When simple is too simple

Thanks to reader Don M, I came across this fascinating chart published in the New York Times Review recently (link). The main article, about gender segregation in job categories, is found here.

This is one of those charts that require a reader's guide.

The chart shows the proportion of women in each job category in year 1980 and in year 2010 (and nothing in between). The jobs are divided into three large chunks: the top chunk (shaded) consists of jobs in which women account for more than 70 percent of the total; the middle chunk (white background) are those jobs with 30 to 70 percent women; the bottom chunk (also shaded) are jobs with more than 70 percent men.

The designer then uses the red, green and gray colors (apologies to the color-blind folks) to group the jobs into three clusters. This is usually a great idea except that it is poorly executed here. Don is very annoyed with this because these colors lead the readers to the wrong conclusion, and I agree.

***

The color scheme is unnecessarily convoluted. Here is an alternative I prefer:

• if the change is 5 percent or less, color as gray no matter where the line is. (It is insane to color the line for housekeepers "red" for going from 87 to 89 percent in 30 years). 
• if the change is over 5 percent in the female direction, color it red to indicate the occupation is becoming more female. (There would be many red lines, such as for managers in education, HR staff, social workers, architects, etc.)
• if the change is over 5 percent in the male direction, color it blue to indicate the occupation is becoming more male (There would be only one blue line, and that is for welfare service aides.)

This would mean the lines for dentists and architects would be labelled progress. So too with most of the jobs that were predominantly male in 1980. In fact, there really isn't any occupation that went backwards--all those red lines in the bottom shaded chunk indicate shifts of only 1 to 4 percent, over 30 years!

This conclusion usurps the premise of the column in which the author claims that the conventional wisdom is wrong.

***

The other precaution in reading this chart is to realize that each occupation is put on equal footing in this chart even though some job categories employ a lot more people than others. Also confounded with this data is the differential growth/decline in job categories over the 30-year period. Further, the proportion of women entering the labor force must be accounted for.

This is a case in which less is less. The structure of the problem is complex, and it requires a more sophisticated approach.

Comments

[Baptiste Hautdidier]

I really disagree.

Why need your revised red/blue color scheme when it says hardly anything more than just the slopes? You can question their choice of setting a [30%;70%] range to define a job as "integrated". But if you accept it, then the chart is not poorly executed: it serves well the purpose of the authors.

The improvements I could think of would be: a colorblind-friendly color scheme, a 50% horizontal dotted line (as it defines how to read the slopes to measure "progress"), a softer shade for the changes that are not statistically significant.

And when you say that "there really isn't any occupation that went backwards--all those red lines in the bottom shaded chunk indicate shifts of only 1 to 4 percent, over 30 years"... Well the problem lies also in the upper part of the chart, unless you're happy with 'predominantly female jobs' becoming even more so in the future.

[Jörgen Abrahamsson]

Yes I like this chart as well. Not at all bad as these things go.
The data is disturbing of course.

[Rick Wicklin]

For technical readers, a scatter plot does a better job of showing trends in the data. For a redesign of the graphic and a discussion, see http://blogs.sas.com/content/iml/2012/10/15/women-and-jobs/ .