Thanks for your many excellent comments on Guns and States. I’ve made it through most of the comments thread and learned a lot. Here are some (hopefully) final thoughts on the matter:
1. In yesterday’s post, I suggested that the difference in homicide rates between America and other First World countries were about two-thirds cultural, one-third gun-related. That’s sort of true, but people have reminded me to think of it as an interaction. Without the cultural factors in place, guns are pretty harmless. That’s why Wyoming can be the highest-gun-ownership state in the country, with 60% of households having a weapon, and still have a murder rate equal to Canada’s and lower than 45 other states’. With the cultural factors in place, guns make a bad situation worse. This is why the robbery rate and the gun homicide rate correlate at about 0.9 (0.75 if DC is removed as an outlier) even though most robberies do not involve firearms.
2. But some robberies do involve firearms, making them an imperfect control. I looked at rape, a crime that sounds like it should be associated with other violent crimes but which almost never involves firearms. Unfortunately, the results were really weird:
As you can see, rape is negatively correlated with robbery and %black, insignificantly correlated with homicide and urbanness, and positively correlated with gun ownership and suicide rate (which itself is heavily correlated with gun ownership). I also added one more variable in an attempt to explain some of this: MFRatio, which is the number of men per 100 women in the state. That certainly mattered a lot, but doesn’t take away from the confusingness of the other variables. My first theory was that rape is more common in Red Tribe rural white culture, which would explain the guns and %black but really doesn’t fit with the PUrban or Southernness issues, nor the fact that the correlation with guns was much higher than the racial correlation. My second was that people bought guns to defend against rape, but that doesn’t explain the negative correlation with robbery, and surely people would buy guns based on crime in general (and since it’s mostly men doing the gun-buying, rape would be the least likely to affect the gun rate). Overall I admit I am confused.
3. Commenter eccdogg can’t replicate the Southern culture of violence thing after controlling for other factors. I acknowledge that he seems to be doing the right statistics, but can’t square that with commonsense eyeballing of the data.
4. I did a reanalysis that found that (after adjusting for confounders) the gun data from 2002 successfully predicted the murder rate in 2002, but did not successfully predict the murder rate today, even though most other relevant variables did. I’m going to attribute this to gun ownership patterns changing faster than (say) whether a state is Southern or not, and probably if I had gun ownership data from today (which I don’t) it would work, but it seemed important to mention that and get it out there.
5. I was shocked to see that between 20 to 30 percent of people in most European/Anglosphere countries owned guns, including the ones like Canada that gun control advocates hold up as an example of what they want. That makes it very strange the degree to which people expect gun control to mean “the government confiscates all your guns, after which no guns are left”. I mean, I understand why (for example) the NRA would promote that story in order to get people angry, but it’s very strange how often liberals nod along and say “Yup, that’s what we want to do!”. I wish more of the debate could be about waiting periods, required training classes, and background checks – none of which would prevent people from using guns to defend against a potentially tyrannical government. (I don’t know enough about guns to be know whether restricting handguns would make it harder to defend against tyranny. Could you shoot the tyrants equally well with a rifle?)
6. Commenter Elias brings up this meta-analysis by Gary Kleck claiming that the 40 guns-and-states style studies he could find were split almost exactly half and half in terms of whether they found a significant guns-homicide correlation or not. He further claimed that the better the study, the less likely it was to find a significant correlation. The study I cited, plus my own analyses, passed two of Kleck’s tests – used good gun ownership numbers and controlled for confounders – but failed the third, which was distinguishing forward causation from reverse causation. I admitted this was a problem and that it would take very powerful methods beyond my abilities to solve, but I also said that I suspected forward causation since the homicide causation remained independent of robbery and I would expect people to buy guns based on a general crime rate. I stick to that suspicion. Kleck has some studies that he thinks establish reverse causation instead, and I encourage you to read them.
This reminds me a lot of my post Beware The Man Of One Study – I can at least say in my own defense that I did the analysis on my own, which should obviate exactly which studies I did or didn’t look at, but the media articles that just presented MA&H and said “Look! Here’s what Science says, discussion over!” really should have mentioned that there were also nineteen studies, 48% of the total, that disagreed with them.
On the other hand, Kleck said that only three studies met his quality standards, and all of them were his own, so take that with however much salt you feel like it deserves.