I think I noticed this because I’ve been thinking recently about crime and punishment . . . anyway, Gary Wills in this article in the New York Review of Books makes a basic statistical error. Wills writes:
In the most recent year for which figures are available, these are the numbers for firearms homicides:
Ireland 54
Japan 83
Sweden 183
Great Britain 197
Australia 334
Canada 1,034
United States 30,419
But, as we always tell our students, what about the denominator? Ireland only has 4 million people (it had more in the 1800s, incidentally). Yeah, 30,000/(300 million) is still greater than 54/(4 million), but still . . . this is basic stuff. I’m not trying to slam Wills here–numbers are not his job–but shouldn’t the copy editor catch something like that? Flipping it around, if a scientist had written an article for the magazine and had messed up on grammar, I assume the copy editor would’ve fixed it.
The problem, I guess, is that there are a lot more people qualified to be grammar-copy-editors than to be statistics-copy-editors. But still, I think that magazines would do well to hire such people.
P.S. I also noticed the Wills article linked here, although they didn’t seem to notice the table.
While they're at it, they should do TV. Mythbusters rather famously called a yawning experiment significant when they got 27% under treatment and 25% under control with something like 50 subjects.
The Australian count doesn't look right to me. The latest figure I can find is for 1995: 479 firearm deaths in total (accidents, suicides and homicides). Of these, 67 were homicide.
If I look at the numbers of firearm homicides from 1980 to 1995 they are trending down.
So, without knowing what year they are quoting, 334 firearm homicides in Australia seems unlikely to me. Looks more like total firearm deaths.
The Canadian figures also appear to be total deaths from firearms, not just homicides.
The George Wills article does indeed say "homicides". This isn't surprising, since in my experience (most recently a newspaper article just a few weeks ago), gun control advocates quite deliberately quote the figure for total deaths in a way designed to make the reader think it is homicide deaths. They know that most people are a lot more worried about being killed than about killing themselves. They also know that most people consider the idea that restricting guns would substantially reduce homicides to be more a priori plausible than that it would substantially reduce suicides.
It appears that the US numbers are for total firearm deaths (at least according to this site).
http://polyticks.com/polyticks/beararms/liars/wor…
Keep in mind that newspapers often do carry an agenda, and the editors may simply choose to accept numbers that are accurate but not meaningful. In this case, the numbers are definitely gun deaths, not gun homicides, so the heading is absolutely wrong.
But asking every editor in the New York Review of Books to refer stories with numbers to one person before publication is horribly inefficient. Instead, what we seem to need are not statistical editors, but statistical ombudsmen (ombudspersons?) who can receive and properly direct complaints from people like us to keep these violators in line, and isolate the worst cases.
Andrew T.,
I see what you're saying, but my point is that I suspect the do have a copy editor who reads over the articles and finds errors of writing. So why not also have a numbers editor (or share one with a few other publications)? One reason is that the editors probably don't think numerical errors are as important as verbal errors; another reason is that it's probably much more difficult to find an affordable "numbers editor" than a comparable "words editor."
A more charitable interpretation of Wills would be that since he is writing for a fairly well-educated audience, he is counting on them to supply the denominators themselves, and just presenting the raw numbers for their dramatic effect which would be diluted by giving rates instead. A rate is often perceived as just a statistic, while most people can actually imagine what 200 people look like (all the people at a local restaurant for me, on a busy evening) or 30,000 people (all the people at a sports stadium, roughly speaking. I agree, though, that in general numbers should be checked more carefully in jounalism.
Abbas,
Well, yeah, but that's exactly why I don't like the raw numbers: the "dramatic effect" is partly an unearned product of the denominator. Why not really go for the dramatic effect and compare the number of murders in the U.S. in 1995 to the number of murders in Ireland in the past 5 minutes–that would really sharpen the comparison!
Yes, I agree that people can have difficulty interpreting rates. (For me, I find "rates per 100,000" very confusing–I'd much rather have rates per million.) So I don't know that there's an easy answer. But comparing raw numbers in this case can't be right.
It's not like it's easy to find a competent regular editor either … Anybody who is qualified to do the job properly (not just wield the Strunk&White, but also checking background facts, assessing the integrity of the sources, etc) would hopefully also catch this kind of error. But they don't exist.