Eric Tassone sent me the this graph from a U.S. Treasury Department press release. The graph is so ugly I put it below the fold.
Anyway, it got me to thinking about a famous point that Orwell made, that one reason that propaganda is often poorly written is that the propagandist wants to give a particular impression while being vague on the details. But we should all be aware of how we write:
[The English language] becomes ugly and inaccurate because our thoughts are foolish, but the slovenliness of our language makes it easier for us to have foolish thoughts.
Tufte made a similar point about graphs for learning vs. graphs for propaganda.
Here are (some) of the errors in the Treasury Department graph:
1. Displaying only 3 years gives minimal historical perspective. This is particularly clear, given that the avg rate of unemployment displayed is from 1960 to 2005.
2. Jobs are not normalized to population. Actually, since unemployment rate is being displayed, it’s not clear that anything is gained by showing jobs also.
3. Double-y-axis graph is very hard to read and emphasizes a meaningless point on the graph where the lines cross. Use 2 separate graphs (or, probably better, just get rid of the “jobs” line entirely).
4. Axes are crowded. x-axis should be labeled every year, not every 2 months. y-axis could just have labels at 4%, 5%, 6%. And if you want to display jobs, display them in millions, not thousands! I mean, what were they thinking??
5. Horizontal lines at 129, 130, etc., add nothing but clutter.
To return to the Orwell article, I think there are two things going on. First, there’s an obvious political motivation for starting the graph in 2003. And also for not dividing jobs by population. But the other errors are just generic poor practice. And, as Junk Charts has illustrated many times, it happens all the time that people use too short a time scale for their graphs, even when they get no benefit from doing so. So, my take on it: there are a lot of people out there who make basic graphical mistakes because they don’t know better. But when you’re trying to make a questionable political point, there’s an extra motivation for being sloppy. This sort of graph is comparable to the paragraphs that Orwell quoted in “Politics and the English Language”: the general message is clear, but when you try to pin down the exact meanings of the words, the logic becomes less convincing.
OK, here’s the graph:
P.S. Following Orwell, I’m happy to criticize left or center as well as the right, so feel free to nominate other examples of propaganda graphs. (As I recall, “How to Lie with Statistics” has a nice example from some old Soviet propaganda.)
P.P.S. I’m not trying to say that the people who made this graph are “bad guys” who are “lying with statistics.” Rather, I’m making the Tufte-esque point that standard practice in statistical graphics is so bad that cluttery, low-information graphs are the norm, thus making it easy for people to mislead themselves (and others).
P.P.P.S. Brendan Nyhan made some nice plots in Stata of the employment and unemployment series. I still think it’s weird that employment is not adjusted for population (or for population aged 18-65, or whatever) but maybe that’s just the way people do it. As Bill James said of batting averages: for the purpose they were designed, other formulas (e.g., on-base average) are better, but people are so used to batting average, that these numbers have the power of words. Maybe that’s true (among economists) for absolute non-farm employment, I dunno.