A New Study on Guns and Homicide
There's new research (paywall) from Boston University's Michael B. Siegel and Emily F. Rothman analyzing the relationship between gun ownership and various types of homicide in U.S. states. Covering the period 1981-2013, the study argues that higher gun ownership goes with higher rates of lethal violence among non-strangers, but not among strangers, and the results hold for male and female victims alike. It's the first study to look at gender and the stranger/non-stranger distinction simultaneously.
I can't pretend to be unbiased; I'm skeptical of this line of research and have not been the least bit quiet about it. But this isn't one of those simple studies that a journalism major like me can effectively pick apart. The true test will come when more qualified researchers dig in. Once that happens, the rest of us can make an informed decision about how this work should change our views.
In the meantime, I’ll stick to laying out how the study worked, detailing what it found, and pointing out a few potential pitfalls and policy implications. Fair warning: The following goes into a lot of detail.
The outcome of interest here is homicide, broken out according to the victim's gender, whether or not the killing was accomplished with a gun, and the relationship between the parties. That last part is the tricky one: The FBI does classify offenders as strangers and non-strangers to the deceased — with non-strangers defined to include those "casually acquainted," as well as groups of attackers where even one person knows the victim — but a lot of the data are missing. In no small part, this is because more than 30 percent of murders go unsolved. So the authors use a version of the FBI data where the missing information has been statistically "imputed." Across the entire time period studied, about 78 percent of homicides were committed by non-strangers.
The other key variable is gun ownership, which is also tricky. Most surveys are too small to provide state-level data; the main exception has numbers only for 2001, 2002, and 2004. To address this problem, the authors instead use a "proxy" for gun ownership, a combination of two other variables — the percentage of suicides committed with a gun, and the number of hunting licenses issued in each state.
This is a dramatic improvement over the percentage-of-suicides variable by itself, which has been popular as a proxy for decades. That measure has a nasty habit of correlating with homicide rates even in situations where the actual survey data do not. The new measure has a much tighter correlation with the survey data (technically speaking, r=0.95 vs. 0.80), suggesting it's a closer fit to the real numbers.
There's no denying that hunting licenses represent only a particular type of gun ownership, though. Handguns, as opposed to the long guns favored for hunting, make up an increasing proportion of gun sales — in the late 1990s and early 2000s, the FBI conducted more than twice as many background checks for long-gun sales as for handgun sales, but the gap has disappeared since. Since the proxy was tested against older survey data, it may be a worse fit in more recent years. One gun study incorporated the FBI's background-check data into its gun-ownership proxy, but these data begin in the late 1990s.
Armed with their best approximations of homicide rates and gun ownership, the authors take a whole bunch of other variables into account: "age, gender, race/ethnicity, region, urbanization, poverty, unemployment, income, education, income inequality, divorce rate, alcohol use, nonviolent crime rate, hate crime rate, suicide rate, and incarceration rate." (Some previous studies, including this one that Siegel coauthored, have controlled for broader measures of violent crime, such as the overall violent-crime rate or the robbery rate, instead of including just nonviolent crime and hate crimes.) They also adjust for national trends in homicide rates and use a statistical technique that "accounts for the correlation of data within the same state across time."
Here are their results:
According to this model, when you increase the gun-ownership rate by 10 percentage points (e.g., from 30 percent to 40 percent), you increase total firearm homicides by about 10 percent (e.g., from 4 to 4.4 per 100,000). The results for strangers and non-gun homicides are all statistically insignificant. This jibes with some previous studies, though others disagree. (Here is a very critical literature review from the criminologist Gary Kleck, including a whole section dissecting a previous study coauthored by Siegel.)
A qualm: It's odd to me that, while statistically insignificant individually, all nine of the non-gun results are positive. One might expect higher gun ownership to go with lower non-gun homicide: Guns can't cause non-gun homicides, but guns can deter attackers with inferior weapons, and as guns become more available, murderers may substitute guns for other weapons. There are, no doubt, also cases where the presence of a gun causes a homicide to occur that wouldn't have otherwise — raising the key question of whether the good outweighs the bad. But I tend to think these other effects exist, and this study fails to pick up on them.
