Guns are (probably) not responsible for the high murder rates in the US: how to understand the stats
by Alan Cohen
As I stated in my last post, I support gun control, and I am really appalled by what happened in Newtown. But this is one (rare) case where I think the people on the right may understand the facts better than those in the center and the left. Two of my favorite columnists, Charles Blow at the NY Times and Fareed Zakaria at the Washington Post, had recent columns (here and here) on how the US is the clear exception in the developed world on both gun ownership and murder rates, and suggesting that solution is simple: limit guns, have less murders.
These are nice, well-argued columns. Unfortunately, they appear to be wrong. I’ve been digging in the data a bit, and I think both of these columnists have taken correlation to be causality. (Note: Correlation can sometimes imply causality, contrary to popular belief – one needs to know how to do the right analyses and have the right data. But this is not the case here.) So I will work you through the data, as an example of how to do a more thorough statistical analysis of a question like this. Don’t worry, I won’t get too technical, and I’ll keep it intuitive…
The first step is to decide what data to collect. We are comparing across countries here, specifically, the 31 OECD countries. So we need country-level data. We want to understand what influences total murder rates. This is the important number, since it doesn’t do us much good if people start murdering each other with knives instead of guns, but at the same rate. We need rates, not totals, because population sizes are different. It’s not really fair to compare 10,000 US murders with 70 in Estonia, since Estonia only has 1.3 million compared to 300 million in the US. Rates are expressed as the number of homicides per 100,000 people per year.
What we want to understand is the factors influencing total homicide rate. We will further break this down into gun homicides and non-gun homicides. We also need data on gun ownership. I got all these data from Charles Blow’s column, calculating non-gun homicides by subtracting them from the total (which is actually measured as assault deaths). However, gun ownership isn’t the only thing that might influence homicide rates across countries. Poorer countries appear to have more homicides, as do ones with more income inequality. So I looked up two measures of income inequality (the Gini coefficient and the income ratio of the 10% richest to the 10% poorest, from Wikipedia). I also looked up per capita GNI (Gross National Income, the new word for GDP, from the World Bank).
There are lots of other things that could affect murder rates – culture, detailed aspects of gun laws, alcohol consumption – but many of these are hard to measure or hard to get data for. They might be important and we shouldn’t forget them, but I’m not enough of an expert to be able to include them here. And I can show, with just the gun, crime, and economic data, that just reducing gun ownership in the US would be unlikely to solve our problems.
The first step is just to look at a histogram of each variable. This shows us where the US is relative to the other countries:
The height of each bar indicates the number of countries in that range. We clearly see that the US is an extreme outlier for the gun murder rate, the total murder rate, and gun ownership. This is what was shown by Zakaria and Blow in their columns, and from which they drew the conclusion that simply getting rid of guns would solve the problem. But look: The US is also at the extreme high end of income inequality. We have pretty much the largest gap between rich and poor of any developed country. Lots of poor people, especially marginalized poor people in ghettos, leads to crime. So it’s not clear if it’s the guns, the income inequality, or both.
For per capita GNI, the US is in the middle of the range, though not many countries are at the upper end. But the really interesting histogram is the non-gun murder rate: look, the US is still in second place (that’s Estonia up there in first), even for all the murders committed without guns. This is a pretty strong argument against the simplistic idea that getting rid of guns would pull our murder rates down to those of other OECD countries.
We can go deeper. Let’s look at how total murder rates correlate with non-gun murder rates:
This plot makes it clear just how much of an anomaly the US is. We are way outside the range of what’s happening in other countries. There, the correlation is almost perfect: the total murder rate allows us to predict with great accuracy the rate of non-gun murders, and vice-versa. So the question is, why is the US different? Is it because a lot of murders that would have been committed anyway are committed with guns (thus driving down the non-gun murder rate artificially), or because a lot more total murders happen because of guns (thus driving up the total murder rate):
In the former case, were we to dramatically reduce gun availability so as to fall in line with other countries, we would have little change in the total murder rate, but we would force up the non-gun murder rate (the vertical arrow above). In the latter case, dramatically reducing gun availability would drive down the total murder rate, while the non-gun murder rate stayed constant (the horizontal arrow). Or the truth could lie in between – these data alone don’t tell us.
