Less than zero
There’s been a bit of publicity about a recent study of the effects of the Australian gun buyback. The central finding of the authors was that, while gun homicides declined after the buyback this was merely a continuation of a pre-existing trend.
I’m dubious about the whole approach. In the absence of a well-founded explanation for the trend, there’s no reason to treat maintenance of the trend, rather than the level, as the null hypothesis. The rate of gun homicides has clearly fallen (the authors find the same for suicides), so the data supports the policy, contrary to the claims.
And eyeballing the data, I’m doubtful that it’s even sufficient to establish the existence of a declining trend for the period up to and including 1996. It might be argued that the Port Arthur massacre in 1996 should be excluded and that a downward trend would then emerge, but, given that this was the even that precipitated the buyback, this seems like begging the question to me.
In any case, Andrew Leigh has the ultimate knockdown objection. If you look at the confidence intervals, the only way the gun buyback could have been shown to work, on the authors’ tests is if gun homicides fell below zero by 2004. Clearly, even if you buy the declining trend story, a linear trend is just wrong.
Mark Bahnisch has more, though quite a few commenters don’t seem to appreciate how conclusive Leigh’s refutation has been.