Content uploaded by César Alonso-Borrego
Author content
All content in this area was uploaded by César Alonso-Borrego
Content may be subject to copyright.
Does Immigration Cause Crime? Evidence
from Spain
Ce
´sar Alonso-Borrego, Universidad Carlos III, Nuno Garoupa, University
of Illinois, and Pablo Va
´zquez, Universidad Complutense and FEDEA
Send correspondence to: Ce
´sar Alonso-Borrego, Department of Economics, Universidad
Carlos III, Av. Madrid, 126, E-28903, Getafe, Spain; Tel: 34-96249749; Fax: 34-
916249875; E-mail: cesar.alonso@uc3m.es.
The relationship between crime and immigration has been a matter of controversy in
the United States and around the world. This paper investigates empirically the case
of Spain. From 1999 to 2009, Spain had a large wave of immigration from different
areas of the globe. At the same time, crime rates increased. However, in comparison
with other European countries that received similar massive immigration waves
during the same period, crime rates in Spain increased less considerably. We show
that there is a significant relationship between crime and immigration. Nevertheless,
the explanation is found in the specific characteristics of the different immigrant
groups, particularly in the amount and type of human capital, which result is largely
in tune with the previous studies on U.S. immigration and crime. (JEL K42, J15,
C23, C25)
This paper is based on a larger project developed by FEDEA on the economics and
sociology of immigration in Spain (2008). Nuno Garoupa acknowledges the financial
support of the European Commission, MMECC (EC Project 044422), and Ce
´sar
Alonso-Borrego acknowledges the financial support of the Spanish Ministry of Science
and Innovation (Grant ECO2009-11165). We are grateful to John Donohue, an anony-
mous referee, and Michele Boldrin, Antonio Cabrales, Raquel Carrasco, Marco Celentani,
Marcelo Perera, and Giulio Zanella for helpful comments on previous drafts. Mario
Alloza, Brindusa Anghel, Yeny C. Estrada, and Roya H. Samarghandi provided excellent
research assistance. The usual disclaimers apply.
American Law and Economics Review
doi:10.1093/aler/ahr019
Advance Access publication January 4, 2012
ÓThe Author 2012. Published by Oxford University Press on behalf of the American Law and Economics
Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
165
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
1. Introduction
Immigration and crime are related very frequently within the media, po-
litical discussions, and even in daily conversation. The opinion polls in many
countries show an increasing concern for the perceived correlation between
immigration and crime.
This paper contributes to the growing empirical literature on immigration
and crime, with a particular application to Spanish data. Although the rela-
tionship between crime and immigration has been the focus of much political
debate, the economic and empirical literature is not vast. Some studies have
looked into U.S. data, but they seem to point out that immigration and crime
do not present a consistent pattern.
1
The situation of the European Union (EU) is different from that of the
United States. Until the late 1990s, immigration from outside of the Union
was not very significant. Furthermore, in the context of the EU member
states, Spain might be regarded as a particular case. Spain had an important
but not a dramatic increase in crime rates, while there was a very significant
growth in immigration in a short time span. The interesting puzzle is to rec-
oncile such a finding with the alleged fact that there is a strong correlation
between crime and immigration.
The most obvious response to this puzzle is that there is no causal rela-
tionship between immigration and crime in Spain. However, this relation-
ship does exist and is quite significant in our econometric analysis.
Immigration partially explains the evolution of the crime rates in Spain
in the last decade or so, demonstrating conclusively that both phenomena
are related to a significant extent.
A second possible interpretation relies on the period of economic
growth in Spain observed in the last 10 years and the consequent devel-
opment of the labor market for immigrants in Spain. Unlike other EU coun-
tries (such as Italy, France, or Portugal), Spain has experienced a long
period of economic growth since the mid-1990s. Economic growth has
provided better legitimate economic opportunities in the labor market
1. See, among others, Butcher and Piehl (1998a, 1998b), Card (2001), Butcher
and Piehl (2005), Borjas, Grogger, and Hanson (2006), and Moehling and Piehl
(2007).
166 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
for everyone. At the same time, it has led to an increase in new opportu-
nities for both natives and immigrants to commit crimes (in particular,
property and economic crimes). There is a potential substitution effect with
an ambiguous outcome. Cross-country comparisons are not helpful to de-
termine how this substitution effect should operate. For example, Spain
and Ireland share a high rate of economic growth during this period but
differ in terms of their increase in crime rates. On the contrary, Italy
and Greece had significantly lower entry of immigration and worse growth
rates in the period 2000–06, and yet they present similar patterns to Spain
with respect to crime rates.
On the other hand, and focusing exclusively on the case of Spain, the
evolution of the different types of crime does not have a consistent relation-
ship with the rate of unemployment, as shown in Figure 1. Additionally, one
has to remember that a major portion of the immigrant population has en-
tered the country illegally and has remained illegal until an official
Figure 1. Variation of Crime and Unemployment in Spain, 2000–06.
Does Immigration Cause Crime? 167
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
regularization has taken place. These regularizations have occurred twice but
have apparently not significantly affected crime rates.
2
We need a more sophisticated explanation. We suggest that our puzzle
can be solved by concentrating on the types of immigrants coming into the
country. The bulk of the thesis is that Spain has received immigrants that,
because of their characteristics (i.e., age, gender, education, and culture),
are less likely to commit crimes. We refer to this explanation as cultural
proximity.
Immigration in Spain is a phenomenon that is extremely heterogeneous.
By that we mean that immigration to Spain encompasses very different re-
alities, from a retired German woman who spends her retirement pension in
Spain to a young Ecuadorian man who works in construction to help his
family back in his native country.
Our hypothesis is that the specific composition of the immigrant popu-
lation determines the degree of correlation between immigration and crime
to a great extent. In the particular case of Spain, a large proportion of the
immigrant population has characteristics that make them less likely to com-
mit crimes than otherwise.
Obviously, it could be that the process of integration is particularly ad-
equate and successful in Spain. An adequate and successful integration pol-
icy could substantially reduce crime rates that are explained by immigration
elsewhere. Depending on their cultural upbringing, immigrants usually
could be more or less inclined to criminal activities depending on institu-
tional characteristics of each native country. Nevertheless, in those countries
where immigrants are better integrated, it is expected that fewer criminal
offenses are committed.
In Table 1, we show the rate of homicides per thousand inhabitants—a
crime easily comparable for the purpose of criminal law specifics—of people
with different nationalities in their country of origin and in Spain. If we look
at each continent individually, one has very different rates of homicide even
2. Two regularization amnesties took place in 2000 and 2005 in order to solve the
situation of illegal immigrants. The amnesty of 2000 required entry into Spain before June
1999, an expired permit of residence or working permit, and no criminal record. Almost
a quarter of a million people benefited from this legal reform. The amnesty of 2005 only
required having a job or an offer of a job in Spain (an employment contract for at least six
months). Almost six hundred thousand people benefited from this second amnesty.
168 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
though the proportions in Spain and in the countries of origin are fairly
similar, as seen in Column 3. At the same time, the differences across con-
tinents of origin can indicate that different nationalities represent a distinct
attitude toward crime that might not have been eliminated by the process of
integration.
3
In Table 2, we present some statistics about different crimes and the origin
or nationality of the immigrants arrested for such crimes. We can immedi-
ately see that there are significant variations across nationalities and across
crimes.
In this study, we present an empirical analysis that aims at responding
conclusively to these issues.
4
In Section 2, we discuss the economic model
of criminal behavior that provides the theoretical foundation for the main
hypothesis. In Section 3, we test econometrically the hypothesis, and in
Section 4, we discuss the empirical results. Section 5 concludes.
Table 1. Homicide Rates in Spain and the Country of Origin
Homicide Rate
in Spain (A)
Homicide Rate
in the Native
Country (B) Ratio (A)/(B)
Europe United Kingdom 14.8 1.5 9.6
Romania 26.7 2.4 10.9
Africa Algeria 98.7 1.7 56.9
Morocco 30.8 0.5 64.8
America Colombia 29.1 59.3 0.5
Ecuador 13.1 17.0 0.8
Peru 5.5 4.9 1.1
Asia China 14.3
Notes: Sources: MIR (Spain), INE (Spain), and UN Office on Drugs and Crime.
The homicide rate (A) is the ratio of number of homicides committed by immigrants in Spain per 1,000,000
population (average 2000–06); the homicide rate (B) is the ratio of the number of homicides in the native
country per 1,000 population (1999).
3. In the period 2000–09, the nationalities with the highest annual immigration
rates were, by order of magnitude, Moroccans, Ecuadorians, Romanians, and Colombians.
The nationalities with the highest annual crime rates were, by order of magnitude, Algerians,
Romanians, and Moroccans. Colombians, Ecuadorians, and Peruvians follow significantly
below.
4. Previous work with Spanish provincial data on crime, but disregarding immigra-
tion, includes Buonanno and Montolio (2008). For the Italian case, see Bianchi,
Buonanno, and Pinotti (2008) and Buonanno, Montolio, and Vanin (2009).
Does Immigration Cause Crime? 169
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
2. Economic Models of Criminal Behavior
The economic model of crime (Becker, 1968; Garoupa, 1997; Polinsky
and Shavell, 2000) is based on the rational comparison between the ben-
efits and the costs of committing an offense. Undercompliance with
the law arises when the benefits of compliance are lower than the costs.
Usually the benefits of a crime include the illegal gain obtained by the
criminal, whereas the costs include the probability and severity of pun-
ishment (although the economic model has been extended to include
other factors). The approach in the economic model is to hypothesize
that everyone is a potential criminal, however only a subset of the pop-
ulation does indeed commit crimes. As a result, the economic model does
not provide any particular argument for why immigrants should under-
comply with the law more or less frequently than natives, except that
their profile of benefits and costs of crime could be different from that
of natives.
In this context, from a theoretical perspective, there are countervailing
arguments about the higher or lower propensity of immigrants toward crime
with respect to natives, based on their profile of costs and benefits. However,
the strength of these arguments depends obviously on empirical confirma-
tion. Only a serious and rigorous empirical analysis can decide which rea-
sons do actually prevail in reality.
The first argument is that, due to difficulties in the labor market and
for lack of economic opportunities, the illegal gain obtained by immi-
grants is relatively higher than that of natives. In other words, it could
Table 2. Crime Rates in Spain and the Country of Origin, Average 2000–06
Homicide Battery Burglary Theft Drug Traffic
United Kingdom 0.8 5.8 5.6 2.8 6.0
Romania 0.4 3.5 24.2 21.8 0.9
Algeria 1.1 5.6 57.7 33.1 12.6
Morocco 0.6 4.6 11.9 12.0 11.7
Colombia 0.8 3.0 5.0 3.3 11.0
Ecuador 0.3 2.8 1.5 2.6 0.7
Peru 0.2 1.9 2.0 5.8 0.7
China 0.4 1.5 0.5 0.6 0.1
Notes: Sources: MIR and INE.
The crime rate is the ratio of arrested persons per 1,000 population of age 20–50.
170 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
be that immigrants have less to lose than natives from noncompliance
with the law. The argument is based on the income/employment status
of the individual, that is to say, it has nothing to do with the particular
attribute of being an immigrant but with the fact that a higher propor-
tion of the immigrant population has economic problems or has fewer
economic opportunities.
5
A second argument looks at cooperative crimes that require a close
network of trust and coordination. The economic literature has discussed
how organized crime can develop mechanisms of control and quality as-
surance (Garoupa, 2007) and has identified ethnic homogeneity as a very
powerful mechanism to guarantee the success of a criminal organization.
Ethnic and family ties reduce opportunism and holdups in the criminal
network, therefore achieving the necessary levels of trust for a profitable
and lucrative criminal enterprise. In that respect, homogeneous and closed
immigrant communities can be an attractive focal point for these activ-
ities.
6
Hence, it would be no surprise if criminal enterprises or gangs fo-
cused on prostitution, drugs, money laundering, and traffic of weapons,
women or babies were dominated by immigrants. The closeness of
immigrant communities provides the necessary and trustworthy link be-
tween the home and the resident countries. These effects will necessarily
be reduced when the immigrant community becomes better integrated
into the resident country and grows larger. It is true that larger cities (such
as Barcelona or Madrid in Spain, or New York and Chicago in the United
States) facilitate the search of contacts that might help the expansion
of criminal activities. However, an integrated and large immigrant com-
munity will lose the homogeneity and closeness that make it attractive for
criminal enterprises of this type. In summary, for criminal activities that
5. In a fully rational model, labor market arguments could explain overall crime
rates. However, in a bounded rational model such as the one endorsed by behavioral
law and economics, labor market arguments should be more relevant to explain property
crime and those crimes more directly pursued for economic reasons.
6. The same argument also explains why certain (legal) business activities and enter-
prises that require trust and coordination attract immigrant communities. Hence, we would
expect immigrant communities to be more engaged in criminal activities and certain en-
trepreneurial activities.
Does Immigration Cause Crime? 171
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
require coordination and trust, we should expect to find a disproportional
representation of immigrant communities.
7
A third argument relies on the cost of undercompliance. It has been docu-
mented by legal economists (McAdams, 2000) that compliance with the law
is easier to achieve when it embodies social norms shared by most individ-
uals in a community and when it triggers psychological reactions that limit
criminal opportunism (because individuals are socialized in a culture that
inculcates since childhood that certain behaviors are just wrong). We can
talk of undercompliance due to different cultural perceptions. Examples
could include terrorist acts, gender violence, or antisocial behavior, where
asymmetries of social norms and cultural values could generate different
perceptions about not only the law but also the fairness and general accep-
tance of the law. As perceptions converge, either because the immigrant
community internalizes the national social norms due to successful integra-
tion or because the law is reformed to accommodate the diversity of percep-
tions, this type of undercompliance will tend to be mitigated.
