Let me begin by saying I’m not trying to make a point about gun control. I want to show what happens when you infer a linear relationship where one probably doesn’t exist.
Via Facebook, I got a link to this article by gun control opponent John Lott. The headline sums up his argument: “More Guns = More Murders? A Myth. More Guns = Fewer Murders”
Here’s a snippet:
. . . [T]he New York Times earlier this month put forward the notion: “Generally, if you live in a civilized society, more guns mean more death.” The claim is all over the news from CNN to various “Fact Check” articles.
It would be nice if things were that simple.
The evidence — and there is plenty of it — points to the very opposite, that cutting access to guns mainly disarms law-abiding citizens, making criminals’ lives easier. Guns let potential victims defend themselves when the police aren’t there.
In his article Lott’s evidence is the chart you see here.
Countries like Switzerland and Finland, at the bottom right, have lots of firearms per 100 residents, with low homicide rates. And Estonia, way at the top, has a fairly low firearm ownership rate, but a high homicide rate. The line is a fitted regression of the underlying data. It slopes downward, leading Lott to conclude that the more firearms you have, the fewer homicides.
But that regression line didn’t look right to me, mainly because there are only eight data points above the line, while there are 23 below the line. So I tried to reproduce the chart. I don’t have the underlying data, so I interpolated the points on Lott’s chart. (There’s software that does this, but I don’t have a copy.
My version is what you see below:
I’ve also included the R-square value, 0.0773. Of course, that’s a poor fit for the data. Speaking loosely, it says that handgun density only predicts about 8% of the murder rate.
There’s really not a linear relationship here, and I think Lott is being a bit disingenuous suggesting there is one.
Then I got curious about the title of Lott’s chart, “Gun ownership and annual homicide rates for developed countries (excluding US).” He was writing an editorial about owning guns in the United States. Why would he leave out data regarding the United States?
I’ll acknowledge up front my data point may cover a different year than Lott has modeled. And there may be a mismatch between my two sources. But both are consistent with what I’ve read elsewhere. I’m aware of no trends that would cause my estimate to be significantly off the mark.
Inserting a U.S. data point yields the following graph:
When you include the United States, the slope of the line changes. Now the regression indicates that the more handguns per capita, the higher the homicide rate.
But note the R-square: 0.0255. (So that handgun prevalence only explains 2.5% of the homicide rate.) That’s a worse fit than before.
Realistically, you can’t conclude, based on this data that handguns lead to more crime, or less.
I’m not trying to persuade anyone in any direction on this issue. And I’m sure there are stronger cases to be made on either side.
However, Lott disserves the reader and his cause by relying on this data to make his case.
Incidentally, a much better analysis appears here. (I found it when I was wrapping this post up.) The author appears to be working with the same data Lott worked with. He incorporates a couple of other variables, like income inequality, and reaches a similar conclusion to mine.