Axes of evil: How to lie with graphs
Hopefully-unnecessary disclaimer: This post is very much tongue-in-cheek. I do NOT condone anyone intentionally misleading their audiences. Rather, these graphs are presented as examples of what not to do. Use your powers for good, people.
As Mark Twain once said, “Never let the truth get in the way of a good story.” Here are a few techniques to hide those pesky numbers and tell the story you feel, not the one you can prove.
Don your handlebar mustache and practice your evil laugh — we’re going in.
Drawing a bar graph? For heaven’s sake, don’t start at zero!
You know, I find it… interesting that a Google Image search for “misleading graph” results in an immediate suggestion to refine my search with “Fox News.” There are easily enough Fox News examples for the network to warrant its own results category — most of the readily available real-world examples of blatantly misleading graphs do, in fact, come from Fox.
I’m trying to keep this site as politically neutral as possible (we get enough of that from other venues, I believe), so I’ll only show the one graph. But come on, Fox News. Pretend you care.
Okay, back to the graph. First, note that this is a bar graph. Bar graph y-axes always start at zero. Readers assume that the relationship between column heights is the same as the relationship between numbers in your data. But this graph has instead been lopped off midway down, which distorts the relationship between the two bars.
Not all graphs need to start at zero, however — line graphs in particular. This excellent video from Vox explains why:
Numbers showing an embarrassing trend? Turn that frown upside down!
Do those embarrassing numbers just keep going up? Try flipping the axis! Suddenly that rising rate takes a dive and everything is a-okay.
Seriously though, zero only gets to be at the top of your y-axis if your numbers are negative. As the Floridian guns were not bringing people back to life as perforated zombies, this was a clearly bad choice.
Double up to double down
Oh double axis, what can’t you do?
In this case, an unfortunate overlay leads one to believe that in 1990–1991, not only were all New Yorkers killed, but the perpetrators actually found a few extra unfortunates from neighboring cities and offed them too.
Yes, the dual axes are properly labeled and yes, everything is technically displayed correctly, but the graph is still misleading. By putting two elements together, you imply that those elements are on the same scale. At first glance, NYC seems to be a rather dangerous place to live, what with a minimum of 20% of its residents getting killed each year.
This is why I never use dual y-axes.
The author could have put both numbers on the same scale (all data shown in millions, for example), but that would have made the murders nigh invisible along the bottom of the graph.
Another option would have been to instead combine both numbers and display the data as a rate — the number of murders per 100,000 people, for example — giving readers a single accurate trendline to follow.
Labels are always optional
Good news, everyone! Illinois is up to 0.06 hawks!
Okay, this one probably wasn’t intentional. But it does illustrate the need to label your y-axis. Is that a percent increase? Or fractions of a thousand?
As it turns out, this was actually meant to represent the number of hawks observed during the spring bird count, per party hour. …Yeah, that didn’t spring to my mind either.
Bonus points for putting a bird on it, though!
No one reads numbers anyway
Just to prove it isn’t all Yankees making dodgy graphs, here’s a set of charts fresh from the Great White North.
At first blush, it looks like Trudeau is walloping Trump in all categories except “Strong,” where Trump’s bar extends like a verdant middle finger. Yet if you look at the datapoints printed on each bar, you’ll see that Trump is actually four points behind Trudeau.
Once again, the axis is the culprit.
As you know, paired graphs need to be on the same scale to make valid visual comparisons. But these graphs have two different maxima: Trudeau’s graph goes up to 65%, while Trump’s only reaches 50%. This distortion makes Trump’s results look better than they actually are.
I have a few other complaints about these graphs (separate graphs make comparisons more difficult, etc.), but they pale in comparison. Nothing should get in the way of your readers drawing accurate conclusions.
Thanks to Jon for spotting this one in the wild!
Want to revel in more dataviz malarkey?
Check out http://viz.wtf/, a monument to presumably good intentions.
Originally published at hypsypops.com on February 28, 2017.