29
specification.
18
In fact, this was essentially the message of the NRC report. Small changes made the
estimates bounce around so much that it was difficult to reach any conclusion about the true
causal impact of RTC laws. Perhaps it might have been helpful to Wilson if the majority had
gone one step further and presented something like the alternative results from Table 2. As we
will see in the ensuing sections, there are many additional avenues that could have been explored
to probe the robustness of the Table 1b findings that Wilson had accepted so unquestioningly.
We will explore these factors in subsequent sections: Section VI will explore whether one
should control for individual state trends in crime, section VII will look at additional years of
data (adding data beyond 2000 to 2006), section VIII will alter the Lott and Mustard
specification (beyond the already mentioned correction for the contemporaneous, crime-specific
arrest rates), section IX will go beyond the county data to look at state data, and Section X will
18
In the process of reviewing our previous published models and data from ADZ (2011), we discovered some errors
in the two data sets that we had constructed (the so-called updated 2009 county data and updated 2009 state data).
For the county data set, we miscoded the state trend variable for Arkansas. Second, we incorrectly coded Oregon's
year of adoption as 1989 instead of 1990. Third, Kansas counties have been incorrectly coded as belonging to
Kentucky for years 1997-2006. Additionally, there was an error in our spline and hybrid models, which we have
now corrected (we had included a counter variable to capture the effect of a post-passage trend, but we inadvertently
omitted the overall trend variable off of which this post-passage trend was to be estimated).
In addition to these errors that we discovered, Moody, Lott, Marvell, and Zimmerman (2012) identified two
other errors: observations for county 2060 for Alaska are duplicated 73 times for 1996 and Kansas' year of adoption
is coded incorrectly as 1996 instead of 2006. All of these errors have been corrected in the tables prepared for this
paper.
Moody, Lott et al also claimed that Florida's year of adoption is coded incorrectly as 1989 instead of 1987,
and South Dakota's year of adoption is coded incorrectly as 1986 instead of 1985. We disagree on these two points.
First, our county data does not provide crime category information for Florida counties for 1988 (this is evident in
the NRC data set as well), so we elected to drop the observations for this year for all Florida counties. Thus, it may
seem that our first year of adoption is erroneously coded as 1989. However, this simply reflects the fact that we have
not included observations for 1988. Note that we had maintained consistency with our other trend variables by
beginning the post-passage variable counter with a value of "2" in year 1989 to demonstrate 2 years since the
passage of RTC legislation. The second issue – regarding South Dakota's year of adoption as 1986 instead of 1985 –
is disputed. Our research indicates that the law was passed in 1986 (and thus has been appropriately coded with
1987 as the first full year of implementation).
For our state data set that we will employ below, we note the following changes from ADZ (2011): North
Dakota should show RTC adoption in year 1985 with a post-passage trend variable beginning in 1986; South Dakota
should show RTC adoption in year 1986 with a post-passage trend variable beginning in 1987.