A fourth argument could be less knowledge of local laws. Immigrants
could violate the law by mistake more frequently simply because they ignore
the law or are unaware of the specific enforcement choices of national au-
thorities. Ignorance of the law is no excuse in a court of law, but it may
nevertheless lead to more frequent criminal behavior among immigrants than
among natives. Presumably, this argument can only be relevant for those
areas of the law that require a degree of sophistication that make it more
difficult for the immigrant community to immediately grasp the substance
and the procedure of local laws. This could include, for example, antisocial
behavior, traffic offenses, consumption of illegal substances, and so on.
A final argument has to do with risk aversion. If criminals have a less risk-
averse utility profile (Becker, 1968), it may be that immigrants are on
7. There is no detailed evidence we can use in our data set to test this hypothesis.
According to the Spanish Office of the Chief Prosecutor, there were 482 criminal organ-
izations in 2006 (compared to 594 in 2002 and 494 in 2004). Three groups are quite sig-
nificant: Colombians (142 organizations), Moroccans (108 organizations), and Romanians
(92 organizations). The Office of the Chief Prosecutor points out that these ethnic organ-
izations are quite consolidated and are responsible for a large proportion of the crimes
committed by illegal organizations in Spain. While the Moroccan and Romanian organ-
izations largely operate in all criminal markets, the Colombians focus on the market for
narcotics and related activities (money laundering, corruption, and violence).
172 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
average closer to such profile than natives because there is a selection
effect due to immigration. The latter being, intrinsically, a risky activity;
presumably only people with lower risk aversion become immigrants.
However, notice that the underlying risks, for immigration and crime,
are different and hence the selection effect could well not be very relevant
(Bianchi, Buonanno, and Pinotti, 2008).
So far we have identified reasons for why immigrants could be more
prone to commit certain crimes. They include the labor market and general
economic conditions, need for cooperation and trust, costs of undercompli-
ance due to different cultural perceptions, less knowledge of local laws, and
(less convincingly) lower risk aversion.
However, there are also good economic reasons for why immigrants
could be less prone to commit other types of crime. One obvious reason
has to do with opportunities. First, earning potentials could be higher for
immigrants, so that the lost legal earnings due to criminal activities could
be more significant as a deterrent. Second, if immigrants tend to be located
in more depressed economic neighborhoods, then they have fewer oppor-
tunities for property crime. Third, even if economic well-being determines
that immigrants could be more likely to commit certain economic crimes, it
is also likely that those crimes that require specific levels of human capital or
technology will be committed less by immigrants (e.g., regulatory and ad-
ministrative crimes). Another line of reasoning could be that if immigrants
have certain distinct characteristics that make them easier to be targeted by
the enforcement authorities, then they could be more deterred since the
expected severity of punishment could be higher. Furthermore, the risk
of deportation could make punishment more costly. Finally, less knowledge
of local laws could drive them to comply due to overestimating punishment.
As we have seen, there is no economic theory that provides support to the
hypothesis that there is a strong correlation between crime and immigration.
In fact, the economic arguments about differential behavior toward crime
among natives and immigrants are mixed. Furthermore, the sign and the de-
gree of correlation between crime and immigration are expected to differ by
crime types. If there exists a linkage between immigration and certain crime
types, it is probably better explained by incentives that factor into opportun-
ism rather than by any specific attribute of the immigrant community.
Given the relevant explanations and the available data for the Spanish
economy and society, we suggest that the economic theory at its best
Does Immigration Cause Crime? 173
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
proposes a very weak positive correlation between general crime rates and
immigration. The theoretical reasons for more crime (labor market and
general economic conditions, need for cooperation and trust, costs of
undercompliance due to different cultural perceptions, less knowledge
of local laws, and lower risk aversion) might weakly dominate the theo-
retical reasons for less crime (opportunity costs, information available, ac-
cess to technology). In summary, we suggest that, focusing on the criminal
data published by the Spanish authorities, the arguments for a positive cor-
relation are likely to be more relevant than the arguments for a negative
correlation.
8
Notice, however, that we hypothesize a weak positive correlation, the
argument being cultural proximity. Immigrants with cultural proximity to
the Spanish society will have a profile of preferences akin to natives. Therefore,
the economic model should predict that differences in criminal behavior
cannot be significant.
The possible testable hypothesis we suggest is that crimes committed by
immigrants can be explained by labor market conditions and economic op-
portunities. Education might also be used as a proxy for knowledge of local
laws and reliance on closed ethnical groups; hence, higher levels of educa-
tion should reduce crime. Age and gender are also relevant for the (uncon-
ditional) probability of undercompliance; in particular, young males are
responsible for most law infractions. Urban areas could have an important
impact because they reduce not only the probability of detection but also the
likelihood of strong ethnic ties. At the same time, following the criminal
proximity argument, the presence of Spanish-speaking immigration should
decrease the incidence of criminal behavior.
8. Immigrants can also be disproportionally affected by crime as victims. There is no
similar sophisticated economic model for victimization of crime. In fact, a major criticism
to the economic literature of crime is that it is oriented to deterrence and seriously neglects
the role of the victim (Garoupa, 1997). However, from a theoretical perspective, we can
also consider possible reasons for why immigrants could be the targets of crimes more
often than natives. One obvious example is hate crimes, crimes motivated by racism or
xenophobia (Dharmapala and Garoupa, 2004). Another example is crimes related to the
legal status or regulation of immigration such as corruption in obtaining residence permits
or other type of extortions for social benefits. Finally, crimes committed within closed
ethnic homogeneous groups could also affect immigrants in a disproportional way.
The available data for Spain does not allow us to test any of these possible explanations.
174 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
3. Econometric Approach
As discussed in the previous section, the baseline economic model
we use lies on the traditional choice-theoretic approach at the individual
level (Becker, 1968; Ehrlich, 1973). At each period t, each individual i
decides whether or not to commit a crime. Each individual lives in a par-
ticular location l, being assumed that a person only commits crimes
within the location of residence. Defining Y
it
as a binary variable that
takes the value of 1 when individual icommits a crime and 0 otherwise,
the probability that individual icommits a crime at period tcan thus be
written as
PrðYit ¼1jXlt ;Xit Þ¼FðX0
ltbþX0
itcþlit þhit Þ;ð1Þ
where X
lt
is a vector of observable location-specific characteristics and X
it
is a vector of observable individual-specific characteristics; such location-
specific and individual-specific characteristics might affect the opportunity
costs of committing a crime with respect to engaging in legal economic
activities. In addition, there can be further individual-specific and location-
specific characteristics, denoted respectively by l
it
and h
lt
, potentially af-
fecting the individualÕs decision of crime commission, which nonetheless
are unobserved by the researcher. Among the individual variables, it is of
particular interest to consider whether the individual is an immigrant or
not, even including his specific nationality. We should also consider
the level of education, age, gender, etc. As to the location-specific varia-
bles, we consider the Spanish province in which the individual lives and
whether the individual resides in an urban or rural area (or the size of the
municipality in which the individual resides). Also, economic character-
istics of the location, such as the unemployment rate and the per capita
income, provide a measure of the economic prospects of their residents
in the legal labor market.
In addition, we are concerned with an individual’s criminal experience,
a relevant aspect already discussed in the literature (Sah, 1991; Grogger,
1995; Glaeser, Sacerdote, and Scheinkman, 1996; Fajnzylber, Daniel,
and Loayza, 2002). This factor has a potential relevance regarding the rel-
ative cost of entry in criminal activity, the learning curve, and the economic
opportunities for recidivists in the legal labor market. Consequently, the
probability of crime commission can be written as
Does Immigration Cause Crime? 175
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
PrðYit ¼1jXlt ;Xit Þ¼FdYi;t1þX0
ltbþX0
itcþlit þhit :ð2Þ
To estimate this model, we would ideally use a representative sample
of the population; yet, as is usually the case, there is not individual-level
data available. Instead, we must rely on aggregate data by geographic
locations. Our location units are the fifty Spanish provinces.
9
The use
of province-level data involves data aggregation among individuals
within each location unit, as well as several statistical assumptions
which are treated in detail for the empirical studies about crime by
Durlauf, Navarro, and Rivers (2010).
10
The underlying conditional
probability for individual crime action is implicitly characterized by
a linear probability model.
The consequence of using aggregate data at the province level is that the
dependent variable is no longer the probability that an individual will com-
mit crime. Instead, it turns out to be the crime rate, that is, the number of
criminal acts relative to the province population. Also, when aggregating
whether each individual within a province committed a crime in the pre-
vious year, we will get the lagged crime rate. Each of the explanatory
variables related to a particular individual characteristic will shape the
proportion of people living in the province with such characteristic after
aggregating.
The explanatory variables will consist of the lagged crime rate, the pro-
portion of immigrants in the province (which will be further broken down by
country or geographic zone of origin), the province rate of unemployment of
9. At the administrative level, Spain is broken down into fifty provinces and
two autonomous North-African cities, Ceuta and Melilla, which were excluded from
the analysis because of their different idiosyncrasy due to their geography and
history.
10. In addition to the usual parameter homogeneity assumption, for the model
to preserve the interpretation of the baseline behavioral model at the individual level,
it is required: (i) the implicit individual utility function be linear; (ii) that there be
constraints on the dependence between the two sources of unobservable terms and
the observed individual-specific and location-specific variables; (iii) that there be
constraints on the dependence between unobservable terms and the observed
individual-specific and location-specific variables. Also, the underlying conditional
probability for individual crime commission is implicitly characterized by a linear
probability model, which also introduces further constraints on the implicit random
utility function.
176 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
these groups (including that of the nationals), the proportion of individuals in
each collective with a given level of education, the proportion of the pop-
ulation that lives in urban areas, and the proportion of individuals in a pre-
determined range of ages in each collective, and the gross domestic product
(GDP) per capita, among other things.
Our empirical specifications, using land tto index provinces and years,
respectively, are written as follows:
Cl;t¼qCl;t1þb0Zl:tþglþatþul:t;ð3Þ
where Crepresents the crime rate and Zis the set of covariates affecting
this rate. The last three terms represent unobserved variables that capture
province-level unobserved heterogeneity, aggregate shocks common to all
provinces, and a term that comprises idiosyncratic shocks, measurement
errors in the dependent variable, and aggregation errors. Even though
in the individual-level model the covariates might be uncorrelated with
the unobserved factors, the right-hand-side (RHS hereafter) variables in
the aggregate regression are potentially correlated with the unobservable
variables.
Given the features of the different unobserved components, we
propose the following strategy. First, assuming that province-level
unobserved heterogeneity is invariant over time, we can exploit the lon-
gitudinal data structure and apply a fixed-effects transformation to
remove this unobserved component. Namely, denoting Das the first-
difference operator,
11
the fixed-effects transformation yields the follow-
ing model:
DCl;t¼qDCl;t1þb0DZl;tþDatþDul;t;ð4Þ
where all the variables are now expressed in first differences, but the param-
eters of interest are kept invariant. Second, the potential endogeneity be-
tween the covariates and the unobserved idiosyncratic term u
l,t
remains
after the fixed-effects transformation. Consequently, the covariates must
be instrumented for. It is not obvious, though, whether there exist external
11. For any variable V
l,t
,DV
l,t
¼V
l,t
V
l,t1
.
Does Immigration Cause Crime? 177
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
valid instruments, consisting of variables not included in the model but un-
correlated with the unobserved components.
12
However, we can exploit the availability of panel data to get appropriate
instruments. The most obvious instruments are the lagged values of the RHS
variables. We can reasonably assume that the unobserved error term in the
fixed-effects transformation (4), Du
l,t
, is uncorrelated with past realizations
of the RHS variables, which can then be used as instruments. Formally, for
s2 and for each jth covariate, Z
jl,t
, we have the moment condition of zero
correlation with the fixed-effects transformation of the error:
CovZjl;tsDul;t[EZjl;tsDui;t¼0;ð5Þ
where E() denotes the expectation operator. It must be noted that we need at
least 3 years (T¼3) of data. In fact, when T¼3, estimation of the fixed-
effects model (4) using lags at t2 of the RHS variables becomes two-stage
least squares (2SLS) on a cross section (see Wooldridge, 2002; Arellano,
2003). However, when T>3, we have further moment conditions, which
can be optimally exploited period by period (t¼3, ... ,T) by means of
a generalized method of moments (GMM) estimator (see Arellano and
Bond, 1991).
13
In addition, Arellano and Bover (1995) showed that under
certain conditions,
14
we can also exploit additional moment conditions:
12. In the search for appropriate instruments external to the model, in the case of
immigration variables, which are of our main concern, we should look for variables highly
correlated with immigration while exogenous to our model. Among the potential candi-
dates, we can think of the province-level weight of services and construction sectors in
GDP, the investment expenditure in public infrastructures, and the share of employment in
low-skilled industries. Given the high level of participation of immigrants in such eco-
nomic activities, such potential instruments are positively correlated with the proportion of
immigrants in province populations. However, its exogenous nature is under question, to
the extent that the higher these variables, the higher the immigrantsÕopportunities in the
legal sector, what might affect the immigrantsÕpropensity to be involved in the criminal
sector.
13. For instance, if we restrict s¼2, we have EðZjl;t2Dul;tÞ¼0;for any t¼3, ...,T.
The availability of panel data allows us to consider a system of T3 equations, one for each
available period t¼3, ...,T, where the parameters are equal for every period. The 2SLS
estimator, though, only exploits one moment condition for each jth covariate as it aggregates
the moment conditions over time, thus considering PT
t¼3EðZjl;tsDul;tÞ¼0:
14. Namely, it suffices that for any time period, CovðZjl;t;glÞ¼CovðZjl;t1;glÞ,
that is, the correlation between the RHS variables and the unobserved heterogeneity
be constant over time.
178 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
CovDZjl;t1;gl[EDZjl;t1gl¼0:ð6Þ
Such moment conditions imply that first differences of the RHS variables
can also be used as instruments for the original untransformed model (3).
The GMM estimator that exploits such conditions in addition to the earlier
moment conditions for the fixed-effects transformation is known as system
GMM estimator. Such estimator combines the use of the fixed-effects model
(4) and the original model (3), using the corresponding moment conditions
(5) and (6), respectively. In comparison with the estimator based on the
transformed model only, the system GMM estimator has been proved to pro-
vide more precise estimates of the parameters of interest (see Arellano and
Bover, 1995; Blundell and Bond, 1998). Given the small sample size, the
standard errors must be appropriately corrected from potential finite-sample
biases following Windmeijer (2004).
4. Discussion of Results
In Table 3, we present the descriptive statistics of our endogenous var-
iables. We use four measures of criminal activity. The first variable is total
crime infractions, including felonies and misdemeanors.
15
The second
Table 3. Descriptive Statistics: Crime Rates (number of crimes per 10,000
inhabitants—provincial data)
Observations Mean
Standard
Deviation Minimum Maximum
Felony rate 450 200 112 53 722
Property crime rate 450 155 90 36 496
Misdemeanors rate 450 204 79 56 494
Crime rate 450 399 180 110 1,139
15. We produce results for the total crime rate because it has been used very often in
the previous related studies. However, we believe that the total crime rate is not very in-
formative since it aggregates too many different activities concerning undercompliance
with the law: violent crimes, organized crime, property crime, administrative infractions,
and misdemeanors. When using the total crime rate, we are uniformly weighting all these
activities, irrespective of their type (in fact, we could argue that it is unclear how the
weights should be determined). Also we are implicitly imposing that a change in any
RHS variable of the econometric model leads to a certain increase/decrease in the number
of infractions, irrespective of their nature.
Does Immigration Cause Crime? 179
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
variable is felonies only.
16
The third variable is misdemeanors only (offenses
less serious than a felony). Property crimes, the last variable, involve the
taking of money or property without violence or threat of violence against
a victim (included in felonies).
17
We can see that misdemeanors (which are
mostly punished with administrative penalties) are more numerous on av-
erage, whereas property crimes exhibit a larger relative dispersion.
There are good cautious reasons to use four measures of criminal activity.
All capture criminal behavior although they might exhibit a different pattern.
Felonies have increased only slightly during this period, whereas the mis-
demeanors have rapidly increased. Moreover, there might be a substitution
effect in property crime, where immigrants could have replaced natives. Fur-
thermore, property crimes could adhere to economic models more easily
than other crimes. Finally, there have been legal changes during the sample
period that have transformed misdemeanors into felonies, which justify why
we should look at the total crime rate just to make sure our results are not
contaminated by legal reform.
18
As explanatory variables related to immigration, we consider the propor-
tion of immigrants to the provinceÕs total population. Additionally, we con-
sider the proportion of immigrants that speak Spanish as a native language
and those whose nationality is from the EU before the expansions that oc-
curred during this decade (EU15). These last two variables were considered
because of their quantitative importance (although their share of total immi-
gration has been decreasing in the sample period). The specific consideration
16. Under Spanish criminal law, it includes property crimes, crimes against people
(homicides, injuries), crimes against freedom and collective security (drugs, road traffic),
crimes against the government and the courts (mainly noncompliance with judicial
decisions), and public order.
17. This is probably the cleanest measure of crime for statistical purposes out of the
four we use and the one most easily related to the economic model. However, since our
hypothesis is also related to noneconomic variables, in particular cultural proximity, it
seems to us more comprehensive and technically more competent to test also the other
two variables, felonies and misdemeanors.
18. The reason why we do not use the homicides rate, the best feasible measure of
crime for regression analysis, as a possible dependent variable is due to the small numbers
as evident from Table 2 and small annual variations. According to the Spanish Ministry of
Interior, in 2006 there were 570 homicides, 436 committed by Spanish citizens and 134 by
foreigners. These numbers compare with 470 homicides in 2000, 373 committed by
Spanish citizens and 97 by foreigners.
180 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
of immigrants from different nationalities should control for different atti-
tudes among these groups.
19
To account for the effect of living in large urban agglomerations, we have
considered the percentage of province population living in cities with at least
one hundred thousand and five hundred thousand inhabitants. Another char-
acteristic that we use is the GDP per capita and the rate of unemployment in
the province. These two variables provide a measure of legitimate economic
opportunities. The hypothesis is that better economic opportunities in the
legal labor market make criminal activity less attractive.
We also use measures to control for differences among immigrants and
natives. We use the corresponding proportions of immigrants and natives,
which have completed at least secondary education. Unfortunately, given the
lack of representativeness of these collectives in provinces with small pop-
ulations, we have used data at the regional level from the Spanish Labor
Force Survey. Consequently, we cannot exploit the variability across prov-
inces of the same region. This same data source has been exploited to cal-
culate, at the regional level, the proportion of young male natives and
immigrants, respectively, for which we have considered separately two
age cohorts, 15–24 and 25–34 years old. Table 4 summarizes the descriptive
statistics for the explanatory variables.
For every measure of crime rate, we have considered three different es-
timation methods, which are presented in Table 5.
20
As a benchmark, we consider the ordinary least squares (OLS) estima-
tions of the untransformed model. For the aforementioned reasons, such
estimates are expected to be inconsistent because of the correlation be-
tween unobserved province-level characteristics and the covariates. We
also report the within-groups estimations, which consist of least squares
estimates of a fixed-effects transformation of the aggregate model. Even
though the within-group transformation removes the unobserved
19. Unfortunately, the lack of representativeness of other immigrants of different
countries of origin in low-populated provinces has precluded us from controlling for more
disaggregated measures of immigration.
20. With respect to misdemeanors, a legal reform, performed in 2003, turned a great
part of the misdemeanors related to physical aggression and threats, particularly among
family members, to felonies with criminal consequences. The inclusion of year binary
dummies allows us to control for this legal change.
Does Immigration Cause Crime? 181
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
time-invariant province effects, the resulting estimates might not be
consistent if the variables are not strictly exogenous, which is not held
for the reasons explained earlier. Either way, we report both OLS and
within-group estimates for the purpose of comparison. Last, we report
the system GMM results in which the instrument set is composed of
the second and third lags of the explanatory variables.
21
In order to assess the quality of the GMM estimates, we have used two
types of specification tests, the Hansen-Sargan test to check for the val-
idity of the instrument set and the AR(2) test of error autocorrelation. The
latter test does not show any evidence of error within the specifications in
any of the cases. However, the Hansen-Sargan test seems to reject the
specifications in the case of total crime. With respect to total crimes,
Table 4. Descriptive Statistics of Explanatory Variables
Observations Mean
Standard
Deviation Minimum Maximum
% Immigrants 450 4.7 4.3 0.3 21.5
% Spanish-speaking
immigrants
450 28.2 21.3 0 76.0
% EU15 immigrants 450 5.4 15.3 0 71.4
% Population in cities
>100,000 inhabitants
450 17.7 4.6 9.2 34.0
% Population in cities
>500,000 inhabitants
450 10.9 5.2 2.2 33.3
GDP per capita 450 17.7 4.6 9.2 34.0
Unemployment rate 450 10.9 5.2 2.2 33.3
% Natives with secondary
education
450 59.0 6.0 46.7 74.3
% Immigrants with secondary
education
450 68.3 9.8 24.4 92.2
% Spanish-speaking immigrants
with secondary education
450 76.4 8.4 41.8 100.0
% Male natives aged 15–24 years 450 13.3 1.9 9.5 18.6
% Male natives aged 25–34 years 450 16.1 1.5 12.5 20.7
% Male immigrants aged 15–24
years
450 15.5 2.8 8.1 22.4
% Male immigrants aged 25–34
years
450 30.3 5.7 11.1 40.4
21. System GMM estimations have been implemented using the Stata command
‘‘xtdpd’’ and the third-party Stata command ‘‘xtabond2’’ (see Roodman, 2006).
182 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
which include both felonies and misdemeanors, the aggregation of re-
markably different infractions is the most likely reason for the rejection
of the specification. Namely, we are aggregating offenses that are very
different in nature, implicitly supposing that the effect of the explanatory
variables is simply proportional to the number of different illegal activ-
ities. On the other hand, the specification tests for felonies and property
crimes behave well.
In the case of total crimes, OLS estimates exhibit that the presence of
immigrants is positive and significant but being a Spanish-speaking (Latin
American) immigrant offsets the positive effect of being an immigrant on
crime. However, when we consider the within-group and, more importantly,
the GMM estimates, the coefficients on the immigrant shares keep their signs
but are no longer significant. In fact, the lagged total crime rate is the only
significant variable in the GMM estimates. This fact, together with the
Hansen-Sargan test, whose low P-value casts doubt on the validity of the
specification, underlies the aggregation of remarkably different crime infrac-
tions, with very different attitudes and behavioral issues explaining them.
This result suggests a focus on different types of infractions, namely, the
three that we have proposed earlier.
The main estimation results obtained for the different crime types can be
summarized as follows. First, variations in the presence of immigrants at the
different provinces have a significant impact. Second, there is a differential
attitude toward crime across different immigrants by geographical area of
origin and personal characteristics. Third, previous crime history, captured
by the lagged crime rate, has a substantial effect.
Regarding the estimated effect of total immigration, OLS estimates of the
share of immigrants in the total population are positive and significant for
every type of crime considered. This implies that the rise in immigration
increases crime, even after controlling for observable characteristics like
the economic conditions and the composition of the population. However,
when considering the within-group estimates, the estimated effect becomes
nonsignificant. However, neither the OLS nor the within-group estimators
account properly for the potential endogeneity problems. Our GMM estima-
tions confirm the importance of immigration when explaining crime during
this period, yet the effects are different both at the qualitative and at the quan-
titative levels. In the case of misdemeanors, the share of immigrants is still
positive but not significant.
Does Immigration Cause Crime? 183
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
Table 5. Estimations for Crime Rates
Total Crime Rates Felony Rates Property Crime Rates Misdemeanor Rates
OLS Within GMM OLS Within GMM OLS Within GMM OLS Within GMM
% Immigrants 4.38
y
9.09 6.86 3.43
§
5.38* 8.69
§
2.97
§
4.17 7.05
y
1.48
y
4.55 4.11
2.32 7.30 8.51 1.53 3.64 3.62 1.27 3.36 3.60 0.88 4.12 3.29
% Spanish-speaking immigrants 7.24
y
8.61 16.06 5.98
§
4.89 17.09
§
5.50
§
5.14 14.86
§
2.27* 5.81 14.42*
3.84 12.77 18.65 2.58 6.49 6.67 2.24 6.46 7.42 1.67 7.46 9.56
% EU15 immigrants 1.14 26.89 6.72 0.47 19.18
§
6.43* 0.09 20.21
§
5.09 0.15 4.19 4.02
2.78 14.09 11.16 1.62 7.67 4.53 1.53 6.74 4.19 1.33 7.62 5.24
% Population in cities >100,000
inhabitants
0.09 2.01 1.65 0.08 1.46 1.72* 0.07 1.65 1.24 0.04 0.34 0.70
0.11 2.49 1.99 0.07 1.23 1.16 0.07 1.29 0.98 0.05 1.27 0.77
% Population in cities >500,000
inhabitants
0.42
§
8.79 0.18 0.19
y
1.67 0.29 0.20
§
0.27 0.21 0.20
§
7.21* 0.19
0.19 7.71 0.43 0.10 3.84 0.37 0.10 3.41 0.33 0.09 4.50 0.21
GDP per capita 2.48 13.45 3.47 0.75 4.99
y
5.56 0.66 4.69
y
4.47 1.75
§
7.80
§
0.98
0.97 5.08 7.38 0.60 2.73 3.73 0.55 2.60 4.12 0.43 2.59 2.68
Unemployment rate 0.08 2.03 1.26 0.19 0.83* 0.97 0.09 0.50 0.42 0.61
y
0.99* 0.16
0.68 1.06 2.02 0.40 0.50 1.21 0.51 0.59 0.83 0.34 0.62 0.75
% Natives with secondary
education
0.02 1.34 0.89 0.56* 0.36 1.44 0.58 0.50 0.78 0.54
y
1.05* 0.63
0.57 1.23 1.87 0.38 0.67 1.13 0.37 0.82 0.88 0.29 0.68 0.87
% Immigrants with secondary
education
0.38 0.42 0.95 0.11 0.20 0.78
y
0.17 0.24 0.94
y
0.23
§
0.22 0.47
§
0.22 0.29 0.80 0.13 0.15 0.44 0.19 0.25 0.52 0.11 0.17 0.22
% Spanish-speaking immigrants
with secondary education
0.00 0.05 0.22 0.03 0.04 0.09 0.10 0.15 0.19 0.05 0.01 0.12
0.23 0.23 0.44 0.14 0.13 0.20 0.20 0.17 0.29 0.12 0.11 0.18
% Male natives aged 15–24 years 7.97 26.57 1.13 3.44
§
12.87
§
6.01 3.07
§
14.27
§
6.79 3.09
§
9.98
§
1.37
2.42 10.56 18.05 1.49 5.51 6.29 1.56 5.28 8.21 1.18 5.68 6.52
% Male natives aged 25–34 years 0.93 11.88 18.78 0.83 5.28 11.65* 0.22 6.78* 7.97
y
0.88 3.21 8.57
3.22 9.52 18.61 2.04 5.28 9.06 1.62 5.04 5.78 1.24 5.21 8.78
% Male immigrants aged 15–24
years
0.42 2.45 2.36 0.13 1.75 1.33 0.02 1.51* 0.33 0.17 0.93 1.06
0.81 1.92 2.81 0.52 1.18 1.59 0.50 1.05 1.91 0.44 0.97 1.35
Continued
184 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
Table 5. Continued
Total Crime Rates Felony Rates Property Crime Rates Misdemeanor Rates
OLS Within GMM OLS Within GMM OLS Within GMM OLS Within GMM
% Male immigrants aged 25–34
years
0.02 3.49 1.38 0.05 1.83 0.33 0.12 2.13
§
0.64 0.09 1.63* 2.06
y
0.42 1.94 1.48 0.26 0.99 0.81 0.30 1.02 0.65 0.23 1.04 1.06
Lagged dependent variable 0.86 0.45 1.00
§
0.84
§
0.40 0.88
§
0.84
§
0.37
§
0.90
§
0.91
§
0.56
§
0.95
§
0.06 0.08 0.12 0.07 0.09 0.16 0.06 0.08 0.13 0.04 0.06 0.09
Hansen-Sargan test (% P-value) 3.4 22.8 25.6 15.1
AR(2) test (% P-value) 32.7 46.0 37.7 29.7
Notes: *, y, and § denote significance at the 20%, 10% and 5%, respectively. Year dummies included in all estimates. Standard errors are reported in italics below each estimated coefficient.
Both the Hansen test and the AR(2) test are specification tests that help to evaluate the validity of the estimates. The P-values indicate the significance level below which the null hypothesis is
rejected. The Hansen-Sargan test evaluates the null hypothesis of validity of the overidentifying restrictions, and it is asymptotically distributed (under the null) as a chi-square with as many
degrees of freedom as the number of overidentifying restrictions. In our context, this test of overidentifying restriction can be viewed as a test for instrument validity. If the instruments used in
GMM estimation are valid, then the Hansen-Sargan test ought to be statistically equal to zero (see Arellano, 2003). The AR(2) test is asymptotically distributed as a standard normal under the
null hypothesis of no second-order autocorrelation in the error term of the first-difference transformed model. This test is based on the fact that if the model is properly specified, the
transformed error term cannot exhibit second-order autocorrelation (see Arellano and Bond, 1991).
Does Immigration Cause Crime? 185
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
The estimations also corroborate our hypothesis concerning different be-
havior of the different groups of immigrants based on geographical area of
origin.
22
The models that account for the shares of immigrant population
who speak Spanish as a native language (mostly Latin American immi-
grants), on the one hand, or come from one of the EU15 countries, on
the other hand, are particularly relevant. These variables ought to be inter-
preted as an additional effect given the characteristics already controlled,
specially being an immigrant. For all crime types, being a Spanish-speaking
immigrant reduces the likelihood of criminal activity in an important and
significant way, although the precision of such an estimate is reduced for
misdemeanors. To a lesser extent, being an EU15 immigrant also appears
to reduce the propensity to commit crime, but both the magnitude and
the precision of the estimated effect for this group of immigrants are much
lower.
We have also controlled for the percentage of immigrants and natives who
have completed at least secondary education. Its significance and interpre-
tation appear ambiguous. Given the importance of Latin American immi-
grants, we have also accounted for the percentage in this group who has
completed at least secondary education, finding a negative but clearly non-
significant effect for property crimes and misdemeanors. Together with the
general result concerning Spanish-speaking immigrants, we can interpret our
results as showing that the cultural proximity of this group to the native pop-
ulation has played an outstanding role in explaining the exceptional case of
Spain.
The lagged crime rate exhibits a significantly positive effect, being close
to unity (but clearly below one for all crime types), which suggests a sub-
stantial inertia in crime dynamics. We interpret this result as the importance
of the learning curve in crime propensity. Interestingly, the highest inertia is
found for misdemeanors and, to a lesser extent, property crimes, being lower
for felonies, which are precisely the crime types for which economic reasons
might not play a major role.
22. We have concentrated on these two groups of immigrants by geographical area
of origin since they are very representative in most of the provinces considered. The fact
that the representativeness of many other groups of immigrants from other countries is
very unequal has prevented us from controlling for further areas of origin.
186 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
Concerning the remaining variables, we observe that, in line with the eco-
nomic model of crime, the GMM estimates for GDP per capita exhibit a neg-
ative effect, although not significant. On the other hand, the estimates of
province unemployment rate are very imprecise.
23
We have also controlled for the proportion of people living in highly pop-
ulated areas, by means of the variables indicating the proportion of the prov-
ince population living in cities above one hundred thousand and five
thousand inhabitants. These variables are intending to capture the difficulty
of detection and the less likelihood of strong ethnic ties, as measured by the
size of the population. Although not significant, if anything, we find that
such effect is positive, especially for felonies, in line with the evidence that
population density favors crime as it hardens crime prosecution.
Finally, given that most crimes are committed by young males, we have
considered the percentage of male natives and immigrants aged between 15
and 24 years and between 25 and 34 years. We find that, in the case of prop-
erty crimes and misdemeanors, the percentage of male immigrants aged be-
tween 25 and 34 years is statistically significant, indicating that age and
gender may affect an individualÕs propensity to commit crimes. It must
be noted, as seen in Table 4, that the share of young males is much higher
for immigrants than for natives.
We have also considered alternative specifications in order to assess the
robustness of our results. First, we have considered alternative instrument
sets to gauge the sensitivity of our results to the inclusion of different lags
of the covariates in the instrument set. Second, we have considered a non-
linear transformation of the crime rate to evaluate the robustness of our
results to departures of the underlying linear probability model at the indi-
vidual level. Third, we have considered a static specification, which ex-
cluded the lagged crime rate.
The results, not reported here (but available upon request), can be sum-
marized as follows. First, changes in the instrument set, particularly the lags
of the explanatory variables included, mostly affected the precision of the
23. Ideally, we would like to consider unemployment rates in the province for
natives and immigrants. Nevertheless, the level of representativeness for immigrants
was very poor for underpopulated provinces and even for some underpopulated regions.
Therefore, we have opted out of using province unemployment rates as a measure of em-
ployment opportunities for both natives and immigrants.
Does Immigration Cause Crime? 187
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
estimates while the main qualitative results remain. Second, defining a non-
linear transformation of the crime rate, which is consistent with an under-
lying logistic distribution of the individual probability of committing crime,
the qualitative results are kept. However, we have opted out of our speci-
fication since the interpretation of estimates is clearer. Last, regarding the
static model, ignoring the lagged crime rate led to a clear rejection of the
specification, except for misdemeanors, and the precision of the estimates,
particularly for the immigrant shares, was substantially reduced.
In summary, our results largely confirm the hypothesis that we have de-
veloped before and are consistent with the economic literature.
5. Conclusions
The present work introduces a first comprehensive analysis of the relation-
ship between immigration and crime in Spain. In the context of the EU, Spain
is not a country with high crime rates. During the last decade of significant
immigration, Spain has seen an important rise in crime at a similar pace with
its immigrant population but to a lower rate than other countries. We hypoth-
esize that the explanation of this behavior is related to the individual char-
acteristics of the immigrants that Spain has received. We argue that it is not so
much the number of immigrants but the specific characteristics that seem to
explain the relationship between crime and immigration in Spain.
In this respect, cultural proximity and education must be specially noted.
Immigrants from some populous groups, such as those who are Spanish na-
tive speaking, present a substantial proportion of people with at least sec-
ondary education, for whom criminality is much lower. This effect has
contributed decisively to avoid any kind of explosion of criminality. Gender
has also positively contributed to this effect. In fact, even after controlling for
gender and education, we can still conclude that Latin-American immigra-
tion has probably undermined the potential rise of criminal rates in Spain.
This result also happens, to a lesser extent, with EU15 immigrants. Our result
is fully consistent with the evidence for the United States regarding Mexican
immigrants (known as the Latino Paradox), where immigration from Mexico
has lowered crime rates in some areas.
24
24. See, for example, the discussion by Sampson (2008) and references therein.
188 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
Other immigrant groups with lower education levels have contributed sig-
nificantly to the rise in crime rates. It must also be noted that these immi-
grants started at arrival with a crime rate significantly higher than Spanish
nationals but have been converging as their size and composition changed. It
must be noted the specific case of the Romanian immigrant group, now one
of the most numerous in Spain. Even though it started with high crime rates,
nowadays it presents lower rates than nationals in the 20–50 age group.
This work also provides for a good example of standard discussions in the
econometrics of crime. We have observed how the booming stage of the
business cycle in Spain reduced crime because it increased the opportunity
cost; we have seen that population density favors crime because it makes
harder the identification and tracking of criminals; we have concluded that
crime is concentrated among young males and lesser educated individuals.
The implications of this work for designing public policies are clear. Immi-
gration is not a simple homogeneous phenomenon and must not be treated as
such. Public authorities should respond with differentiated policies, depend-
ing on the relevant characteristics.
References
Arellano, Manuel. 2003. Panel Data Econometrics. Oxford, UK: Oxford University
Press.
Arellano, Manuel, and Stephen Roy Bond. 1991. ‘‘Some Tests of Specification for
Panel Data: Monte Carlo Evidence and Application to Employment Equation,’’
58 Review of Economic Studies 277–97.
Arellano, Manuel, and Olympia Bover. 1995. ‘‘Another Look at the Instrumental-
Variable Estimation of Error Components Models,’’ 68 Journal of Econometrics
29–52.
Becker, Gary. 1968. ‘‘Crime and Punishment: An Economic Approach,’’ 76 Journal
of Political Economy 169–217.
Bianchi, Milo, Paolo Buonanno, and Paolo Pinotti. 2008. ‘‘Do Immigrants Cause
Crime?,’’ Working Paper No. 2008-05, Paris School of Economics.
Blundell, Richard, and Stephen Roy Bond. 1998. ‘‘Initial Conditions and Moment
Restrictions in Dynamic Panel Data Models,’’ 87 Journal of Econometrics
115–43.
Borjas, George J., Jeffrey Grogger, and Gordon H. Hanson. 2006. ‘‘Immigration
and African-American Employment Opportunities: The Response of Wages,
Employment, and Incarceration to Labor Supply Shocks,’’ Working Paper No.
12518, NBER.
Does Immigration Cause Crime? 189
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
Buonanno, Paolo, and Daniel Montolio. 2008. ‘‘Identifying the Socioeconomic
and Demographic Determinants of Crime Across Spanish Provinces,’’ 28
International Review of Law and Economics 89–97.
Buonanno, Paolo, Daniel Montolio, and Paolo Vanin. 2009. ‘‘Does Social Capital
Reduce Crime?,’’ 52 Journal of Law and Economics 145–70.
Butcher, Kristin F., and Anne Morrison Piehl. 1998a. ‘‘Cross-city Evidence on the
Relationship between Immigration and Crime,’’ 17 Journal of Policy Analysis and
Management 457–93.
———. 1998b. ‘‘Recent Immigrants: Unexpected Implications for Crime and Incar-
ceration,’’ 51 Industrial and Labor Relations Review 654–79.
———. 2005. ‘‘Why Are Immigrants’ Incarceration Rates So Low? Evidence on
Selective Immigration, Deterrence, and Deportation,’’ Working Paper No. 05-
19, Federal Reserve Bank of Chicago.
Card, David. 2001. ‘‘Immigrant Inflows, Native Outflows, and the Local Labor Mar-
ket Impacts of Higher Immigration,’’ 19 Journal of Labor Economics 22–64.
Dharmapala, Dhammika, and Nuno Garoupa. 2004. ‘‘Penalty Enhancements for Hate
Crimes: An Economic Analysis,’’ 6 American Law and Economics Review
185–207.
Durlauf, Steven N., Salvador Navarro, and David A. Rivers. 2010. ‘‘Understanding
Aggregate Crime Regressions,’’ 158 Journal of Econometrics 306–17.
Ehrlich, Isaac. 1973. ‘‘Participation in Illegal Activities: A Theoretical and Empirical
Investigation,’’ 81 Journal of Political Economy 521–65.
Fajnzylber, Pablo, Lederman Daniel, and Norman Loayza. 2002. ‘‘Inequality and
Violent Crime,’’ 45 Journal of Law and Economics 1–40.
Garoupa, Nuno. 1997. ‘‘The Theory of Optimal Law Enforcement,’’ 11 Journal of
Economic Surveys 267–95.
———. 2007. ‘‘Optimal Law Enforcement and Criminal Organization,’’ 63 Journal
of Economic Behavior and Organization 461–74.
Glaeser, Edward L., Bruce Sacerdote, and Jose A. Scheinkman. 1996. ‘‘Crime and
Social Interactions,’’ 111 Quarterly Journal of Economics 507–48.
Grogger, Jeffrey. 1995. ‘‘The Effect of Arrest on the Employment and Earnings of
Young Men,’’ 110 Quarterly Journal of Economics 51–72.
McAdams, Richard. 2000. ‘‘A Focal Point Theory of Expressive Law,’’ 86 Virginia
Law Review 1649–729.
Moehling, Carol, and Anne M.orrison Piehl. 2007. ‘‘Immigration and Crime in Early
20th Century America,’’ Working Paper No. 13576, NBER.
Polinsky, A. Mitchell, and Steven Shavell. 2000. ‘‘The Economic Theory of Public
Enforcement of the Law,’’ 38 Journal of Economic Literature 45–76.
Roodman, David. 2006. ‘‘How to Do xtabond2: An Introduction to ÔDifferenceÕ
and ÔSystemÕGMM in Stata,’’ Working Paper No. 103, Center for Global
Development, Washington, DC.
190 American Law and Economics Review V14 N1 2012 (165–191)
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
Sah, Raaj K. 1991. ‘‘Social Osmosis and Patterns of Crime,’’ 99 Journal of Political
Economy 1271–95.
Sampson, Richard J. 2008. ‘‘Rethinking Crime and Immigration,’’ 7 Contexts 28–33.
Windmeijer, Frank. 2004. ‘‘A Finite Sample Correction for the Variance of Linear
Efficient Two-Step GMM Estimators,’’ 12 Journal of Econometrics 25–51.
Wooldridge, Jeffrey M. 2002. Econometric Analysis of Cross Section and Panel
Data. Cambridge, MA: The MIT Press.
Does Immigration Cause Crime? 191
by guest on May 28, 2012http://aler.oxfordjournals.org/Downloaded from
