So starting big scope and zooming in, our first issue is that inferences made from the dataset by the author without articulating those assumptions. We do not have crime stats, rather we have REPORTED crime stats in America. More directly, we do not have gun stats in America, but rather reported gun stats.
We don’t even have those. Gun ownership rates in US studies are back-calculated from looking at the ratio of gun suicide to total suicide, since we do have that. A lot of study has gone in to making this back-calculation as accurate as possible, but the fact that ownership rates in general are back-calculated is a point of contention within the pro-gun folks.
I’ve spoke with Michael Siegel, one of the pioneers of this approach, and I tend to think it’s mostly valid. He’s done some good studies about how to control it and make it more accurate. Google around.
That said, these maps weren’t based on ownership rates at all, they were merely reported homicide and suicide stats via the CDC. FWIW, almost all gun deaths fall into one of those two categories. While gun accidents are a thing, they are not a very large piece of the pie.
Basically…..missing persons, unknown causes of death, etc are more prevalent in less populated areas per capita.
I would like to see this figure to determine how important it is in the analysis. You may have a point, but you may not, and how big of a point will depend on the numbers. It’s an intriguing thought though.
Take a “nation” of three people, how many possible relationships are there? Person A, B,C, can each be in their own world, there is two sets of two person relationships, and a single set where everyone knows everyone. That is six possible ways. Taking a linear approach to this, a person may very create an equation in their head that you simply take the number of people and multiply by 2….that is…….number of people (3), times 2, equals 6. The problem is that this equation working is a spurious correlation and in fact, the real equation is a factorial one. So if you have 5 people, the number of combinations is not 10, but rather 5!, or 120.
How does this relate? Expected incidence rates. You should expect as the number of individuals increases, an increased rate in outcomes, not a linear comparison in outcomes. So when you compare rural to urban areas, you should automatically expect increased rates.
I’m not sure I buy this. Suicides happen in a nation of 1. Homicides almost always happen between people who know each other. The factoral math would apply a lot more readily to the spread of contagious disease or similar, but people’s “nation” for the gun space basically caps out at the limit of their social connections. There’s an upper limit that’s experienced no matter where you live.
You can observe this when public events that are attended by mainly individuals from rural areas with low populations….multi-day country music festivals, NASCAR events (this is a great example), you find that these people coming from areas with low crime incident rates congregate, the outcomes are in fact similar to similar sized events consisting of individuals from suburban and urban areas. You can almost think of this like a “critical mass” situation, where a certain amount of people creates a certain outcome….rather than the “types” of people contained within the outcome.
This is another interesting theory, but is it supported by the data? I would like to see the homicide/suicide incidence rate of NASCAR events as compared to the general population pool. I’d also like to see it for Burning Man and for other instances. Even then, you’d have selection bias issues, since the sample set of NASCAR attendees is by its nature limited to the people who can afford NASCAR tickets and often an RV.
The flip side that the author should have investigated is the question as to whether the two groups they selected, the poor urban blacks, and the poor rural whites, how does their behavior compare to the antipose? What is the behavior of the poor rural black, and the poor urban white? These analyses would possibly confirm or refute the authors statements of situations being for example “a poor urban black problem”, or a “poor rural white” situation. It could very well prove that these are steps away removed proxies (instruments) for other more direct causes….for example employment opportunity.
I will ABSOLUTELY GRANT that the analysis in this article, which was done in about three hours on a Saturday afternoon, did not control for socioeconomics at a multivariate level. Those conclusions were my impressions, not only from the mapping but from a lot of other reading into the situation, and my personal experiences. That said, I think I wouldn’t be going too far out on a limb in saying that employment opportunity in rural South Carolina and employment opportunity in rural Kentucky aren’t too different, yet we see a notable differential in suicide rates between those two regions. I highly suspect this is related to culture, “11 Nations” stuff as presented towards the end of the article.
If we truly wanted to show there was no evidence that guns have no impact on aggregate outcomes, we would do the opposite, attempt to show that guns DO have an impact. This could be presented by examining where gun ownership is high to where gun ownership is low, and holding all other things constant (ethnicity makeup, population rates) see is the suicide and gun crime rates are similar.
Other academics have done this sort of thing using state level statistics and found that there is a multivariate correlation, but the correlations are generally weak. Michael Siegel, my favorite gun policy researcher, (he’s anti gun btw) does a good job of this. I critiqued one of his homicide studies here:
I used his study to back-figure a stab at the efficacy and cost of a gun buyback here:
And I looked at one of his studies on suicide here:
Siegel is a good person to follow in this space, in my opinion, whether you’re pro or anti gun.
