Right‐to‐Carry Laws and Violent Crime: A Comprehensive Assessment Using Panel Data and a State‐Level Synthetic Control Analysis

Date01 June 2019
Published date01 June 2019
DOIhttp://doi.org/10.1111/jels.12219
Journal of Empirical Legal Studies
Volume 16, Issue 2, 198–247, April 2019
Right-to-Carry Laws and Violent Crime:
A Comprehensive Assessment Using
Panel Data and a State-Level Synthetic
Control Analysis
John J. Donohue, Abhay Aneja, and Kyle D. Weber*
This article uses more complete state panel data (through 2014) and new statistical tech-
niques to estimate the impact on violent crime when states adopt right-to-carry (RTC)
concealed handgun laws. Our preferred panel data regression specification, unlike the sta-
tistical model of Lott and Mustard that had previously been offered as evidence of crime-
reducing RTC laws, both satisfies the parallel trends assumption and generates statistically
significant estimates showing RTC laws increase overall violent crime. Our synthetic control
approach also finds that RTC laws are associated with 13–15 percent higher aggregate vio-
lent crime rates 10 years after adoption. Using a consensus estimate of the elasticity of
crime with respect to incarceration of 0.15, the average RTC state would need to roughly
double its prison population to offset the increase in violent crime caused by RTC
adoption.
I. Introduction
For two decades, there has been a spirited academic debate over whether “shall-
issue” concealed carry laws (also known as right-to-carry or RTC laws) have an impor-
tant impact on crime. The “More Guns, Less Crime” hypothesis originally articulated
by John Lott and David Mustard (1997) claimed that RTC laws decreased violent
*Address correspondence to John J. Donohue, Stanford Law School, 559 Nathan Abbott Way, Stanford, CA 94305;
email: donohue@law.stanford.edu. Abhay Aneja, Haas School of Business, 2220 Piedmont Avenue, Berkeley, CA
94720; email: aaneja@law.stanford.edu; Kyle D. Weber, Columbia University, 420 W. 118th Street, New York, NY
10027; email: kdw2126@columbia.edu.
We thank Phil Cook, Dan Ho, Stefano DellaVigna, Rob Tibshirani, Trevor Hastie, Stefan Wager, Jeff Strnad,
and participants at the 2011 Conference of Empirical Legal Studies (CELS), 2012 American Law and Economics
Association (ALEA) Annual Meeting, 2013 Canadian Law and Economics Association (CLEA) Annual Meeting,
2015 NBER Summer Institute (Crime), and the Stanford Law School faculty workshop for their comments and
helpful suggestions. Financial support was provided by Stanford Law School. We are indebted to Alberto Abadie,
Alexis Diamond, and Jens Hainmueller for their work developing the synthetic control algorithm and program-
ming the Stata module used in this paper and for their helpful comments. The authors would also like to thank
Alex Albright, Andrew Baker, Jacob Dorn, Bhargav Gopal, Crystal Huang, Mira Korb, Haksoo Lee, Isaac Rabbani,
Akshay Rao, Vikram Rao, Henrik Sachs and Sidharth Sah who provided excellent research assistance, as well as
Addis O’Connor and Alex Chekholko at the Research Computing division of Stanford’s Information Technology
Services for their technical support.
198
crime (possibly shifting criminals in the direction of committing more property
crime to avoid armed citizens). This research may well have encouraged state legisla-
tures to adopt RTC laws, arguably making the pair’s 1997 paper in the Journal of
Legal Studies one of the most consequential criminological articles published in the
last 25 years.
The original Lott and Mustard paper as well as subsequent work by John Lott in his
1998 book More Guns, Less Crime used a panel data analysis to support the theory that
RTC laws reduce violent crime. A large number of papers examined the Lott thesis, with
decidedly mixed results. An array of studies, primarily those using the limited data ini-
tially employed by Lott and Mustard for the period 1977–1992 and those failing to adjust
their standard errors by clustering, supported the Lott and Mustard thesis, while a host
of other papers were skeptical of the Lott findings.
1
It was hoped that the 2005 National Research Council report Firearms and Violence:
A Critical Review (hereafter the NRC Report) would resolve the controversy over the
impact of RTC laws, but this was not to be. While one member of the committee—James
Q. Wilson—did partially endorse the Lott thesis by saying there was evidence that mur-
ders fell when RTC laws were adopted, the other 15 members of the panel pointedly criti-
cized Wilson’s claim, saying that “the scientific evidence does not support his position.”
The majority emphasized that the estimated effects of RTC laws were highly sensitive to
the particular choice of explanatory variables and thus concluded that the panel data evi-
dence through 2000 was too fragile to support any conclusion about the true effects of
these laws.
