Saturday, March 3, 2012

Guns and crime

With the increase in crime in our neighborhood my thoughts turned to whether it might be worthwhile to procure a gun. As part of my research on the costs and benefits of gun ownership I deferred to John Lott’s “More Guns, Less Crime”. (link is Wikipedia version - I read the 2nd edition)

I found the book and its research pretty comprehensive. Included were many sensitivity analysis which I appreciated. The effects were large and require some explanation. Lott’s explanation is that more guns means less crime. Unfortunately I was not quite convinced and like Gary Kleck of University of Florida whom Lott cites in the book, I feel like something is missing in the regression specification but I can’t quite put my finger on it.

Perhaps the first problem with the results was the lack of published R-squares. If his model explained more than 80 percent of the variation in crime I may be more persuaded.

The second “problem” is that this is not really a study of guns and crime but the passage of concealed carry laws in states. Lott’s analysis is at the state and county level using longitudinal data. I have serious issues with instrumental variables but most economists are in love with it so I’m in the minority. In my book, the year the concealed carry law is passed is an instrumental variable. When we use different instruments we would naturally get different size effects. The implication here is that the passage of concealed carry laws led to higher gun ownership although this chart seems to indicate that it might not be the case. (I may take this with a grain of salt since its in Mother Jones - and I would have to convince myself by actually doing my own tabulation.) Anyway it would be whole lot different if the regressions were run on actual gun ownership so that the results would be more easily interpretable as in the following manner: “Increasing one gun in the county reduces 3.2 robberies” or something to that effect. There is no indication that the year concealed carry is passed is a correlated with gun ownership nor as far as I can see the use of an alternative instrument.

Lott rejects the use of trend terms in his model because including a quadratic term might capture most of the effects in his model. In his charts, after the passage of a concealed carry law, the number of robberies, for instance would decline and continue falling. He objects to using the trend term because then it would capture the intended effects of the year concealed carry is passed. His models use year as dummies so he claims that this effect is partially captured and including a trend would obviously invalidate the results. If I read it correctly, some other economists who have tried to replicate his results have included a trend term and have found that the effects are not statistically significant. I am undecided at this point whether including turning points as part of the regression is a mispecification or an omitted variable.

The other larger substantive issue is that even though Lott claims that more guns would lead to less crime he does not say how much more (although he does say that it is more than what it is now). Introspection will indicate that 100% gun ownership will not drive crime down to 0. If everyone owns a gun there is no uncertainty when a criminal intends to rob someone. Since he knows for certainty the the other person is armed he will shoot first and then rob the victim while the victim is incapacitated.

Likewise 0% gun ownership is not a stable equilibrium either unless we can stuff the knowledge of gun making and guns into a vacuum. There will always be an incentive to own a gun because then - well - I would outgun you and can rob you easily.

More speculatively, I would also suggest that anything above 50% gun ownership would have the same effect as 100% ownership. The story that Lott tells is that a criminal is uncertain of whether the potential victim has a gun and therefore this uncertainty drives the criminal away toward other ‘safer’ crimes such as burglary.

I have my doubts as to how substitutable different types of crimes actually are for instance one doesn’t think that rape and robbery are substitutes and it isn’t clear to me that robberies and burglaries are highly substitutable either. In the latter case someone who commits robberies is likely to be impatient and hence just sticks some up rather than have the patience to case a place to determine whether he should break in or not.

In sum I thought that the research on guns and crime is important but I also think that attention needs to shift toward finding other instruments or measuring actual gun ownership directly such as gun registrations, etc. (Yes, this will be biased since there are stolen, unregistered guns which is why I say different instruments so that gun ownership while biased would presumably still be a valid instrument.)

P.S. I also flipped through The Bias Against Guns but it didn’t really hold my attention. Most of the book was about whether guns increased other types of crimes such as mass shootings and such.

No comments: