Crime, Deterrence, and Right-to-Carry Concealed Handguns
John R. Lott, Jr.
School of Law
University of Chicago
Chicago, Illinois 60637
and
David B. Mustard
Department of Economics
University of Chicago
Chicago, Illinois 60637
July 26, 1996
* The authors would like to thank Gary Becker, Phil Cook, Clayton
Cramer, Gertrud Fremling, Ed Glaeser, Hide Ichimura, Don Kates, Gary
Kleck, David Kopel, William Landes, David McDowall, Derek Neal, Dan
Polsby, and Douglas Weil and the seminar participants at the
University of Chicago, American Law and Economics Association
Meetings, and the Western Economic Association Meetings for their
unusually helpful comments.
Crime, Deterrence, and Right-to-Carry Concealed Handguns
Abstract
Using cross-sectional time-series data for U.S. counties from 1977 to
1992, we find that allowing citizens to carry concealed weapons deters
violent crimes and it appears to produce no increase in accidental
deaths. If those states which did not have right-to-carry concealed
gun provisions had adopted them in 1992, approximately 1,570 murders;
4,177 rapes; and over 60,000 aggravate assaults would have been
avoided yearly. On the other hand, consistent with the notion of
criminals responding to incentives, we find criminals substituting
into property crimes involving stealth and where the probabilities of
contact between the criminal and the victim are minimal. The largest
population counties where the deterrence effect on violent crimes is
greatest are where the substitution effect into property crimes is
highest. Concealed handguns also have their greatest deterrent effect
in the highest crime counties. Higher arrest and conviction rates
consistently and dramatically reduce the crime rate. Consistent with
other recent work (Lott, 1992b), the results imply that increasing the
arrest rate, independent of the probability of eventual conviction,
imposes a significant penalty on criminals. The estimated annual gain
from allowing concealed handguns is at least $6.214 billion.
I. Introduction
Will allowing concealed handguns make it likely that otherwise law
abiding citizens will harm each other? Or, will the threat of citizens
carrying weapons primarily deter criminals? To some, the logic is
fairly straightforward. Philip Cook argues that, "If you introduce a
gun into a violent encounter, it increases the chance that someone
will die."[1] A large number of murders may arise from unintentional
fits of rage that are quickly regretted, and simply keeping guns out
of people's reach would prevent deaths.[2] Using the National Crime
Victimization Survey (NCVS), Cook (1991, p. 56, fn. 4) further states
that each year there are "only" 80,000 to 82,000 defensive uses of
guns during assaults, robberies, and household burglaries.[3] By
contrast, other surveys imply that private firearms may be used in
self-defense up to two and a half million times each year, with
400,000 of these defenders believing that using the gun "almost
certainly" saved a life (Kleck and Gertz, 1995, pp. 153, 180, and
182-3).[4] With total firearm deaths from homicides and accidents
equaling 19,187 in 1991 (Statistical Abstract of the United States,
1995), the Kleck and Gertz numbers, even if wrong by a very large
factor, suggest that defensive gun use on net saved lives.
While cases like the 1992 incident where a Japanese student was shot
on his way to a Halloween party in Louisiana make international
headlines (Japan Economic Newswire, May 23, 1993 and Sharn, USA TODAY,
September 9, 1993), they are rare. In another highly publicized case,
a Dallas resident recently became the only Texas resident so far
charged with using a permitted concealed weapon in a fatal shooting
(Potok, March 22, 1996, p. 3A).[5] Yet, in neither case was the
shooting found to be unlawful.[6] The rarity of these incidents is
reflected in Florida statistics: 221,443 licenses were issued between
October 1, 1987 and April 30, 1994, but only 18 crimes involving
firearms were committed by those with licenses (Cramer and Kopel,
1995, p. 691).[7] While a statewide breakdown on the nature of those
crimes is not available, Dade county records indicate that four crimes
involving a permitted handgun took place there between September 1987
and August 1992 and none of those cases resulted in injury (pp.
691-2).
The potential defensive nature of guns is indicated by the different
rates of so-called "hot burglaries," where residents are at home when
the criminals strike (e.g., Kopel, 1992, p. 155 and Lott, 1994).
Almost half the burglaries in Canada and Britain, which have tough gun
control laws, are "hot burglaries." By contrast, the U.S., with laxer
restrictions, has a "hot burglary" rate of only 13 percent. Consistent
with this, surveys of convicted felons in America reveals that they
are much more worried about armed victims than they are about running
into the police. This fear of potentially armed victims causes
American burglars to spend more time than their foreign counterparts
"casing" a house to ensure that nobody is home. Felons frequently
comment in these interviews that they avoid late-night burglaries
because "that's the way to get shot."[8]
The case for concealed handgun use is similar. The use of concealled
handguns by some law abiding citizens may create a positive
externality for others. By the very nature of these guns being
concealed, criminals are unable to tell whether the victim is armed
before they strike, thus raising criminals' expected costs for
committing many types of crimes.
Stories of individuals using guns to defend themselves has helped
motivate thirty-one states to adopt laws requiring authorities to
issue, without discretion, concealed-weapons permits to qualified
applicants.[9] This constitutes a dramatic increase from the nine
states that allowed concealed weapons in 1986.[10] While many studies
examine the effects of gun control (see Kleck, 1995 for a survey), and
a smaller number of papers specifically address the right-to-carry
concealed firearms (e.g., Cook, et al., 1995; Cramer and Kopel, 1995;
McDowall, et. al., 1995; and Kleck and Patterson, 1993), these papers
involve little more than either time-series or cross-sectional
evidence comparing mean crime rates, and none controls for variables
that normally concern economists (e.g., the probability of arrest and
conviction and the length of prison sentences or even variables like
personal income).[11] These papers fail to recognize that, since it is
frequently only the largest population counties that are very
restrictive when local authorities have been given discretion in
granting concealed handgun permits, "shall issue" concealed handgun
permit laws, which require permit requests be granted unless the
individual has a criminal record or a history of significant mental
illness (Cramer and Kopel, 1995, pp. 680-707), will not alter the
number of permits being issued in all counties.
Other papers suffer from additional weaknesses. The paper by McDowall,
et. al. (1995), which evaluates right-to-carry provisions, was widely
cited in the popular press. Yet, their study suffers from many major
methodological flaws: for instance, without explanation, they pick
only three cities in Florida and one city each in Mississippi and
Oregon (despite the provisions involving statewide laws); and they
neither use the same sample period nor the same method of picking
geographical areas for each of those cities.[12]
Our paper hopes to overcome these problems by using annual
cross-sectional time-series county level crime data for the entire
United States from 1977 to 1992 to investigate the impact of "shall
issue" right-to-carry firearm laws. It is also the first paper to
study the questions of deterrence using these data. While many recent
studies employ proxies for deterrence ---- such as police expenditures
or general levels of imprisonment (Levitt, 1996) ----, we are able to
use arrest rates by type of crime, and for a subset of our data also
conviction rates and sentence lengths by type of crime.[13] We also
attempt to analyze a question noted but not empirically addressed in
this literature: the concern over causality between increases in
handgun usage and crime rates. Is it higher crime that leads to
increased handgun ownership, or the reverse? The issue is more
complicated than simply whether carrying concealed firearms reduces
murders because there are questions over whether criminals might
substitute between different types of crimes as well as the extent to
which accidental handgun deaths might increase.
II. Problems Testing the Impact of "Shall Issue" Concealed Handgun
Provisions
on Crime
Starting with Becker (1968), many economists have found evidence
broadly consistent with the deterrent effect of punishment (e.g.,
Ehrlich (1973), Block and Heineke (1975), Landes (1978), Lott (1987),
Andreoni (1995), Reynolds (1995), and Levitt (1996)). The notion is
that the expected penalty affects the prospective criminal's desire to
commit a crime. This penalty consists of the probabilities of arrest
and conviction and the length of the prison sentence. It is reasonable
to disentangle the probability of arrest from the probability of
conviction since accused individuals appear to suffer large
reputational penalties simply from being arrested (Lott, 1992b).
Likewise, conviction also imposes many different penalties (e.g., lost
licenses, lost voting rights, further reductions in earnings, etc.)
even if the criminal is never sentenced to prison (Lott, 1990b, 1992a
and b).
While this discussion is well understood, the net effect of "shall
issue" right-to-carry, concealed handguns is ambiguous and remains to
be tested when other factors influencing the returns to crime are
controlled for. The first difficulty involves the availability of
detailed county level data on a variety of crimes over 3054 counties
during the period from 1977 to 1992. Unfortunately, for the time
period we study, the FBI's Uniform Crime Report only includes arrest
rate data rather than conviction rates or prison sentences. While we
make use of the arrest rate information, we will also use county level
dummies, which admittedly constitute a rather imperfect way to control
for cross county differences such as differences in expected
penalties. Fortunately, however, alternative variables are available
to help us proxy for changes in legal regimes that affect the crime
rate. One such method is to use another crime category as an exogenous
variable that is correlated with the crimes that we are studying, but
at the same time is unrelated to the changes in right-to-carry firearm
laws. Finally, after telephoning law enforcement officials in all 50
states, we were able to collect time-series county level conviction
rates and mean prison sentence lengths for three states (Arizona,
Oregon, and Washington).
