51

The Korean Economic Review

Volume 28, Number 1, Summer 2012, 51-67.

Mobility-Based Explanation of Crime Incentives*

Yoonseok Lee** · Donggyun Shin***

The canonical economic model of crime is extended to include individuals’ expectation of

future income mobility as an additional crime determinant. The model predicts that with
all else being held constant including net gain from current criminal activity, reduced
upward mobility among the poor increases crime rate whereas enhanced downward
immobility among the rich decreases crime rate. These predictions are empirically supported
by country-level panel data. In addition, a typical change in income distribution was
implemented, such that both the poor and the rich groups contribute to crime rate with a
greater contribution among the poor.

JEL Classification: C10, D63, K42

Keywords: crime, immobility, income distribution

8
I. Introduction

Many economic and sociological studies have attempted to determine the

relationship between inequality and crime, as well summarized by Kelly (2000);
Fajnzylber, Lederman, and Loayza (2002); and Demombynes and Özler (2005).
The canonical economic model of crime1 states that the more unequal income
distribution results in a greater gap between benefits and costs of crime committed
by low income earners and thus a higher (property) crime rate, which is largely
based on the differential returns from legal and illegal activities. However, economic
and sociological theories vary depending on which aspect of social phenomena is
____________________
Received: April 25, 2012. Revised: May 8, 2012. Accepted: June 11, 2012.
* This work was supported by the National Research Foundation of Korea Grant funded by the
Korean Government (NRF-2011-330-B00040). The authors acknowledge research assistance given by
Jongsoon Yu and valuable comments of James Hines, Lutz Kilian, and Daniel Silverman on an earlier
version of this paper. The usual disclaimer applies.
** Department of Economics, University of Michigan, Ann Arbor, Michigan, U.S.A.
*** Corresponding author: Department of Economics, College of Political Science and Economics,
Kyung Hee University, Seoul, 130-701, South Korea, dgshin@khu.ac.kr.
1
See Becker (1963).
52 The Korean Economic Review Volume 28, Number 1, Summer 2012

operative in these theories. According to sociological theories, inequality makes

individuals in lower strata feel alienated from society, generally undermines the
ability of society to control its members, and reduces demand for public safety, all of
which contribute toward increasing crime rate. However, as argued by Fajnzylber,
Lederman, and Loayza (2002, p. 2), whether the positive crime-inequality relation,
if any, results from economic incentives-disincentives of crime or from social strain
or disorganization is difficult to identify empirically.
Based on empirical evidence,2 most cross-sectional comparisons across states and
cities in the United States or those across countries conclude that inequality leads to
property and/or violent crime (e.g., the survey by Demombynes and Özler, 2005).
However, findings in these studies may be subject to an omitted variable bias
problem because they do not control for unobserved fixed effects that are specific to
the cross-sectional unit and possibly correlated with the unit’s inequality. Evidence
remains insufficient and contradicting even for panel data models with fixed effects.
For example, Lee (1993, cited in Freeman, 1996) regressed changes in crime in a
metropolitan area between 1970 and 1980 against changes in inequality and
discovered insignificant coefficient estimates. The first-difference model of Doyle,
Ahmed, and Horn (1999) also produced insignificant coefficients of the Gini index.
On the contrary, based on international panel data for 39 countries, Fajnzylber,
Lederman, and Loayza (2002) reported significantly positive coefficients for both
homicide rates and robbery rates even after country-specific fixed-effects were
controlled for.
Despite the repeated theoretical and empirical efforts to connect the crime rate
with cross-sectional inequality, little effort has been exerted to understand the
relationship between the crime rate and inequality in the distribution of lifetime
income. This situation is surprising considering that many studies such as that of
Backer and Solon (2003) have emphasized that long-term inequality is even more
consequential than instability of transitory earnings. From a theoretical viewpoint,
the decision of an individual regarding whether to supply his labor to the legal or
illegal labor market is based on his relative position in the distribution of the
expected lifetime income rather than in the current distribution. The latter is related
to the conventional static inequality measure such as the Gini coefficient, while the
former is determined by the mobility of the person to the income group other than
the one to which he belongs. Existing measures of cross-sectional inequality do not
sufficiently capture these mobility aspects of lifetime income distribution (e.g.,
Yitzhaki and Wodon, 2002) and therefore may not be adequate in explaining
individual or collective violence.
This paper is the first that attempts to present a mobility-based explanation of
crime incentives of individuals. We first extend the canonical crime model by
____________________
2
The first empirical study dates back to Ehrlich (1973).
Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 53

including expectation of individuals regarding their future mobility as an additional

crime determinant. Each individual forms his expectation of future mobility based
on the information obtained from the current distribution. In particular, mobility
aspects of income distribution, which are not adequately captured by conventional
inequality measures, are important in decision-making; individuals feel a greater
extent of income class separation and so expect less mobility in the future either
when between-group income distance is longer or when within-group income
distribution is more clustered around its local mean, and therefore they recognize
less of the income bridge to the other income group. Then, using country-level
panel data, we directly estimate the equation generated by the utility maximization
process that connects the crime rate with a measure of the subjective mobility of
individuals along with other crime determinants. As such, we attempt to provide a
more structural explanation of the crime incentives of individuals.
Organization of this paper is as follows: Section II presents a theoretical
framework that helps explain how expectation of individuals regarding their future
mobility affects their decision to engage in criminal activity. Given the paucity of
reliable data, theoretical reasoning is necessary to analyze the relationship between
crime and income distribution (e.g., Bourguignon, 1999). The canonical economic
crime model is enriched by encompassing the subjective feeling of mobility of
individuals as an additional crime determinant. The model is further extended to
include crime incentives of the rich as an additional model element, which also is a
unique feature of the current model. Section III presents evidence from country-
level panel data. Consistent with our theoretical prediction, with all else being
controlled for, the crime rate increases in relation to measured immobility among
the poor. Moreover, enhanced immobility among the rich reduces the crime rate. As
our measure of the feeling of immobility increases among the poor but decreases
among the rich, both groups make positive contributions to the crime rate, albeit
much greater contribution is observed among the poor. Section IV concludes.

II. Model

When a person decides whether to supply his labor in the legal or illegal labor
market, the person considers his relative position in the distribution of expected
lifetime income rather than in the current distribution. The latter is related to
conventional static inequality measures such as the Gini coefficient, while the
former is determined by the entire lifetime income stream of the person, which in
turn depends on, among others, the mobility of the person to the income group
other than the one to which he belongs. Given this perspective, even a current low-
income earner would not have high crime incentive if he had better prospects in the
54 The Korean Economic Review Volume 28, Number 1, Summer 2012

future. If one has higher expectation of upward mobility, then expected lifetime
income is high and so is the marginal cost of the current crime action; therefore, one
has lower crime incentive.
More precisely, we consider a utility maximization problem to explain individual
crime incentive. We assume that the individual has utility function
u( yi ,η i ) = V ( yi ) + η i , where yi is the income (or wealth) level of person i,
V ( yi ) > 0 , ∂V ( y) / ∂y > 0 , ∂ 2V ( y) / ∂y2 < 0 , and ηi is the psychological crime
disincentive factor of person i. In the literature, the psychological crime disincentive
factor ηi is normally specified as the individual level of honesty, which is
independent of income level. We emphasize, however, that the expected lifetime
income or expectation of future income mobility is another important determinant
for the psychological crime disincentive. To model such idea, we assume that
society is divided into two income groups, Gr (rich) and Gp (poor), whose
average income and population sizes are denoted as μ r , μ p , and nr , np ,
respectively. We then define

ηi = ηi ( hi , IMi ) = η( hi , IMip ) if i ∈ Gp and = η( hi , IMir ) if i ∈ Gr ,

where hi represents the individual level of “honesty” as disincentive for crime,

which is independent of income level. hi ~ iidU[0, H ] with some H > 0. IMik
for k = r , p captures heterogeneous and subjective immobility that each individual
feels toward the current income distribution of society. Values of IMik differ
between groups and across individuals. For a given level of objective immobility,
individuals in the poor group feel a greater extent of immobility than those in the
rich group, resulting in greater crime incentive among the poor. ∂η / ∂h > 0 ,
∂η / ∂IMip < 0 , and ∂η / ∂IMir > 0 , which are satisfied by a simple specification
that

ηi = hi − IMip for i ∈ G p , and ηi = hi + IMir for i ∈ Gr . (1)

Consequently, the rich generally have a greater extent of crime disincentive than
the poor do.
The crime pays x with probability (1 − q) and − F (i.e., fine once the crime
is detected) with q . Similar to the standard crime model (e.g., Bourguignon, 1999),
individual i (assuming he is the only person to commit a crime; i.e., ceteris paribus)
opts for criminal activity if

(1 − q)V ( yi + x ) + qV ( yi − F ) > V ( yi ) + η i . (2)

Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 55

In this setup, crime action is an all-or-nothing decision. Existing studies assume

that F is proportional to the income level yi , and thus the rich for whom x is
relatively small are not tempted to commit crimes even when they are sufficiently
dishonest ( hi = 0) . This assumption simplifies the discussion by assuming that
only the poor commit crimes, although this assumption may not be justified by the
judicial system of each society. By contrast, our specification (1) of the disincentive
factor works toward making the crime incentive greater for the poor than the rich
even without the “proportionality” assumption; even with an equal amount of fine
between the rich and the poor, the crime disincentive ( ηi ) is always smaller for the
poor because of the feeling of blocked upward mobility, other things being held
constant. Similarly, even when the fine is independent of the income level and even
when the rich are sufficiently dishonest, they may not have strong crime incentive
when they feel sufficiently secure about the future. In other words, compared with
existing crime models, crime incentive of the poor is stronger than that of the rich in
the current model. For the purpose of comparison of our crime model with existing
ones, we first follow existing studies and assume that H , x , q , F , and IMip
are such that only the poor engage in criminal activity. Observing the crime rate
between 0 and 1, equation (2) is satisfied for ηi = − IMip in group p (sufficiently
dishonest) but not for ηi = H − IMip (sufficiently honest).3
Given this simple specification, the crime rate (CR) in a society can be obtained
as

CR = n−1 ∑ i∈Gp Pr( hi < {V ( yi + x) − V ( yi )} − q{V ( yi + x) − V ( yi − F )} + IMip }

= H −1 n−1 ∑ i∈Gp [{V ( yi + x) − V ( yi )} − q{V ( yi + x) − V ( yi − F )}]
+ H −1(n−1 ∑ i∈Gp IMip ) (3)

In equation (3), {V ( yi + x) − V ( yi )} and {V ( yi + x) − V ( yi − F )} represent the

utility change when the crime is not detected and the utility change when it is
detected, respectively. The second term implies that, other things being held
constant, the crime rate increases as the feeling of immobility escalates among the
poor.
Empirical implementation of equation (3) requires specification of the average
group-specific subjective immobility as a function of observable variables. We
postulate that individuals evaluate their future income mobility based on the
properties of the current income distribution. Specifically, individuals in the poor
group feel a greater degree of immobility to the rich group either when the between-

____________________
3
For the rich class, even dishonest people usually do not involve themselves in criminal activity
because of the feeling of economic safety. Equation (2) may not be satisfied even when ηi = IMir .
56 The Korean Economic Review Volume 28, Number 1, Summer 2012

group income gap is greater or when the within-group income distribution is less
dispersed. When income levels of individuals are more clustered around a local
mean, each individual feels a greater extent of group identity to members of their
own group, which enhances income class separation and lessens the subjective
probability of switching to the other group. To the extreme, if all individuals in the
lower income group had the same income level as μ p and all income levels of
individuals in the higher group were identical to μ r such that no bridge income
level exists in the society, the crime incentive of the poor would be high even for a
relatively small between-group income gap. Given these conditions, the average
feeling of immobility of the poor group is specified as follows.

IM p = n p −1 ∑ i∈G p IM ip = β p μ −1( μ r − μ p )υ p−1 , (4)

where μ −1( μ r − μ p ) is a measure of income distance between the two groups

normalized by the population mean ( μ ), and υ p represents a measure of within-
group income dispersion. Specification (4) implies that the average immobility
feeling is proportional to the product of the between-group income distance and the
degree of within-group income clustering, with β p being the psychological
proportionality factor. Consequently, the psychological crime incentive of the poor
generated by their expectation of future mobility is assumed proportional to some
observable statistical characteristics of the current distribution. We further assume
that υ p is represented by Gini −1Gini p , the within-group Gini coefficient relative
to the overall Gini.4 As each society has a different degree of income (or wealth)
inequality, the feeling of identity of the poor group is determined by the relative
inequality. Even given the same extent of within-group inequality among the poor,
the poor feel more identification to their group members as the within-group
income distribution among the rich (thus the entire distribution) becomes more
unequal.
Inserting (4) into the parenthesis of the second term of equation (3) produces an
expression of the group-size-adjusted crime incentive of the poor generated by their
immobility feeling.

IM _ Poor = β p μ −1( μ r − μ p )υ p−1π p (5)

Equation (5) describes how the expectation of future mobility of the poor group
affects the overall crime rate. To repeat, first, other things being equal, the between-
group income distance raises the subjective immobility of the poor group, which
____________________
4
In the empirical execution, we also attempt to use the ratio of within-group standard deviation to
overall standard deviation as an alternative measure of υ p and find little difference in the results.
Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 57

increases the crime incentive of the poor. Second, as income levels of individuals are
more clustered around μ p , the poor feel a greater extent of income class separation
(or group identity), which enhances the crime incentive of the poor (called the
“clustering effect”). In addition, equation (5) emphasizes the group size as an
additional crime determinant. Given the clustering effect, the poor feel a greater
extent of group identity as the group size increases, which is positively associated
with the crime incentive (called the “scale effect”). The poor tend to view the
income class separation more as a structural problem when the population
proportion of their income group increases; otherwise, income class separation is
viewed as a personal problem.
Finally, by substituting (5) into (3), the crime rate is determined by the following
equation.

CR = f ( IM _ Poor , H , ng( x, F , q)) , (6)

where ng( x, F , q| yi ) = n−1 ∑ i∈Gp [{V ( yi + x) − V ( yi ) − q{V ( yi + x) − V ( yi − F )}]. This

result implies that the crime rate is a function of the feeling of immobility of the
poor group, a unique feature of our crime model, as well as the extent of honesty in
the society H , which in turn may depend on various economic and demographic
observable characteristics such as education, age, and gender as well as unobservable
characteristics specific to that society and the net gain from the current crime action
ng .5 The net gain of crime depends on the crime premium ( x ), which depends on
the average income (or wealth) level (if the victim is randomly selected from the
society), the current income level of the potential criminal ( yi ), fine F , which is
determined by the judicial system, and the effort level of crime prevention, q (e.g.,
police size, arrest rate, and conviction rate, among others).
We can further enrich the above crime model by including the feeling of
immobility of the rich group as an additional psychological crime disincentive factor,
as in equation (1). That is, we consider a more general situation where both the rich
and the poor groups have a positive level of crime incentives, although the level of
the rich is presumably lower than that of the poor. For both groups to have a
positive level of crime activity, values of H , x , q , F , IMip , and IMir are such
that equation (2) is satisfied for ηi = − IMip in group p but not for
ηi = H − IMip and for ηi = IMir in group r but not for ηi = H + IMir . Even the
rich have positive crime incentives when they are sufficiently dishonest or when
they expect strong downward income mobility in the future.
Given this addition, the crime rate in a society is now expressed as

____________________
5
Although not explicit in the current model, inclusion of the net gain from the current crime action
partly reflects imperfection of the labor market.
58 The Korean Economic Review Volume 28, Number 1, Summer 2012

CR = n−1 ∑ i∈Gp Pr( hi < {V ( yi + x) − V ( yi )} − q{V ( yi + x) − V ( yi − F )} + IMip }

+ n−1 ∑ i∈Gr Pr( hi < {V ( yi + x) − V ( yi )} − q{V ( yi + x) − V ( yi − F )} − IM ir }
= H −1 n−1 ∑ i =1[{V ( yi + x) − V ( yi )} − q{V ( yi + x) − V ( yi − F )}]
n

+ H −1(n−1 ∑ i∈Gp IMip − n−1 ∑ i∈Gr IMir ) . (7)

Finally, by obtaining comparable expressions for equation (4) and (5), the crime
rate in a society is determined by the following equation.

CR = f ( IM _ Poor , IM _ Rich , H , ng( x , F , q)) , (8)

where IM _ Rich = β r μ −1( μ r − μ p )υ r−1π r , and

g( x , F , q) = n−1 ∑ i =1[{V ( yi + x) − V ( yi )} − q{V ( yi + x) − V ( yi − F )}] .
n

