6 Right-to-Carry Laws

К оглавлению
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 
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.