The FBI crime reports include seven categories of crime: murder, rape,
aggravated assault, robbery, auto theft, burglary, and larceny.[14]
Two additional summary categories were included: violent crimes
(including murder, rape, aggravated assault, and robbery) and property
crimes (including auto theft, burglary, and larceny). Despite being
widely reported measures in the press, these broader categories are
somewhat problematic in that all crimes are given the same weight
(e.g., one murder equals one aggravated assault). Even the narrower
categories are somewhat broad for our purposes. For example, robbery
includes not only street robberies which seem the most likely to be
affected by "shall issue" laws, but also bank robberies where the
additional return to having armed citizens would appear to be
small.[15] Likewise, larceny involves crimes of "stealth," but these
range from pick pockets, where "shall issue" laws could be important,
to coin machine theft.[16]
This aggregation of crime categories makes it difficult to separate
out which crimes might be deterred from increased handgun ownership,
and which crimes might be increasing as a result of a substitution
effect. Generally, we expect that the crimes most likely to be
deterred by concealed handgun laws are those involving direct contact
between the victim and the criminal, especially those occurring in a
place where victims otherwise would not be allowed to carry firearms.
For example, aggravated assault, murder, robbery, and rape seem most
likely to fit both conditions, though obviously some of all these
crimes can occur in places like residences where the victims could
already possess firearms to protect themselves.
By contrast, crimes like auto theft seem unlikely to be deterred by
gun ownership. While larceny is more debatable, in general ---- to the
extent that these crimes actually involve "stealth" ---- the
probability that victims will notice the crime being committed seems
low and thus the opportunities to use a gun are relatively rare. The
effect on burglary is ambiguous from a theoretical standpoint. It is
true that if "shall issue" laws cause more people to own a gun, the
chance of a burglar breaking into a house with an armed resident goes
up. However, if some of those who already owned guns now obtain
right-to-carry permits, the relative cost of crimes like armed street
robbery and certain other types of robberies (where an armed patron
may be present) should rise relative to that for burglary.
Previous concealed handgun studies that rely on state level data
suffer from an important potential problem: they ignore the
heterogeneity within states (e.g., Linsky, et. al., 1988 and Cramer
and Kopel, 1995). Our telephone conversations with many law
enforcement officials have made it very clear that there was a large
variation across counties within a state in terms of how freely gun
permits were granted to residents prior to the adoption of "shall
issue" right-to-carry laws.[17] All those we talked to strongly
indicated that the most populous counties had previously adopted by
far the most restrictive practices on issuing permits. The implication
for existing studies is that simply using state level data rather than
county data will bias the results against finding any impact from
passing right-to-carry provisions. Those counties that were unaffected
by the law must be separated out from those counties where the change
could be quite dramatic. Even cross-sectional city data (e.g., Kleck
and Patterson, 1993) will not solve this problem, because without time
series data it is impossible to know what impact a change in the law
had for a particular city.
There are two ways of handling this problem. First, for the national
sample, we can see whether the passage of "shall issue" right-to-carry
laws produces systematically different effects between the high and
low population counties. Second, for three states, Arizona, Oregon,
and Pennsylvania, we have acquired time series data on the number of
right-to-carry permits for each county. The normal difficulty with
using data on the number of permits involves the question of
causality: do more permits make crimes more costly or do higher crimes
lead to more permits? The change in the number of permits before and
after the change in the state laws allows us to rank the counties on
the basis of how restrictive they had actually been in issuing permits
prior to the change in the law. Of course there is still the question
of why the state concealed handgun law changed, but since we are
dealing with county level rather than state level data we benefit from
the fact that those counties which had the most restrictive permitting
policies were also the most likely to have the new laws exogenously
imposed upon them by the rest of their state.
Using county level data also has another important advantage in that
both crime and arrest rates vary widely within states. In fact, as
Table 1 indicates, the standard deviation of both crime and arrest
rates across states is almost always smaller than the average within
state standard deviation across counties. With the exception of
robbery, the standard deviation across states for crime rates ranges
from between 61 and 83 percent of the average of the standard
deviation within states. (The difference between these two columns
with respect to violent crimes arises because robberies make up such a
large fraction of the total crimes in this category.) For arrest
rates, the numbers are much more dramatic, with the standard deviation
across states as small as 15 percent of the average of the standard
deviation within states. These results imply that it is no more
accurate to view all the counties in the typical state as a homogenous
unit than it is to view all the states in the United States as one
homogenous unit. For example, when a state's arrest rate rises, it may
make a big difference whether that increase is taking place in the
most or least crime prone counties. Depending upon which types of
counties the changes in arrest rates are occurring in and depending on
how sensitive the crime rates are to changes in those particular
counties could produce widely differring estimates of how increasing a
state's average arrest rate will deter crime. Aggregating these data
may thus make it more difficult to discern the true relationship that
exists between deterrence and crime.
Perhaps the relatively small across-state variation as compared to
within-state variations is not so surprising given that states tend to
average out differences as they encompass both rural and urban areas.
Yet, when coupled with the preceding discussion on how concealed
handgun provisions affected different counties in the same state
differently, these numbers strongly imply that it risky to assume that
states are homogenous units with respect to either how crimes are
punished or how the laws which affect gun usage are changed.
Unfortunately, this focus of state level data is pervasive in the
entire crime literature, which focuses on state or city level data and
fails to recognize the differences between rural and urban counties.
However, using county level data has some drawbacks. Frequently,
because of the low crime rates in many low population counties, it is
quite common to find huge variations in the arrest and conviction
rates between years. In addition, our sample indicates that annual
conviction rates for some counties are as high as 13 times the offense
rate. This anomaly arises for a couple reasons. First, the year in
which the offense occurs frequently differs from the year in which the
arrests and/or convictions occur. Second, an offense may involve more
than one offender. Unfortunately, the FBI data set allows us neither
to link the years in which offenses and arrests occurred nor to link
offenders with a particular crime. When dealing with counties where
only a couple murders occur annually, arrests or convictions can be
multiples higher than the number of offenses in a year. This data
problem appears especially noticeable for murder and rape.
One partial solution is to limit the sample to only counties with
large populations. For counties with a large numbers of crimes, these
waves have a significantly smoother flow of arrests and convictions
relative to offenses. An alternative solution is to take a moving
average of the arrest or conviction rates over several years, though
this reduces the length of the usable sample period, depending upon
how many years are used to compute this average. Furthermore, the
moving average solution does nothing to alleviate the effect of
multiple suspects being arrested for a single crime.
Another concern is that otherwise law abiding citizens may have
carried concealed handguns even before it was legal to do so. If shall
issue laws do not alter the total number of concealed handguns carried
by otherwise law abiding citizens but merely legalizes their previous
actions, passing these laws seems unlikely to affect crime rates. The
only real effect from making concealed handguns legal could arise from
people being more willing to use handguns to defend themselves, though
this might also imply that they more likely to make mistakes using
these handguns.
It is also possible that concealed firearm laws both make individuals
safer and increase crime rates at the same time. As Peltzman (1975)
has pointed out in the context of automobile safety regulations,
increasing safety can result in drivers offsetting these gains by
taking more risks in how they drive. The same thing is possible with
regard to crime. For example, allowing citizens to carry concealed
firearms may encourage people to risk entering more dangerous
neighborhoods or to begin traveling during times they previously
avoided. Thus, since the decision to engage in these riskier
activities is a voluntary one, it is possible that society still could
be better off even if crime rates were to rise as a result of
concealed handgun laws.
Finally, there are also the issues of why certain states adopted
concealed handgun laws and whether higher offense rates result in
lower arrest rates. To the extent that states adopted the law because
crime were rising, ordinary least squares estimates would underpredict
the drop in crime. Likewise, if the rules were adopted when crimes
rates were falling, the bias would be in the opposite direction. None
of the previous studies deal with this last type of potential bias. At
least since Ehrlich (1973, pp. 548-553), economists have also realized
that potential biases exist from having the offense rate as both the
endogenous variable and as the denominator in determining the arrest
rate and because increasing crime rates may lower the arrest if the
same resources are being asked to do more work. Fortunately, both
these sets of potential biases can be dealt with using two-stage
least-squares.
III. The Data
Between 1977 and 1992, 10 states (Florida (1987), Georgia (1989),
Idaho (1990), Maine (1985), Mississippi (1990), Montana (1991), Oregon
(1990), Pennsylvania (1989), Virginia (1988), and West Virginia
(1989)) adopted "shall issue" right-to-carry firearm laws. However,
Pennsylvania is a special case because Philadelphia was exempted from
the state law during our sample period. Nine other states (Alabama,
Connecticut, Indiana, Maine, New Hampshire, North Dakota, South
Dakota, Vermont, and Washington) effectively had these laws on the
books prior to the period being studied.[18] Since the data are at the
county level, a dummy variable is set equal to one for each county
operating under "shall issue" right-to-carry laws. A Nexis search was
conducted to determine the exact date on which these laws took effect.
For the states that adopted the law during the year, the dummy
variable for that year is scaled to equal that portion of the year for
which the law was in effect.
While the number of arrests and offenses for each type of crime in
every county from 1977 to 1992 were provided by the Uniform Crime
Report, we also contacted the state department of corrections, State
Attorney Generals, State Secretary of State, and State Police offices
in every state to try to compile data on conviction rates, sentence
lengths, and right-to-carry concealed weapons permits by county. The
Bureau of Justice Statistics also released a list of contacts in every
state that might have available state level criminal justice data.