In the present setup, the hypothesis is that β p > 0 and β r < 0 ; a unit increase
in the group-specific measured immobility increases the crime rate among the poor
and decreases it among the rich. Whether the overall crime rate will increase or
decrease following a change in the current income distribution also depends on how
the change affects within-group dispersions as well as group sizes. Specifically,
given the hypothesis that β p > 0 and β r < 0 , if the change in the income
distribution enhances the size-adjusted group identity of the poor, υ p−1π p , and
weakens that of the rich, υ r−1π r , then the overall crime rate increases, other things
being equal. If the distribution change enhances the identification feelings of both
groups, the overall crime rate increases if and only if β p Δ(υ p−1π p ) > β r Δ(υ r−1π r ) .
This would be called weak asymmetry in the mobility-generated crime incentive
between the rich and the poor. Even given an equal increase in the extent of the
group-specific identification feeling, the crime rate increases if and only if
β p > β r , which is called strong asymmetry in the crime incentive. The main goal
of the next section is to verify that with all else being controlled for, our immobility
variable increases the crime rate among the poor in equation (6), and enhanced
immobility increases the crime incentive of the poor but reduces that of the rich in
equation (8) (or in the following equation (9)). Our sample also was tested for the
existence of asymmetry.
Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 59

III. Empirical Findings

1. Data, empirical specifications, and estimation method

The most challenging aspect of this empirical study is to compute our immobility
variable using micro data on individual households, which requires a large sample
size for each unit of analysis. In particular, for each unit of analysis in which crime
is examined, our immobility variable requires dividing the sample into the poor and
the rich and computing the Gini coefficient as well as various group-level statistics
for each group. Thus, a unit smaller than the country is difficult to consider in the
analysis.6 On the contrary, differences in the definition of household income often
make the comparison of various inequality and income mobility measures across
different countries difficult. Thus, we adopt country data from the Luxembourg
Income Study (LIS) database, which includes micro income data obtained from
many countries at different times. The LIS uses the same definition and
components of household income across countries, which renders more significance
to a cross-country comparison of income-related statistics. We use household
disposable income adjusted by the “OECD equivalency scale” (e.g., Atkinson,
Rainwater, and Smeeding, 1995) and household sample weights.
One more advantage of using country data is that when the unit of analysis is
very small, the local crime rate does not necessarily reflect the economic conditions
of the region. Criminals travel to neighborhoods in search of higher returns (e.g.,
Demombynes and Özler, 2005), or those who are frustrated in one region transfer to
another region where they have better prospects and thus decide to supply labor to
the legal labor market. This increased geographical interdependence makes the
analysis extremely complicated. By contrast, the crime market is closed at the
country level. The original LIS includes a large number of countries in its data set,
while our final regression sample has 100 country-year observations for 24 countries,
where the sampling frequency is approximately five years on average.7 The LIS
serves as our main data set, although we also supplement the results with another
country-level data set, the Cross-National Equivalent File (CNEF). The CNEF
uses sources of income data different from the LIS. Unlike the LIS, the CNEF
contains only longitudinal micro data sets for all member countries, enabling
longitudinal cross-country research on topics like employment and earnings
____________________
6
This may explain why none of the studies that investigate the relationship between crime and
inequality using the unit of observation smaller than the level of country include inequality measures
in their regression models (e.g., Cornwell and Trumbell, 1994; Glaeser, Sacerdote, and Scheinkman,
1996; Wilson and Daly, 1997; Kelly, 2000).
7
They are Australia, Austria, Belgium, Canada, Czech Republic, Denmark, Estonia, Finland,
France, Germany, Greece, Hungary, Ireland, Italy, Mexico, Netherlands, Norway, Poland, Slovenia,
Spain, Sweden, Switzerland, the United Kingdom, and the United States.
60 The Korean Economic Review Volume 28, Number 1, Summer 2012

dynamics. This exercise is meaningful because the crime rate is affected by

measured characteristics of an income distribution such as the Gini coefficient and
our immobility variable, and those measured characteristics are often different
depending on the data sets used in the analysis, even when the same income
definition is adopted in a country. The main disadvantage of the CNEF is that it
includes only five countries (Australia, Canada, Germany, the U.K., and the U.S.).
The final regression sample, which contains all valid observations on all variables
used in the regression analysis, includes 30 country-year observations, where the
sample frequency is approximately two years on average.8 As in the LIS, we use
household disposable income adjusted by the “OECD equivalency scale” and
household sample weights. In the LIS, most observations pertain to the 1980s and
the 1990s, while those of the CNEF refer to the 1990s and the 2000s.
Our primary goal is to not distinguish empirically between the economic and
sociological explanations of the relationship between crime and income
distribution; our immobility measure reflects both economic and psychological
motives of crime. Thus, the overall crime rate is the relevant variable used to test the
hypotheses. The crime data we use are obtained from the United Nations Surveys
on Crime Trends and the Operations of Criminal Justice Systems (CTS).
Regarding other control variables, police size (POLICE) is imported from the CTS.
Average years of education (EDUCATION), percentage of urban population
among the total population (URBAN POPULATION), and population density
(DENSITY) are obtained from the World Bank. The proportion of men aged 15 to
29 among the total population (YOUNG MEN), unemployment rates
(UNEMPLOYMENT), and the percentage of tax revenue among gross domestic
product (TAX-GDP-RATIO) to be used as an instrumental variable for POLICE,
are obtained from the Organization for Economic Cooperation and Development
(OECD).
Due to lack of data, penalties ( F ) are neglected, and the probability of
apprehension ( q ) is instead determined by the police size. The net gain from the
current criminal activity, if succeeded, is measured by (1 − μ −1 μ p ) given that the
potential victim and potential criminal are randomly selected from the entire
population and the poor group, respectively. The dependent variable is the
logarithm of the crime rate as measured by the number of reported crimes per
100,000 people.
Our expanded regression based on equation (8) has the following form.

log CRit = β p immobility _ poorit + β r immobility _ richit +

δ1 netgainit + δ 2 log( police)it + γ ′Xit + λ ′α i + ε it , (9)
____________________
8
The small sample problem in the final regression is also attributed to large missing values of some
control variables with crime- and police- related variables being the most important sources missing.
Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 61

where log CRit represents the logarithm of the crime rate of country i in year t ,
immobility _ poor = μ −1(μr − μ p )υ p−1π p , immobility _ rich = μ −1(μr − μ p )υr−1π r ,
and netgain = (1 − μ μ p ) . log( police) is the logarithm of the police size. Xit
−1

represents a vector of time-varying economic and demographic variables. α i is a

vector of country-specific time-invariant characteristics, observable or not. ε it is
the error term. In the estimation, the simultaneity between the crime rate and the
crime deterrent variable, police size, is addressed using instrumental variables
described below. Unobservable unit-specific fixed effects are included in every
regression model, and two-step fixed-effect Generalized Method of Moments
(GMM) estimation is applied with Heteroskedasticity and Autocorrelation
Consistent (HAC) standard error estimates.
In the fixed-effect GMM, the endogenous right hand side variable, log( police) ,
is instrumented by two excluded variables, TAX-GDP-RATIO and CRIME
COMPOSITION. First, as argued by Cornwell and Trumbull (1994), countries
with residents who are more agreeable toward law enforcement will reveal their
preferences by voting for higher tax rates to finance larger police forces. Such
countries would have larger police sizes for reasons not directly related to the crime
rate. As for CRIME COMPOSITION, defined by the ratio of the number of violent
crimes to that of property crimes,9 while this ratio is not directly related with the
total number of crime, the crime composition is related with police size. For
example, with the total number of crimes fixed, a greater proportion of violent
crimes calls for more police activity and for more policemen involved.

2. Findings

Based on the LIS data, Figure 1 displays the sample correlation between the total
crime rate and our measure of immobility of the poor group, while Figure 2 shows
the result for the rich group. We normalize the total crime rate and the immobility
variable by the country-specific mean of each variable, exploiting within-variation of
each variable. Apparently, crime is positively correlated with the feeling of
immobility of the poor group and negatively correlated with the immobility of the
rich group. Although not reported for brevity, the CNEF also reveals similar
patterns.10 Technically speaking, these observed correlations would be obtained by
applying fixed-effect estimation of the crime rate with only the immobility variable
included in the right hand-side variable. In the following exercises, we test if these
observed patterns survive controlling for other variables.

____________________
9
A similar instrumental variable is suggested by Cornwell and Trumbull (1994). Information on
the numbers of violent and property crimes is obtained from the CTS.
10
The negative association of the crime rate and the immobility of the rich group is even more
apparent in the CNEF.
62 The Korean Economic Review Volume 28, Number 1, Summer 2012

