Nobody is ever going to mistake me for Edward Tufte, but whenever I run across a chart like this one:
(from Matt Yglesias, who got it from Justin Fox where it was merely one of many equally horrible plots), I find myself distracted from the actual point of the graph by the awfulness of the presentation.
I mean, look at this thing. The numerical labels for the horizontal axis are up at the top, rather than at the bottom where they usually go. The label that states what's actually plotted on that axis is down at the bottom of the graph, where it appears to be just a stray bit of text labelling nothing, while the explanation of what is on the vertical axis is presented as a sort of subtitle (while the axis itself gets only the uninformative label "Percent"), directly above the numerical labels of the horizontal axis, making it look like the percent change in nonfarm employment is plotted on the horizontal axis. As if that weren't enough, the horizontal axis labels are sideways, for no earthly reason I can think of, so you have to turn your head to read them.
And you know what the worst part of this graph is? The little "Time" graphic down at the bottom right. Because that means that somebody from a major news media organization had a hand in making or approving this graphic. A graph that I would jump all over in a first-year undergraduate lab report has the official stamp of approval of a major media organization.
And people wonder why the public isn't better informed.
- Log in to post comments
It looks like Excel was involved at some stage in making this graph. I don't have the latest version, but earlier versions of Excel were notable for the abysmal quality of their plots. (No offense intended to ocean trenches.) Putting the X axis labels where they are here is a standard Stupid Excel Trick, as is using a Y axis label which was a shorthand column title. It looks like some graphic artist tried to clean this up, but the plot was not salvageable.
That's aside from the changes in how unemployment is measured between the 1930s and today. This comparison is not merely apples to oranges; it is apples to Frisbees.
That is indeed a steaming pile of awful. Every time I look at it it gets worse. The eye just shies away from it, and no doubt most readers would just instinctively skip over it on the page.
It also doesn't help that it is portraying an odd and inherently confusing data set. 'Percent change in nonfarm employment vs. Months from employment peak'? There must be a better set of data to illustrate this trend then that.
Regarding the "Time" logo - when I read Huff's "How to Lie with Statistics", from 1954, I noticed that a lot of his bad examples came from magazines like Time and Newsweek. I have never seen any evidence that the newsmagazines have gotten any better about their graphs since then.
Why people still think those magazines actually represent any sort of reputable source, is beyond me.
Aside from the axis label issues that you describe, it's really not that bad. It has nice color-coded lines, it has a good sense of foreground vs background, and it's comparing (more or less) apples to apples by using percents instead of millions of jobs. I've seen a lot worse...
Aside from the axis label issues that you describe, it's really not that bad. It has nice color-coded lines, it has a good sense of foreground vs background, and it's comparing (more or less) apples to apples by using percents instead of millions of jobs. I've seen a lot worse...
Wow, that's bad! At first glance, the pink line is something to do with September '29, but the x-axis is time, so it only *starts* in September '29. Also the initial unemployment is subtracted out from both curves, making it impossible to compare the actual values.
ClayB
What the graph is comparing are the two time series of the percentage drop in nonfarm EMPLOYMENT from the peak.
The actual value of initial nonfarm employment isn't important to the comparison.
(I agree with Chad's criticisms of the formatting of the plot. I think it still gets the point across reasonably well, if only because it's hard to totally botch a display of that big an effect.)
A hateful, and egregious case of Pixel abuse. Won't someone protect the innocent pixels? If this can be done to millions of innocent pixels are any of us safe?
Two small steps make this chart much better:
1. Move the horizontal axis ticks and labels to the bottom of the chart (in Excel, format the Y axis so the category axis crosses at -14).
2. Rotate the horizontal axis labels so that they are horizontal.
The legend clearly says 'Month when employment peaked' right above the legend labels that show the month and year. This might be improved by using the same font as used for the legend labels.
There are many utterly horrendous charts published every day, but this isn't really one of them.