Eric Helland and Alexander Taborrak have a new paper “Using Placebo Laws to Test More Guns, Less Crime : A Note”. I commented on their paper back in May, but here is Lott’s take (in full, 8/22/03 blog entry):
A new research paper has an new important approach towards estimating statistical significance. Professors Eric Helland and Alex Tabarrok conclude that:
“the cross equation restrictions implied by the Lott-Mustard theory are strongly supported.”
“Surprisingly, therefore, we conclude, that there is considerable support for the hypothesis that shall-issue laws cause criminals to substitute away from crimes against persons and towards crimes against property.”
Compare the parts Lott quoted with their abstract, so you can see what Lott left out:
A boomlet has occurred in recent years in the use of quasi-natural experiments to answer important questions of public policy. The intuitive power of this approach, however, has sometimes diverted attention from the statistical assumptions that must be made, particularly regarding standard errors (Bertrand, Duflo and Mullainathan 2002, Donald and Lang 2001). Failing to take into account serial correlation and grouped data can dramatically reduce standard errors suggesting greater certainty in effects than is actually the case. We reexamine Mustard and Lott’s important and controversial study on the affect of shall-issue gun laws on crime using an empirical standard error function randomly generated from placebo laws. We find that the in some specifications the effect of shall-issue laws on specific types of crimes is much less well-estimated than the Mustard and Lott (1997) and Lott (2000) results suggest (i.e. placebo shall-issue laws produce estimated real effects at greater rates than suggested by the standard errors in the original studies.) We also find, however, that the cross equation restrictions implied by the Lott-Mustard theory are strongly supported.
Taking into account grouped data is what the clustering correction does. Helland and Tabbarok show that even this correction isn’t enough and it is wrong for Lott to remove it. No wonder Lott didn’t quote that bit. In fact, the only statistically significant decrease that they found was that on murder using the trend model. However, they only used the data up to 1992 for this. Include the data up to 1997 and that result goes away too.
As for the cross equation restriction, that only occurs with the dummy variable model, which Lott abandoned when the addition of more data meant that it stopped supporting his thesis. Is he going to pick it up again?