Unfortunately, county data on the total number of outstanding
right-to-carry pistol permits were available for only Arizona,
California, Florida, Oregon, Pennsylvania, and Washington, though time
series county data before and after a change in the permitting law was
only available for Arizona (1994 to 1996), Oregon (1990 to 1992) and
Pennsylvania (1986 to 1992). Since the Oregon "shall issue" law passed
in 1990, we attempted to get data on the number of permits in 1989 by
calling up every county sheriff in Oregon, with 25 of the 36 counties
providing us with this information. (The remaining counties claimed
that records had not been kept.)[19] For Oregon, data on the county
level conviction rate and prison sentence length was also available
from 1977 to 1992.
One difficulty with the sentence length data is that Oregon passed a
sentencing reform act that went into effect in November 1989 causing
criminals to serve 85 percent of their sentence, and thus judges may
have correspondingly altered their rulings. Even then, this change was
phased in over time because the law only applied to crimes that took
place after it went into effect in 1989. In addition, the Oregon
system did not keep complete records prior to 1987, and the
completeness of these records decreased the further into the past one
went. One solution to both of these problems is to interact the prison
sentence length with year dummy variables. A similar problem exists
for Arizona which adopted a truth-in-sentencing reform during the fall
of 1994. Finally, Arizona is different from Oregon and Pennsylvania in
that it already allowed handguns to be carried openly before passing
its concealed handgun law, thus one might expect to find a somewhat
smaller response to adopting a concealed handgun law.
In addition to using county dummy variables, other data were collected
from the Bureau of the Census to try controlling for other demographic
characteristics that might determine the crime rate. These data
included information on the population density per square mile, total
county population, and detailed information on the racial and age
breakdown of the county (percent of population by each racial group
and by sex between 10 and 19 years of age, between 20 and 29, between
30 and 39, between 40 and 49, between 50 and 64, and 65 and over).
(See Table 2 for the list and summary statistics.) While a large
literature discusses the likelihood of younger males engaging in crime
(e.g., Wilson and Herrnstein, 1985, pp. 126-147), controlling for
these other categories allows us to also attempt to measure the size
of the groups considered most vulnerable (e.g., females in the case of
rape).[20] Recent evidence by Glaeser and Sacerdote (1995) confirms
the higher crime rates experienced in cities and examines to what
extent this arises due to social and family influences as well as the
changing pecuniary benefits from crime, though this is the first paper
to explicitly control for population density. The data appendix
provides a more complete discussion of the data.
An additional set of income data was also used. These included real
per capita personal income, real per capita unemployment insurance
payments, real per capita income maintenance payments, and real per
capita retirement payments per person over 65 years of age.[21]
Including unemployment insurance and income maintenance payments from
the Commerce Department's Regional Economic Information System (REIS)
data set were attempts to provide annual county level measures of
unemployment and the distribution of income.
Finally, we recognize that other legal changes in penalties involving
improper gun use might also have been changing simultaneously with
changes in the permitting requirements for concealed handguns. In
order to see whether this might confound our ability to infer what was
responsible for any observed changes in crimes rates we read through
various editions of the Bureau of Alcohol, Tobacco, and Firearms'
State Laws and Published Ordinances - Firearms (1976, 1986, 1989, and
1994). Excluding the laws regarding machine guns and sawed-off
shotguns, there is no evidence that the laws involving the use of guns
changed significantly when concealed permit rules were changed.[22]
Another survey which addresses the somewhat boarder question of
sentencing enhancement laws for felonies committed with deadly weapons
(firearms, explosives, and knives) from 1970-1992 also confirms this
general finding with all but four of the legal changes clustered from
1970 to 1981 (Marvell and Moody, 1995, pp. 258-261). Yet, controlling
for the dates supplied by Marvell and Moody still allows us to examine
the deterrence effect of criminal penalties specifically targeted at
the use of deadly weapons during this earlier period.[23]
IV. The Empirical Evidence
A. Using County Data for the United States
The first group of regressions reported in Table 3 attempt to explain
the natural log of the crime rate for nine different categories of
crime. The regressions are run using weighted ordinary least squares.
While we are primarily interested in a dummy variable to represent
whether a state has a "shall issue" law, we also control for each type
of crime's the arrest rate, demographic differences, and dummies for
the fixed effects for years and counties. The results imply that
"shall issue" laws coincide with fewer murders, rapes, aggravated
assaults, and rapes.[24] On the other hand, auto theft and larceny
rates rise. Both changes are consistent with our discussion on the
direct and substitution effects produced by concealed weapons.[25]
Rerunning these specifications with only the "shall issue" dummy, the
arrest rates, and the fixed year and county effects produces even more
significant effects for the "shall issue" dummy and the arrest rates.
The results are large empirically. When state concealed handgun laws
went into effect in a county, murders fell by 8.5 percent, and rapes
and aggravated assaults fell by 5 and 7 percent. In 1992, there were
18,469 murders; 79,272 rapes; 538,368 robberies; and 861,103
aggravated assaults in counties without "shall issue" laws. The
coefficients imply that if these counties had been subject to state
concealed handgun laws, murders in the United States would have
declined by 1,570. Given the concern that has been raised about
increased accidental deaths from concealed weapons, it is interesting
to note that the entire number of accidental gun deaths in the United
States in 1992 was 1,409. Of this total, 546 accidental deaths were in
states with concealed handgun laws and 863 were in those without these
laws. The reduction in murders is as much as three times greater than
the total number of accidental deaths in concealed handgun states.
Thus, if our results are accurate, the net effect of allowing
concealed handguns is clearly to save lives. Similarly, the results
indicate that the number of rapes in states without "shall issue" laws
would have declined by 4,177; aggravated assaults by 60,363; and
robberies by 11,898.[26]
On the other hand, property crime rates definitely increased after
"shall issue" laws were implemented. The results are equally dramatic.
If states without concealed handgun laws had passed such laws, there
would have been 247,165 more property crimes in 1992 (a 2.7 percent
increase). Thus, criminals respond substantially to the threat of
being shot by instead substituting into less risky crimes.[27]
A recent National Institute of Justice study (Miller, Cohen, and
Wiersema, 1996) provides estimates the costs of different types of
crime based upon lost productivity; out-of-pocket expenses such as
medical bills and property losses; and losses for fear, pain,
suffering, and lost quality of life. While there are questions about
using jury awards to measure losses such as fear, pain, suffering, and
lost quality of life, the estimates provide us one method of comparing
the reduction in violent crimes with the increase in property crimes.
Using the numbers from Table 3, the estimated gain from allowing
concealed handguns is over $6.214 billion in 1992 dollars. The
reduction in violent crimes represents a gain of $6.6 billion ($4.75
billion from murder, $1.4 billion from aggravated assault, $374
million from rape, and $98 million from robbery), while the increase
in property crimes represents a loss of $417 million ($342 million
from auto theft, $73 million from larceny, and $1.5 million from
burglary). However, while $6.2 billion is substantial, to put it into
perspective, it equals only about 1.33 percent of the total aggregate
losses from these crime categories. These estimates are probably most
sensitive to the value of life used (in the Miller et. al. study this
was set at $1.84 million in 1992 dollars). Higher estimated values of
life will increase the net gains from concealed handgun use, while
lower values of life will reduce the gains.[28] To the extent that
people are taking greater risks towards crime because of any increased
safety produced by concealed handgun laws (again see Peltzman (1975)),
these numbers will underestimate the total savings from concealed
handguns.
The arrest rate produces the most consistent effect on crime. Higher
arrest rates imply lower crime rates for all categories of crime. A
one standard deviation change in the probability of arrest accounts
for 3 to 17 percent of a one standard deviation change in the various
crime rates. The crime most responsive to arrest rates is burglary (11
percent), followed by property crimes (10 percent); aggravated assault
and violent crimes more generally (9 percent); murder (7 percent);
rape, robbery, and larceny (4 percent); and auto theft (both 3
percent).
For property crimes, a one standard deviation change in the percent of
the population that is black, male, and between 10 and 19 years of age
explains 22 percent of these crime rates. For violent crimes, the same
number is 5 percent. Other patterns also show up in the data. For
example, more black females between the ages of 20 and 39, more white
females between the ages of 10 and 39 and those over 65, and other
race females between 20 and 29 are positively and significantly
associated with a greater number of rapes occurring. Population
density appears to be most important in explaining robbery, burglary,
and auto theft rates, with a one standard deviation change in
population density being able to explain 36 percent of a one standard
deviation change in auto theft. Perhaps most surprising is the
relatively small, even if frequently significant, effect of income on
crime rates. A one standard deviation change in real per capita income
explains no more than 4 percent of a one standard deviation change in
crime and in seven of the specifications it explains 2 percent or less
of the change. If the race, sex, and age variables are replaced with
variables showing the percent of the population that is black and the
percent that is white, 50 percent of a standard deviation in the
murder rate is explained by the percent of the population that is
black. Given the high rates that blacks are arrested and incarcerated
or are victims of crimes, this is not unexpected.
Rerunning the regressions by adding a dummy variable to control for
state laws that increase sentencing penalties when deadly weapon are
used (Marvell and Moody, 1995, pp. 259-260) has no noticeable effect
on the concealed handgun coefficients. The enhanced sentencing law
dummy is negative and statistically significant only for aggravated
assaults, with the coefficient implying that adopting this type of law
reduces aggravate assaults by 4 percent. Otherwise these laws
generally appear to have little effect on crime rates.