There are also other interesting clues in the data:
If we put all the countries together, there is a relatively strong correlation between gun ownership and total murder rate, concordant with Blow and Zakaria’s argument and suggesting but not proving a relationship. But if we take the US out, there is absolutely no relationship whatsoever. And as we know, the US is anomalous in a bunch of ways, so we probably should not include it. The evidence from the other countries alone would suggest that gun ownership has no relationship whatsoever to homicide rates. Any correlation that depends on one data point is not really a correlation, just an anecdote.
The next step in a thorough analysis is a regression model. This allows us to see if the correlations we see above are mediated by some of the other variables. For example, if murder rates are lower in rich countries and gun ownership is low in rich countries, maybe gun ownership has no effect of its own – it just appears to have one because it is associated with per capita GNI. Regression models help us answer this kind of question.
I won’t bore you with the details, but the results are pretty clear: when we include the US in the model, total murder rates are predicted by gun ownership AND income inequality AND per capita GNI – each one of these things appears to play a role, even when controlling for the others. But when we exclude the US, the overall model performs poorly and only per capita GNI appears to have a (weak) association with total murder rates.
So this is telling us the same thing again: when the US is included, all these things seem important, but when it’s not they don’t. The US is anomalous in many ways: in addition to those listed above, there is more racial tension, more intergenerational poverty in ghettos and in some rural areas, more a culture of mistrusting government, more a culture of violence than in the other OECD countries. Based only on the lessons we can learn from the other countries without the US, there is no reason to suspect that limiting gun ownership would decrease overall murder rates.
But what the data are really telling us is that we don’t know: for the things that are relevant to murder rates, the other countries don’t provide much guidance because the US is too different. (I’m not an American exceptionalist, and it’s rare that I would utter that phrase!) It’s possible that by limiting guns we would change the culture of violence and drive down both gun and non-gun murder rates. It’s possible that we would increase paranoia and gun hoarding and drive up mass shootings without decreasing normal murders. My best guess? Limiting guns would make a small dent in murder rates, but would not come anywhere close to bringing us down to the level of other countries. If it were that simple, US non-gun murder rates would not be so abnormally high.
Is gun control still a good idea? Yup. Why not save some lives if we can. But let’s not pretend that we can get the US down to several hundred homicides per year just by passing a few gun laws. Instead, let’s invest in health care, education, and anti-poverty initiatives, the long-term solutions to poverty and crime.
Thoughts? Comments? Disagree? Let me know in the comments… Here are the raw data, in case you want them.
Country | Firearm homicide rate | Assault death rate | Guns per 100 | Gini coefficient Late 2000s Wikipedia | UN RP 10% Wiki | per capita GNI 2010 (world bank) |
Australia | 0.1 | 1 | 15 | 336 | 12.5 | 46200 |
Austria | 0.2 | 0.5 | 30.4 | 261 | 6.9 | 46920 |
Belgium | 0.7 | 1.4 | 17.2 | 259 | 8.2 | 45780 |
Canada | 0.5 | 1.6 | 30.8 | 324 | 9.4 | 43250 |
Czech Republic | 0.2 | 0.8 | 16.3 | 256 | 5.2 | 18490 |
Denmark | 0.3 | 0.7 | 12 | 248 | 8.1 | 59410 |
Estonia | 0.2 | 4.5 | 9.2 | 315 | 10.8 | 14180 |
Finland | 0.4 | 1.9 | 45.3 | 259 | 5.6 | 47460 |
France | 0.1 | 0.8 | 31.2 | 293 | 9.1 | 42190 |
Germany | 0.2 | 0.6 | 30.3 | 295 | 6.9 | 42970 |
Greece | 0.3 | 1.4 | 22.5 | 307 | 10.2 | 26890 |
Hungary | 0.1 | 1.4 | 5.5 | 272 | 5.5 | 12860 |
Iceland | 0 | 0.3 | 30.3 | 301 | NA | 33890 |
Ireland | 0.5 | 0.8 | 8.6 | 293 | 9.4 | 41720 |
Israel | 0.1 | 2.1 | 7.3 | 371 | 13.4 | 27270 |
Italy | 0.7 | 0.8 | 11.9 | 337 | 11.6 | 35530 |
Japan | 0 | 0.3 | 0.6 | 329 | 4.5 | 42050 |
Luxembourg | 0.6 | 1.8 | 15.3 | 288 | NA | 76820 |
Netherlands | 0.3 | 0.9 | 3.9 | 294 | 9.2 | 48920 |
New Zealand | 0.2 | 1.3 | 22.6 | 330 | 12.4 | 29350 |
Norway | 0.1 | 0.7 | 31.3 | 250 | 6.1 | 86390 |
Poland | 0.1 | 0.9 | 1.3 | 305 | 8.8 | 12450 |
Portugal | 0.4 | 1.2 | 8.5 | 353 | 15 | 21830 |
Slovakia | 0.2 | 1.2 | 8.3 | 257 | 6.7 | 16030 |
Slovenia | 0.1 | 0.4 | 13.5 | 236 | 2.9 | 23910 |
South Korea | 0 | 1.2 | 1.1 | 315 | 7.8 | 19720 |
Spain | 0.2 | 0.7 | 10.4 | 317 | 10.3 | 31460 |
Sweden | 0.4 | 1 | 31.6 | 259 | 6.2 | 50580 |
Switzerland | 0.8 | 0.6 | 45.7 | 303 | 9 | 71590 |
United Kingdom | 0.1 | 0.3 | 6 | 345 | 13.8 | 38140 |
United States | 3.2 | 5.7 | 88.8 | 378 | 15.9 | 47350 |
Very nice initial analysis! I should adopt a more casual attitude myself about doing these kinds of things. You make a really important point that almost never comes out in the usual discussion, and it’s a great example of outliers and influential points.
You may have some ideas on how to account for the opposite effect: people buying guns precisely because crime is high.
Interesting point about controlling for purchases caused by high crime (reverse causality here). This suggests the possibility of a vicious cycle: lots of guns leads to more crime, leads to more guns, leads to more crime… I have no great ideas on how to deal with this, unless there is a longitudinal data set that lets us follow the numbers over time. In this case, we could apply structural equations models and truly infer causation from correlation… I think the data set would have to be at a more local scale, though, because the national stats will be influenced by too many other things.
I agree completely. I think there’s way too much national-level data in social science, and this is a good example of why.
I stumbled upon this and appreciate the time, effort, and the math! It would be interesting to add some more variables and see how they play out. Other things that make the US unique is a limited access to healthcare and mental healthcare, and ironically the over abundance of medicated children. I am very interested in those two statistics and how they relate to homicides, after all one of the side effects of ADHD medication is “hearing voices” and mass shootings at least do correlate well with the popularity of ADHD and anti-drepressants, i.e. 1990s and on…
Interesting, and I fully agree about the possibility for health care and mental health related causes. However, I can’t undertake the analyses myself, for a few reasons. First, mass shootings are a drop in the bucket compared to total homicides, and we wouldn’t have enough sample size. Total homicides have been going down over this time, so that doesn’t seem like a fruitful avenue to pursue. Second, I don’t think the country-level data will help – there are too many confounding factors, and once again I expect the US will be an outlier and so we won’t be able to make much out. We’d need prescriptions by region of the US, and I don’t know how to get that data easily. Even if we had it, we’d have to carefully control for cultural factors that associate with prescription level and gun ownership across regions… Not an easy analysis to undertake, but an interesting one.
[…] 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 […]
Your charts are all broken, at least in my browser (updated Chrome).
Thanks for the head up. I think it’s fixed now. Please let me know if you continue to have problems! Jpegs are low quality, but the beautiful EMFs produced by R are not compatible with WordPress, and I’m not a graphics format whiz…
How about .png?
PNG was the format that was causing the problems – I wasn’t even able to try EMF. It seems OK with JPG now, so that’s fine.
The number of households in Canada that have guns is roughly the same as the US. There are more guns per household in the US.
And yet the homocide rate is 3x higher in the US.
This adds more weight to the socioeconomic and gang related factors.
Canada’s homocide rate is similar to England, where gun ownership is very low. Mr. Pearce Morgan may want to comment on that the next time he concludes that its all about the guns.
Responsible gun owing Canadian
I certainly agree that gun ownership does imply high crime rates in all cases, as supported by the general lack of association between ownership rates and murder rates across countries (excluding the US). Whether reducing gun ownership in the US would reduce murder rates there is a more complicated question. I think it likely that a very large reduction in guns available in the marketplace would have a very small impact on the murder rate, but the legislation recently rejected by the US congress wouldn’t have done nearly enough to make a noticeable difference. it was just symbolic.
