6 Right-to-Carry Laws
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101
102 103 104 105 106 107 108 109 110 111 112 113 114 115
This chapter is concerned with the question of whether violent crime is
reduced through the enactment of right-to-carry-laws, which allow individuals
to carry concealed weapons.1 In all, 34 states have right-to-carry
laws that allow qualified adults to carry concealed handguns. Proponents of
these laws argue that criminals are deterred by the knowledge that potential
victims may be carrying weapons and therefore that the laws reduce crime.
However, it is not clear a priori that such deterrence occurs. Even if it does,
there may be offsetting adverse consequences. For example, increased possession
of firearms by potential victims may motivate more criminals to carry
firearms and thereby increase the amount of violence that is associated with
crime. Moreover, allowing individuals to carry concealed weapons may increase
accidental injuries or deaths or increase shootings during arguments.
Ultimately, it is an empirical question whether allowing individuals to carry
concealed weapons generates net social benefits or net social costs.
The statistical analysis of the effects of these laws was initiated by John
Lott and David Mustard (1997) and expanded by Lott (2000) and Bronars
and Lott (1998) (hereinafter referred to simply as Lott). Lott concludes that
the adoption of right-to-carry laws substantially reduces the prevalence of
violent crime. Many other researchers have carried out their own statistical
analyses using Lott’s data, modified versions of Lott’s data, or expanded
1The laws are sometimes called shall-issue laws because they require local authorities to
issue a concealed-weapons permit to any qualified adult who requests one. A qualified adult
is one who does not have a significant criminal record or history of mental illness. The
definition of a nonqualified adult varies among states but includes adults with prior felony
convictions, drug charges, or commitments to mental hospitals.
data sets that cover the more recent time period not included in the original
analysis.2
Because the right-to-carry issue is highly controversial, has received much
public attention, and has generated a large volume of research, the committee
has given it special attention in its deliberations. This chapter reviews the
existing empirical evidence on the issue. We also report the results of our own
analyses of the data. We conclude that, in light of (a) the sensitivity of the
empirical results to seemingly minor changes in model specification, (b) a
lack of robustness of the results to the inclusion of more recent years of data
(during which there are many more law changes than in the earlier period),
and (c) the imprecision of some results, it is impossible to draw strong conclusions
from the existing literature on the causal impact of these laws. Committee
member James Q. Wilson has written a dissent that applies to Chapter 6
only (Appendix A), and the committee has written a response (Appendix B).
DESCRIPTION OF THE DATA AND METHODS
Researchers studying the effects of right-to-carry laws have used many
different models. However, all of the analyses rely on similar data and methodologies.
Accordingly, we do not attempt to review and evaluate each of the
models used in this literature. Instead, we describe the common data used and
2Two other general responses to Lott’s analysis deserve brief mention. First, some critics
have attempted to discredit Lott’s findings on grounds of the source of some of his funding
(the Olin Foundation), the methods by which some of his results were disseminated (e.g.,
some critics have claimed, erroneously, that Lott and Mustard, 1997, was published in a
student-edited journal that is not peer reviewed), and positions that he has taken on other
public policy issues related to crime control. Much of this criticism is summarized and responded
to in Chapter 7 of Lott (2000). The committee’s view is that these criticisms are not
helpful for evaluating Lott’s data, methods, or conclusions. Lott provides his data and computer
programs to all who request them, so it is possible to evaluate his methods and results
directly. In the committee’s view, Lott’s funding sources, methods of disseminating his results,
and opinions on other issues do not provide further information about the quality of his
research on right-to-carry laws.
A second group of critics have argued that Lott’s results lack credibility because they are
inconsistent with various strongly held a priori beliefs or expectations. For example, Zimring
and Hawkins (1997:59) argue that “large reductions in violence [due to right-to-carry laws]
are quite unlikely because they would be out of proportion to the small scale of the change in
carrying firearms that the legislation produced.” The committee agrees that it is important for
statistical evidence to be consistent with established facts, but there are no such facts about
whether right-to-carry laws can have effects of the magnitudes that Lott claims. The beliefs or
expectations of Lott’s second group of critics are, at best, hypotheses whose truth or falsehood
can only be determined empirically. Moreover, Lott (2000) has argued that there are
ways to reconcile his results with the beliefs and expectations of the critics. This does not
necessarily imply that Lott is correct and his critics are wrong. The correctness of Lott’s
arguments is also an empirical question about which there is little evidence. Rather, it shows
that little can be decided through argumentation over a priori beliefs and expectations.
focus on the common methodological basis for all of them. In particular, we
use the results presented in Tables 4.1 and 4.8 of Lott (2000) to illustrate the
discussion. We refer to these as the “dummy variable” and “trend” model
estimates, respectively. Arguably, these tables, which are reproduced in Table
6-1 and Table 6-2, contain the most important results in this literature.
Data
The basic data set used in the literature is a county-level panel on
annual crime rates, along with the values of potentially relevant explanatory
variables. Early studies estimated models on data for 1977-1992, while
more recent studies (as well as our replication exercise below) use data up
to 2000. Between 1977 and 1992, 10 states adopted right-to-carry laws.3 A
total of 8 other states adopted right-to-carry laws before 1977. Between
1992 and 1999, 16 additional states adopted such laws.
