1. This compromise seems to add confusion. The share of women CEOs has scale to the left, but data to the right. This is an unusual way to show data and I was fooled by it until I was preparing to tear my shirt in the comments over the confusion of data between the two graphs.

    I'll have to side with the Keller version. In the real world, the rise is barely visible and that's what his graph shows. I wouldn't want readers to go away with the idea that it's rising real fast (what about another 3% increase in the next 18 years?). I'll quote in support this entry from "mathematically literate" headlines.

    Our World: Market Share for Electric Cars Triples
    Mathematically Literate World: Market Share for Electric Cars Rises to 0.4%


    1. I thought about putting the scale of the line graph on the left, and also making its slopes less steep (perhaps following Cleveland's banking to 45 degrees rule) but that would involve re-drawing the entire graph, which I don't have time to do. Would all that improve it?

      And there's a rise; that's my point: Percentage goes from 0.3% to 3.5%, which is clearly insufficient, but still worth showing.

    2. I'd suggest that the scale of the two graphs should be 50%. While men or women can theoretically reach 100% in both cases (and men came quite close as CEOs of Fortune 500 companies), it can be assumed that the target is 50% in a gender-neutral world. This would increase the visibility of the slope for the rise, without exaggerating it.

  2. That's an interesting idea. It's worth trying. If someone wants to redesign it, please post the link here

  3. My approach with this sort of thing would tend to be to visualise the proportion and detail separately - so in this case and for the "Women CEOs" graph, I'd probably put the title in the centre of a doughnut graph and visually link (by using the same colour/pattern) the 4% shown there to the graph showing the 1995-2013 detail. The essential quandary here is that you're trying to show two conflicting pieces of information with the same graph - you have to suppress one in order to exaggerate the other. I don't believe that there's a "right scale" here. Parsimony has to go out of the window under such circumstances, in my view - it is better to have two different graph types in the same space than to fail to convey the key information clearly.

    The same approach goes for "Women in durable goods manufacturing", although you'd need some hard numbers in order to understand the total proportion of women in that group.

    1. Took a two quick stabs at this: http://imgur.com/EFc7Te5

      A couple of things struck me while I was doing this. One was that the actuals were more interesting than the percentages - that is, the idea that there were 20 companies with a woman at the helm among Fortune 500 companies seemed more impactful than the abstraction of 4%. The other was that the area graph didn't sit well with me. Given that you've got two key factors that are going to cause a change in the numbers YoY - a company changes CEO and that includes a gender change, and the companies on the list change - the implication of a continuum that you get with an area graph didn't seem right. So, columns it is.

  4. Chris, your phrase "company changes CEO" triggered an interesting idea for me - though I'm not sure if the data is available. I would love to see the gender split of new CEOs. The established base of male CEOs will not go away overnight, so knowing how many new CEOs are named each year could give us a better yardstick.

  5. To be honest, I think there's a value judgement coming in - do you want to point out that the proportion of women in CEO positions is pitifully small, or that it has increased 20 fold in the time under consideration. Mostly, I would take the former. The shape of the graph which you have put on the right seems to suggest that huge improvements have been made but when you're starting from an extremely low base of 0.3%, an improvement up to 4% or so is actually not a huge improvement.

    The headline on the original piece is that the glass ceiling still persists. From that point of view, I'd take the view that Jonathon Keller's rework is a more accurate reflection both of reality and that headline; you can see that there is some progress but it is pitifully small. The Reuters graph, because of the y-scale does not reflect that it is pitifully small unless you are wise enough to check the scale. Most general readers will mostly likely remember the shape of the graph than the scale and the shape suggest that women have seen a huge improvement in their access to CEO roles. Maybe a scale up to 50% might be a reasonable compromise but the ultimate issue is that if you compare the proportions of CEO roles held by men versus those held by women on something like a pie chart or a proportionally split block, the simple truth is that yes, the glass ceiling exists and changes over the last 20 years aren't really all that great.

  6. Francis, this worries me: "There seems to be two schools of thought: those who want to represent the numbers and those who want to represent the reality." Even sharing your societal concerns, the design philosophy —based on a false dichotomy— implied at the beginning of your new post may lead you to consciously and willingly let biases drive design choices. That’s dangerous.

