Sam Harris, crimes, and statistics

As an addendum to the recent flurry of posts about Sam Harris and gun control, one lesson relates to the dangers in using crime statistics. Those dangers do affect gun control advocates, but I think they have a much greater effect on the arguments of opponents of gun control. That is because the statistical information used by gun control advocates tends to be relatively straightforward – the number of unlawful homicides involving guns in the US is very high compared to other economically advanced countries (though lower by, say, Central American standards). No one seriously disputes this, though different explanations are offered.

If the arguments got into issues of the percentage of murders compared with, say, manslaughters, there might be problems, as there are different crimes, definitions, partial defences, etc., in different jurisdictions. However, unlawful homicides are not the sorts of crimes where overall statistics can be dramatically skewed by such things as over-reporting or under-reporting. In the upshot, the annual number of deaths by unlawful homicide with firearms in the US is very large indeed, and many times larger in per capita terms than those of other economically advanced countries.

Perhaps different murder weapons are favoured in countries where guns are strictly regulated? Perhaps, but the overall unlawful homicide rate is still much higher in the US than in countries that the US might normally want to itself to, and I don’t think anyone seriously disputes this.

In his recent FAQ on the subject of violence and guns, Harris suggests that rape and assault rates are higher in the the UK, Australia, and Sweden than in the US, and even that the much higher reported rape rates in Australia and Sweden suggest we might to well to provide women in those countries with guns for self-defence!

To his credit, Harris has (since?) added a footnote in which he concedes the point that it is difficult to compare figures between countries for rape: “One reader has pointed out that cultural differences in how often rapes get reported, and how they are recorded by different police departments, makes comparing rates of sexual violence between countries problematic. I tend to agree.”

Fair enough – there is no point in berating Harris over an issue where he has made such a concession, and that is not the intention of this post. But it does underline the difficulties of using comparitive statistics with a crime such as rape. There are, notoriously, sociological studies that suggest a very high incidence of rape in the US, together with massive under-reporting of this crime. Even if some of those studies are methodologically flawed and the tendency is to produce exaggerated data, it is widely agreed that rape is an under-reported crime, so much so that differences between jurisdictions in reported rates may have far more to do with police attitudes, public awareness, and the like than with the actual danger of experiencing sexual violence (or even, I hypothesise, than with problems from different definitions).

Unfortunately, something similar applies to assault. Harris correctly observes that assaults can be very serious and harmful crimes, so we should not underplay the importance of assault rates. Fair enough.

But international comparisons are difficult with such a crime. An assault does not leave a body to be accounted for, and it does not even have to leave injuries. Traditionally, if I so much as shake my fist at you, instilling fear that I might attack you, that is a common law assault. Definitions can vary enormously between jurisdictions, as can cultural attitudes to reporting such things as domestic violence and simple fighting, as well as attitudes among police to how they handle these things.

If you doubt that there’s a methodological problem, you can browse the various statistical series provided by the United Nations Office on Drugs and Crime. While homicide rates using firearms are uniformly very low across Europe and consistent over time, assault rates vary enormously over Europe and over time. The latter might mean that some countries which we’d think of as similar are much more dangerous than others, but this really doesn’t pass the smell test.

Perhaps when we look at what appear to be anomalies they sometimes represent real cultural differences. For example, maybe there really is a culture in Scotland that generates a huge number of assaults per year. I’d need to know a lot more to draw a conclusion one way or the other.

But how many anomalies does it take? Do we really think that there are nearly twenty times as many assaults (per capita) in Sweden as in Norway – assuming a common definition of “assault”? Or that the number (per capita) changed to the extent of becoming only one sixth what it had been in Malta between 2006 and 2007? Or that it fell, recently and suddenly, to one twentieth what it previously was in Switzerland? Or even that there are such dramatic cultural differences between Belgium and France as to make the per capita rate of assault in Belgium more than twice as high as in France? Or what about the fact that France’s rate is also only about half of Germany’s, or that Germany’s is currently almost fifteen times as high as Austria’s?

Or to use Australia again, our rate has (supposedly) fallen to much less than half what it was only a decade ago.

When you go through these figures, you have to conclude that international statistical comparisons are, without much deeper criminological research, almost meaningless for a crime like assault.

The lesson is that we need to do more triangulating and digging before we place evidentiary weight on international crime statistics. As long as definitions have not changed within a jurisdiction, they might tell us something about trends over time in that jurisdiction, and they can certainly be more reliable with some crimes than others – more reliable with crimes that are likely to be reported and taken seriously than with crimes that may be under-reported or under-recorded (or, depending on your views about policing of minor assaults, even over-recorded).

More generally, evidence from statistical surveys is useful… but it can also be misleading. It always has to be handled with care, and it is best when triangulated with other evidence.