After Martin Bryant murdered 35 people at Port Arthur, Prime Minister John Howard got new laws enacted that banned semi-automatic long guns. At the time, I felt that was bad policy. Since almost all most gun killings involve just one death it didn’t matter whether whether the gun was a semi-auto or not, writing:
I think that the new laws are mostly a stupid waste of money.
In the years that followed, homicide, firearm homicide, suicide and firearm suicide all fell, but a paper published last year by Baker and McPhedran argued that the decline was not statistically significant. However, Andrew Leigh found a flaw in their analysis. Look at this graph of the gun homicide rate (the solid line):
For the reduction to be significant, the rate would have had to have dropped below the bottom dotted line. In other words, to below zero. There is no possible outcome that their method would have found as significant.
Now he’s done some more analysis and written a short paper that shows that if you use more data than Baker and McPhedran or change their method slightly, you find significant effects.
There is never going to be conclusive evidence in a case like this because it is possible that the decline was coincidental, but it seems likely that the laws were at least partly responsible, so I must admit that I was wrong when I predicted that they would have no effect.