[Figure 1] Crime and Immobility of the Poor Group

[Figure 2] Crime and Immobility of the Rich Group

Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 63

[Table 1] Determinants of Total Crime: Basic Model

LIS CNEF
Variables
Specification 1 Specification 2 Specification 3 Specification 3
0.630*** 0.562** 1.607**
Immobility_Poor -
(0.238) (0.236) (0.643)
1.044 0.907 7.302**
Net gain of crime -
(0.812) (0.774) (3.058)
0.906 0.855 0.784 0.443**
Log(police)
(0.782) (0.787) (0.710) (0.195)
0.060*** 0.050*** 0.051*** 0.076
Urban population
(0.011) (0.010) (0.010) (0.082)
1.110 0.815 1.114 -15.760*
Young men
(0.902) (0.809) (0.952) (8.748)
0.020 0.034*** 0.023* -0.035
Unemployment rate
(0.016) (0.012) (0.014) (0.033)
0.052 -0.012 0.023 -1.136***
Education
(0.042) (0.035) (0.043) (0.434)
0.009*** 0.013*** 0.010*** 0.115***
Population density
(0.003) (0.003) (0.002) (0.042)
F-test of excluded instrument:
0.220 0.260 0.238 0.096
Prob>F
Hansen J-test 0.244 0.589 0.541 0.139
N: country-year 100 30
[country] [24] [5]
The dependent variable is the logarithm of the number of total crimes per 100,000. Our
immobility variable is measured based on household income data from the LIS and the CNEF.
Fixed-effect GMM estimation with HAC standard error estimates. The police size is
instrumented by the TAX-GDP-RATIO and CRIME COMPOSITION. *, **, *** =
significance at 10%, 5%, and 1%, respectively.

Table 1 presents estimation results of equation (6). As in existing studies, Table 1

assumes that only the poor are involved in criminal activity. What has changed in
our model is the inclusion of the feeling of immobility of the poor group as an
additional crime determinant. Estimates in the first three columns are obtained
from the LIS, and those in the last column are obtained from the CNEF. Focusing
on the effects of the two distribution-related variables, immobility and net gain of
crime, when only the net gain is included in the regression (conventional
specification), the estimated coefficient is positive but insignificant (column 1).
However, when only the feeling of immobility of the poor group is included in the
regression, it increases the crime rate significantly (column 2). When both variables
are included in the regression (our preferred specification), the estimated
immobility effect is still positive and significant (column 3); the feeling of
immobility of the poor group significantly increases the crime rate whether or not
64 The Korean Economic Review Volume 28, Number 1, Summer 2012

the net gain is controlled for. Evidence from the CNEF also suggests that although
the crime rate increases in the net gain, the income immobility also increases the
crime rate significantly even after the net gain is controlled for. The effects of other
control variables are generally consistent with those in existing studies. Overall,
other things held constant, the crime rate increases in the population share of those
living in urban areas, the population density, and the unemployment rate.

[Table 2] Asymmetry between the Rich and the Poor in the Immobility Effect

Variables LIS CNEF

0.534* 0.959*
Immobility_Poor
(0.287) (0.613)
-0.138 -3.026**
Immobility_Rich
(0.924) (1.417)
0.954 11.374***
Net gain of crime
(0.741) (3.362)
0.890 0.473***
Log(police)
(0.650) (0.169)
0.052*** 0.136*
Urban population
(0.011) (0.080)
1.158 -13.662
Young men
(0.986) (8.464)
0.022* -0.009
Unemployment rate
(0.013) (0.036)
0.025 -0.682*
Education
(0.040) (0.354)
0.010*** 0.122***
Population density
(0.003) (0.037)
F-test of excluded
0.205 0.066
instrument: Prob>F
Hansen J-test 0.554 0.222
N: country-year 100 30
[country] [24] [5]
Refer to the caption of Table 1.

Table 2 shows estimation results of equation (9). Consistent with our hypothesis
that enhanced immobility increases crime incentive of the poor but decreases that of
the rich, both data sets show that estimated β p is positive and estimated β r is
negative. As a minor difference between the two data sets, while the LIS reveals that
the estimated coefficient of the immobility of the rich group is negative but
insignificant, the corresponding estimate obtained from the CNEF data is negative
both statistically and substantially. Also consistent with the conventional belief, the
crime rate increases when criminals expect higher returns net of their opportunity
costs. Estimated coefficients of other control variables remain similar to those in
Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 65

