Incorrect claims about the data in More Guns, Less Crime
1) There are no coding errors in the data used in “More Guns, Less Crime.” The book used crime data from 1977 to 1996, and as far as I know, no academics have claimed that there were any coding errors in that data. An interesting and useful Stanford Law Review article by Plassmann and Whitley added an additional four years to the data and the problem arose in this additional data. Overall, less than 200 cells out of 7.5 million cells were accidentally left blank. More important, the results that Plassmann and Whitley noted were the results which they thought were the correct ones were not affected at all by this minor change in the data.
2) I had put the data set on my website as a favor to Plassmann because it was very large and he did not have the ability to put it up for people to download. The data was corrected as soon as the missing cells were discovered and it was available for people to download. A note was added to the site to alert people to the correction. While I had helped them out on their paper, I let Plassmann and Whitley make their own statements about their results. My website reported the regressions in the same way that I had done them in all my previous estimates.
3) In statements here and here, I have discussed what was involved in Levitt's errors. The regressions where they left out the fixed effects were essentially the only ones that were at all useful in testing their hypothesis because they were the only ones that didn't require large degrees of aggregation to get to their "effective abortion rate" that were related to total murder rates in a state. The work that I did with Whitley directly related the age of the murder to the number of abortions when that murderer was born.
Thanks to Bob Thomas for his email.