    I insist on what I suggested before: The huge differences between men and women in positions of power can be represented by a bar graph. That's what bar graphs are useful for. Line charts exist to display change within a certain period of time. I don’t necessarily mean change from the starting point to the end point of that period, but also (here’s the key idea) variations between each specific point on the time scale and the next one. Therefore the banking to 45 degrees principle, which I still think is a sensible rule of thumb:


    I also like the line chart you made, with the scale going up to 50%. That may be a good compromise, as Treasa has pointed out. But my guess is that Cleveland's rule may work better. These are all hunches, though. As fun as endless theorizing may be, all these graphics should be tested with readers to see what messages they get.

    Thanks everyone for a great discussion.

  7. I like what Chris has proposed and I think too it is better to say that there are only 20 out of 500 rather than say 4%
    I tried a slightly different solution:

  8. For reference, here is how Catalyst, the organization whose study was the basis for the Reuters graphic, chose to represent the data:


    The top of the graphic is roughly where the 10% line would be.

  9. I've got a few further thoughts that I'd like to share on this.

    One of the problems that we're facing here is that the only number that actually needs to be visualised is the 4%. The graph is interesting enough, and it does serve a purpose, but the primary reason for that graph being there is that those are the numbers that were available to the original designer. It's interesting, but it's basically filler. If you were to only produce a single visualisation to illustrate the point, then it would have to be a visualisation of that 4% specifically.

    Probably a better way to do this, if you wanted to add texture, would be to use more interesting numbers. Benchmarking against women in top-of-field roles in other areas might be revealing. How do the CEO numbers compare to female occupancy of senior roles in medicine, science, politics, engineering, law? (It's interesting that SCOTUS - the epitome of membership inertia - has better figures than Fortune 500 CEOships).

    I mentioned in an earlier post that I was uneasy with the area graph specifically, and attributed it to the fact that the data set had two distinct contributing factors and so oughtn't to be shown as a continuum. I stand by that view, but, on reflection, there's a more significant reason why it doesn't work, and it relates to the issue of scaling that's formed the meat of this whole discussion.

    The primary purpose of an area graph is to show magnitude/proportionality along a timeline. Because we're showing a percentage here, and it's a binary dataset (woman/man), there's an invisible "man" set that we automatically derive from the space above the visualised percentages. And this is why the only way to present this data, IF you're using an area graph, is to set the scale at 100%, because you're implying that the leftover space is "the rest". To put it another way, if you were doing a male CEOs area graph, then you wouldn't put the minimum value to 50%.

    John: I agree that it'd be interesting, and I'd also be keen to see how many companies actually make the Fortune 500 with a woman in the top spot. The rule of thumb is that a member of any minority in any given group that excels is likely to be exceptional in comparison to their peers - if the field is against you and you still distinguish yourself, then you really must be something special. So I wonder if the numbers of new companies entering the 2000/1000/500 bear that out (company with exceptional CEO performs better --> enjoys success --> moves up the list).

    For what it's worth, the number of women in CEO positions is currently 21, not 20 (so 4.2%), and Mary Barra and Jacqueline Hinman will make it 4.6% in January (source: http://www.catalyst.org/knowledge/women-ceos-fortune-1000).

  10. You might be interested in a thorough study on gender diversity in company leadership published last week by Fenwick & West (a Silicon Valley law firm). We worked with them to create about 70 different data graphics for the report.

    We chose to use a 50% scale for most of the line graphs that show percentages of women in different leadership positions, with the idea that 50% represents parity. In some places, we showed line graphs of the average number of women over time alongside the percentage figures, which follow more of the 45 degree principle. There is usually more than just one question people are trying to answer when they look at things like this, so it's helpful to look at the data from multiple angles.

    In addition to the line graphs, there are histograms, box and whisker plots, a couple of bubble charts and even a lovely pair of slopegraphs toward the end :-)

    You can see a few of graphs from the report on the Threestory Studio blog, or you can download the PDF of the report here: