Youth gangs as pseudo-governments: implications for violent crime.
Sobel, Russell S. ; Osoba, Brian J.
1. Introduction
In the early 1970s, fewer than 300 cities cited having problems
with youth gangs. (1) Since then, gangs have been identified in all 50
states, with over 2500 cities reporting problems by the late 1990s. (2)
Anecdotal evidence, along with casual empiricism, has led many people to
hold a strong belief that youth gangs are a serious problem because
areas with more gang activity also tend to have higher rates of violent
crime committed by youths. Simply put, the commonly accepted wisdom is
that gangs cause violence.
In this paper, we propose and test a hypothesis suggesting that the
causal relationship between youth violence and gang activity might flow
in the exact opposite direction of what is commonly accepted. We propose
that the failure of government to protect the rights of individuals from
violence committed by youths has led to the formation of gangs as
protective agencies among those populations who are most victimized by
unpunished juvenile offenders in areas with high preexisting rates of
violent crime. By banding together under the threat of mutual
retaliation, potential victims of youth violence can secure increased
safety. This same phenomenon also explains the widespread prevalence of
gangs within prisons, where the rights of individuals are largely
unenforced. While gangs, like governments, use coercion and violence to
enforce their rules through retaliation, the net impact of gangs (like
governments) is likely to lower the overall amount of violence. (3)
Generally, for an equilibrium to exist in which gang-type agencies
prevail, the deterrence effect must reduce violence by more than the
amount of violence used by the enforcement agency. (4)
Our analysis is solidly founded in the economic literature on the
formation and evolution of "governments" from a situation of
anarchy developed by Nozick (1974) and Buchanan (1975). (5) These
authors, particularly Nozick, explain how and why infant governments
evolve as protection firms in the anarchistic "Hobbesian
jungle," (6) characterized by violence and theft. Assuming that
protection firms already exist, Sutter (1995) uses a game-theoretic
model to address the behavior of and relations between individual
protection firms and their respective clients when there are varying
levels of symmetry between the former and the latter. While other
authors, such as Bandiera (2003), have previously applied this
theoretical framework to the evolution of specific protection firms,
like the Sicilian mafia, very little has been done on applying this
model to youth gangs, with the exception of a purely theoretical model
by Skaperdas and Syropoulos (1995). Our analysis of youth gangs also
relies on several recent theories developed in the literature on anarchy
and whether it is welfare improving relative to a predatory state (for
example, Moselle and Polak 2001; Leeson 2006).
In this paper, we develop this youth gang application of government
evolution and anarchy theory to a much greater extent than has
previously been done in the literature, and then conduct empirical
testing. Our hypothesis--that gangs form in areas where there is a high
rate of preexisting violence as a protection agency substituting for the
lack of government enforcement of rights--is an alternative explanation
for the well-documented cross-sectional correlation between gang
activity and violent crime. In particular we show that our model
predicts an exactly opposite direction of causality between youth gang
activity and the rate of violent crime from what is commonly accepted.
Because our hypothesis relies on the causality flowing from crime to
gang membership rather than vice versa, we use empirical causality
models to test our hypothesis. We indeed find a one-way causal
relationship that violent crime causes gang membership, and we can
reject the hypothesis that gang membership causes violent crime.
Our results have significant implications for government policy
directed toward youth gangs. Just as the overthrowing or dissolving of a
government in a geographic area might result in more violence because of
a lack of rights enforcement in the resulting anarchy, government policy
aimed at dissolving youth gangs will not be successful in reducing
violent crime, and may in fact increase it. (7) By failing to adequately
punish youth offenders when they violate the rights of other
individuals, the current government legal system has created an
environment where there is a significant demand for these private
protective agencies (youth gangs). While gangs do use violence to
enforce their rules and protect the rights of their members, the net
result of gangs, according to our results, is to reduce the amount of
violent crime because of mutual deterrence. Because there will always be
a market for private protection when government fails to protect
individual rights, the implications are clear for how public policy
reform can reduce gang activity: more effective enforcement of laws that
protect the rights of individuals from violent crimes committed by
youthful offenders. Breaking up and destabilizing gangs within our model
is violence increasing, rather than violence reducing.
Our findings suggest policy implications that are sometimes
contrary to the existing deterrence literature. The older theoretical
literature on the deterrence effect indicates that stricter law
enforcement is less effective than better detection (Davis 1988; Leung
1995). These models, however, deal with single offenders and not gangs.
Garoupa (2007), on the other hand, actually addresses punishment as it
relates to gangs. The author finds that government may actually increase
the criminal effectiveness of gangs by using stricter enforcement
measures. Also, Garoupa (2000) stipulates that government should impose
less severe punishment on relatively nonviolent gangs because those
gangs are likely more efficient in controlling criminal activity than is
government. Government should instead focus its efforts on controlling
violent gangs that use "costly extortion." The results of this
paper suggest that Garoupa's (2000) solution is indeed the
appropriate policy response to gang activity.
This does not mean to say that all gang activity is explained by
these differentials in enforcement. Sociological explanations and other
factors surely play a major role. We also note that in some other
countries with softer juvenile punishments, gangs are less prevalent.
However, the fact that many gang members report joining a gang for
protection, both among prison and youth gangs, suggests that the effect
we examine is important nonetheless. What is important for deterring
crime is whether the individual committing the crime is punished. When
offenders go unpunished by the formal legal system, informal gangs help
to fill this role, providing protection for those who would be potential
victims of the unpunished offenders.
2. The Traditional View of Gangs
Sociologists and criminologists have weighed in most heavily on the
debates regarding gang formation. Spergel et al. (1996a) theorize that
youth gang problems are brought about by several community-level
factors, including a lack of both social opportunities and social
organization, institutional racism, and failures of social policy. They
claim that, especially in black neighborhoods, the street gang provides
control and employment opportunities that are not provided by legally
recognized institutions. The popular perception is that gangs, like the
infamous Bloods and Crips, seek out new markets in which to franchise
their names. However, the empirical literature has found results that
reject this hypothesis. For example, Spergel et al. (1996b) note that
most new gangs are not franchises. This is later reaffirmed by Maxson
(1998).
Other authors have hypothesized that gangs are little more than
organized drug-dealing firms, and that the main reason for gang
existence is the fact that drugs are illegal. This claim is widely made
by law enforcement officials. While it is true that some gangs use the
drug trade to help finance their activities, the empirical literature
has uniformly provided results that reject the view that drug activity
is the main reason for gang formation and existence. Maxson (1995) tests
the connection between street gangs, illicit drug sales, and violence
and finds that street gangs are far less likely to be involved in the
illegal drug trade and the associated violence than the law enforcement
literature suggests. The author finds that only a few gangs seemed to
specialize in drug sales. Levitt and Venkatesh (2000) describe the inner
workings of a gang that, in fact, does sell drugs. However, the gang
charges an additional membership lee for those who wish to sell drugs.
One reason to be very skeptical of these claims by law enforcement
officials of the gang relationship with the drug trade is because if law
enforcement exaggerates the extent to which gangs are involved in drug
trade, they are more likely to get bigger budgets. The fact that
budgetary considerations play a major role in the decisions and actions
of police departments is now widely demonstrated in the literature by
authors such as Rasmussen and Benson (1994).
That gang activity is present to a greater extent in areas with
higher rates of violent crime has been well demonstrated in the cross
section. Based on this strong correlation, it is widely accepted that
the way to reduce violent crime is to reduce gang activity. Inherent in
this statement is an underlying assumption about the direction of
causation between violent crime and gang activity. An intervention that
reduces gang activity will only reduce violent crime if gangs cause
violent crime. All protection firms and organizations, from the mafia to
private security to traditional governments, use coercion, retaliatory
violence, and predatory violence to enforce certain rules of conduct and
to enforce and protect the rights of their members. However, saying that
gangs cause violence based on this observed behavior is identical to
claiming that governments that use coercion and violence as a means to
provide protection services are causing more total violence than would
exist without any government in place. The gang's use of
retaliatory violence against someone who aggresses against a gang member
actually results in a lower level of total violence as it creates a
strong incentive for individuals not to initiate violence to begin with
because of the fear of retaliation by the gang. Within the economic
model of protection services, an intervention that resulted in weakened
gangs (or weakened governments) would result in more violence, not less.
Any theory of gangs should be accepted or rejected based on its
ability to explain real-world empirical observations. As we have already
discussed, the real-world evidence rejects the hypothesis that gangs
franchise and that they primarily form to participate in the drug trade
(although it may be a secondary function performed by the gang once it
is organized). Perhaps the most useful empirical observation that must
be explained by a good theory of gang formation is why these gangs are
primarily present among youths and not adults. (8) The most widely
accepted reason within law enforcement and in sociology is that gangs
employ and recruit youth members because these members can commit crimes
virtually without punishment because of their age. In this framework,
youths are employed to coerce other individuals and commit violent acts
to obtain resources for the gang leaders. Data on the age distribution
of gang members, however, is notably inconsistent with this view.
Figure 1 presents the age distribution of gang members from the
1998 National Youth Gang Survey (Office of Juvenile Justice and
Delinquency Prevention 2000). If gang members are employed based upon
their ability to commit illegal acts without punishment from law
enforcement, there should be a large, discrete decline in gang
membership beginning at age 18. The age distribution of youth gang
membership, however, does not show a significant drop at exactly age 18,
but rather tapers off through the mid-twenties. This is a widely
recognized puzzle in the standard theory.
Our hypothesis--that repeated violence committed by youths who then
go unpunished causes gangs to subsequently form among the potential
victims of this violence--does a better job of explaining the true age
profile of gang membership. Because social groups and interactions do
tend to be stratified by age, our theory also predicts that gangs would
form among youths more than adults. However, within our model the age
distribution of gang membership should begin to smoothly decline after
age 18 as individuals move into new social groups as they age. The fact
that both a 16-year-old and an 18-year-old are just as likely to be the
victim of a 17-year-old criminal explains why our model fits this data
better than the existing, and more commonly accepted, view of youth
gangs. Both our later finding, that gangs lower net violence, and the
fact that there is a gradual erosion of gang membership with age also
provide some empirical support for the model by Sutter (1995). In that
model, when exit is easier for members, the gang will tend to be more
protective rather than predatory.
3. The Economic Model of Gang Formation
Much of the economic literature on gang formation developed from
earlier works on government formation. Lane (1958) describes how
government, in its role as a "protection firm," became a
monopolist over the protection industry and then the entire market.
Carneiro (1970) confirms that coercive force, not enlightened
self-interest, led to the formation of states throughout history.
Despite these two early contributions, the modern economic literature on
the formation and evolution of protection firms from a situation of
anarchy is generally attributed to the influential works of Nozick
(1974) and Buchanan (1975). Holcombe (2004), extending this literature,
argues that in the absence of government, people will organize
protection firms, which will grow into mafias and eventually gain
monopoly power and establish themselves as governments. Skaperdas and
Syropoulos (1995) describe how, in a state of anarchy, those with a
comparative advantage in violence grab and maintain power through
coercion of those who have a comparative advantage in production.
The economic model of gang formation also can be derived from
literature on weak governments. For example, Konrad (1999) explains how,
even in the presence of a legitimate modern government, gangs can
develop in situations where there is a regional power vacuum. Mehlum,
Moene, and Torvik (2002) develop a theoretical model to show how violent
gangs may become monopoly producers of violence and protection. This
occurs when legitimate government fails to protect rights. The gangs may
extort protection fees both from individuals that operate in the
underground economy and from others that lack legal protections.
Bandiera (2003) shows that the Sicilian mafia emerged due to a lack of
strong governmental protection of property rights. Similarly, according
to Anderson (1995), certain conditions encourage the formation of a
mafia: a loss of legitimate state power and the presence of illegal
markets. Although a street gang is not a mafia, the two criteria are
typically present in poor inner cities, potentially aiding in gang
formation, just as they do in mafia formation. Meanwhile, Skaperdas
(2001) lists certain contributing factors for the birth and growth of
gangs and organized crime, again including illegal markets along with
ethnic and/or social distance from mainstream society.
Most of the literature on the evolution of private protection firms
focuses on a situation of anarchy within a certain geographical area,
where there is little or no government provision of rights protection.
(9) However, Hirshleifer (1995) notes that some degree of anarchy is
present in every social order. As government law enforcement cannot
completely keep violence and theft from occurring, markets for
specialized private protection firms may develop. Even within one
specific society, the extent of rights protection can differ
dramatically among subsets of the population. This is a particularly
realistic description for the case of youths. As anyone who has attended
a public school knows, individual rights are generally not enforced
except in situations of extreme violence. Bullying, theft of lunch
money, physical coercion, and other types of violence or threats of
violence are not only commonplace but widely accepted and tolerated even
by school administrators. The same is true in adult prisons. Our
hypothesis is that it is precisely these areas in which the failure of
the government to protect the rights of individuals results in the
formation of gangs as substitute protective agencies. So, we postulate
that, in areas where there is less government rights protection,
individuals will be more prone to become members of gangs. By being part
of the gang, one obtains the protection and "law enforcement"
services of the gang within one's own community. If violence is
inflicted upon a gang member, the gang will retaliate against the
perpetrator. In neighborhoods with little violence, there is much less
need for these protective services, and thus the likelihood of gang
formation is reduced.
There is a common misperception in society that only government can
provide peace and order. In fact, according to Reuter (1983), conflicts
in an organized crime setting are usually settled peacefully. Leeson
(2007) shows that even under anarchy, private trading arrangements can
(and will) evolve to prevent or minimize violence. Dowd (1997) reminds
us that the "Wild West" was usually more peaceful and orderly
than generally perceived, due to the occasional creation of citizen
vigilante groups that enforced the legal attitudes of the day. According
to Sutter (1995), the level of violence (and degree of protection of
private property vs. predation) will depend on how the firms interact
with each other. In a world of quasi-anarchy and competing protection
firms, the potential for violence will depend on the returns to scale in
violence and whether firms are near the minimum efficient scale to
protect their members. Sutter (1995) also concludes that the ability of
individuals to exit from a gang (either to join another gang or to
withdraw from gang membership altogether) will affect whether gangs
would be overly predatory or primarily provide protection services to
members. Our paper provides evidence to show that, in the inner city,
where government protection of rights is limited at best, gangs form to
provide more safety and order than would otherwise be available.
4. The Economic Model of Gang Operation
It is important to understand not only how gangs form, but also how
they operate once formed. Spergel et al. (1996b) describe the symbiotic
nature of street and prison gangs, with both organizations acting to
maintain control and order in their respective geographic areas.
Interestingly, the relatively high degree of gang activity within
prisons also supports our hypothesis. Like inner cities and public
schools, prisons are well known for being places where the rights of
individuals are not very well protected. Prison inmates can expect to
get more protection of their rights from belonging to a prison gang than
from the prison security guards. (10) The limited and incomplete
protection of rights in prisons, like in inner cities, leads to an unmet
demand for rights protection that private organizations fulfill.
Skaperdas (2001) argues that organized crime is more like a state
than a firm. However, it more closely resembles predatory states.
Skaperdas and Syropoulos (1995) show how those ruled by the gangs will
tend to devote fewer resources toward production, resulting in a lower
level of economic activity and growth. This is especially true in the
impoverished inner cities, where the gangs already compete for
increasingly scarce resources. However, gangs may still be better for
the overall social and economic performance of an area than anarchy
without gangs. Moselle and Polak (2001) show that both organized
banditry and anarchy can be welfare enhancing relative to government, if
government is predatory enough. (11) Their argument could be applied to
gangs to suggest that gangs might also be welfare enhancing relative to
a highly corrupt and predatory government.
According to Baumol (1990), governments throughout history have for
the most part behaved like gangs, being tyrannical and self-serving.
However, he proposes that gangs are less like governments and more like
firms serving clients. Gambetta (1993) also disagrees that the mafia is
another form of a state. He argues that mafias neither are centralized
nor maintain undisputed control over a certain geographic region. They
are also not accountable and not universal. Instead, Gambetta (1993)
indicates that the mafia is probably closer to a business firm that
provides protection services to paying clients only, and not to all
citizens in their region. (12) For example, as Anderson (1995) points
out, existing businesses that are victims of extortion may actually
support the gangs because they exhibit some control on the entry of
rival firms. Like firms, gangs develop and display the gang's
logos, which are then worn by gang members. This is consistent with our
hypothesis that gangs form primarily as protective firms. Identifying
oneself visibly as belonging to a gang communicates the threat of gang
retaliation to anyone looking to commit violent aggression, regardless
of whether they personally know the individual.
Actually, some theoretical papers do show that violence decreases
with fewer competing coercive organizations. Buchanan (1973) explains
that a monopoly on violence is better than a more competitive market in
the protection industry because monopolists tend to underproduce. So,
society should experience a lower output of violence when one gang has
monopoly power over a certain geographical area. In this case, as
Buchanan indicates, by following its own self-interested goals, a
monopoly producer of an "economic bad" makes society better
off by underproducing violence. Similarly, in the models presented in
Skaperdas (2001) and Cothren (2002), increased competition (or
instability) in the violence industry results in decreased social
welfare.
5. Data and Empirical Analysis
While there is a strong theoretical literature on anarchy, gangs,
and organized crime, the empirical testing of these theories is almost
nonexistent. In the case of gangs, this lack of empirics is primarily
due to the lack of data on gang-related activity. Although some
cross-sectional survey data on gang prevalence across cities exist,
cross-sectional data cannot be used to infer the direction of causation.
Previous research in criminal justice has shown a strong positive
correlation between violence and gang activity in this existing
cross-sectional data areas like Los Angeles have both more violence and
more gangs than areas such as rural Nebraska (with low violence and low
gang activity). To test the direction of causality, time-series data are
necessary. From those data, it is possible to see, on average, which
series moves "first" and which moves "second." The
method that allows us to conduct such an analysis is the Granger
causality test. While technically a test to measure if one series
forecasts another, Granger causality has been used to investigate links
between oil price shocks and recessions (Hamilton 1983), exports and
economic growth (Ahmad and Harnhirun 1996), and local government
revenues and grants (Dahlberg and Johansson 2000). So, we are careful to
use the term "Granger-cause" to describe any significant
results from this model. Another caution of using this test is that it
is sensitive to the lag length of the vector autoregression (VAR)
(Hamilton 1994). We utilize the Akaike information criterion (AIC) to
determine the lag length, since the AIC usually recommends a longer lag
length than the Schwarz information criterion. Longer lag lengths
coincide with the final major caution of using Granger causality: The
test is sensitive to omitted variables (Hamilton 1994). While utilizing
more degrees of freedom, longer lag lengths are considered more
conservative because a VAR with a higher number of lagged endogenous
variables includes more potentially relevant variables.
To test our hypothesis, we obtained nine years of monthly gang
membership data directly from the Los Angeles Police Department (LAPD)
Special Operations Support Division. (13) This is the only available
time series of gang membership data that is long enough to use for
causality testing. In compiling the data, the LAPD employs undercover
gang intelligence officers to infiltrate gangs in order to identify each
gang's respective members. In addition, these officers report on
whether each member is currently active in gang activities or is
currently absent from the gang. The gang membership data contain
information on total gang membership and membership for several
individual gang categories. Again, while these data may have some
limitations, they are the only data of their type currently in
existence. Also, the LAPD's reliance on these data for internal
decisions gives them an incentive to make the data as accurate as
possible. (14)
We supplement the LAPD's gang membership data with violent
crime data from the LAPD's annual Statistical Digest. It consists
of the following Type 1 offenses: homicide, aggravated assault, and
robbery. (15) These data are useful because they provide a long
time-series of monthly data that extends for the majority of the time
that the gang membership data cover. While the gang membership data span
from April 1998 to July 2007, the monthly crime data span from January
1997 to December 2004. (16) Because the time periods differ, we use only
the overlapping 81 months of data, from April 1998 to December 2004.
(17)
To test the robustness of our model, we conduct our tests not only
for total gang membership but also individually for membership in the
three largest individual gang classes reported by the LAPD: Hispanic
gangs, Crips, and Bloods. Because of the well-documented seasonal nature
of criminal activity, we employ a 12-month seasonal difference of all
variables and use the indication [[DELTA].sub.12].
Because we are interested in employing the Granger causality test
to determine the direction of causality, we need to first ensure that
our series are stationary. Otherwise, results from nonstationary data
series will lead to spurious causality results. To test for unit roots,
we employ the augmented Dickey-Fuller (ADF) test.
Table 1 shows the ADF test results for all of our data series.
Looking at the first column of results, only one of the three crime
variables, homicide, is stationary (in the 12-month difference form).
Both aggravated assault and robbery are nonstationary. The first
difference is taken for these two series, and the ADF tests for the
transformed series (shown in the second column of results) shows that
they are now stationary and can be used for the causality regressions
(the first difference of the 12-month seasonally differenced series).
For the four gang membership variables, three are stationary in their
level form, leaving only Hispanic gang membership nonstationary. Once
the first difference is taken and the transformed series retested, it is
now stationary. (18) We adopt the standard convention of using
[[DELTA].sub.12] to refer to the series that are only seasonally
differenced to remove the 12-month seasonal pattern (and they were
stationary in that form), and [[DELTA].sub.1]([[DELTA].sub.12]) to refer
to any series for which the first (monthly) difference was taken to make
the series stationary based on the results of the ADF test.
With all of the variables of interest now in a form that is
stationary, we can proceed to test our hypothesis regarding the causal
direction between gang membership and our crime variables. Our initial
Granger causality tests are conducted as indicated in Equations 1 and 2
below:
[V.sub.t,j] = [[beta].sub.1,j] + [r.summation over (i-
1)]([[beta].sub.1i,j][V.sub.t-i,j]) + [s.summation over
(i=1)([[alpha].sub.i,k][G.sub.t-i,k]) + [[epsilon.sub.1t,j], (1)
[G.sub.t,k] = [[alpha].sub.2,k] + [s.summation over (i-
1)]([[alpha].sub.2i,k][G.sub.t-i,k]) + [r.summation over
(i=1)([[beta].sub.2i,j][V.sub.t-i,j]) + [[epsilon.sub.2t,k], (2)
where V and G are, respectively, violent crimes and gang
membership, j is an indicator for the type of violent crime, and k is an
indicator for the gang category. We set up the null hypotheses that (i)
[[alpha].sub.1i,k] = 0, (ii) [[alpha].sub.2i,k] = 0, (iii)
[[beta].sub.1i,j] = 0, and (iv) [[beta].sub.2i,j]= 0, for all i = 1 to
r, s. (19) Again, the optimal lags (r and s) are determined by using the
AIC on the vector autoregressive equations. Using an F-test, we evaluate
these null hypotheses for each gang (k) and crime (j) to determine if
any causal relationship exists. Our Granger causality results are
presented in Table 2.
The first section at the top of Table 2 shows the results of the
causality test for the relationship between homicide and gang
membership. For all cases we cannot reject the null hypothesis that gang
membership does not Granger-cause homicide; however, we can reject the
null hypothesis that homicide does not Granger-cause gang membership
both for total gang members and Hispanic gangs. Thus, for total gang
membership and the largest gang subcategory, the causality tests show
that there is a one-directional causal relationship: Homicide causes
gang membership. In no case do we find that gang membership causes
homicide. As the rate of homicide in Los Angeles increases, so does gang
membership in subsequent months as a result, but not vice versa.
The middle section of Table 2 shows the results of the causality
tests for the relationship between aggravated assault and gang
membership. For aggravated assault, we again cannot reject the null
hypothesis that gang membership does not Granger-cause aggravated
assault in Los Angeles. Consistent with this paper's theory, the
results show that assault does Granger-cause membership in both Blood
and Crip gangs. For Hispanic gangs and total gang membership, there is
no causality in either direction. The evidence here again points to the
conclusion that an increase in violence, in this case aggravated
assault, causes an increase in certain gang membership levels, but not
vice versa.
The bottom section of Table 2 shows the results of the causality
tests for the relationship between robbery and gang membership. Here, we
can reject a causal relationship in both directions. There is neither a
causal relationship flowing from gang membership to robbery nor from
robbery to gang membership. We believe that the results for robbery are
not significant for several reasons. First, the other two crimes we
examine involved the actual infliction of physical harm and are thus
much more likely to be reported in the first place. As Neumayer (2003)
points out, homicide is the most accurately reported crime. As is well
known, sexual crimes are severely underreported due to the social stigma
the victim faces. However, the underreporting of crimes such as robbery
is likely strongest among youth victims, due both to their historical
distrust of police and to the stigma of "tattling," which very
well may earn oneself additional future violence of a more serious
nature. (20) If a robbery victim joins a gang the following month after
being robbed but not reporting it, gang membership will increase while
the robbery will be absent from the data. Because homicide is the most
accurately reported of the violent crimes, we believe our results for
that variable are the most trustworthy.
Perhaps most noteworthy is that in only one case did we uncover a
causal relationship showing that gang membership causes violent crime.
When causality does exist, it is almost exclusively violent crime
causing gang membership. Our results strongly show that an exogenous
change in violent crime, particularly homicide and aggravated assault,
results in a subsequent increase in gang membership as additional people
seek the protective services offered by gangs. Again, in only a single
case do we find evidence that an exogenous change in gang membership
results in a corresponding increase in violent crime.
To ensure the robustness of our results, we conducted our Granger
causality tests using an exogenous measure of the unemployment rate
(both seasonally adjusted [SA] and nonseasonally adjusted [NSA]). This
variable allows us to control to some degree for regional economic
conditions that may impact the decisions to commit crimes or join gangs.
(21) There is a significant literature that addresses the association
between certain crime rates and unemployment rates. (22) While there is
still substantial debate over this association, the unemployment rate is
the best available control variable for our monthly crime and gang data.
All unemployment rate data was obtained from the U.S. Bureau of Labor
Statistics.
The results of the NSA and SA Granger causality tests are located
in Tables 3 and 4, respectively. Based on our augmented models, the
results show an even clearer relationship of crime leading to gang
membership. Homicides Granger-cause total gang membership, along with
Crip and Hispanic gang membership. This is true for both the SA
unemployment rate and the NSA unemployment rate. Recall that homicide is
likely the best measure of violence for this study. So, the results lend
additional and strong confirmation that crime Granger-causes gangs.
When controlling for the unemployment rate, our results for
aggravated assault are clear for both Blood and Hispanic gangs: Assault
Granger-causes gang membership. For Crip gangs, the results are mixed.
When using the NSA unemployment rate, the results show assault
Granger-causing Crip membership. With the SA unemployment rate as a
control variable, we encounter bidirectional causality. This result
indicates that Crip gang membership Granger-causes more assault while
assault Granger-causes greater gang membership. It provides some
evidence that gangs increase crime levels; however, since it occurs only
in one instance, that evidence is quite weak.
In the final section of our tables, we find that, when controlling
for the unemployment rate (both SA and NSA), robbery Granger-causes Crip
gang membership. Again, no other causal relationship exists between
robbery and the other gangs. So, there is some evidence that robbery
could lead certain individuals to join the Crips. Again, since this
occurs only twice across our various model specifications, the evidence
for this finding is relatively weak.
The LAPD occasionally purges inactive members from the gang member
databases. This results in some drops in membership that could
potentially affect our results. To account for these purges, we added an
exogenous binary variable to account for each purge. We then ran all
three of our previous models. The results are very similar. For the
original bivariate causality tests, we find that homicide Granger-causes
total gang membership, along with Blood and Hispanic gang membership. In
the models that adjust for unemployment, there is no longer any
statistically significant causal relationship between homicide and
Bloods. Aggravated assault again consistently Granger-causes Blood gang
membership across all models. However, assault Granger-causes Crip and
Hispanic gang membership in only one of three models. Finally, there is
no causality between robbery and gangs except in one model that controls
for the SA unemployment rate. As before, robbery Granger-causes Crips.
In none of the "purge adjusted" models is there any causation
from gangs to violence.
We did try to perform a more micro-neighborhood-level analysis
using rough measures from the Website www.streetgangs.com and limiting
our data to only crimes that were classified by the police as gang
crimes. Gangs that appeared to be most active in a given police
subbureau were allocated all gang crimes within that sub-bureau. Then,
each gang's crime and gang membership data were aggregated across
all sub-bureaus. For example, Hispanic gangs were most active in several
areas, including the Rampart and Van Nuys sub- bureaus. We aggregated
Hispanic gang membership and total gang crime from those two areas
(along with other sub-bureaus where Hispanic gangs were most active). As
evidenced by the greatest number of gangs, Hispanic gangs also tend to
dominate many more geographical areas within the city of Los Angeles
than do Blood and Crip gangs. The Blood and Crip gangs seem to be most
dominant in the South and Central sections of the city, along with
neighboring cities (e.g., Compton and Inglewood) to the south. Hispanic
gangs are more prevalent in the remaining areas (Alonso 2007a, b, c).
Causality tests were then conducted using the same six-model variations,
and the results are presented in Appendix A. These models were much more
suspect in their explanatory power, with significantly lower R-squares
due to our having to assign all crimes to a specific gang and counting
only gang-related crimes. The results of these regressions did not find
any causal relationships between gangs and crime.
6. Conclusion
The popular perception that gangs cause violent crime is based on
tenuous casual observations. Although gangs and violence do seem to
frequently coexist, such cross-sectional correlations do not imply
causality. Our results provide strong evidence that violent crime causes
an increase in gang membership, and not vice versa. Thus, areas with
higher rates of violent crime will also experience higher rates of gang
membership as a result of the increased violence.
We extend the models of government formation out of anarchy
developed by Nozick (1974) and Buchanan (1975) and apply them to the
relative anarchy faced by inner-city youths both at school and in their
neighborhoods. Our analysis is based on the observation that government
does not adequately protect the rights of individuals from violent crime
committed by youths. Based on past violence or perceived future
violence, these youths seek protection by forming organizations to
provide safety where government public safety agencies have failed.
Our results are important because they uncover a situation where
public policy, implemented with the best possible intentions, may in
fact be harming those it was intended to help. As we have shown in all
but one instance, violent crime leads to an increase in gang membership,
not vice versa. If policies are enacted to break up gangs, the resulting
increased anarchy should in fact lead to more violence among youths.
This is because the gangs serve as a net deterrent of violence. In
addition, as theorized by Buchanan (1973) and later by Konrad (1999) and
Skaperdas (2001), increased competition between gangs will lead to
additional violence. Unless the already existing violence is mitigated,
youths from the previous gangs will again form gangs. However, as these
new gangs are smaller and more fragmented, more violence will occur.
Our main policy implication is that governments should try harder
to protect the rights of individuals who are the victims of violence or
coercion by juvenile offenders. Youths form and join gangs to secure
protection primarily because of the inability or unwillingness of police
and school administrators to protect their rights by punishing those
juveniles who commit or threaten violence. When schools and inner cities
are Hobbesian jungles, with little rights protection, it is only natural
for individuals to seek protection in the private sector by forming
gangs. While law enforcement likely is active in many of these city
neighborhoods, the emphasis may be too heavily focused on prosecuting
those participating in the illicit drug trade, in lieu of more directly
protecting public safety and individual rights. These same implications
apply to prison gangs in that they exist due to the lack of formal
enforcement of the rights of inmates against aggression and violent acts
committed by other inmates. Unless government improves on the protection
it provides to individuals who are the potential victims of crimes,
others will continue to join gangs to purchase these missing protective
services currently underprovided by the government sector.
We do, however, want to provide a caveat that while our data are
the only time- series data available that allow us to directly examine
our hypothesis regarding causality, they are aggregated to the city
level. An analysis at a more detailed sub-city level complemented with
individual survey data might afford a more detailed estimate of some of
the factors we were unable to directly uncover, such as why some areas
have higher preexisting crime rates to begin with and whether there are
significant differences among sub-areas within a specific city.
Appendix
Table A1. Gangs and Violent Crime Causality Tests: Bivariate Models
with Exogenous Binary "Purge Adjustment" Variable
Homicide
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Bloods [[DELTA].sub.12] 6
Crips [[DELTA].sub.12] 7
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Homicide
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang 3.7373 ** 0.2979
members 4.3942 ** 0.8793
Bloods
Crips 1.9646 1.3718
Hispanic 3.5195 * 0.2962
Gang Finding
Total gang Homicide Granger-causes gangs
members
Bloods Homicide Granger-causes Blood gangs
Crips No causality
Hispanic Homicide Granger-causes Hispanic gangs
Aggravated Assault
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.12] 12
Crips [[DELTA].sub.12] 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Aggravated Assault
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.5024 0.8954
Bloods 2.8272 ** 0.6931
Crips 2.1251 * 1.0143
Hispanic 1.3675 0.8900
Aggravated Assault
Gang Finding
Total gang
members No causality
Bloods Assault Granger-causes Blood gangs
Crips Assault Granger-causes Crip gangs
Hispanic No causality
Robbery
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.12] 1
Crips [[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 1
Robbery
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.0729 0.8845
Bloods 0.6998 0.3198
Crips 1.7498 0.0105
Hispanic 0.0048 3.6179
Robbery
Gang Finding
Total gang
members No causality
Bloods No causality
Crips No causality
Hispanic No causality
Lags for the Granger causality tests were determined using the Akaike
information criterion. All variables were twelfth-differenced in order
to correctly deal with the seasonality present in the crime data.
Additionally, some of the twelfth-differenced series had unit roots.
To correct for this situation, those series were first-differenced.
All resultant series are stationary. A binary variable, to control for
the presence of structural breaks within the gang membership data, was
added to the vector autoregressions as an exogenous variable. The
structural breaks were the result of periodic purging of inactive
members from the gang membership database. Generalized least squares
estimation was used in situations where serial correlation was
present. Asterisks indicate statistical significance at the following
levels: ** = 1%, * = 5%.
Table A2. Gangs and Violent Crime Causality Tests: Non-Seasonally
Adjusted County Unemployment Rate and Binary "Purge Adjustment"
Variable
Homicide
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Bloods [[DELTA].sub.12] 6
Crips [[DELTA].sub.12] 7
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Homicide
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 3.6698 ** 0.2980
Bloods 2.2145 0.9385
Crips 1.9050 1.5045
Hispanic 3.4579 * 0.2932
Homicide
Gang Finding
Total gang
members Homicide Granger-causes gangs
Bloods No causality
Crips No causality
Hispanic Homicide Granger-causes Hispanic gangs
Aggravated Assault
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.12] 12
Crips [[DELTA].sub.12] 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 2
Aggravated Assault
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.4160 0.9786
Bloods 2.6315 * 0.7775
Crips 2.0297 1.1083
Hispanic 1.1602 0.9000
Aggravated Assault
Gang Finding
Total gang
members No causality
Bloods Assault Granger-causes Blood gangs
Crips No causality
Hispanic No causality
Robbery
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 3
Bloods [[DELTA].sub.12] 12
Crips [[DELTA].sub.12] 3
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 1
Robbery
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 0.2472 1.2562
Bloods 0.8155 1.8281
Crips 0.5594 1.2064
Hispanic 0.0056 3.6451
Robbery
Gang Finding
Total gang
members No causality
Bloods No causality
Crips No causality
Hispanic No causality
Lags for the Granger causality tests were determined using the Akaike
information criterion. All variables were twelfth-differenced in order
to correctly deal with the seasonality present in the crime data.
Additionally, some of the twelfth-differenced series had unit roots.
To correct for this situation, those series were first-differenced.
All resultant series are stationary. A binary variable, to control for
the presence of structural breaks within the gang membership data, was
added to the vector autoregressions as an exogenous variable. The
structural breaks were the result of periodic purging of inactive
members from the gang membership database. Generalized least squares
estimation was used in situations where serial correlation was
present. Asterisks indicate statistical significance at the following
levels: ** = 1%, * = 5%.
Table A3. Gangs and Violent Crime Causality Tests: Seasonally Adjusted
County Unemployment Rate and Binary "Purge Adjustment" Variable
Homicide
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Bloods [[DELTA].sub.12] 6
Crips [[DELTA].sub.12] 7
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Homicide
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 3.6683 * 0.2938
Bloods 2.2489 0.9335
Crips 1.8700 1.5473
Hispanic 3.4554 * 0.2907
Homicide
Gang Finding
Total gang
members Homicide Granger-causes gangs
Bloods No causality
Crips No causality
Hispanic Homicide Granger-causes Hispanic gangs
Aggravated Assault
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.12] 12
Crips [[DELTA].sub.12] 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 2
Aggravated Assault
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.3099 0.9221
Bloods 2.5993 * 0.7384
Crips 1.8873 1.0536
Hispanic 3.6996 * 0.8758
Aggravated Assault
Gang Finding
Total gang
members No causality
Bloods Assault Granger-causes Blood gangs
Crips No causality
Hispanic Assault Granger-causes Hispanics gangs
Robbery
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.12] 1
Crips [[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Robbery
[H.sub.0]: Violence Does [H.sub.0]: Gangs Do Not
Not Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.2338 0.8735
Bloods 0.2272 2.8868
Crips 5.8606 * 0.6069
Hispanic 1.2920 0.9462
Robbery
Gang Finding
Total gang
members No causality
Bloods No causality
Crips Robbery Granger-causes Crip gangs
Hispanic No causality
Lags for the Granger causality tests were determined using the Akaike
information criterion. All variables were twelfth-differenced in order
to correctly deal with the seasonality present in the crime data.
Additionally, some of the twelfth-differenced series had unit roots.
To correct for this situation, those series were first-differenced.
All resultant series are stationary. A binary variable, to control for
the presence of structural breaks within the gang membership data, was
added to the vector autoregressions as an exogenous variable. The
structural breaks were the result of periodic purging of inactive
members from the gang membership database. Generalized least squares
estimation was used in situations where serial correlation was
present. Asterisks indicate statistical significance at the following
levels: ** = 1 %, * =5%.
Table A4. Police Sub-Bureau Imputed Data Results
Bivariate Models
Gang Model Lags
Bloods [[DELTA].sub.1]([[DELTA].sub.12] 12
Crips [[DELTA].sub.1]([[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1]([[DELTA].sub.12] 12
Bivariate Models
[H.sub.0]: Gang Crime Does Not [H.sub.0]:: Gangs Do Not
Granger-Cause Gangs Granger-Cause Gang Crime
Gang (F-statistic) (F-statistic)
Bloods 0.8776 1.4212
Crips 0.3542 0.1165
Hispanic 3.3340 1.7325
Bivariate Models
Gang Finding
Bloods No causality
Crips No causality
Hispanic No causality
Controlling for the Non-Seasonally Adjusted County
Unemployment Rate
Gang Model Lags
Bloods [[DELTA].sub.1]([[DELTA].sub.12] 12
Crips [[DELTA].sub.1]([[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1]([[DELTA].sub.12] 12
Controlling for the Non-Seasonally Adjusted County
Unemployment Rate
[H.sub.0]: Gang Crime Does Not [H.sub.0]:: Gangs Do Not
Granger-Cause Gangs Granger-Cause Gang Crime
Gang (F-statistic) (F-statistic)
Bloods 0.7666 1.0017
Crips 0.0677 0.2661
Hispanic 2.5940 1.4841
Controlling for the Non-Seasonally Adjusted County
Unemployment Rate
Gang Finding
Bloods No causality
Crips No causality
Hispanic No causality
Controlling for the Seasonally Adjusted County
Unemployment Rate
Gang Model Lags
Bloods [[DELTA].sub.1]([[DELTA].sub.12] 12
Crips [[DELTA].sub.1]([[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1]([[DELTA].sub.12] 12
Controlling for the Seasonally Adjusted County
Unemployment Rate
[H.sub.0]: Gang Crime Does Not [H.sub.0]:: Gangs Do Not
Granger-Cause Gangs Granger-Cause Gang Crime
Gang (F-statistic) (F-statistic)
Bloods 0.8834 1.2580
Crips 0.3779 0.0592
Hispanic 2.8591 1.4985
Controlling for the Seasonally Adjusted County
Unemployment Rate
Gang Finding
Bloods No causality
Crips No causality
Hispanic No causality
Lags for the Granger causality tests were determined using the Akaike
information criterion. All variables were twelfth-differenced in order
to correctly deal with the seasonality present in the crime data. The
twelfth-differenced series are all nonstationary. To correct for this
situation, those series were first-differenced. Except for the Crips
gang membership series, all other resultant series are stationary. The
geographically imputed Crips gang membership series was persistently
nonstationary at the 5% significance level; however, the series was
stationary using the 10% significance level.
Table A5. Police Sub-Bureau Imputed Data Results (with Exogenous
Binary "Purge Adjustment" Variable)
Bivariate Models
Gang Model Lags
Bloods [[DELTA].sub.1]([[DELTA].sub.12] 12
Crips [[DELTA].sub.1]([[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1]([[DELTA].sub.12] 12
Bivariate Models
[H.sub.0]: Gang Crime Does Not [H.sub.0]:: Gangs Do Not
Granger-Cause Gangs Granger-Cause Gang Crime
Gang (F-statistic) (F-statistic)
Bloods 0.9104 1.5190
Crips 0.3090 0.0019
Hispanic 3.3635 1.1881
Bivariate Models
Gang Finding
Bloods No causality
Crips No causality
Hispanic No causality
Controlling for the Non-Seasonally Adjusted County
Unemployment Rate
Gang Model Lags
Bloods [[DELTA].sub.1]([[DELTA].sub.12] 12
Crips [[DELTA].sub.1]([[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1]([[DELTA].sub.12] 12
Controlling for the Non-Seasonally Adjusted County
Unemployment Rate
[H.sub.0]: Gang Crime Does Not [H.sub.0]:: Gangs Do Not
Granger-Cause Gangs Granger-Cause Gang Crime
Gang (F-statistic) (F-statistic)
Bloods 1.0120 1.5360
Crips 0.0363 1.6023
Hispanic 3.0384 1.1603
Controlling for the Non-Seasonally Adjusted County
Unemployment Rate
Gang Finding
Bloods No causality
Crips No causality
Hispanic No causality
Controlling for the Seasonally Adjusted County
Unemployment Rate
Gang Model Lags
Bloods [[DELTA].sub.1]([[DELTA].sub.12] 12
Crips [[DELTA].sub.1]([[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1]([[DELTA].sub.12] 12
Controlling for the Seasonally Adjusted County
Unemployment Rate
[H.sub.0]: Gang Crime Does Not [H.sub.0]:: Gangs Do Not
Granger-Cause Gangs Granger-Cause Gang Crime
Gang (F-statistic) (F-statistic)
Bloods 0.8277 1.2161
Crips 0.3335 0.0116
Hispanic 2.9793 1.1593
Controlling for the Seasonally Adjusted County
Unemployment Rate
Gang Finding
Bloods No causality
Crips No causality
Hispanic No causality
Lags for the Granger causality tests were determined using the Akaike
information criterion. All variables were twelfth-differenced in order
to correctly deal with the seasonality present in the crime data. The
twelfth-differenced series are all nonstationary. To correct for this
situation, those series were first-differenced. Except for the Grips
gang membership series, all other resultant series are stationary. The
geographically imputed Grips gang membership series was persistently
nonstationary at the 5% significance level; however, the series was
stationary using the 10% significance level. A binary variable, to
control for the presence of structural breaks within the gang
membership data, was added to the vector autoregressions as an
exogenous variable. The structural breaks were the result of periodic
purging of inactive members from the gang membership database.
References
Ahmad, Jaleel, and Somchai Harnhirun. 1996. Cointegration and
causality between exports and economic growth: Evidence from the ASEAN
countries. Canadian Journal of Economics 29:S413-6.
Alonso, Alex. 2007a. "Blood Gangs in the City of Los
Angeles." Street Gangs. Accessed 24 May 2007. Available at
http://www.streetgangs.com/bloods/losangeles.html.
Alonso, Alex. 2007b. "Crip Gangs in the City of Los
Angeles." Street Gangs. Accessed 24 May 2007. Available at http://
www.streetgangs.com/crips/losangeles.html.
Alonso, Alex. 2007c. "Hispanic Gangs in the City of Los
Angeles." Street Gangs. Accessed 24 May 2007. Available at
http://www.streetgangs.com/hispanic/cityofla.html.
Alonso, Alex. 2007d. "Street Gangs Resource
Center.'" Street Gangs. Accessed 24 May 2007. Available at
http://www. streetgangs.com.
Anderson, Annelise. 1995. Organised crime, mafia and governments.
In The economies of organised crime, edited by Gianluca Fiorentini and
Sam Peltzman. Cambridge, UK: Cambridge University Press, pp. 33-54.
Bandiera, Oriana. 2003. Land reform, the market for protection, and
the origins of the Sicilian mafia: Theory and evidence. Journal of Law,
Economies, and Organization 19:218-44.
Baumol, William J. 1990. Entrepreneurship: Productive,
unproductive, and destructive. Journal of Political Economy 98:893-921.
Benson, Bruce L. 1995. An exploration of the impact of modern
arbitration statutes on the development of arbitration in the United
States. Journal of Law, Economics, and Organization 11:479-501.
Benson, Bruce L., and Brent D. Mast. 2001. Privately produced
general deterrence. Journal of Law and Economies 44:725-46.
Buchanan, James M. 1973. A defense of organized crime. In The
economics of crime and punishment, edited by Simon Rottemberg.
Washington, DC: American Enterprise Institute for Public Policy
Research, pp. 119-32, reprinted in The economics of crime, edited by
Ralph Andreano and John J. Sigfried. 1980. Cambridge, MA: Schenkmann,
pp. 395-409.
Buchanan, James M. 1975. The limits of liberty: Between anarchy and
leviathan. Chicago: University of Chicago Press.
Carneiro, Robert U 1970. A theory of the origin of the state.
Science 169:733-8.
Cothren, Richard. 2002. A model of military spending and economic
growth. Public Choice 110:121-42.
Curry, G. David, Richard A. Ball, and Scott H. Decker. 1996.
Estimating the national scope of gang crime from law enforcement data.
Research in Brief Washington, DC: U.S. Department of Justice, Office of
Justice Programs, National Institute of Justice.
Dahlberg, Matz, and Eva Johansson. 2000. An examination of the
dynamic behaviour of local governments using GMM bootstrapping methods.
Journal of Applied Econometrics 15:401-16.
Davis, Michael L. 1988. Time and punishment: An intertemporal model
of crime. Journal of Political Economy 96:383-90.
Dowd, Kevin. 1997. Anarchy, warfare, and social order: Comment on
Hirshleifer. Journal of Political Economy 105:648-51.
Ehrlich, Isaac. 1996. Crime, punishment, and the market for
offenses. Journal of Economic Perspectives 10:43-67.
Gambetta, D. 1993. The Sicilian mafia: The business of private
protection. Cambridge, MA: Harvard University Press.
Garoupa, Nuno. 2000. The economics of organized crime and optimal
law enforcement. Economic Inquiry 38:278-88.
Garoupa, Nuno. 2007. Optimal law enforcement and criminal
organization. Journal of Economic Behavior and Organization 63:461-74.
Gordon, Scott. 1976. The new Contractarians. Journal of Political
Economy 84:573-90.
Hamilton, James D. 1983. Oil and the macroeconomy since World War
II. Journal of Political Economy 91:228-48.
Hamilton, James D. 1994. Time series analysis. Princeton, NJ:
Princeton University Press.
Hirshleifer, Jack. 1995. Anarchy and its breakdown. Journal of
Political Economy 103:26-52.
Hobbes, Thomas. 1651 [1996]. Leviathan. Cambridge, UK: Cambridge
University Press.
Holcombe, Randall G. 1994. The economic foundations of government,
chapter 2. New York: New York University Press.
Holcombe, Randall G. 2004. Government: Unnecessary but inevitable.
Independent Review 8:325-42.
Howell, James C. 1998. Youth gangs: An overview. Juvenile Justice
Bulletin. Washington, DC: U.S. Department of Justice, Office of Justice
Programs, Office of Juvenile Justice and Delinquency Prevention.
Konrad, Kai Andreas. 1999. The market for protection and the origin
of the state. Centre for Economic Policy Research Discussion Paper No.
2173.
Lane, Frederic C. 1958. Economic consequences of organized
violence. Journal of Economic History 18:401-17.
Leeson, Peter T. 2006. Better off stateless: Somalia before and
after government collapse. Working paper, West Virginia University.
Leeson, Peter T. 2007. Trading with bandits. Journal of Law and
Economics 50:303-21.
Leeson, Peter T., Christopher J. Coyne, and Peter J. Boettke. 2006.
Converting social conflict: Focal points and the evolution of
cooperation. Review of Austrian Economics 19:137-47.
Leung, Siu Fai. 1995. Dynamic deterrence theory. Economica
62:65-87.
Levitt, Steven D., and Sudhir Alladi Venkatesh. 2000. An economic
analysis of a drug-selling gang's finances. Quarterly Journal of
Economics 115:755-89.
Los Angeles Police Department. 2006. Crime and Arrest Weekly
Statistics: Crimes Year to Date, December 31.
Maxson, Cheryl L. 1995. Street gangs and drug sales in two suburban
cities. Research in Brief Washington, DC: U.S. Department of Justice,
Office of Justice Programs, National Institute of Justice.
Maxson, Cheryl L. 1998. Gang members on the move. Juvenile Justice
Bulletin. Washington, DC: U.S. Department of Justice, Office of Justice
Programs, Office of Juvenile Justice and Delinquency Prevention.
Mehlum, Halvor, Karl Ove Moene, and Ragnar Torvik. 2002. Plunder
& Protection, Inc. Journal of Peace Research 39:447-59.
Miller, Walter B. 2001. The Growth of Youth Gang Problems in the
United States: 1970-1998. Washington, DC: U.S. Department of Justice,
Office of Justice Programs, Office of Juvenile Justice and Delinquency
Prevention.
Moselle, Boaz, and Benjamin Polak. 2001. A model of a predatory
state. Journal of Law, Economics, and Organization 17:1-33.
Neumayer, Eric. 2003. Good policy can lower violent crime: Evidence
from a cross-national panel of homicide rates, 1980-97. Journal of Peace
Research 40:619-40.
Nozick, Robert. 1974. Anarchy, state, and utopia. New York: Basic.
Office of Juvenile Justice and Delinquency Prevention. 2000. 1998
National Youth Gang Survey: Summary. Washington, DC: U.S. Department of
Justice, Office of Justice Programs.
Rasmussen, David W., and Bruce L. Benson. 1994. The economic
anatomy of a drug war: Criminal justice in the commons. Lanham, MD:
Rowman & Littlefield.
Reuter, Peter. 1983. Disorganized crime: The economics of the
visible hand. Cambridge, MA: MIT Press.
Skaperdas, Stergios. 2001. The political economy of organized
crime: Providing protection when the state does not. Economics of
Governance 2:173-202.
Skaperdas, Stergios, and Constantinos Syropoulos. 1995. Gangs as
primitive states. In The economics of organised crime, edited by
Gianluca Fiorentini and Sam Peltzman. Cambridge, UK: Cambridge
University Press, pp. 61-82.
Spergel, Irving, Ron Chance, Kenneth Ehrensaft, Thomas Regulus,
Candice Kane, Robert Laseter, Alba Alexander, and Sandra Oh. 1996a. Gang
suppression and intervention: Community models. OJJDP Summary.
Washington, DC: U.S. Department of Justice, Office of Justice Programs,
Office of Juvenile Justice and Delinquency Prevention.
Spergel, Irving, David Curry, Ron Chance, Candice Kane, Ruth Ross,
Alba Alexander, Edwina Simmons, and Sandra Oh. 1996b. Gang suppression
and intervention: Problem and response. OJJDP Summary. Washington, DC:
U.S. Department of Justice, Office of Justice Programs, Office of
Juvenile Justice and Delinquency Prevention.
Sutter, Daniel. 1995. Asymmetric power relations and cooperation in
anarchy. Southern Economic Journal 61:602-13.
(1) Curry, Ball, and Decker (1996) provide a methodology for
estimating data on gangs, gang membership, and gang related criminal
activity.
(2) Miller (2001) reports that smaller cities, especially those
with populations below 10,000, have seen much more growth in
gang-related activity than their larger counterparts between the 1970s
and the 1990s.
(3) One exception to this would be the case of an extremely
predatory government, which might be welfare reducing relative to pure
anarchy; see Moselle and Polak (2001).
(4) For the representative individual it would not be Pareto
improving to join (as the probability of being violated would be higher
for a member of the gang).
(5) For a good summary and review of this literature, see Gordon
(1976).
(6) Hobbes (1651) argued that prior to government there was a state
of nature characterized by an absence of rules governing ownership and
interaction. Absent rules, Hobbes posited that life would be
"'nasty, brutish, and short." For a more recent
discussion of this concept, with an analysis of when such situations arc
likely to evolve into cooperative outcomes, see Leeson, Coyne, and
Boettke (2006).
(7) A current-day example of this is postwar Iraq, where private
security firms are being hired to compensate for the lack of protection
by U.S. and Iraqi government sources.
(8) Interestingly, gang activity among adults is most prevalent in
prisons, a place where the enforcement and protection of individual
rights is almost nonexistent. This is an observation that would also
reject the commonly accepted wisdom in favor of our theory.
(9) For an examination of the role of private protection in the
presence of government as well, see Benson and Mast (2001). On the
private provision of dispute resolution, also in the presence of
government, see Benson (1995).
(10) Holcombe (1994) presents all economic model of how individuals
secure others' observance of their rights. Those who will secure
the largest degree of observance are those who either have a larger
ability to threaten coercion or violence, or those who are the most
productive and can trade their output in exchange for observance.
(11) For an application of this to stateless Somalia, see Leeson
(2006).
(12) The particular gang studied by Levitt and Venkatcsh (2000)
behaves somewhat like a business, extorting money for protection,
selling drugs, paying a franchise fee, charging a membership fee, and
taking over nearby territory. The gang they study also goes "out of
business." However, as we indicated earlier, such drug-dealing
franchise gangs are more the exception than the rule.
(13) These data were prepared by the LAPD Bureau Special
Enforcement Units and Community Resources Against Street Hoodlums units
and compiled by the Special Operations Support Division. The authors
appreciate the data assistance provided by Detective Chuck Zeglin of the
LAPD.
(14) Los Angeles County has approximately 500 Hispanic gangs, 200
Crip gangs, and 75 Blood gangs. Within the city limits of Los Angeles,
there are approximately 143 Hispanic gangs, 100 Crip gangs, and 24 Blood
gangs (Alonso 2007d). Total gang membership, at 65,578 in April 1998,
declined significantly to 39,119 in July 2006. Hispanic gangs lost the
largest number of members (-18,734), while Asian gangs lost the largest
portion of members (-61.8%).
(15) The only violent Type 1 offense we do not consider is rape
because the Federal Bureau of Investigation only counts rapes committed
on females. Gangs are dominated by males (Howell 1998): however,
male-on-male rape is reported as assault according to the Uniform Crime
Reporting Division of the Federal Bureau of Investigation.
(16) Gang membership data for February 2002 were not reported. For
simplicity, we interpolate February using January and March 2002
membership.
(17) Note these data include all crimes, not just crimes judged by
the police to be gang related. For 2006, the total number of homicides
(481) in Los Angeles comprised 272 (56.5%) gang-related homicides (Los
Angeles Police Department 2006). This portion has fluctuated between
50.1% and 59.1% between 2000 and 2007. Gang felony assaults in 2006
(2877) constituted 20.4% of all 14,118 aggravated assaults in Los
Angeles. Gang assaults have constituted between 16.2% and 20.4% of all
city assaults since 2000. Gang-related robberies also hovered between
16.0% and 19.9% of total robberies during the same time span. In 2006,
of the 14,235 robberies in Los Angeles, 2515, or 17.7%, were gang
related.
(18) For consistency and robustness we also show in the second
column of results in Table 1 the test statistics after taking the first
difference for the four series that were already stationary in their
12-month seasonal difference form. They should remain significant (with
increasing test statistics), which they all do.
(19) Of course we also test the null hypotheses that (v)
[[beta].sub.1,j] = 0 and (vi) [[alpha].sub.2,k] = 0.
(20) However, because of their low incomes, youths may also be less
often the victims of significant reportable robberies.
(21) Many thanks to an anonymous referee for suggesting this
control variable.
(22) Ehrlich (1996) provides a concise summary.
Russell S. Sobel, Department of Economics, P.O. Box 6025, West
Virginia University, Morgantown, WV 26506-6025, USA; Phone: (304)
293-7864; E-mail rsobel2@wvu.edu.
Brian J. Osoba, Department of Economics, 208 RVAC, Central
Connecticut State University, New Britain, CT 06050, USA; Phone: (860)
832-2735; E-mail osobabrj@ccsu.edu; corresponding author.
The authors wish to acknowledge Ronald Balvers, Andrea Dean,
Stratford Douglas, George Hammond, Peter Leeson, Santiago Pinto, two
anonymous referees, and seminar participants at the Annual Meeting of
the Southern Economic Association, the Association for Private
Enterprise Education, and San Jose State University for helpful comments
on previous drafts of this paper.
Received November 2006; accepted April 2008.
Table 1. Augmented Dickey-Fuller Unit Root Test Results
Variable Form: [[DELTA].sub.12]
Variable ADF Test Statistic Lags
Aggravated assault -0.9775 12
Homicide -4.1147 ** 1
Robbery -1.5367 12
Total gang members -2.435 12
Hispanic gang members -2.05 12
Crips gang members -3.3648 * 12
Bloods gang members -3.7834 ** 12
Variable Form: [DELTA]([[DELTA].sub.12])
Variable ADF Test Statistic Lags
Aggravated assault -4.6420 ** 12
Homicide n/a n/a
Robbery -4.7995 ** 12
Total gang members -4.8501 ** 11
Hispanic gang members -4.3693 ** 11
Crips gang members n/a n/a
Bloods gang members n/a n/a
The number of lags was determined using the Akaike information
criterion. Asterisks indicate statistical significance at the
following levels: ** = 1%, * = 5%.
Table 2. Gangs and Violent Crime Causality Tests: Bivariate Models
Homicide
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Bloods [[DELTA].sub.12] 12
Crips [[DELTA].sub.12] 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Homicide
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 3.6873 ** 0.3222
Bloods 1.6822 2.0018
Crips 1.9014 1.0974
Hispanic 3.3927 * 0.2796
Homicide
Gang Finding
Total gang Homicide Granger-
members causes gangs
Bloods No causality
Crips No causality
Hispanic Homicide Granger-
causes Hispanic gangs
Aggravated Assault
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Crips [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Aggravated Assault
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.7011 0.9168
Bloods 3.4864 ** 0.7155
Crips 2.5556 * 1.0319
Hispanic 1.4485 0.9074
Aggravated Assault
Gang Finding
Total gang
members No causality
Bloods Assault Granger-causes
Blood gangs
Crips Assault Granger-causes
Crip gangs
Hispanic No causality
Robbery
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Crips [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Robbery
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause
Gang (F-statistic) (F-statistic)
Total gang
members 1.2338 0.9200
Bloods 0.7151 1.1784
Crips 0.9793 1.0621
Hispanic 1.1859 0.8720
Robbery
Violence
Gang Finding
Total gang
members No causality
Bloods No causality
Crips No causality
Hispanic No causality
Lags for the Granger causality tests were determined using the Akaike
information criterion. All variables were twelfth-differenced in
order to correctly deal with the seasonality present in the crime data.
Additionally, some of the twelfth-differenced series had unit roots.
To correct for this situation, those series were first-differenced.
All resultant series are stationary. Generalized least squares
estimation was used in situations where serial correlation was present.
Asterisks indicate statistical significance tit the following levels:
** = 1%, * = 5%.
Table 3.
Gangs and Violent Crime Causality Tests: Controlling for the
Non-seasonally Adjusted County Unemployment Rate
Homicide
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Bloods [[DELTA].sub.12] 2
Crips [[DELTA].sub.12] 2
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Homicide
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 3.6189 * 0.3193
Bloods 0.2391 1.6495
Crips 3.8996 * 2.1576
Hispanic 3.3303 * 0.2753
Homicide
Gang Finding
Total gang Homicide Granger-
members causes gangs
Bloods No causality
Crips Homicide Granger-
causes Crip gangs
Hispanic Homicide Granger-
causes Hispanic gangs
Aggravated Assault
Gang Model Lags
Total gang
members [[DELTA].sub.12] 2
Bloods [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Crips [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 2
Aggravated Assault
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.0685 0.8153
Bloods 3.1890 ** 0.8037
Crips 2.3608 * 1.1254
Hispanic 3.4689 * 0.7259
Aggravated Assault
Gang Finding
Total gang
members No causality
Bloods Assault Granger-causes
Blood gangs
Crips Assault Granger-causes
Crip gangs
Hispanic Assault Granger-causes
Hispanic gangs
Robbery
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.12] 1
Crips [[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 1
Robbery
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.4387 0.9004
Bloods 1.0342 0.0871
Crips 5.6439 * 0.1246
Hispanic 0.0001 3.2244
Robbery
Gang Finding
Total gang
members No causality
Bloods No causality
Crips Robbery Granger-
causes Crip gangs
Hispanic No causality
Lags for the Granger causality tests were determined using the
Akaike information criterion. All variables were twelfth-differenced
in order to correctly deal with the seasonality present in the crime
data. Additionally, some of the twelfth-differenced series had unit
roots. To correct for this situation, those series were first-
differenced. All resultant series are stationary. Generalized least
squares estimation was used in situations where serial correlation
was present. Asterisks indicate statistical significance at the
following levels: ** = 1%, * = 5%.
Table 4.
Gangs and Violent Crime Causality Tests: Controlling for the
Seasonally Adjusted County Unemployment Rate
Homicide
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Bloods [[DELTA].sub.12] 2
Crips [[DELTA].sub.12] 2
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 4
Homicide
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 3.6197 * 0.3222
Bloods 0.2026 1.7877
Crips 3.3519 * 2.5420
Hispanic 3.3329 * 0.2767
Homicide
Gang Finding
Total gang Homicide Granger-
members causes gangs
Bloods No causality
Crips Homicide Granger-
causes Crip gangs
Hispanic Homicide Granger-
causes Hispanic gangs
Aggravated Assault
Gang Model Lags
Total gang
members [[DELTA].sub.12] 2
Bloods [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Crips [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 2
Aggravated Assault
Aggravated Assault
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.3573 0.5697
Bloods 3.0521 ** 0.7637
Crips 2.1418 * 3.7537 **
Hispanic 3.4951 * 0.7676
Aggravated Assault
Gang Finding
Total gang
members No causality
Bloods Assault Granger-causes
Blood gangs
Crips Bi-Directional causality
between Assault and
Crip gangs
Hispanic Assault Granger-causes
Hispanic gangs
Robbery
Gang Model Lags
Total gang
members [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Bloods [[DELTA].sub.12] 1
Crips [[DELTA].sub.12] 1
Hispanic [[DELTA].sub.1] ([[DELTA].sub.12]) 12
Robbery
[H.sub.0]: Violence Does Not [H.sub.0]: Gangs Do Not
Granger-Cause Gangs Granger-Cause Violence
Gang (F-statistic) (F-statistic)
Total gang
members 1.3142 0.8755
Bloods 0.7417 0.1229
Crips 5.5153 * 0.1032
Hispanic 1.2459 0.9139
Robbery
Gang Finding
Total gang
members No causality
Bloods No causality
Crips Robbery Granger-
causes Crip gangs
Hispanic No causality
Lags for the Granger causality tests were determined using the
Akaike information criterion. All variables were twelfth-differenced
in order to correctly deal with the seasonality present in the crime
data. Additionally, some of the twelfth-differenced series had unit
roots. To correct for this situation, those series were first-
differenced. All resultant series are stationary. Generalized least
squares estimation was used in situations where serial correlation
was present. Asterisks indicate statistical significance at the
following levels: ** = 1%, * = 5%.
Figure 1. Age Distribution of U.S. Gang Members, 1998
Age Range
<15 0.11
15-17 0.29
18-24 0.46
>24 0.14
Note: Table made from bar graph.