What I’ve found in years of researching, discussing, and weighing the findings on this particular topic is that social cues are left out of analyses like this. A country with a stationary population who all know their neighbors, like Switzerland, are very unlikely to use their (fairly high) access to firearms to embark on a crime spree. A country like Mexico with very strict (draconian, even) gun laws which STILL has notable gun ownership does so because the law is corrupt, and the people face multiple prohibition-driven conflicts. Comparing locations with differing laws and availabilities here WITHIN the US allows for greater assurances to the methodology and the sociology.
I would posit that the locations with the most turbulent conflict over malum prohibitum laws here in the US would show intense levels of violence, and those without would show far less. Now, today’s prohibitions are the drug trade and, ironically, guns themselves. The places here in the US with the highest level of restriction, and “all else being equal” (as much as possible else, anyways) have the highest murder rates. Of course, hospitals will tell you that many of them are “repeat clients”, which tends to mean people involved in gang activity (to be expected) or drugs. The good news is that now we begin to have a few locations in which at least SOME drugs (the most common one) and within a few years should have new data to bandy about.
I have my faith in guns still, mostly because it’s too hard to keep the whole cop with you, when you need them.
Seriously, have you seen how much they eat? I think I know why some of them get nicknamed pigs!
Interesting points. I completely agree that culture is much more important than laws in determining gun murder rates. In that sense, crude international (or even inter-state) comparisons are not that useful because the key question is always the counterfactual: what would the murder rate be HERE if we changed the law? (choose your HERE, and what legal change) Culture is much harder to change than laws, so even if the laws only account for a small portion of the difference, it still might be worthwhile to impose a more restrictive gun law (assuming a utilitarian, save-the-most-lives perspective).
You are certainly right that murder rates are higher where guns laws are more restrictive, but this seems to me clearly putting the cart before the horse. In places with high murder rates, people feel compelled to try to do something to stop it, whether or not it is effective. Thus, big cities with gangs, poverty, and high murder rates impose restrictive gun laws. Rural states (even rural blue states like Vermont, if I’m not mistaken) have much looser gun laws.
Your last remark, about having faith in guns because you can’t keep the whole cop with you, ignores the risk of having guns to their owners. Again assuming the utilitarian perspective, it is an empirical question whether owning a gun is more likely to result in harm to the owner (or other innocents) versus to save the owner (or other innocents). Personally, I don’t really care if I die from an accidental gunshot versus as a homicide victim – I just would like to avoid dying. However, the big picture stats that suggest there is more risk to the owner may paint too general a picture. I could imagine that in certain neighborhoods one might be safer owning a gun, and in others, safer not owning one. The irony is that the gun rights advocates mostly live in places where they are safe and thus likely to die from accidents, and the gun control advocates live in places where they might be safer owning guns. But again, these are empirical questions not firmly answered, and we should fun rigorous, unbiased research to tackle the questions.
As far as I can tell gun restrictions are stronger in urban areas. For probably different reasons (maybe lead exposure), urban areas have had more crime in recent decades. So any connection between gun laws and high crime is likely to be an artifact of population density.
I agree, though I have my doubts that lead can explain that all much. I think intergenerational poverty and a lack of cultural institutions to bring people out of it drives the high crime.
If you go to this site :-
http://www.gunpolicy.org/firearms/region/united-states
You can see the quantities of different categories of guns owned in the the USA and Canada (or any other country) as well as the number of murders carried out using the different category of gun. The point I’m slowly getting to (LOL) is that Canada has rifles as a much higher proportion of its firearms, while the USA has a much higher proportion of handguns, and handguns would appear to be the murder weapon of choice in the USA.
Canada total civilian guns : 9,950,000
Of which 1,100,000 are handguns
USA total civilian guns : ~290,000,000
Of which 114,000,000 are handguns
I wasn’t aware of that, but it makes perfect sense. Of course, the type of gun one buys reflects its intended use. Hunters are unlikely to prefer handguns, and street criminals unlikely to prefer rifles, to over-simplify. But causality can go the other direction too: because the handgun is there, it might get used as a murder weapon. Less so for the rifle.
However, the US is so extreme an outlier that I doubt it really makes much difference in the figures above if one uses handguns or total guns (I haven’t rerun the numbers…)
[…] new lobbying group that can stand up to the NRA. Of course, handguns might not be the cause of America’s high murder rate […]