Table 1.
Our model addresses how the poor and the rich groups behave differently in
response to a change in the income distribution. Note that the extent of the feeling
of immobility of each group ( immobility _ poor or immobility _ rich in equation
(9)) is measured by the product of the between-group income distance and the
extent of size-adjusted within-group clustering. Estimates presented in Table 2
indicate the marginal effect of a unit-increase in the feeling of immobility of each
group. Evidence obtained from the LIS data supports the strong asymmetry
hypothesis; given an equal increase in the extent of the feeling of immobility for
both groups, the rise in the crime incentive of the poor is greater than the reduction
in the crime incentive of the rich. Evaluation of whether the crime rate increases
following a change in the income distribution requires information on how the
distribution change affects the expectation of future immobility of each group. Our
LIS sample shows that the average change between two adjacent years (on average,
five years apart) in the measured immobility feeling is 0.015 and -0.001 for the poor
and the rich group, respectively. Therefore, the feeling of immobility of the rich
group has actually weakened albeit by a small extent, which resulted in an increase
in the crime rate even among the rich. Finally, the estimated overall immobility
effect generated by a distribution change is approximately 0.008 for the poor, while
the comparable figure for the rich is 0.0001. Evidence obtained from the CNEF
shows similar results. As previously stated, estimates obtained from the CNEF
sample do not support the strong asymmetry hypothesis, as the estimated coefficient
of the immobility variable is larger in an absolute value for the rich than the poor.
However, even the CNEF sample observes a reduction in our measure of the
feeling of immobility among the rich; the average change between two adjacent
years (on average, two years apart) in the measured feeling of immobility is 0.007
and -0.0015 for the poor and the rich group, respectively. Consequently, the
estimated overall immobility effect of the poor is approximately 0.007, while the
comparable figure for the rich is 0.003.
In sum, estimates show that the feeling of income class separation or between-
group immobility is an important determinant of the decision of individuals to
engage in crime, and enhanced immobility increases the crime incentive of the poor
but decreases the crime incentive of the rich. Evidence obtained from the LIS data
also supports the hypothesis of strong asymmetry between the poor and the rich in
the crime incentive/disincentive generated by enhanced feeling of immobility.

IV. Conclusion

Both our theoretical reasoning and empirical execution suggest that the decision
66 The Korean Economic Review Volume 28, Number 1, Summer 2012

of an individual on whether or not to become involved in criminal activities is based

not just on the net gain of crime from the current action but also on expectation of
future income mobility. Labor and education market policies that mitigate
permanent income inequalities and ensure greater upward mobility among the poor
are effective measures of reducing the crime rate. In this regard, future research
could investigate determinants of income mobility rather than inequality.
Yoonseok Lee · Donggyun Shin: Mobility-Based Explanation of Crime Incentives 67

References

Atkinson, A. B., L. Rainwater, and M. Smeeding (1995), “Income Distribution in OECD

Countries: Evidence from the Luxembourg Income Study,” OECD Social Policy
Studies, No. 18, Paris: OECD.
Baker, M., and G. Solon (2003), “Earnings Dynamics and Inequality among Canadian Men,
1976-1992: Evidence from Longitudinal Income Tax Records,” Journal of Labor
Economics, 21, 289-321.
Becker, G. S. (1963), “Crime and Punishment: An Economic Approach,” Journal of Political
Economy, 76, 169-217.
Bourguignon, F. (1999), “Crime, Violence, and Inequitable Development,” Annual World
Bank Conference on Development.
Cornwell, C., and W. N. Trumbull (1994), “Estimating the Economic Model of Crime with
Panel Data,” Review of Economics and Statistics, 76, 360-366.
Demombynes, G., and B. Özler (2005), “Crime and Local Inequality in South Africa,”
Journal of Development Economics, 76, 265-292.
Doyle, J. M., E. Ahmed, and R. N. Horn (1999), “The Effects of Labor Markets and Income
Inequality on Crime: Evidence from Panel Data,” Southern Economic Journal, 65, 717-
738.
Ehrlich, I. (1973), “Participation in Illegitimate Activities: A Theoretical and Empirical
Investigation,” Journal of Political Economy, 81, 521-565.
Fajnzylber, P., D. Lederman, and N. Loayza (2002), “Inequality and Violent Crime,”
Journal of Law and Economics, 45, 1-39.
Freeman, R. B. (1996), “Why Do So Many Young American Men Commit Crimes and
What Might We Do About It?” Journal of Economic Perspectives, 10, 25-42.
Glaeser, E. L., B. Sacerdote, and J. A. Scheinkman (1996), “Crime and Social Interactions,”
Quarterly Journal of Economics, 111, 507-548.
Kelly, M. (2000), “Inequality and Crime,” Review of Economics and Statistics, 82, 530-539.
Wilson, M., and M. Daly (1997), “Life Expectancy, Economic Inequality, Homicide, and
Reproductive Timing in Chicago Neighborhoods,” British Medical Journal, 314, 1271-
1274.
Yizhaki, S., and Q. Wodon (2002), “Mobility, Inequality, and Horizontal Equity,” memeo.