Some time ago, my brother linked on Facebook an article called "Shooting Down the More Gun, Less Crime Hypothesis."
Much of the study is very sound and completely valid. But the criticisms of the hypothesis that "more guns means less crime" are criticisms common to almost all of social science. There are simply so many variables that controlling all of them is very difficult if not impossible. The application of time-honored techniques like multiple linear regression often fails as a result.
Under laboratory conditions, one can control the environment and see the degree to which variable A responds to changes in variable B. In my line of work, such statistical methods are very common.
As the number of variables increases, the method begins to fail-- for two primary reasons. The first reason is the inability to keep some variables from-- well, varying. The more variables you have, the weaker is the assumption that all other factors are properly controlled. Think of it is the statistical equivalent of "hiss" in the stereo when you crank up the volume. The data is getting "noisier."
The second reason is something "collinearity"-- namely, when two or more variables are affected by another variable in the same direction. For example, one could survey people walking out of the gym about the number of hours they spend working out each week and their approximate Body Mass Index (BMI). You could run a regression that would predict BMI as a function of time spent working out. But this survey was done in front of the gym, remember? How does this statistical model account for the fact that perhaps those going to the gym might eat a better diet? It can't. What about those who don't workout at all but have low BMIs and those who don't workout at all that have high BMIs? Including either of these groups would radically affect the statistical model. This illustrates how even a very simplistic social science statistical study can easily become statistically worthless.
What then when trying to predict something is incredibly complex as the relationship of guns to crime rates? I submit that it cannot be proven scientifically-- it is simply too complex. That means you cannot prove that more guns lower crime rates-- which is the thrust of the article, and the manner in which it is valid.
But the converse is also true. The same problems prevent one from demonstrating that fewer guns (via regulation) reduce crime. The authors of the article do not emphasize the striking difference in this case between absence of proof and proof of absence. The inability to prove that more guns reduces crime doesn't make it true that fewer guns reduces crime. Nor does it mean that more guns doesn't reduce crime-- it just means that it cannot be proven using the methodologies available or agreed to be valid.
People who work with data all the time are familiar with this problem. It's basic 101 for all those in the hard sciences-- which likely explains why the fallacy is prevalent mostly in social sciences.
My brother included some snarky remark with the link about how those who disagree with him essentially don't believe in science or disparage peer review-- because naturally the paper was the last word to him. I won't fault him for a little confirmation bias-- we are all prone to it.
But I will offer up the fact that the paper he linked has been "peer reviewed" by none other than the US District Court of Appeals for the 7th Circuit in the case of Michael Moore vs Lisa Madigan (embedded bellow). The Court said this (in context-- pages 11 and 12):
A few studies find that states that allow concealed carriage of guns outside the home and impose minimal restrictions on obtaining a gun permit have experienced increases in assault rates, though not in homicide rates. See Ian Ayres & John J. Donohue III, “More Guns,Less Crime Fails Again: The Latest Evidence From 1977–2006,” 6 Econ. J. Watch 218, 224 (2009). But it has not been shown that those increases persist. Of another, similar paper by Ayres and Donohue, “ShootingDown the ‘More Guns, Less Crime’ Hypothesis,” 55 Stan. L. Rev. 1193, 1270–85 (2003), it has been said that if they “had extended their analysis by one more year, they would have concluded that these laws [laws allowing concealed handguns to be carried in public] reduce crime.” Carlisle E. Moody & Thomas B. Marvell, “The Debate on Shall-Issue Laws,” 5 Econ. J.Watch 269, 291 (2008). Ayres and Donohue disagree that such laws reduce crime, but they admit that data and modeling problems prevent a strong claim that they increase crime. 55 Stan. L. Rev. at 1281–82, 1286–87;6 Econ. J. Watch at 230–31.
Emphasis mine. Ayres and Donohue essentially cherry-picked the data by confining their study to a time period that would produce the outcome they intended.
The more important point is that any 'study' is invalid if is so sensitive to the sample period as to produce completely different conclusions with even a slightly different sample. That is the essence of the problem. The question cannot be settled empirically. Those who claim they can prove it one way OR the other or deceiving themselves.Michael Moore v. Lisa Madigan