Given the wide use of state level crime data by economists and the
large within state heterogeneity shown in Table 1, Table 4 provides a
comparison by reestimating the specifications reported in Table 3
using state level rather than county level data. The only other
difference in the specification is the replacement of county dummies
with state dummies. While the results in these two tables are
generally similar, two differences immediately manifest themselves: 1)
all the specifications now imply a negative and almost always
significant relationship between allowing concealed handguns and the
level of crime and 2) concealed handgun laws explain much more of the
variation in crime rates while arrest rates (with the exception of
robbery) explain much less of the variation.[29] Despite the fact that
concealed handgun laws appear to lower both violent and property crime
rates, the results still imply that violent crimes are much more
sensitive to the introduction of concealed handguns, with violent
crimes falling three times more than property crimes. These results
imply that if all states had adopted concealed handgun laws in 1992,
1,777 fewer murders and 7,000 fewer rapes would have taken place.[30]
Overall, Table 4 implies that the estimated gain from the lower crime
produced by handguns was $10.3 billion in 1992 dollars (see Table 5).
Yet, at least in the case of property crimes, the concealed handgun
law coefficients' sensitivity to whether these regressions are run at
the state or county level suggests caution in aggregating these data
into such large units as states.
Table 6 examines whether changes in concealed handgun laws and arrest
rates have differential effects in high or low crime counties. To test
this, the regressions shown in Table 3 were reestimated first using
the sample above the median crime rate by type of crime and then
separately using the sample below the median. High crime rates may
also breed more crime because the stigma from arrest may be less when
crime is rampant (Ramusen, 1996). If so, any change in apprehension
rates should produce a greater reputational impact and thus greater
deterrence in low crime than high crime counties.
The results indicate that the concealed handgun law's coefficient
signs are consistently the same for both low and high crime counties,
though for two of the crime categories (rape and aggravate assault)
concealed handgun laws have only statistically significant effects in
the relatively high crime counties. For most violent crimes such as
murder, rape, and aggravated assault concealed weapons laws have a
much greater deterrent effect in high crime counties, while for
robbery, property crimes, auto theft, burglary, and larceny the effect
appears to be greatest in low crime counties. The table also shows
that the deterrent effect of arrests is significantly different at
least at the 5 percent level between high and low crime counties for
eight of the nine crime categories (the one exception being violent
crimes). The results do not support the claim that arrests produce a
greater reputational penalty in low crime areas. While additional
arrests in low and high crime counties produce virtually identical
changes in violent crime rates, the arrest rate coefficient for high
crime counties is almost four times bigger than it is for low crime
counties.
One relationship in these first three sets of regressions deserves a
special comment. Despite the relatively small number of women using
concealed handgun permits, the concealed handgun coefficient for
explaining rapes is consistently comparable in size to the effect that
this variable has on other violent crimes rates. In Washington and
Oregon states in January 1996, women constituted 18.6 and 22.9 percent
of those with concealed handgun permits for a total of 118,728 and
51,859 permits respectively.[31] The time-series data which are
available for Oregon during our sample period even indicates that only
17.6 percent of permit holders were women in 1991. While it is
possible that the set of women who are particularly likely to be raped
might already carry concealed handguns at much higher rates than the
general population of women, the results are at least suggestive that
rapists are particularly susceptable to this form of deterrence.
Possibly this arises since providing a woman with a gun has a much
bigger affect on her ability to defend herself against a crime than
providing a handgun to a man. Thus even if relatively few women carry
handguns, the expected change in the cost of attacking women could
still be nearly as great. To phrase this differently, the external
benefits to other women from a women carrying a concealed handgun
appear to be large relative to the gain produced by an additional man
carrying a concealed handgun. If concealed handgun use were to be
subsidized to capture these positive externalities, these results are
consistent with efficiency requiring that women receive the largest
subsidies.[32]
As mentioned in Section II, an important concern with these data is
that passing a concealed handgun law should not affect all counties
equally. In particular, we expect that it was the most populous
counties that most restricted people's ability to carry concealed
weapons. To test this, Table 7 repeats all the regressions in Table 3
but instead interacts the Shall Issue Law Adopted Dummy with county
population. While all the other coefficients remain virtually
unchanged, this new interaction retains the same signs as those for
the original Shall Issue Dummy, and in all but one case the
coefficients are more significant. The coefficients are consistent
with the hypothesis that the new laws produced the greatest change in
the largest counties. The larger counties have a much greater response
in both directions to changes in the laws. Violent crimes fall more
and property crimes rise more in the largest counties. The bottom of
the table indicates how these effects vary for different size
counties. For example, passing a concealed handgun law lowers the
murder rate in cities two standard deviations above the mean
population by 12 percent, 7.4 times more than a shall issue laws
lowers murders for the mean population city. While the law enforcement
officers we talked to continually mentioned population as being the
key variable, we also reran these regressions using population density
as the variable that we interacted with the shall issue dummy. The
results remain very similar to those reported.
Admittedly, although arrest rates and county fixed effects are
controlled for, these regressions have thus far controlled for
expected penalties in a limited way. Table 8 reruns the regressions in
Table 7 but includes either the burglary or robbery rates to proxy for
other changes in the criminal justice system. Robbery and burglary are
the violent and property crime categories that are the least related
to changes in concealed handgun laws, but they are still positively
correlated with all the other types of crimes. One additional minor
change is made in two of the earlier specifications. In order to avoid
any artificial collinearity either between violent crime and robbery
or between property crimes and burglary, violent crimes net of robbery
and property crimes net of burglary are used as the endogenous
variables when robbery or burglary are controlled for.
Some evidence that burglary or robbery rates will proxy for other
changes in the criminal justice system can be seen in their
correlations with other crime categories. The Pearson correlation
coefficient between robbery and the other crime categories ranges
between .49 and .80, and all are statistically significant at least at
the .0001 level. For burglary the correlations range from .45 to .68,
and they are also equally statistically significant. The two sets of
specifications reported in Table 8 closely bound our earlier
estimates, and the estimates continue to imply that the introduction
of concealed handgun laws coincided with similarly large drops in
violent crimes and increases in property crimes. The only difference
with the preceding results is that they now imply that the affect on
robberies is statistically significant. The estimates on the other
control variables also essentially remain unchanged.
We also reestimated the regressions in Table 3 using first differences
on all the control variables (see Table 9). These regressions were run
using a dummy variable for the presence of "shall issue" concealed
handgun laws and differencing that variable, and the results
consistently indicate a negative and statistically significant effect
from the legal change for violent crimes, rape, and aggravated
assault. Shall issue laws negatively affect murder rates in both
specifications, but the effect is only statistically significant when
the shall issue variable is also differenced. The property crime
results are also consistent with those shown in the previous tables,
showing a positive impact of shall issue laws on crime rates. Perhaps
not surprisingly, the results imply that the gun laws immediately
altered crime rates, but that an additional change was spread out over
time, possibly because concealed handgun use did not instantly move to
its new steady state level. The annual decrease in violent crimes
averaged about 2 percent, while the annual increase in property crimes
average about 5 percent.
All the results in tables 3, 6, and 7 were reestimated to deal with
the concerns raised in Section II over the "noise" in arrest rates
arising from the timing of offenses and arrests and the possibility of
multiple offenders. We reran all the regressions in this section first
by limiting the sample to those counties over 100,000 and then 200,000
people. Consistent with the evidence reported in Table 7, the more the
sample was limited to larger population counties the stronger and more
statistically significant was the relationship between concealed
handgun laws and the previously reported effects on crime. The arrest
rate results also tended to be stronger and more significant. We also
tried rerunning all the regressions by redefining the arrest rate as
the number of arrests over the last three years divided by the total
number of offenses over the last three years. Despite the reduced
sample size, the results remained similar to those already reported.
Not only does this initial empirical work provide strong evidence that
concealed handgun laws reduce violent crime and that higher arrest
rates deter all types of crime, but the work also allows us to
evaluate some of the broader empirical issues concerning criminal
deterrence discussed in Section II. The results confirm some of our
earlier discussion on potential aggregation problems with state level
data. County level data implies that arrest rates explain about six
times the variation in violent crime rates and eight times the
variation in property crime rates that arrest rates explain when we
use state level data. Breaking the data down by whether a county is a
high or a low crime county indicates that arrest rates do not affect
crime rates equally in all counties. The evidence also confirms the
claims of law enforcement officials that "Shall Issue" laws
represented more of a change in how the most populous counties
permitted concealed handguns. One concern that was not borne out was
over whether state level regressions could bias the coefficients on
the concealed handgun laws towards zero. In fact, while state and
county level regressions produce widely different coefficients for
property crimes, seven of the nine crime categories imply that the
effect of concealed handgun laws was much larger when state level data
were used. However, one conclusion is clear: the very different
results between state and county level data should make us very
cautious in aggregating crime data and would imply that the data
should remain as disaggregated as possible.
B. The Endogeniety of Arrest Rates and the Passage of Concealed
Handgun Laws
The previous specifications have assumed that both the arrest rate and
the passage of concealed handgun laws are exogenous. Following Ehrlich
(1973, pp. 548-551), we allow for the arrest rate to be a function of:
the lagged crime rates; per capita and per violent and property crimes
measures of police employment and payroll at the state level (these
three different measures of employment are also broken down by whether
police officers have the power to make arrest); the measures of
income, unemployment insurance payments, and the percentages of county
population by age, sex, and race used in Table 3; and county and year
dummies.[33] In an attempt to control for political influences, we
also included the percent of a state's population that are members of
the National Rifle Association and the percent of the vote received by
the Republican presidential candidate at the state level. Because
presidential candidates and issues vary between elections, the percent
voting Republican is undoubtedly not directly comparable across years.
To account for these difference across elections, we interacted the
percent voting Republican with dummy variables for the years
immediately next to the relevant elections. Thus, the percent of the
vote obtained in 1980 is multiplied by a year dummy for the years from
1979 to 1982, the percent of the vote obtained in 1984 is multiplied
by a year dummy for the years from 1983 to 1986, and so on through the
1992 election. A second set of regressions explaining the arrest rate
also include the change in the natural log of the crime rates to proxy
for the difficulty police forces face in adjusting to changing
circumstances.[34] However, the time period studied in all these
regressions is more limited than in our previous tables because state
level data on police employment and payroll are only available from
the U.S. Department of Justices' Expenditure and Employment data for
the Criminal Justice System from 1982 to 1992.
There is also the question of why some states adopted concealed
handgun laws while others did not. As noted earlier, to the extent
that states adopted the law because crime was either rising or was
expected to increase, ordinary least squares estimates underpredict
the drop in crime. Similarly, if these rules were adopted when crimes
rates were falling, a bias is in the opposite direction. Thus, in
order to predict whether a county would be in a state with concealed
handgun laws we used both the natural logs of the violent and property
crime rates and the first differences of those crime rates. To control
for general political differences that might affect the chances of
these laws being adopted, we also included the National Rifle
Association membership as a percent of a state's population; the
Republican presidential candidate's percent of the statewide vote; the
percentage a state's population that is black and the percent white;
the total population in the state; regional dummy variables for
whether the state is in the South, Northeast, or Midwest; and year
dummy variables.
While the 2SLS estimates shown in the top half of Table 10 again use
the same set of control variables employed in the preceding tables,
the results differ from all our previous estimates in one important
respect: concealed handgun laws are associated with large significant
drops in the levels of all nine crime categories. For the estimates
most similar to Ehrlich's study, five of the estimates imply that a
one standard deviation change in the predicted value of the Shall
Issue Law dummy variable explains at least 10 percent of a standard
deviation change in the corresponding crime rates. In fact, concealed
handgun laws explain a greater percentage of the change in murder
rates than do arrest rates. With the exception of robbery, the set of
estimates using the change in crime rates to explain arrest rates
indicates a usually more statistically significant but economically
smaller effect from concealed handgun laws. For example, concealed
handgun laws now explains 3.9 percent of the variation in murder rates
compared to 7.5 percent in the preceding results. While these results
imply that even crimes with relatively little contact between victims
and criminals experienced declines, the coefficients for violent
crimes are still relatively more negative than the coefficients for
property crimes.
For the first stage regressions explaining which states adopt
concealed handgun laws (shown in the bottom half of Table 10), both
the least square and logit estimates imply that the states adopting
these laws are relatively Republican with large National Rifle
Association memberships and low but rising violent and property crime
rates. The other set of regressions used to explain the arrest rate
shows that arrest rates are lower in high income, sparsely populated,
Republican areas where crime rates are increasing.
We also reestimated the state level data using similar two-stage least
squares specifications. The coefficients on both the arrest rates and
concealed handgun law variables remained consistently negative and
statistically significant, with the state level data again implying a
much stronger effect from concealed handguns and a much weaker effect
from higher arrest rates. Finally, in order to use the longer data
series available for the nonpolice employment and payroll variables,
we reran the regressions without those variables and produced similar
results.
C. Concealed Handgun Laws, the Method of Murder, and the Choice of
Murder
Victims
Do concealed handgun laws cause a substitution in the methods of
committing murders? For example, it is possible that the number of gun
murders rises after these laws are passed even though the total number
of murders falls. While concealed handgun laws raise the cost of
committing murders, murderers may also find it relatively more
dangerous to kill people using nongun methods once people start
carrying concealed handguns and substitute into guns to put themselves
on a more even basis with their potential prey. Using data on the
method of murder from the Mortality Detail Records provided by the
United States Department of Health and Human Services, we reran the
murder rate regression from Table 3 on counties over 100,000 during
the period from 1982 to 1991. We then separated out murders caused by
guns from all other murders. Table 11 shows that carrying concealed
handguns appears to have been associated with approximately equal
drops in both categories of murders. Carrying concealed handguns
appears to make all types of murders realtively less attractive.
There is also the question of what effect does conceal handgun laws
have on determining which types of people are more likely to be
murdered? Using the Uniform Crime Reports Supplementary Homicide
Reports we were able to obtain annual state level data from 1977 to
1992 on the percent of victims by sex and race as well as information
on the whether the victim and the offender knew each other (whether
they were members of the same family, knew each other but were not
members of the same family, strangers, or the relationship is
unknown).[35] Table 12 implies no statistically significant
relationship between the concealed handgun dummy and the victim's sex,
race, or relationships with offenders. However, while they are not
quite statistically significant at the .10 level for a two-tailed
t-test, two of the point estimates appear economically important and
imply that in states with concealed handgun laws victims know their
nonfamily offenders 2.6 percentage points more frequently and that the
percent of victims where it was not possible to determine whether a
relationship existed declined by 2.9 percentage points. This raises
the question of whether concealed handguns cause criminals to
substitute into crimes against those whom they know and presumably are
also more likely to know whether they carry concealed handguns.
The arrest rate for murder variable produces more interesting results.
The percent of white victims and the percent of victims killed by
family members both declined when states passed concealed handgun
laws, while the percent of black victims and the percent that killed
by nonfamily members that they know both increased. The results imply
that higher arrest rates have a much greater deterrence effect on
murders involving whites and family members. One explanation is that
whites with higher incomes face a greater increase in expected
penalties for any given increase in the probability of arrest.
D. Arizona, Pennsylvania, and Oregon County Data
One problem with the preceding results was the use of county
population as a proxy for how restrictive counties were in allowing
concealed handgun permits before the passage of "shall issue" laws.
Since we are still going to control county specific levels of crime
with county dummies, a better measure would have been to use the
actual change in a gun permits before and after the adoption of a
concealed handgun law. Fortunately, we were able to get that
information for three states: Arizona, Oregon, and Pennsylvania.
Arizona and Oregon also provided additional information on the
conviction rate and the mean prison sentence length. However, for
Oregon, because the sentence length variable is not directly
comparable over time, it is interacted with all the year dummies so
that we can still retain any cross-sectional information in the data.
One difficulty with the Arizona prison sentence and conviction data is
that they are available only from 1990 to 1995 and that since the
shall issue handgun law did not take effect until July 1994, it is not
possible for us to control for all the other variables that we control
for in the other regressions. Unlike Oregon and Pennsylvania, Arizona
did not allow private citizens to carry concealed handguns prior to
July 1994, so the value of concealed handgun permits equals zero for
this earlier period. Unfortunately, however, because Arizona's change
in the law is so recent, we are unable to control for all the
variables that we can control for in the other regressions.
The results in Table 14 for Pennsylvania and Table 15 for Oregon
provide a couple of consistent patterns. The most economically and
statistically important relationship involves the arrest rate: higher
arrest rates consistently imply lower crime rates, and in 12 of the 16
regressions the effect is statistically significant. Five cases for
Pennsylvania (violent crime, murder, aggravated assault, robbery, and
burglary) show that arrest rates explain more than 20 percent of a
standard deviation change in crime rates. Automobile theft is the only
crime for which the arrest rate is insignificant in both tables.
For Pennsylvania, rape is the one crime where a one standard deviation
change in per capita concealed handgun permits explains a greater
percentage of a standard deviation in crime rates than it does for the
arrest rate. However, increased concealed handguns usage explains more
than 10 percent of a standard deviation change in murder, rape,
aggravated assualt, and burglary rates. For six of the nine
regressions, the concealed handgun variable for Pennsylvania exhibits
the same coefficient signs that were shown for the national data.
Violent crimes, with the exception of robbery, show that higher
concealed handgun use significantly lowers crime rates, while property
crimes exhibit the opposite tendency. However, concealed handgun use
only explains about half the variation for property crimes that it
explains for violent ones.[36] The regressions for Oregon weakly imply
a similar relationship between concealed handgun use and crime, but
the effect is only statistically significant in one case: larceny,
which is also the only crime category where the negative concealed
handgun coefficient differs from our previous findings.
The Oregon data also show that higher conviction rates consistently
result in significantly lower crime rates. A one standard deviation
change in conviction rates explains 4 to 20 percent of a one standard
deviation change in the corresponding crime rates. However, increases
in conviction rates appear to produce a smaller deterrent effect than
increases in arrest rates for five of the seven crime categories.[37]
The biggest differences between the deterrence effects of arrest and
conviction rates produce an interesting pattern. For rape, increasing
the arrest rate by one percentage point produces more than ten times
the deterrent effect of increasing the conviction rate conditional on
arrest by one percent. The reverse is true for auto theft where a one
percentage point increase in reduces crime by about ten times more
than the same increase in convictions. These results are consistent
with arrests producing large shaming or reputational penalties (e.g.,
see Kahan 1996). In fact, the existing evidence shows that the
reputational penalties from arrest and conviction can dwarf the other
legally imposed penalties (Lott, 1992a and b). However, while the
literature has not separated out whether these drops are occurring due
to arrest or conviction, these results are consistent with the
reputational penalties for arrests alone being significant for at
least some crimes.
The results for the prison sentences are not shown, but the
t-statistics are frequently near zero and the coefficients indicate no
clear pattern. One possible explanation for this result is that all
the changes in sentencing rules produced a great deal of noise in this
variable not only over time but also across counties. For example,
after 1989 whether a crime was prosecuted under the pre or post 1989
rules depended upon when the crime took place. If the average time
between when the offense occurred and when the prosecution took place
differs across counties, the recorded prison sentence length could
vary even if the actual time served was the same.
Finally, the much more limited data set for Arizona used in Table 16
produces no significant relationship between the change in concealed
handgun permits and the various measures of crime rates. In fact, the
coefficient signs themselves indicate no consistent pattern with the
fourteen coefficients being equally divided between negative and
positive signs, though six of the specfications imply that a one
standard deviation change in the concealed handgun permits explains at
least 8 percent of a one standard deviation change in the
corresponding crime rates. The results involving either the mean
prison sentence length for those sentenced in a particular year or the
actual time served for those ending their sentences also imply no
consistent relationship between prison and crime rates. While the
coefficients are negative in 11 of the 14 specifications, they provide
weak evidence of the deterrent effect of longer prison terms: only two
coefficients are negative and statistically significant.
Overall, the Pennsylvania results provide more evidence that concealed
handgun ownership reduces violent crime, murder, rape, aggravated
assault, and burglary; and in the case of Oregon larceny decreases as
well. While the Oregon data implies that the change in handgun permits
is statistically significant at .11 percent level for a one-tailed
t-test, the point estimate is extremely large economically: implying
that a doubling of permits reduces murder rates by 37 percent. The
other coefficients for Pennsylvania and Oregon imply no significant
relationship between the change in concealed handgun ownership and
crime rates. The evidence from the small sample for Arizona implies no
relationship between crime and concealed handgun ownership. All the
results also support the claim that higher arrest and conviction rates
deter crime, though, possibly in part due to the relatively poor
quality of the data, no systematic effect appears to occur from longer
prison sentences.
V. Accidental Deaths from Handguns
Even if "shall issue" hand gun permits lower murder rates, the
question of what happens to accidental deaths still remains. Possibly,
with more people carrying handguns, accidents may be more likely to
happen. Earlier we saw that the number of murders prevented exceeded
the entire number of accidental deaths. As Table 2 showed, while only
a small portion of either accidental deaths are attributable to
handgun laws, there is still the question whether concealed handgun
laws affected the total number of deaths through their effect on
accidental deaths.
To get a more precise answer to this question, Table 17 uses county
level data from 1982 to 1991 to test whether allowing concealed
handguns increased accidental deaths. Data are available from the
Mortality Detail Records (provided by the United States Department of
Health and Human Services) for all counties from 1982 to 1988 and for
counties over 100,000 population from 1989 to 1991. The specifications
are identical to those shown in all the previous tables with the
exceptions that we no longer include variables related to arrest or
conviction rates and that the endogenous variables are replaced with
either a measure of the number of accidental deaths from handguns or
accidental deaths from all other nonhandgun sources.
While there is some evidence that the racial composition of the
population and the level of income maintenance payments affect
accident rates, the coefficient of the shall issue dummy is both quite
small economically and insignificant. The point estimates for the
first specification implies that accidental handgun deaths rose by
about .5 percent when concealed handgun laws were passed. With only
156 accidental handgun deaths occurring in counties over 100,000
population (27 accidental handgun deaths occurred in states with
"shall issue" laws), this point estimate implies that implementing a
concealed handgun law in those states which currently do not have it
would produce less than one more death (.645 deaths).
Given the very small number of accidental handgun deaths in the United
States, the vast majority of counties have an accidental handgun death
rate of zero and thus using ordinary least squares is not the
appropriate method of estimating these relationships. To deal with
this, the last two columns in Table 17 reestimate these specifications
using Tobit procedures. However, because of limitations in statistical
packages we were no longer able to control for all the county dummies
and opted to rerun these regressions with only state dummy variables.
While the coefficients for the concealed handgun law dummy variable is
not statistically significant, with 186 million people living in
states without these laws in 1992,[38] the third specification implies
that implementing the law across those remaining states would have
resulted in about 9 more accidental handgun deaths. Combining this
finding with the earlier estimates from Tables 3 and 4, if the rest of
the country had adopted concealed handgun laws in 1992, the net
reduction in total deaths would have been approximately 1,561 to
1,767.
VI. Conclusion
Allowing citizens without criminal records or histories of significant
mental illness to carry concealed handguns deters violent crimes and
appears to produce an extremely small and statistically insignificant
change in accidental deaths. If the rest country had adopted
right-to-carry concealed handgun provisions in 1992, at least 1,570
murders and over 4,177 rapes would have been avoided. On the other
hand, consistent with the notion that criminals respond to incentives,
county level data provides evidence that concealed handgun laws are
associated with increases in property crimes involving stealth and
where the probability of contact between the criminal and the victim
are minimal. The largest population counties where the deterrence
effect on violent crimes is the greatest is also where the
substitution effect into these property crimes is the highest. The
estimated annual gain in 1992 from allowing concealed handguns was
over $6.21 billion.
The data also supply dramatic evidence supporting the economic notion
of deterrence. Higher arrest and conviction rates consistently and
dramatically reduce the crime rate. Consistent with other recent work
(Kahan, 1996 and Lott, 1992b), the results imply that increasing the
arrest rate, independent of the probability of eventual conviction,
imposes a significant penalty on criminals. Perhaps the most
surprising result is that the deterrence effect of a one percentage
point increase in arrest rates is much larger than the same increase
in the probability of conviction. Also surprising was that while
longer prison lengths usually implied lower crime rates, the results
were normally not statistically significant.
This study incorporates a number of improvements over previous studies
on deterrence, and it represents a very large change in how gun
studies have been done. This is the first study to use cross-sectional
time-series evidence for counties at both the national level and for
individual states. Instead of simply using cross-sectional state or
city level data, our study has made use of the much bigger variations
in arrest rates and crime rates between rural and urban areas, and it
has been possible to control for whether the lower crime rates
resulted from the gun laws themselves or other differences in these
areas (e.g., low crime rates) which lead to the adoption of these
laws. Equally importantly, our study has allowed us to examine what
effect concealed handgun laws have on different counties even within
the same state. The evidence indicates that the effect varies both
with a county's level of crime and its population.
Bibliography
Andreoni, James, "Criminal Deterrence in the Reduce Form: A New
Perspective on Ehrlich's Seminal Study," Economic Inquiry, Vol. 33,
no. 3 (July 1995): 476-483.
Annest, J.L.; J.A. Mercy; D.R. Gibson; and G.W. Ryan, "National
Estimates of NonFatal Firearem-related Injuries, Beyond the Tip of the
Iceberg," Journal of the American Medical Association (June 14, 1995):
1749-54.
Barhnhart, Bob, "Concealed Handgun Licensing in Multnomah County,"
mimeo from the Intelligence/Concealed Handgun Unit: Multnomah County
(October 1994).
Block, Michael K. and John Heineke, "A Labor Theoretical Analysis of
Criminal Choice," American Economic Review, Vol. 65 (June 1975):
314-325.
Cook, P.J., "The Role of Firearms in Violent Crime," In Wolfgang, M.E.
and N.A. Werner (eds.), Criminal Violence, Beverly Hills: Sage
Publishers (1982): 236-291.
________, "The Technology of Personal Violence," Crime and Justice:
Annual Review of Research, Vol. 14 (1991): 57-87.
________, Stephanie Molliconi, and Thomas B. Cole, "Regulating Gun
Markets," Journal of Criminal Law and Criminology, Vol. 86, no. 1
(Fall 1995): 59-92.
Cramer, Clayton E. and David B. Kopel, "`Shall Issue': The New Wave of
Concealed Handgun Permit Laws," Tennessee Law Review, Vol. 62 (Spring
1995): 679-758, and expanded version of this paper dated 1994 is also
available from the Independence Institute, Golden, Colorado.
Ehrlich, Isaac, "Participation in Illegitimate Activities: A
Theoretical and Empirical Investigation," Journal of Political
Economy, Vol. 81, no. 3 (1973): 521-565.
Federal Bureau of Investigation, Crime in the United States, Federal
Bureau of Investigation: Washington, D.C. (editions for 1977 to 1992).
Fort Worth Star-Telegram, "Few Probelms Reported After Allowing
Concealed Handguns, Officers Say," Fort Worth Star-Telegram (July 16,
1996).
Glaeser, Edward L. and Bruce Sacerdote, "Why is There More Crime in
Cities?" Presented at Symposium in Honor of Gary Becker's 65th
Birthday, Harvard University working paper (November 14, 1995).
Greenwald, Bruce C. "A General Analysis of the Bias in the Estimated
Standard Errors of Least Squares Coefficients," Journal of
Econometrics, Vol. 22 (August 1983): 323-338.
Grossman, Michael, Frank J. Chaloupka, and Charles C. Brown, "The
Demand for Cocaine by Young Adults: A Rational Addiction Approach,"
NBER Working Paper (July 1996).
Japan Economic Newswire, "U.S. jury clears man who shot Japanese
student," Kyodo News Service (May 24, 1993).
Kahan, Dan M., "What Do Alternative Sanctions Mean?," University of
Chicago Law Review, Vol. 63, no. 1 (1996): 591-653.
Kleck, Gary, "Guns and Violence: An Interpretive Review of the Field,"
Social Pathology, Vol. 1, no. 1 (January 1995): 12-47.
________ and E. Britt Patterson, "The Impact of Gun Control and Gun
Ownership Levels on Violence Rates," Journal of Quantitative
Criminology, Vol. 9 (1993): 249-287.
________ and Marc Gertz, "Armed Resistance to Crime: The Prevalence
and Nature of Self-Defense with a Gun," Journal of Criminal Law and
Criminology, Vol. 86, no. 1 (Fall 1995): 150-187.
Kopel, David B., The Samuri, the Mountie, and the Cowboy, Prometheus
Books: Buffalo, New York (1992).
________, Guns: Who Should Have Them?, Prometheus Books: Buffalo, New
York (1995).
Landes, William M., "An Economic Study of U.S. Aircraft Hijacking,
1961-1976," Journal of Law and Economics, Vol. 21, no. 1 (April 1978):
1-31.
Levitt, Steven, "The Effect of Prison Population Size on Crime Rates:
Evidence from Prison Overcrowding Litigation," Quarterly Journal of
Economics (1996).
Lipton, Eric, "Virginians Get Ready to Conceal Arms; State's New
Weapon Law Brings a Flood of Inquiries," The Washington Post (June 28,
1995): A1.
Lott, John R., Jr., "Juvenile Delinquency and Education: A Comparison
of Public and Private Provision," International Review of Law and
Economics, Vol.7, no. 2 (December 1987): 163-175.
________, "A Transaction-Costs Explanation for Why the Poor are More
Likely to Commit Crime," Journal of Legal Studies, Vol. 19, no. 1
(January 1990a): 243-245.
________, "The Effect of Conviction on the Legitimate Income of
Criminals," Economics Letters, Vol. 34, no. 12 (December 1990b):
381-385.
________, "An Attempt at Measuring the Total Monetary Penalty from
Drug Convictions: The Importance of an Individual's Reputation,"
Journal of Legal Studies, Vol. 21, no. 1 (January 1992a): 159-187.
________, "Do We Punish High Income Criminals too Heavily?" Economic
Inquiry, Vol. 30, no. 4 (October 1992b): 583-608.
________, "Now That The Brady Law is Law, You Are Not Any Safer Than
Before," Philadelphia Inquirer, Tuesday, February 1, 1994, p. A9.
Marvell, Thomas B. and Carlisle E. Moody, "The Impact of Enhanced
Prison Terms for Felonies Committed with Guns," Criminology, Vol. 33,
no. 2 (May 1995): 247-282.
McCormick, Robert E. and Robert Tollison, "Crime on the Court,"
Journal of Political Economy, Vol. 92, no. 2 (April 1984): 223-235.
McDowall, David; Colin Loftin; and Brian Wiersema, "Easing Conealed
Firearm Laws: Effects on Homicide in Three States," Journal of
Criminal Law and Criminology, Vol. 86, no. 1 (Fall 1995): 193-206.
Miller, Ted R.; Mark A. Cohen; and Brian Wiersema, Victim Costs and
Consequences: A New Look, National Institute of Justice: Washington,
D.C. (February 1996).
Moulton, Brent R., "An Illusration of a Pitfall in Estimating the
Effects of Aggregate Variables on Micro Units," Review of Economics
and Statistics, Vol. 72 (1990): 334-338.
Peltzman, Sam, "The Effects of Automobile Safety Regulation," Journal
of Political Economy Vol. 883, no. 4 (August 1975): 677-725.
Polsby, Daniel D., "Firearms Costs, Firearms Benefits and the Limits
of Knowledge," Journal of Criminal Law and Criminology, Vol. 86, no. 1
(Fall 1995): 207-220.
Potok, Mark, "Texan says gun law saved his life'I did what I thought I
had to do'," USA TODAY (March 22, 1996): 3A.
Rasmusen, Eric, "Stigma and Self-Fulfilling Expectations of
Criminality," Journal of Law and Economics, forthcoming October 1996.
Reynolds, Morgan O., "Crime and Punishment in America," National
Center for Policy Analysis, Policy Report 193 (June 1995).
Sharn, Lori, "Violence shoots holes in USA's tourist image," USA TODAY
(September 9, 1993): 2A.
Southwick, Lawrence, Jr., "Self-defense with Guns: The Consequences,"
SUNY Buffalo working paper (1996).
Uviller, H. Richard, Virtual Justice, Yale University Press: New Haven
(1996).
Will, George F., "Are We `a Nation of Cowards'?" Newsweek (November
15, 1993): 93-94.
Zimring, Franklin, "Is Gun Control Likely to Reduce Violent
Killings?," University of Chicago Law Review, Vol. 35 (1968).
________, "The Medium is the Message: Firearm Caliber as a Determinant
of Death from Assult" Journal of Legal Studies, Vol. 1 (1972): 97-123.
________, "Firearms and Federal Law: The Gun Control Act of 1968"
Journal of Legal Studies, Vol. 4 (1975): 133-198.
Data Appendix
The number of arrests and offenses for each crime in every county from
1977-1992 were provided by the Uniform Crime Report. The UCR Program
is a nationwide, cooperative statistical effort of over 16,000 city,
county and state law enforcement agencies to compile data on crimes
that are reported to them. During 1993, law enforcement agencies
active in the UCR Program represented over 245 million U.S.
inhabitants, or 95% of the total population. The coverage amounted to
97% of the U.S. population living in Metropolitan Statistical Areas
(MSAs) and 86% of the population in non-MSA cities and in rural
counties.[39] The Uniform Crime Reports Supplementary Homicide Reports
supplied the data on the victim's sex and race and whatever
relationship might have existed between the victim andthe
offender.[40]
The regressions report results from a subset of the UCR data set,
though we also ran the regressions with the entire data set. The main
differences were that the effect of concealed handgun laws on murder
were greater than what is shown in this paper and the effects on rape
and aggravated assult were smaller. Observations were eliminated
because of changes in reporting practices or definitions of crimes
(see Crime in the United States (1977 to 1992)). For example, from
1985 to 1994 Illinois adopted a unique "gender-neutral" definition of
sex offenses. Another example involves Cook county, Illinois from 1981
to 1984 where there was a large jump in reported crime because there
was a change in the way officers were trained to report crime. The
additional observations droped from the data set include: Florida
(1988 to 1992); Georgia (1980); Kentucky (1988); Hawaii (1982); Iowa
(1991); Oakland, Ca. (1991 to 1992). The counties with the following
cities were also eliminated: aggravated assult for Steubenville, OH.
(1977 to 1990); aggravated assult for Youngstown, OH (1977 to 1988);
aggravated assult and burglary for Mobile, Al. (1977 to 1985);
aggravated assult for Milwaukee, WI (1977 to 1985); Glendale, AZ (1977
to 1984); aggravated assult for Jackson, MS (1982 and 1983);
aggravated assult for Aurora, CO (1982 and 1983); aggravated assult
for Beaumont, TX (1982 and 1983); aggravated assult for Corpus Cristi,
TX (1982 and 1983); rape for Macon, GA (1977 to 1981); robbery and
larceny for Cleveland, OH (1977 to 1981); aggravated assult for Omaha,
NE (1977 to 1981); Little Rock, Ark. (1977 to 1979); burglary and
larceny for Eau Claire, WI (1977 to 1978); Green Bay, WI. (1977); and
Fort Worth, TX (1977). For all of the different crime rates, if the
true rate equals zero, we added .1 before we took the natural log of
those values. For the accident rates, if the true rate equals zero, we
added .01 before we took the natural log of those values.[41]
The number of police in a state, which of those police have the power
to make arrests, and police payrolls for a state by type of police
officer are available for 1982 to 1992 from the U.S. Department of
Justice's Expenditure and Employment Data for the Criminal Justice
System.
The data on age, sex and racial distributions estimate the population
in each county on July 1 of the respective years. The population is
divided into five year segments and race is categorized as white,
black and neither white nor black. The population data, with the
exception of 1990 and 1992, were obtained from the Bureau of the
Census.[42] The estimates use modified census data as anchor points
and then employ an iterative proportional fitting technique to
estimate intercensal populations. The process ensures that the county
level estimates are consistent with estimates of July 1 national and
state populations by age, sex, and race. The age distributions of
large military installations, colleges, and institutions were
estimated by a separate procedure. The counties for which special
adjustments were made are listed in the report.[43] The 1990 and 1992
estimates have not yet been completed by the Bureau of the Census and
made available for distribution. We estimated the 1990 data by taking
an average of the 1989 and 1991 data. We estimated the 1992 data by
multiplying the 1991 populations by the 1990-1991 growth rate of each
county's populations.
Data on income, unemployment, income maintenance and retirement were
obtained by the Regional Economic Information System (REIS). Income
maintenance includes Supplemental Security Insurance (SSI), Aid to
Families with Dependent Children (AFDC), and food stamps. Unemployment
benefits include state unemployment insurance compensation,
Unemployment for Federal Employees, unemployment for railroad
employees, and unemployment for veterans. Retirement payments include
old age survivor and disability payments, federal civil employee
retirement payments, military retirement payments, state and local
government employee retirement payments, and workers compensation
payments (both federal and state). Nominal values were converted to
real values by using the consumer price index.[44] The index uses the
average consumer price index for July 1983 as the base period.
Data concerning the number of concealed weapons permits for each
county were obtained from a variety of sources. The Pennsylvania data
were obtained from Alan Krug. Mike Woodward of the Oregon Law
Enforcement and Data System provided the Oregon data for 1991 and
after. The number of permits available for Oregon by county in 1989
was provided by the sheriffs departments of the individual counties.
Cari Gerchick, Deputy County Attorney for Maricopa County in Arizona,
provided us with the Arizona county level conviction rates, prison
sentence lengths, and concealed handgun permits from 1990 to 1995. The
National Rifle Association provided data on NRA membership by state
from 1977 to 1992. Information on the dates at which states enacted
enhanced sentencing provisions for crimes committed with deadly
weapons was obtained from Marvell and Moody (1995, pp. 259-260). The
first year where the dummy variable comes on is weighted by the
portion of that first year that the law was in effect.
The Bureau of the Census provided data on the latitude, longitude and
area in square kilometers for each county. The number of total and
firearm unintentional injury deaths was obtained from annual issues of
Accident Facts and The Vital Statistics of the United States. The
classification of types of weapons is in International Statistical
Classification of Diseases and Related Health Problems, Tenth Edition,
Volume 1. The handgun category includes guns for single hand use,
pistols and revolvers. The total includes all other types of firearms.
Table 1: Comparing the
Deviation in Crime Rates
Between States and By
Counties Within States From
1977 to 1992: Does it make
sense to View States as
Relatively Homogenous Units?
Standard Deviation Mean of Within St
ate
of State Means Standard Deviatio
ns
Crime Rates Per 100,000
Population
Violent Crime Rate 284.77 255.57
Murder Rate 6.12 8.18
Murder Rate for Guns 3.9211 6.4756
(from 1982 to 1991)
Variable | Obs
Mean Std. Dev. Min
Max
---------+-------------------
-----------------------------
-----
RATMG | 23278
3.921104 6.475649
.0199036 142.6038
RATMNG | 21908
1.566327 8.675772
-120.2111 502.6832
LRATMUR | 19534
.1921255 2.139152
-2.3 6.22
Rape Rate 16.33 23.55
Aggravate Assault Rate 143.35 172.66
Robbery Rate 153.62 92.74
Property Crime Rate 1404.15 2120.28
Auto Theft Rate 162.02 219.74
Burglary Rate 527.70 760.22
Larceny Rate 819.08 1332.52
Arrest Rates Defined as the
Number of Arrests
Divided By the Number of
Offenses45
Arrest Rate for Violent 23.89 112.97
Crimes
Arrest Rate for Murder 18.58 88.41
Arrest Rate for Rape 19.83 113.86
Arrest Rate for Robbery 21.97 104.40
Arrest Rate for Aggravated 25.30 78.53
Assault
Arrest Rate for Property 7.907 44.49
Crimes
Arrest Rate for Burglary 5.87 25.20
Arrest Rate for Larceny 11.11 71.73
Arrest Rate for Auto Theft 17.37 118.94
Truncating Arrest Rates to
be no greater than one
Arrest Rate for Violent 11.11 25.40
Crimes
Arrest Rate for Murder 10.78 36.40
Arrest Rate for Rape 10.60 31.59
Arrest Rate for Robbery 8.06 32.67
Arrest Rate for Aggravated 11.14 27.08
Assault
Arrest Rate for Property 5.115 11.99
Crimes
Arrest Rate for Burglary 4.63 14.17
Arrest Rate for Larceny 5.91 12.97
Arrest Rate for Auto Theft 8.36 26.66
Table 2:
National Sample
Means and
Standard
Deviations
Variable Obs. Mean Standard Dev.
Gun Ownership
Information:
Shall Issue Dummy 50056 0.164704 0.3
68089
Arrests Rates
are the ratio of
arrests to
offenses for a
particular crime
category:
Arrest Rate for 45108 27.43394 126
.7298
Index Crimes
Arrest Rate for 43479 71.30733 327
.2456
Violent Crimes
Arrest Rate for 45978 24.02564 120
.8654
Property Crimes
Arrest Rate for 26472 98.04648 109
.7777
Murder
Arrest for Rape 33887 57.8318 132
.8028
Arrest for 43472 71.36647 187
.354
Aggravated
Assault
Arrest Rate for 34966 61.62276 189
.5007
Robbery
Arrest Rate for 45801 21.51446 47.
28603
Burglary
Arrest Rate for 45776 25.57141 263
.706
Larceny
Arrest Rate for 43616 44.8199 307
.5356
Auto Theft
Crime Rates are
Defined per
100,000 People:
Crime Rate for 46999 2984.99 336
8.85
Index Crimes
Crime Rate for 47001 249.0774 388
.7211
Violent Crimes
Crime Rate for 46999 2736.59 317
8.41
Property Crimes
Crime Rate for 47001 5.651217 10.
63025
Murder
Murder Rate for 12759 3.9211 6.4
756
Guns
(from 1982 to
1991 in
counties over
100,000)
Crime Rate for 47001 18.7845 32.
39292
Rape
Crime Rate for 47001 44.6861 149
.2124
Robbery
Crime Rate for 47001 180.0518 243
.2615
Aggravated
Assault
Crime Rate for 47001 811.8642 119
0.23
Burglary
Crime Rate for 47000 1764.37 203
6.03
Larceny
Crime Rate for 47000 160.4165 284
.5969
Auto Theft
Causes of
Accidental
Deaths and
Murders per
100,000 People:
Rate of 23278 0.151278 1.2
16175
Accidental
Deaths from Guns
Rate of 23278 1.165152 4.3
42401
Accidental
Deaths from
Sources
Other than Guns
Rate of Total 23278 51.95058 32.
13482
Accidental Deaths
Rate of Murders 23278 0.444301 1.9
30975
Using Handgun
Rate of Murders 23278 3.477088 6.1
15275
Using Other Guns
Income Data (All
$ Values in Real
1983 dollars):
Real Per Capita 50011 10554.21 249
8.07
Personal Income
Real Per Capita 50011 67.57505 53.
10043
Unemployment
Insurance
Real Per Capita 50011 157.2265 97.
61466
Income
Maintenance
Real Per Capita 49998 12328.5 439
7.49
Retirement Per
Over 65
Population
Characteristics:
County Population 50023 75772.78 250
350.4
County 50023 214.3291 142
1.25
Population per
Square Mile
State Population 50056 6199949 534
2068
State NRA 50056 1098.11 516
.0701
membership per
100,000
State
Population
% of votes 50056 52.89235 8.4
10228
Republican in
Pres. Election
% of Pop. Black 50023 0.920866 1.5
56054
Male Between
10-19
% of Pop. Black 50023 0.892649 1.5
45335
Female Between
10-19
% of Pop. White 50023 7.262491 1.7
47557
Male Between
10-19
% of Pop. White 50023 6.820146 1.6
73272
Female Between
10-19
% of Pop. Other 50023 0.228785 0.7
69633
Male Between
10-19
% of Pop. Other 50023 0.218348 0.7
42927
Female Between
10-19
% of Pop. Black 50023 0.751636 1.2
14317
Male Between
20-29
% of Pop. Black 50023 0.762416 1.2
783
Female Between
20-29
% of Pop. White 50023 6.792357 1.9
91303
Male Between
20-29
% of Pop. White 50023 6.577894 1.7
96134
Female Between
20-29
% of Pop. Other 50023 0.185308 0.5
57494
Male Between
20-29
% of Pop. Other 50023 0.186327 0.5
59599
Female Between
20-29
% of Pop. Black 50023 0.539637 0.8
79286
Male Between
30-39
% of Pop. Black 50023 0.584164 0.9
86009
Female Between
30-39
% of Pop. White 50023 6.397395 1.4
60204
Male Between
30-39
% of Pop. White 50023 6.318641 1.4
22831
Female Between
30-39
% of Pop. Other 50023 0.151869 0.4
56388
Male Between
30-39
% of Pop. Other 50023 0.167945 0.4
54721
Female Between
30-39
% of Pop. Black 50023 0.358191 0.5
71475
Male Between
40-49
% of Pop. Black 50023 0.415372 0.6
90749
Female Between
40-49
% of Pop. White 50023 4.932917 1.0
86635
Male Between
40-49
% of Pop. White 50023 4.947299 1.0
38738
Female Between
40-49
% of Pop. Other 50023 0.105475 0.3
02059
Male Between
40-49
% of Pop. Other 50023 0.115959 0.3
04423
Female Between
4049
% of Pop. Black 50023 0.43193 0.7
08241
Male Between
50-64
% of Pop. Black 50023 0.54293 0.9
21819
Female Between
50-64
% of Pop. White 50023 6.459038 1.4
10181
Male Between
50-64
% of Pop. White 50023 6.911502 1.5
4784
Female Between
50-64
% of Pop. Other 50023 0.101593 0.3
67467
Male Between
50-64
% of Pop. Other 50023 0.11485 0.3
74837
Female Between
50-64
% of Pop. Black 50023 0.384049 0.6
71189
Male Over 65
% of Pop. Black 50023 0.552889 0.9
80266
Female O65
% of Pop. White 50023 5.443062 2.0
82804
Male Over 65
% of Pop. White 50023 7.490128 2.6
9476
Female Over 65
% of Pop. Other 50023 0.065265 0.2
86597
Male Over 65
% of Pop. Other 50023 0.077395 0.2
64319
Female Over 65