This article answers the call of the NRC Report for more and better data and
new statistical techniques to be brought to bear on the issue of the impact of RTC laws
on crime. First, we revisit the state panel data evidence to see if extending the data for
an additional 14 years, thereby providing additional crime data for prior RTC states as
well as on 11 newly adopting RTC states, offers any clearer picture of the causal impact
of allowing citizens to carry concealed weapons. We distill from an array of different
panel data regressions for various crime categories for two time periods using two
major sets of explanatory variables—including our preferred specification (DAW) and
that of Lott and Mustard (LM)—a subset of regressions that satisfy the critical parallel
trends assumption. All the statistically significant results from these regressions show
RTC laws are associated with higher rates of overall violent crime, property crime, or
murder.
Second, to address some of the weaknesses of panel data models, we undertake an
extensive synthetic control analysis in order to present the most complete and robust
1
In support of Lott and Mustard (1997), see Lott’s 1998 book More Guns, Less Crime (and the 2000 and 2010 edi-
tions). Ayres and Donohue (2003) and the 2005 National Research Council report Firearms and Violence : A Critical
Review dismissed the Lott/Mustard hypothesis as lacking credible statistical support, as did Aneja et al. (2011) (and
Aneja et al. (2014) further expanding the latter). Moody and Marvell (2008) and Moody et al. (2014) continued to
argue in favor of a crime-reducing effect of RTC laws, although Zimmerman (2014) and McElroy and Wang
(2017) find that RTC laws increase violent crime and Siegel et al. (2017) find RTC laws increase murders, as dis-
cussed in Section III.B.
Right-to-Carry Laws and Violent Crime 199
results to guide policy in this area.
2
This synthetic control methodology—first introduced
in Abadie and Gardeazabal (2003) and expanded in Abadie et al. (2010, 2014)—uses a
matching methodology to create a credible “synthetic control” based on a weighted aver-
age of other states that best matches the prepassage pattern of crime for each “treated”
state, which can then be used to estimate the likely path of crime if RTC-adopting states
had not adopted an RTC law. By comparing the actual crime pattern for RTC-adopting
states with the estimated synthetic controls in the postpassage period, we derive year-by-
year estimates for the impact of RTC laws in the 10 years following adoption.
3
To preview our major findings, the synthetic control estimate of the average impact
of RTC laws across the 33 states that adopt between 1981 and 2007
4
indicates that violent
crime is substantially higher after 10 years than would have been the case had the RTC
law not been adopted. Essentially, for violent crime, the synthetic control approach pro-
vides a similar portrayal of RTC laws as that provided by the DAW panel data model and
undermines the results of the LM panel data model. According to the aggregate synthetic
control models—regardless of whether one uses the DAW or LM covariates—RTC laws
led to increases in violent crime of 13–15 percent after 10 years, with positive but not sta-
tistically significant effects on property crime and murder. The median effect of RTC
adoption after 10 years is 12.3 percent if one considers all 31 states with 10 years worth of
data and 11.1 percent if one limits the analysis to the 26 states with the most compelling
prepassage fit between the adopting states and their synthetic controls. Comparing our
DAW specification findings with the results generated using placebo treatments, we are
able to reject the null hypothesis that RTC laws have no impact on aggregate violent
crime.
The structure of the article proceeds as follows. Section II begins with a discussion
of the ways in which increased carrying of guns could either dampen crime (by thwarting
or deterring criminals) or increase crime by directly facilitating violence or aggression by
permit holders (or others), greatly expanding the loss and theft of guns, and burdening
the functioning of the police in ways that diminish their effectiveness in controlling
crime. We then show that a simple comparison of the drop in violent crime from
2
Abadie et al. (2014) identify a number of possible problems with panel regression techniques, including the dan-
ger of extrapolation when the observable characteristics of the treated area are outside the range of the
corresponding characteristics for the other observations in the sample.
3
The accuracy of this matching can be qualitatively assessed by examining the root mean square prediction error
(RMSPE) of the synthetic control in the pretreatment period (or a variation on this RMSPE implemented in this
article), and the statistical significance of the estimated treatment effect can be approximated by running a series
of placebo estimates and examining the size of the estimated treatment effect in comparison to the distribution of
placebo treatment effects.
4
Note that we do not supply a synthetic control estimate for Indiana, even though it passed its RTC law in 1980,
owing to the fact that we do not have enough pretreatment years to accurately match the state with an appropriate
synthetic control. Including Indiana as a treatment state, though, would not meaningfully change our results. Simi-
larly, we do not generate synthetic control estimates for Iowa and Wisconsin (whose RTC laws went into effect in
2011) or for Illinois (2014 RTC law), because of the limited postpassage data.
200 Donohue et al.

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT