Overview - Guns and Crime

Like so many other far-right special interests, the gun lobby isn't known for the quality of its scholarship, and their attempts to "scientifically" justify their beliefs regarding firearms and crime are a case in point. The United States has what is indisputably the highest per-capita rate of armed crime in the Developed World, even surpassing that of many unstable Third World nations. One of the most central battlefields in public debate about guns and crime is whether guns fuel this trend or mitigate it. Both sides claim "indisputable" evidence for their cases. In fact, for the most part, criminal and demographic data are equivocal as to whether guns increase or decrease crime rates. One or two high profile studies by gun opponents and proponents have attempted to argue otherwise and gained vocal support from one side or the other. In virtually every case, these studies have been found to be based on flawed analyses and/or datasets. To date, the most notable of these is probably that of John Lott, former Olin fellow and professor of Sociology at Yale and as of this writing, a resident scholar at the American Enterprise Institute.

In 1997, Lott and co-author David Mustard published the results of a statistical analysis in which they claimed to have demonstrated that gun ownership in America (as tracked by "Shall Issue" laws) correlated negatively with violent crime rates. Shortly thereafter, Lott published a book titled "More Guns, Less Crime" based on this research in which he argued that a one percent increase in U.S. gun ownership results in a 3.3 percent decrease in homicide rates. Lott's analysis was based on an "econometric" math model.

Econometric models are built from large datasets of variables that are directly or indirectly related (or thought to be related) to a trend, or quantity of interest. These datasets are then subjected to a type of statistical analysis called "multiple regression" that tests for correlations between variables that might be related to underlying causal relationships. The mathematics behind these methods are sound, but their reliability is directly tied to how well they cover the range of independent variables used, the quality the data used as input, and the degree to which the independent variables tested are truly independent of each other. The last point is particularly relevant to Lott’s work.

Econometric models are popular for the study of many social trends, including crime. But the sheer complexity of these makes it extraordinarily difficult to build econometric models that truly meet both of these requirements. Almost without exception, this type of analysis results in "clustered" data—that is, data in which hidden correlations between variables make the number of "independent" data point used seem much larger than it actually is. This in turn leads to an underestimate of how "noisy" the measurements are and makes statistically insignificant (random chance) results seem like real signals of a correlation.

Shortly after Lott and Mustard published their econometric analysis was found to be riddled with data clustering of this sort, leading them to treat connections between gun ownership and crime rates that were random as though they were real correlations. They had simply experimented with different dataset descriptions and gathered data inputs until they hit on a combination that produced the negative "correlation" they wanted. Within 12 months of its publication another team of researchers reran Lott and Mustard's model using the same data they had used, and showed that the removal of a single data point (namely, a county in Florida) caused their results to disappear. As if this weren't enough, it was then discovered that the Lott/Mustard model had basic computer coding errors in it and was not even calculating its regressions correctly.

Lott is still actively turning out impassioned editorials in NRA publications and Far-Right media outlets, raging about "liberals" who are allegedly driving up crime rates with gun restrictions. In virtually every case the data he cites in these has been carefully cherry-picked to give the results he wants, and his conclusions evaporate when the larger picture is examined—and of course, he is still loudly denying that there is anything wrong with his models.

In 2003 Lott became embroiled in scandal when he was caught (by a pro-gun blogger) using Internet “sock puppets” to promote his ideas and work. Shortly after the discovery he confessed to having created the online persona of “Mary Rosh” (whose last name he said, was an acronym derived from the initials of his children). For several years previous Rosh had actively promoted Lott’s work in numerous online and print forums. She purported to be a former student of his and a dainty 115 lb woman (who was perpetually at the risk of muggers and rapists were it not for her firearms), penned numerous 5-start reviews of his books at Amazon.com, and encouraged readers to download copies of Lott’s papers as often as possible from scholarly sources (to increase his citation ratings in peer-review circles).

After Rosh was exposed Lott was discovered to have been using numerous other sock puppets as well, including one or two that I had been having running dialogs with at one blog! He was even caught using sock puppets after he confessed to Mary Rosh. On at least one occasion he is known to have deliberately fabricated a survey of several hundred gun-wielding crime victims who claimed to have foiled potential crime with concealed handguns. To date, he refuses to produce the survey data despite having been repeatedly challenged to do so for several years. No one has ever been able to confirm even a single person who allegedly responded to it.

Needless to say, none of this has affected Lott's popularity in Far-Right circles, where to this day his work is joyously embraced as though it were divine revelation. It seems that he is the closest thing they have to a "scientific" source that supports their claims.