Due to the recent shooting in Colorado, it seemed appropriate to look at an older post from the Atlantic on the Geography of Gun Deaths by Richard Florida.
These emotional moments are often points when people rally for policy changes, and I thought it might be interesting to look at the discussion surrounding data about gun related deaths.
Richard Florida’s mapping exercise is interesting for a number of reasons, most of which go beyond the data presented. The responses from readers both on the original post and the Daily Beast’s recent summary make for some interesting and somewhat troubling reading, partially due to the tenor of the comments and the general lack of statistical acumen. Understanding math is important!
The author, Noah Kristula-Green, of the Daily Beast post tried to leave the political bits out, but the comments very quickly descend into political mud-slinging. Conservatives decry the maps as being liberal misdirection and the liberals claim the conservatives are being racist and trying to blame shootings on people of color, especially African-Americans. Richard Florida doesn’t go into the political aspect much in the original post, noting astutely that it is likely to infuriate people (possibly why Noah didn’t go out of his way to point it out), but it is striking that the strongest correlation Richard found was between deaths by gun and states that voted for McCain in 2008. Richard’s description of the correlation:
What about politics? It’s hard to quantify political rhetoric, but we can distinguish blue from red states. Taking the voting patterns from the 2008 presidential election, we found a striking pattern: Firearm-related deaths were positively associated with states that voted for McCain (.66) and negatively associated with states that voted for Obama (-.66). Though this association is likely to infuriate many people, the statistics are unmistakable. Partisan affiliations alone cannot explain them; most likely they stem from two broader, underlying factors – the economic and employment makeup of the states and their policies toward guns and gun ownership.
The most important thing to note about the comments is that most of the commenters whether liberal or conservative seemed to not get correlation, and fall into the trap of equating it to causation. Many saying that the correlations are drawing liberal conclusions. Statistics are tricky things, as a statistical guru, Dr. Hadley, once told me…
“The thing about statistics is that 100% of water drinkers die.”
That is one of my favorite quotes and serves as a constant reminder that just because two things are connected, does not mean one causes the other. Richard Florida perhaps put too much faith in his readership by saying “As usual, I point out that correlation does not imply causation, but simply points to associations between variables.” Even those who know the definitions of correlation and causation often conflate them.
In statistics, correlation is the degree to which attributes of a group vary together, while causation is when a variable produces a specific effect. Simply put, correlations are clues to larger truths and connections, they are not the truth and should not be assumed to be. Correlations point the way for further investigation and untangling of issues.
As an an example, in this instance one could use the correlation of political affinity to deaths by gun to argue many things:
- that people tired of being menaced by violent crime favored McCain because he was perceived as supporting more social control
- that these states idealize a cowboy attitude exemplified in lax gun control laws and supported McCain because he was perceived as being more likely to protect their gun-toting ways
- that Republicans own guns in those states and Democrats were too afraid to vote for fear of being shot on their way to the voting centers
- that Democrats in those states that voted republican all own guns and were too busy out shooting to vote
These assumptions all can be supported by the correlation, are very different, and are most likely wrong. There are probably much more complex connections between politics and shooting deaths. Based on the limited information available, the myriad reasons people choose who to vote for, and the potential number of reasons for purchasing guns and committing acts of violence, Richard was correct to not make assumptions about why the correlation is there.
The other thing that I found interesting about the commenters is that there is a city/country divide, with many people calling out the authors for not breaking down the data by county, so that we could all see that gun related deaths occur in urban areas (where all the liberals live or where all the people of color live according to the responders). I find this city mouse /country mouse divide interesting because it has been around for soooo long, and seems to be back in the zeitgeist because of the 2010 census showing a shift in population flow (now toward cities, except that isn’t actually the whole story with significant numbers of people of color moving out of cities and into suburbs). Also the data is based on the location of the crime, which is not necessarily the location of the home of the person committing the crime, which indicates an assumption that persons committing these violent crimes do so only in their home counties. Perhaps then it would make more sense to see a county level map that indicates the location of the home of the shooter next to a map of the location of the crimes. I am not sure that this specific data is tabulated and might require considerable time and effort to gather.
Boundaries are mostly artificial. Fussing over them in the comments appears to indicate a strong desire amongst the readers to make strict sense of a dynamic system that is constantly changing. Which I can sympathize with, but also think is a bit naive (to put it nicely). County boundaries themselves possibly made sense at the time they were made, but it is rare and gratifying when they follow a natural or logical path. For instance, the County of New York which should just be an island, actually includes Marble Hill, a tiny bit of the mainland, because the natural boundaries (rivers) changed. In some cases the lines between states are not even clear.
It would be interesting to see some comparisons across similar population densities nationally (comparison of all urban areas or all non-urban areas). However, once one starts making decisions about what constitutes an urban area or other designation there is going to be some bias in where boundaries are drawn and limitations on data available and how it is collected. Does one include Newark, NJ in the NYC metro area or not, and does Newark tabulate the same data the same way that New York City does? Ultimately the choice of state over county boundaries has a lot to do with how much and how accurate the data available is, how much time you have to collect the data and do all the calculations, and how arbitrary or biased the boundaries are.
The commenter-identified purpose of this finer grained analysis is puzzling to me: To argue against a generalization (showing data by state) with a different generalization (that liberals live in cities and conservatives in the country) is not logical. There are conservatives in cities, and there are liberals that live in the country. Since gun control laws are generally enacted on a state, not county, level, adding county level information is not necessarily clarifying unless one wanted to look at trends within a state. While looking at county level data might eliminate some of the urban bias commenters were concerned with, it will not account for the non-urban bias: it is less likely that violent deaths by gun will occur if there is no one around to shoot. Looking at state level data may therefore be more desirable since both biases are included and can generally balance each other out, plus the data follows the same boundaries as the laws, making it the least arbitrary.
Sociology is hard because it is not mathematically straightforward in a scientific way like Newtonian physics is: there are just too many variables when it comes to humans. So when reading these types of articles, one should take it for what it is, a general picture of general trends and tendencies that need further analysis to really be understood. When taken with more in-depth studies, there is a limited role for this type of information to play in policy development. That does not make this or similar articles useless, it makes them starting points for asking questions and further studies, which is certainly extremely valuable.
As a final note, hands down the most striking thing about this data was that they did not test for correlation to gun ownership. Which might seem overly obvious, but, hey, it is usually the most obvious things that make the most sense.