Female CEOs Keller

Scaling An Axis to Make A Point

A clever chart redesign last week got a lot of people talking about which one is “right.” What is more interesting to me is not which one is (supposedly) the better representation of the truth, but which purpose each one serves.

The original chart is the following, which shows the number of female CEOs in Fortune 500 companies (i.e., the top 500 U.S. companies by revenue).

Female CEOs Original

It’s a reasonable graphic showing the number of CEOs over time, as well as compared to the labor force, boards of directors, and executive officers.

The main issue is that the area chart on the left has an aspect ratio that is a bit optimistic. In a way, there has been an increase of over 800% since 2000 – but that is from a remarkably small base of less than 0.5%. Using the entire height of the chart is the sensible thing to do to show as much detail as possible. Technically, there is nothing wrong with that. It is showing the data correctly.

Conveying A Message

But there are two redesigns that not only show the data, but make a point. The first one, and the one that got the whole thing rolling, is by JK Keller.

Female CEOs Keller, half

Scaling the axis to 100% shows how much is missing and how little progress has been made. This is a much better representation of the situation, and one that is based on a clear point of view. However, it has a flaw: the goal is not 100% women CEOs, but equality: that’s 50%.

Alberto Cairo splits the difference by only scaling the bars to 100%, leaving the area chart intact, but adding some annotations. It’s a compromise, but I think it mostly clutters up the chart rather than making the message clearer. It also falls into the 100% trap.

A Clear Story

Francis Gagnon created the version that makes the most sense to me.

Female CEOs Gagnon

By choosing 50% as the maximum on the axis, he points out what is missing. At the same time, that scaling makes it possible to see some of the structure of the data. The overall structure of the graph is pretty simple, but it’s nice to see the details.

I also like Gagnon’s layout much better than the original (though I have some issues with his axis labeling). This is a much clearer graphic, because it leads you through the data, rather than hitting you over the head with the giant red area chart.

The only thing that is still bugging me is the pretense that the percentage actually means anything. 4% of 500 CEOs is 20 women. Why not show them as little stacked squares or figures or similar, so you can actually count the number? Given how small that is, and how much of a difference each new female CEO makes, I think that would create a clearer image of what is actually going on. The Reuters story the original graphic appeared in was triggered by a single new female CEO, after all: Mary Barra of General Motors.

Analysis vs. Presentation

This is a great example of how a few simple choices can make a big difference in the message a graphic conveys. All these charts show the data, and none is obviously flawed. But while the first shows the numbers in a way that makes the most use of the area, the second and third ones actually make a point.

While the data is simple and doesn’t allow much deep analysis, there is still the question of how to present it: take a position and argue it, or present the data and leave the interpretation to the user. None of these graphs really do the latter (the original chart’s title alone is a clear giveaway), but they provide different amounts of guidance and context, in particular to judge the percentage of women CEOs against a reasonable ideal.

Published by

Robert Kosara

Robert Kosara is Senior Research Scientist at Tableau Software, and formerly Associate Professor of Computer Science. His research focus is the communication of data using visualization. In addition to blogging, Robert also runs and tweets.

4 thoughts on “Scaling An Axis to Make A Point”

  1. Thanks a lot for including my graph in your review; it’s very interesting to have your opinion on this.

    I’m curious to know what a neutral graph (“present the data and leave the interpretation to the user”) would look like? Do I understand that to you the original Reuters graph is neutral, excluding its title, that a reader would not be drawn to a conclusion by its appearance?

    Also, I thought the second graph (women by level) was also interesting to analyze, but I understand that it doesn’t have the built-in outrage of the first one. I can’t understand how Reuters chose the order of the data. Putting CEOs at the end seems unhelpful given that Boards of directors are above CEOs. Also, this is a hierarchy and seems better represented in a vertical format, with the top job being on top. To me, the data suggests that it’s still more common to include women as an agent of diversity, since Boards have several members, than to be truly neutral and recruit an equal number of men and women. The vertical redesign made me see this.

  2. Interesting points. As an analyst, everyone has to look into what is being presented and how it is perceived by the eyes of the audience. Always set the area chart scaling to 100% to send a clear message.

    I am sure somewhere we have Do’s and Don’t’s for every graph type and every analyst has to have it in their cube.


Leave a Reply