The data on crime rates were obtained from the FBI’s Uniform Crime
Reports (UCR). Explanatory variables employed in studies include the arrest
rate for the crime category in question, population density in the county,
real per capita income variables, county population, and variables for the
percent of population that is in each of many race-by-age-by-gender categories.
The data on explanatory variables were obtained from a variety of
sources (Lott, 2000: Appendix 3).
Although most studies use county-level panels on crime rates and demographic
variables, the actual data files used differ across studies in ways
that sometimes affect the estimates. The data set used in the original Lott
study has been lost, although Lott reconstructed a version of the data,
which he made available to other researchers as well as the committee. This
data set, which we term the “revised original data set,” covers the period
1977-1992.4 More recently, Lott has made available a data set covering the
3There is some disagreement over when and whether particular states have adopted rightto-
carry laws. Lott and Mustard, for example, classify North Dakota and South Dakota as
having adopted such laws prior to 1977, but Vernick and Hepburn (2003) code these states
as having adopted them in 1985. Likewise, Lott and Mustard classify Alabama and Connecticut
as right-to-carry states adopting prior to 1977, yet Vernick codes these states as not
having right-to-carry laws. See Ayres and Donohue (2003a:1300) for a summary of the
coding conventions on the adoption dates of right-to-carry laws.
4There are 3,054 counties observed over 16 years in the revised original data. In the basic
specifications, there are a number of sample restrictions, the most notable of which is to drop
all counties with no reported arrest rate (i.e., counties with no reported crime). This restricts
the sample to approximately 1,650 counties per year (or approximately 26,000 county-year
observations). In specifications that do not involve the arrest rate, Lott treats zero crime as
0.1 so as not to take the log of 0. Black and Nagin (1998) further restrict the sample to
counties with populations of at least 100,000, which limits the sample to 393 counties per
year. In some regressions, Duggan (2001) and Plassmann and Tideman (2001) estimate models
that include data on the over 2,900 counties per year with nonmissing crime data.
period 1977-2000 that corrects acknowledged errors in data files used by
Plassmann and Whitley (2003). We term this file the “revised new data
set.” 5 We make use of both of these data sets in our replication exercises.
Dummy Variable Model
For expository purposes it is helpful to begin by discussing the dummy
variable model without “control” variables.6 The model (in Lott, 2000:
Table 4.1) allows each county to have its own crime level in each category.
Moreover, the crime rate is allowed to vary over time in a pattern that is
common across all counties in the United States. The effect of a right-tocarry
law is measured as a change in the level of the crime rate in a jurisdiction
following the jurisdiction’s adoption of the law. Any estimate of a
policy effect requires an assumption about the “counterfactual,” in this
case what would have happened to crime rates in the absence of the change
in the law. The implicit assumption underlying this simple illustrative
dummy variable model is that, in the absence of the change in the law, the
crime rate in each county would, on average, have been the county mean
plus a time-period adjustment reflecting the common trend in crime rates
across all counties.
Dummy variable models estimated in the literature are slightly more
complicated than the above-described model. First, they typically include
control variables that attempt to construct a more realistic counterfactual.
For example, if crime rates vary over time with county economic conditions,
then one can construct a more credible estimate of what would have
happened in the absence of the law change by including the control variables
as a determinant of the crime rate. Most estimates in the literature
use a large number of control variables, including local economic conditions,
age-gender population composition, as well as arrest rates.
Second, some estimates in the literature model the time pattern of
crime differently. In particular, some studies allow each region of the
country to have its own time pattern, thereby assuming that in the absence
of the law change, counties in nearby states would have the same
time pattern of crime rates in a crime category. We term this the “regioninteracted
time pattern model,” in contrast to the “common time pattern”
dummy variable model above.
5These data were downloaded by the committee from www.johnlott.org on August 22,
2003.
6This no-control model is often used as a way to assess whether there is an association
between the outcome (crime) and the law change in the data. The committee estimates and
evaluates this model below (see Tables 6-5 and 6-6, rows 2 and 3).
Mathematically, the common time pattern dummy variable model takes
the form
(6.1) ,
where Yit is the natural logarithm of the number of crimes per 100,000
population in county i and year t, YEARt = 1 if the year is t and YEARt = 0
otherwise, Xit is a set of control variables that potentially influence crime rates,
LAWit = 1 if a right-to-carry law was in effect in county i and year t and LAWit
= 0 otherwise, gi is a constant that is specific to county i, and eit is an
unobserved random variable. The quantities at, b, and d are coefficients that
are estimated by fitting the model to data. The coefficient d measures the
percentage change in crime rates due to the adoption of right-to-carry laws.
For example, if d = –0.05 then the implied estimate of the adoption of a rightto-
carry law is to reduce the crime rate by 5 percent. The coefficients at
measure common time patterns across counties in crime rates that are distinct
from the enactment of right-to-carry laws or other variables of the model.
The vector Xit includes the control variables that may influence crime
rates, such as indicators of income and poverty levels; the density, age
distribution, and racial composition of a county’s population; arrest rates;
and indicators of the size of the police force. The county fixed effect gi
captures systematic differences across counties that are not accounted for
by the other variables of the model and do not vary over time. The values of
the parameters at, b, and d are estimated separately for each of several
different types of crimes. Thus, the model accounts for the possibility that
right-to-carry laws may affect different crimes differently.