Importantly, as the authors note, these results are fairly difficult to reconcile with a theory of "reverse causality" — the idea that homicides in a community cause people to arm themselves, rather than the reverse. (See the above-linked Kleck review for a discussion.) It's certainly true that this happens; the most recent example is the explosion of concealed-carry applications and gun purchases in the San Bernardino area. But one would expect this effect to be strongest with stranger homicides, while the new data show the strongest correlation between gun ownership and non-stranger homicides. Even allowing for the facts that "non-stranger" is defined broadly and there are relatively few stranger homicides in the data, that's a challenge to researchers on the other side of the issue.
The authors also provide a simpler analysis with an interesting chart to go with it. As numerous critics (including David Freddoso and yours truly) have pointed out, if you look at the raw state-level data, you see there's actually no correlation between homicide rates and gun ownership. This isn't the end of the conversation, because a relationship can pop out once you account for other important variables, as the authors do here. (And no, Vox, I never pretended otherwise.) But it is a pretty striking illustration of how complicated the topic is: Whatever effect guns have, it's subtle enough that it doesn't show up as a recognizable pattern in the overall data.
The new study confirms this result for men. Setting aside all the aforementioned statistical wizardry, gun-ownership rates explain a paltry 1.5 percent of the variance in male firearm-homicide rates. But for women it's a different story, with 41 percent of the variance explained:
The obvious criticism is that, even in this simple analysis, they shouldn't just be looking at firearm-homicide rates; they should be looking at total homicide rates. Again: Guns can substitute for other weapons, and they can deter gunless attackers, so gun homicides and non-gun homicides are not independent of each other.
For what it's worth, though, I did a poor man's version of the above analysis and still found a significant correlation between gun ownership and total female homicide rates. (I averaged the 2001-2004 surveys for gun ownership, and matched them up with CDC homicide data from the period 1999 to 2006 — because some states are small and female homicide rates are low, I had to cast a wide net to get usable numbers from the CDC.) In my data, gun ownership explains 15 percent of the variation in female gun homicides and 9 percent of the variation in total female homicides. There was no correlation between gun ownership and female non-gun homicide.
So, those are the results. What are the policy implications? If you believe all of the above claims and see no other reason to value the right to bear arms, a possible conclusion is that the government should deliberately reduce gun ownership somehow. Good luck with that. A milder suggestion would be to require background checks on all gun transfers — including between private parties — though, despite popular support, this has proven difficult to enact at the national level.
If the results for women specifically stand out to you, or if you'd like a politically modest agenda, a more focused approach recommends itself. Men who kill intimate partners, for example, usually don't just do it out of the blue; they have a history of criminal behavior. There are ideas for reform that would target these men specifically. As I wrote in National Review last month:
Hillary Clinton ... would expand the list of misdemeanors that legally disqualify people from gun ownership to include domestic violence against a non-cohabiting partner, as well as stalking. A related proposal, heavily promoted by the Center for American Progress and enacted in some states, would strip gun rights from those under a temporary restraining order, which raises some due-process concerns because such orders are issued without a hearing.
Still another reform is to make sure those convicted of domestic violence actually give up their guns. You can read CAP's 2013 report on these proposals here.
Only about 20 percent of American homicide victims are women — and among female homicide victims, only 40 percent are killed by intimate partners, and only about half are killed with guns. So this won't make a huge dent in the total homicide rate. But every little bit helps, and, importantly, it's hard to argue against disarming domestic abusers.
No one study will end the gun debate, but this latest contribution is an interesting and challenging one. It deserves careful consideration by those in both ideological camps.
Robert VerBruggen is editor of RealClearPolicy. Twitter: @RAVerBruggen