FTA: In 2014, the most recent year that a county level breakdown is available, 54% of counties (with 11% of the population) have no murders. 69% of counties have no more than one murder, and about 20% of the population. These counties account for only 4% of all murders in the country. The worst 1% of counties have 19% of the population and 37% of the murders. The worst 5% of counties contain 47% of the population and account for 68% of murders.

When you break it down that way it's a less strong correlation, but tends to support a model based on 'network effects' (opportunity) and/or social pressures (density).

It would be really interesting to slice things and compare relative rates for neighborhood level housing density, poverty/affluence, and education.

As a potential control study though, I'd like to propose focusing on quality of housing and 'tranquility' factors. How well does that housing prevent problematic interaction? (Noisy / smelly (air pollution of various kinds) / resource conflicts) My hypothesis is that higher quality housing and housing codes can contribute to decreasing stress. The need to build more housing to fulfill the more lofty living requirements would also have a positive effect on the affordability of housing (and naturally for the experiment would need to be targeted across all levels of need). Of course to eliminate the effect of housing merely being new at least some units would have to be new but built to the existing standard's minimums rather than the study's re-defined minimums.

What would the statistics look like if they used equal-area regions? And what would they look like if they used equal-population regions (maybe Congressional districts)?

I guess - mind, this is an uneducated guess - that population density is more of a marker here than sheer population. (Or area!)

After all, high population density leads to more interactions with other people, a small portion of which escalate to violence.

Also, if I was aiming for a career in violent crime, I'd pick a place with lots of potential victims. After all, mugging the other inhabitant of East Nowhere, Alaska pretty much means you've cornered that market and need to move elsewhere.

(The underlying assumption is that the percentage of people murdered by their partner or close friends is pretty much even across the country, and that what varies is the - hm - opportunistic murder rate.)

> mugging the other inhabitant of East Nowhere, Alaska

You might also find yourself having an "accident" that makes you bear food pretty quick.

To your point, being a career criminal in a dense urban area also probably increases the likelihood you won't be caught after committing a crime and be able to go on to commit others.

Population density being correlated with crime rate is more mysterious than it would seem at first. After all, people live in cities because many things benefit from economies of scale. Why doesn't policing?

First of all, it seems to make sense that crime also benefits from the same economies of scale. This could offset the improvement in policing.

Secondly, policing becomes harder in a city due to increased anonymity and the ease of hiding in plain sight. As a city grows, more and more people don't know each other. This reduces the effectiveness of social policing.

Is policing the thing that stops crime? There are obviously many parts to it including policing but a lot of things have gone wrong before the police were needed.

It's much less mysterious when you consider socioeconomic factors. In particular, they concentrate in neighborhoods which are vestiges of midcentury economies of scale (i.e. factory jobs), and are now hopeless. Urban police are pretty good at stopping violent crime from leaking into white professional neighborhoods, which is arguably city institutions' true priority.

Police tactics are only going to get you so far in communities that (justifiably) see the police as an oppressive enemy faction, where witnesses don't talk to detectives and juries don't convict.

Also, residents of East Nowhere tend to carry guns and shorten the careers of professional muggers dramatically.

Doing stats per county is useful when you have a gun control debate as gun issuance policy is specific to each county.

0.05% would be 1:2000, or 320M/2k = 160.000 murders a year. The USA is violent, but not _that_ violent. “4.4 per 100,000 people” is closer to 999.995% and 0.005%.

The peak counties (New Orleans, for example. See https://en.wikipedia.org/wiki/Crime_in_the_United_States#Hom...) do get that many murders.

That’s at least partly because “by population” isn’t a good metric. People who sleep or aren’t in the company of others don’t murder people, so you have to look where they are when they are awake and in the company of others; that happens at home, but also at work, in bars, on holiday, etc. and the latter predominantly is in larger cities.

You're right, I dropped a zero.

Also, this looks like a case of Zipf's Law, which tends to crop up a lot if you express statistical distributions as rankings [1].

[1] https://en.wikipedia.org/wiki/Zipf%27s_law#Statistical_expla...

Except that over half of those 0.05% of murder victims lived in close proximity to one another Uber similar socioeconomic conditions. It is very much correlated with geography, which is why people pay absurd prices to live in areas with reduced crime.